Article

Low-Cost Raman Spectroscopy Setup Combined with a Machine

Learning Model
Catarina Domingos 1 , Alessandro Fantoni 1,2, * , Miguel Fernandes 1,2 , Jorge Fidalgo 1 and Sofia Azeredo Pereira 3

1 Department of Electronics, Telecommunication and Computers, Lisbon School of Engineering (ISEL),

Polytechnic University of Lisbon (IPL), Rua Conselheiro Emídio Navarro, n◦ 1, 1959-007 Lisbon, Portugal;
[email protected] (C.D.); [email protected] (M.F.); [email protected] (J.F.)
2 Center of Technology and Systems (UNINOVA-CTS) and Associated Lab of Intelligent Systems (LASI),
2829-516 Caparica, Portugal
3 iNOVA4Health, NOVA Medical School, Faculdade de Ciências Médicas, NMS, FCM, Universidade NOVA de
Lisboa, 1169-056 Lisbon, Portugal; [email protected]
* Correspondence: [email protected]

Abstract: The diagnosis of kidney diseases presents significant challenges, including the
reliance on variable and unstable biomarkers and the necessity for complex and expensive
laboratory tests. Raman spectroscopy emerges as a promising technique for analyzing com-
plex fluids, like urine, and detecting important disease biomarkers. However, its complexity,
high cost and limited accessibility outside clinical contexts complicate its application. More-
over, the analysis of Raman spectra is a challenging and intensive task. In response to these
challenges, in this study, we developed a portable, simplified and low-cost Raman system
designed to acquire high-quality spectra of liquid complex samples. Using the “Starter
Edition” methodology from the OpenRAMAN project, the system was optimized through
laser temperature adjustments, by evaluating the laser emission spectrum under different
temperatures with a spectrometer, and through adjustment of the acquisition parameters of
the software used, by acquiring the ethanol spectra. The system validation was performed
through the acquisition of Raman spectra from five urine samples, demonstrating its con-
sistency and sensitivity to composition variations in urine samples. Additionally, a neural
network was designed and trained using methanol and ethanol solutions. The model’s
Academic Editors:
hyperparameters were optimized to maximize its precision and accuracy, achieving 99.19%
Nicole Jaffrezic-Renault, Krishna Kant accuracy and 99.21% precision, with a training time of approximately 3 min, underlining
and Min Park the model’s potential for classifying simple Raman spectra. While further system validation
Received: 5 December 2024 with more samples, a more in-depth analysis of the biomarkers present in urine and the
Revised: 11 January 2025 integration with more sophisticated elements are necessary, this approach demonstrates
Accepted: 15 January 2025 the system characteristics of affordability and portability, making it a suitable solution for
Published: 23 January 2025 point-of-care applications and offering simplified accessibility for assessing the diseases
Citation: Domingos, C.; Fantoni, A.; risk outside clinical contexts.
Fernandes, M.; Fidalgo, J.; Pereira, S.A.
Low-Cost Raman Spectroscopy Setup Keywords: sensor; point of care; Raman spectroscopy; instrumentation; diagnosis;
Combined with a Machine Learning
kidney disease
Model. Sensors 2025, 25, 659. https://
doi.org/10.3390/s25030659

Copyright: © 2025 by the authors.

Licensee MDPI, Basel, Switzerland.
1. Introduction
This article is an open access article
distributed under the terms and The term acute kidney injury (AKI) reflects a set of heterogeneous conditions, leading
conditions of the Creative Commons to an abrupt decline in renal excretory function, causing azotemia and alterations in urinary
Attribution (CC BY) license flow. AKI is an increasing global concern and a costly clinical syndrome, affecting nearly
(https://creativecommons.org/
one-quarter of all hospitalized patients worldwide [1].
licenses/by/4.0/).

Sensors 2025, 25, 659 https://doi.org/10.3390/s25030659

Sensors 2025, 25, 659 2 of 22

Serum creatinine (sCr) levels and the volume of urine production, known as diuresis,
are commonly used for the diagnosis of acute kidney injury (AKI); however, these have
limitations. Serum creatinine is a biomarker with poor performance in detecting AKI since
creatinine levels only increase when around half of baseline renal function has been lost. At
this stage, AKI has already progressed to Chronic Kidney Disease (CKD) and the diagnosis
is considered late. In addition, creatinine concentration fluctuates with muscle metabolism
and with variations in extracellular volume, two very instable parameters in people with
AKI. Similarly, urine production is influenced by several factors including the volume of
fluids ingested and the use of diuretic medications [2–4]. Therefore, reliance solely on
changes in the serum creatinine level results in a delay in AKI management, leading to
unfavorable outcomes for patients [5].
The classical biomarker paradigm is that one test detects one disease; however, AKI
is a complex disease with multiple causes, and one biomarker is not sufficient for early
diagnosis. Thus, a panel of biomarkers may be necessary for the correct detection of
AKI [6]; however, as most of the AKI biomarkers are measured by ELISA tests, such parallel
simultaneous screening is not possible in a typical clinical procedure.
The classification of renal malfunctioning and its correlation with other diseases is of
overwhelming importance for assessing the risk of kidney failure, addressing a safe lifestyle
and allowing a gentle and safe aging process. Moreover, there are some specific situations,
like hospital emergency departments, extreme sports competitions, or military operations,
where it is not possible to send collected urine to a laboratory for a standard analysis. There
is a clear need and strong request for a fast screening of urine to support the decision of
the medical doctor about the need for immediate medical onsite care. Taking advantage
of the recent technological advances and affordable commercial devices, optoelectronic
and spectral analysis of urine can be performed onsite by low-cost portable detection
technology interfaced with signal processing and machine learning algorithms. These
algorithms, trained on a set of previously known diagnostic outputs, can correlate the urine
optoelectronic measured parameters with the person’s health status [7].
Raman spectroscopy is presently proposed as a novel approach to a multiplexed mea-
surement of specific AKI biomarkers and has become an increasingly valuable and compre-
hensive tool. This technique stands out for not requiring any prior sample preparation and
its compatibility with complex and aqueous samples; since the Raman signal produced by
water is relatively weak, it does not interfere with the signal of the other components. In
addition, due to its non-destructive nature, it can be used as a diagnostic technique and in
situ measurements. Thus, Raman spectroscopy allows for a complete characterization of
samples, including complex biological matrices like urine, which consolidates the relevant
role of this technology in clinical analysis and in the field of biomedicine [8], offering several
advantages over conventional methods for detecting kidney diseases.
Raman spectroscopy reduces the time required for evaluation and provides com-
prehensive information on the urine constituents, enabling early detection of molecular
markers associated with kidney diseases. This is accomplished by using only a small
amount of sample and reducing the need for exhaustive laboratory tests [9].
Due to the low intensity of the Raman light scattered by molecules, one of the main
challenges of this technique is the proper detection of its signal and obtaining spectra
with analyzable peaks. The Raman system must be able to eliminate the intense Rayleigh
radiation while amplifying the weak Raman radiation. To achieve this, the excitation
radiation needs to have high power, be as monochromatic as possible and have high
coherence and stability [10]. Currently, lasers are considered the ideal radiation source for
Raman spectroscopy experiments as they emit highly collimated and directional beams,
with high power and coherence, both spatially and temporally [11].
Sensors 2025, 25, 659 3 of 22

The fluorescence emitted by a sample, or by the impurities it may contain, is one

of the aspects that require particular attention in Raman spectroscopy. If the analyzed
material interacts with the incident radiation and emits fluorescence, the bands associated
with this type of radiation can overlap the sample’s Raman spectrum. Fluorescence is
significantly more intense than Raman radiation, resulting in high-intensity bands that
can contaminate the acquired spectrum and complicate the detection of Raman peaks. To
minimize fluorescence interference, a preliminary study should be carried out to analyze
the sample’s fluorescence response to various wavelengths. This approach allows for the
identification of a spectral region with minimal fluorescence emission, which corresponds
to the optimal wavelength of that sample [12].
The laser’s emission range, also referred to as its spectral amplitude, also plays an
important role in the spectral resolution of the Raman spectroscopy technique. When the
laser’s emission range covers a wide range of wavelengths, the Raman spectral resolution is
compromised, making it difficult to interpret the results and correctly assign the vibrational
modes of the molecules present in the sample. Therefore, the laser chosen needs to have
a wavelength that minimizes the fluorescence emitted by the sample and needs to be as
monochromatic and stable as possible.
Raman spectrum analysis can quickly become difficult to interpret due to its complex
nature and the large amount of information it contains, leading to the risk of losing im-
portant data and underseeing subtle concentration variations in biological markers. For
this reason, automating the process of urine spectra interpretation appears to be a viable
solution for their effective analysis and classification. A supervised learning model, based
on a neural network, could recognize characteristic patterns in these spectra, aiding in the
future possibility of diagnosis. If the supervised model is correctly trained with sufficient
data and information, it can reduce the analysis time and ensure greater consistency in
the results.
The primary objective of this study is to develop a portable and low-cost Raman
spectroscopy-based system, capable of acquiring high-quality spectra, for potential future
diagnosis applications. Furthermore, we aim to evaluate the possibility of using a super-
vised learning model to classify the Raman spectra acquired by the developed system.
While the goal is to support kidney disease diagnostics, the versatility of the proposed
system and the machine learning model extends its applicability to a wide range of diseases
and biological samples. Once validated, this system could serve as a foundation tool for
diagnosing other conditions where biochemical alterations in biofluids play a crucial role.
Additionally, its affordability and portability make this system suitable for use in both
clinical and emergency settings, where complex laboratory analyses are unfeasible.
By proposing a concurrent development of the hardware and signal processing archi-
tecture, this prototype attempts to move the system complexity from the hardware layer to
the software layer. It targets the development of a Cyber–Physical System that can assist in
overcoming the physical limitations of health services access, allowing health monitoring
independently based on geographic location and economical condition.

2. Raman System Development

The methodology adopted for the development of the Raman system was based on
the modular Starter Edition, available on the OpenRAMAN website, with the necessary
adaptations to meet the specific needs of this project. The Starter Edition spectrometer
was chosen because it is the simplest and most economical edition suitable to achieve the
low-cost criterion of the system we intend to build [13].
Sensors 2025, 25, 659 4 of 22
Sensors 2025, 25, x FOR PEER REVIEW 4 of 24

2.1. Description of System Components

2.1. Description of System
For simplicity’s Components
sake, the system will be divided into its main components: radiation
For simplicity’s sake,
source, optical elements and the system will beIn
detector. divided into its
addition, wemain
willcomponents:
refer to theradiation
data collection
software used. All the other important components can be founddata
source, optical elements and detector. In addition, we will refer to the collection
on the soft-
OpenRAMAN site.
ware used. All the other important components can be found on the OpenRAMAN site.
Most of the components used were purchased from the supplier Thorlabs. However,
Most of the components used were purchased from the supplier Thorlabs. However,
some parts of the system needed to be modified or designed from scratch to properly
some parts of the system needed to be modified or designed from scratch to properly fit
fit the specifics of the system. In Figure 1, we present the optical layout of the system
the specifics of the system. In Figure 1, we present the optical layout of the system devel-
developed
oped [14]. [14].

Figure 1.
Figure 1. Overview
Overview of the developed
of the developed system,
system,adapted
adaptedfrom
from[14].
[14].(Copyright
(Copyright 2025
2025Society
Societyof of Photo-
Photo-Optical
Optical Instrumentation
Instrumentation Engineers
Engineers (SPIE).
(SPIE). OneOne print
print or or electroniccopy
electronic copymay
may be
be made
made forforper-
personal use
sonal use only. Systematic reproduction and distribution, duplication of any material in this
only. Systematic reproduction and distribution, duplication of any material in this publication for apubli-
cation
fee for commercial
or for a fee or for commercial
purposes,purposes, and modification
and modification of the contents
of the contents of theofpublication
the publication are
are prohibited).
prohibited).
2.1.1. Radiation Source
2.1.1. Radiation Source
The main factors to take into consideration in laser selection are wavelength and
The main
emission range.factors to takethe
To identify into consideration
optimal wavelengthin laser selectionanalysis,
for Raman are wavelength and
urine’s fluorescence
emission range. To identify the optimal wavelength for Raman analysis, urine’s fluores-
emission was quantified under irradiation at various wavelengths using the FP-8300
cence emission was quantified under irradiation at various wavelengths using the FP-8300
spectrofluorometer from JASCO Corp, Tokyo, Japan. One sample of urine was irradiated
spectrofluorometer from JASCO Corp, Tokyo, Japan. One sample of urine was irradiated
with
with three wavelengths:
three wavelengths: 405405
nm,nm,
532532
nm nm
and and 635 These
635 nm. nm. These wavelengths
wavelengths weretochosen to
were chosen
evaluate theurine’s
evaluate the urine’sfluorescent
fluorescent response
response across
across a wide
a wide range.range.
The
The frequency of the incident radiation is the parameter thatthat
frequency of the incident radiation is the parameter most
most significantly
significantly im- impacts
Raman intensity.
pacts Raman Higher
intensity. Higherradiation frequencies
radiation frequencies result
result in ain a greater
greater intensity
intensity of Ramanof Raman
scattered radiation
scattered radiation from
from aa sample.
sample.Since
Sincefrequency
frequencyisisinversely
inverselyproportional
proportional to to
wave-
wavelength,
length, wavelengths
shorter shorter wavelengths
lead tolead to an increase
an increase in Raman in Raman intensity
intensity [10,15].
[10,15]. That said,
That said, wavelengths
wavelengths larger than 635 nm
larger than 635 nm were not evaluated. were not evaluated.
The urine’s fluorescence response at each wavelength is presented in Figure 2, with
The urine’s fluorescence response at each wavelength is presented in Figure 2, with the
the x-axis representing the wavelength range and the y-axis representing the fluorescence
x-axis representing the wavelength range and the y-axis representing the fluorescence intensity.
intensity.
As shown, urine exhibits a significant fluorescence when excited at 405 nm. Con-
sequently, a laser with this wavelength is not suitable for Raman measurements due to
extensive peak overlap and interference with the Raman signals. At 532 nm, the fluores-
cence intensity is lower, starting at 450 nm and rapidly decreasing after 532 nm. Despite
the considerable fluorescence, Raman signals are stronger when radiation sources with
shorter wavelengths are used. In contrast, excitation at 635 nm minimizes the fluorescence
emitted but reduces the Raman signal intensity, making the spectral analysis more complex,
especially while using low-power lasers.
Due to resource constraints, we were unable to investigate the impact of different
wavelengths on the noise in Raman spectra. Nevertheless, the 532 nm wavelength was
selected as the optimal choice, offering a balance between minimizing fluorescence inter-
Sensors 2025, 25, 659 5 of 22

ference
Sensors 2025, 25, x FOR PEER REVIEW and maximizing Raman signal intensity. Techniques such as spectral filtering
5 of 24 and
signal processing will be employed to further reduce fluorescence effects.

Figure 2.
Figure 2. Urine
Urinefluorescence
fluorescencespectra at 3 at
spectra excitation wavelengths
3 excitation [14]. (Copyright
wavelengths 2025 Society
[14]. (Copyright 2025ofSociety of
Photo-Optical Instrumentation Engineers (SPIE). One print or electronic copy mayper-
Photo-Optical Instrumentation Engineers (SPIE). One print or electronic copy may be made for be made for
sonal use only.
personal Systematic
use only. reproduction
Systematic and distribution,
reproduction duplication duplication
and distribution, of any materialofinany
thismaterial
publi- in this
cation for a fee
publication forora for
feecommercial purposes,
or for commercial and modification
purposes, of the contents
and modification of the
of the publication
contents of theare
publication
prohibited).
are prohibited).

As shown,
Several urine exhibitsavailable
commercially a significant
532fluorescence
nm lasers have whennarrow
excited emission
at 405 nm.ranges
Conse-and high
power, but they can be highly expensive, exceeding the budget proposed fortothe
quently, a laser with this wavelength is not suitable for Raman measurements due ex- low-cost
tensive peak overlap and interference with the Raman signals. At 532 nm, the fluorescence
system we intend to develop. Considering this analysis, the CPS532 compact low-power
intensity is lower, starting at 450 nm and rapidly decreasing after 532 nm. Despite the
laser from Thorlabs was selected as the most cost-effective option for the Raman system.
considerable fluorescence, Raman signals are stronger when radiation sources with
Its compact design and small size make it an ideal candidate for integration into portable
shorter wavelengths are used. In contrast, excitation at 635 nm minimizes the fluorescence
systems. Operable within a temperature range of 10 ◦ to 40 ◦ C, it has a typical power
emitted but reduces the Raman signal intensity, making the Cspectral analysis more com-
of 4.5 mW.
plex, especially while using low-power lasers.
Due to
Due toresource
heat dissipation during
constraints, we wereoperation,
unable tothe laser canthe
investigate reach highoftemperatures,
impact different po-
tentially
wavelengthscompromising
on the noiseitsinstability and spectral
Raman spectra. resolution.
Nevertheless, the Considering this, a was
532 nm wavelength temperature
selected system
control as the optimal choice, offering
was integrated intoathe
balance
laserbetween
to ensureminimizing
consistentfluorescence
operating inter-
conditions,
ference and maximizing Raman signal intensity. Techniques
guaranteeing the laser’s stability, durability and spectra stability.such as spectral filtering and
signalAprocessing will be employed
custom support structure to
tofurther reduce fluorescence
accommodate the CPS532 effects.
laser and its control system
Several commercially available 532 nm lasers have narrow emission ranges and high
was fabricated using 3D printing/milling processes. A Peltier element in the setup ensures
power, but they can be highly expensive, exceeding the budget proposed for the low-cost
the heat transfer between the laser module and a dissipator. A picture of the6laser
Sensors 2025, 25, x FOR PEER REVIEW of 24 support
system we intend to develop. Considering this analysis, the CPS532 compact low-power
is reported
laser in Figure
from Thorlabs was3. selected as the most cost-effective option for the Raman system.
Its compact design and small size make it an ideal candidate for integration into portable
systems. Operable within a temperature range of 10 °C to 40 °C, it has a typical power of
4.5 mW.
Due to heat dissipation during operation, the laser can reach high temperatures, po-
tentially compromising its stability and spectral resolution. Considering this, a tempera-
ture control system was integrated into the laser to ensure consistent operating conditions,
guaranteeing the laser’s stability, durability and spectra stability.
A custom support structure to accommodate the CPS532 laser and its control system
was fabricated using 3D printing/milling processes. A Peltier element in the setup ensures
the heat transfer between the laser module and a dissipator. A picture of the laser support
is reported in Figure 3.

Figure 3.
Figure 3. Detail
Detailofofthe support
the manufactured
support to accommodate
manufactured the CPS532
to accommodate laser and
the CPS532 the temperature
laser and the temperature
control system.
control system.

2.1.2. Optical Elements

This system includes a series of filters and mirrors to achieve high-quality and high-
resolution Raman spectra. All optical elements were purchased from Thorlabs.
The green radiation emitted by the laser begins its optical path by striking the direc-
Sensors 2025, 25, 659 6 of 22

2.1.2. Optical Elements

This system includes a series of filters and mirrors to achieve high-quality and high-
resolution Raman spectra. All optical elements were purchased from Thorlabs.
The green radiation emitted by the laser begins its optical path by striking the direc-
tional mirror, model PF10-03-G01, which directs the beam towards the dichroic mirror.
The dichroic mirror acts as a wavelength-selective filter, transmitting and reflecting light
according to its wavelength. For this system, the dichroic mirror chosen was DMLP550,
with a cutoff wavelength of 550 nm. This mirror reflects radiation below 550 nm, ensur-
ing that the 532 nm laser radiation reaches the sample, and transmits all radiation above
550 nm. Both mirrors are mounted on KM100 kinematic brackets, allowing precise align-
ment of the laser beam with the sample and the spectrometer’s entrance.
Before reaching the sample, the laser beam passes through a lens integrated into the
cuvette holder. This lens focuses the beam onto a small area of the cuvette, concentrating
all the radiation on a specific region of the sample, and it also works as a collimator for the
Raman radiation coming out from the sample.
Upon interacting with the sample, both Raman and non-Raman scattered radiation
are emitted in all directions. A portion of this scattered radiation is captured by the lens
of the cuvette holder and collimated towards the dichroic mirror. The Raman scattered
radiation, with wavelengths in the yellow–red range (above 550 nm), passes through the
dichroic mirror and is directed to the edge-pass filter.
The edge-pass filter, model FELH0550, removes any residual non-Raman radiation,
with wavelengths below 550 nm, ensuring that only Raman radiation proceeds to the
detector. The cutoff wavelength of the filter is precisely adjusted to extend the Raman
spectrum while avoiding the excitation wavelength by slightly tilting it.
As the radiation passes through the dichroic mirror and edge-pass filter, it undergoes
some horizontal displacement due to the thickness of these components. To correct this,
the beam path is corrected using the WG11050-A compensation window.
After the realignment, the radiation passes through a set of lenses and a slit. The first
achromatic lens, AC127-019-A, collects the Raman radiation focusing on an S50K slit with a
50 µm aperture, mounted in a CRM1T/M rotation cage. This slit ensures precise control of
the amount of radiation that passes to the subsequent steps of the process, mainly defining
the spectrometer resolution. A second achromatic lens, AC254-050-A, collimates the light
exiting the slit and adjusts it to the grating surface.
The beam then reaches the GR25-1205 diffraction grating, with a density of
1200 lines per millimeter. This grating disperses the light into various directions based on
its wavelength, allowing the analysis of the radiation spectrum.
Ultimately, the dispersed radiation is focused by an objective on the detector, which
records the intensity at each position corresponding to specific wavelengths in the spectrum.

2.1.3. Detector/Spectrometer
A FLIR Blackfly GigE camera is used as the detector, i.e., model BFLY-PGE-31S4M-C.
This choice was based on its full compatibility with the selected data collection software,
ensuring perfect integration of the camera into the developed system. This PointGrey
model offers a resolution of 2048 × 1536 pixels with a CMOS-type sensor (Teledyne Vision
Solutions, Richmond, BC, Canada). The focusing lens chosen to collect the Raman radiation
was the 50 mm lens MVL50M23 from Thorlabs.

2.1.4. Data Collection Software

The software chosen for Raman spectra acquisition is the Spectrum Analyzer, available
on the OpenRAMAN website. This software was designed to operate with the type of
 radiation was the 50 mm lens MVL50M23 from Thorlabs.

2.1.4. Data Collection Software

Sensors 2025, 25, 659 7 of 22
The software chosen for Raman spectra acquisition is the Spectrum Analyzer, avail-
able on the OpenRAMAN website. This software was designed to operate with the type
of system
system developed,
developed, minimizing
minimizing risks
risks ofof incompatibilityand
incompatibility andensuring
ensuringconsistency
consistencyininthe
the
spectra obtained.
spectra obtained. Additionally,
Additionally, the
the software
software isisfree
freeand
andwas
waspreviously
previouslyvalidated
validatedfor
foruse
use
withthe
with theselected
selectedcamera
cameramodel,
model,guaranteeing
guaranteeingcompatibility
compatibilitywith
withthe
thesystem
system[16].
[16].

2.2.
2.2. System
System Assembly
Assembly
The
Theassembly
assemblyprocess
processisisdetailed
detailedon onthe
theOpenRAMAN
OpenRAMANwebsite website[13].
[13].
The
The baseplate
baseplate was
was fabricated
fabricated from
from anan aluminum
aluminum plate plate using
using computer
computer numerical
numerical
control
control (CNC)
(CNC) machining,
machining, according
according to to the
the distances
distances andand dimensions
dimensions specified
specified on
on the
the
OpenRAMAN platform. The cuvette holder used in the system is
OpenRAMAN platform. The cuvette holder used in the system is the CVH100/M model,the CVH100/M model,
chosen
chosenfor
forits
itsversatility,
versatility,allowing
allowingthetheintegration
integrationofof various
variousanalysis techniques
analysis techniques to examine
to exam-
the same sample.
ine the same sample.
To
Toprotect
protectthe
theRaman
Raman system
system instrumentation
instrumentation from external
from interference,
external twotwo
interference, custom-
cus-
made covers were designed to enclose the entire optical system. These covers
tom-made covers were designed to enclose the entire optical system. These covers were were manufac-
tured using a 3Dusing
manufactured printer,
a 3Densuring
printer,effective
ensuringprotection against light
effective protection and external
against light and impurities.
external
Below, Figure 4 presents an image of the top view of the developed Raman
impurities. Below, Figure 4 presents an image of the top view of the developed Raman system without
the covers.
system The alignment
without the covers.and Thecalibration
alignmentofand thecalibration
Raman system were
of the performed
Raman systemfollowing
were per-
the OpenRAMAN guidelines.
formed following the OpenRAMAN guidelines.

Figure4.4.Top
Figure Topview
viewof
ofthe
thedeveloped
developedRaman
Ramansystem.
system.

2.3. Optimization of System Components

2.3.1. Laser Operating Temperature Optimization
The implementation of the temperature control system also aimed to optimize the
emission spectrum of the CPS532 laser. Adjusting the laser’s operating temperature can
enhance its spectral resolution and narrow its emission range, making it more efficient for
Raman analysis.
To evaluate the laser’s behavior at various temperatures, two experiments were con-
ducted. The first focused on understanding the effect of temperature on the full width
at half maximum (FWHM) of the laser peak. The laser spectrum was acquired using the
CCS200/M spectrometer and the SMA905 fiber optic cable, which captured and transmitted
the laser beam to the spectrometer. The fiber was integrated into the CVH100/M cuvette
holder. The spectrometer has its own software, ThorSpectra. After setting up the equip-
ment and connecting the spectrometer to the software, the laser spectrum was acquired at
operating temperatures of 20 ◦ C, 25 ◦ C, 30 ◦ C, 35 ◦ C and 40 ◦ C. To prevent signal saturation,
all spectra were acquired using an integration time of 0.1 ms.
At an operating temperature of 20 ◦ C (cf. Figure 5), the CPS532 laser spectrum
exhibits two peaks: a primary peak at 532.8 nm, with an intensity of 0.90 and an FWHM
of 1013.7 pm, and a secondary peak at 537.3 nm, with an intensity of 0.04. The secondary
Sensors 2025, 25, 659 8 of 22

peak is attributed to the use of the SMA905 optical fiber, which, despite introducing this
interference, is essential for ensuring the reproducibility and comparability of the results.
Sensors 2025, 25, x FOR PEER REVIEW 9 of 24
Notably, the optical fiber will not be used for Raman spectra acquisition, so this interference
will not affect the quality or accuracy of the spectra obtained with the system.

Sensors 2025, 25, x FOR PEER REVIEW 9 of 24

◦ C, using the “Peak Track” tool from ThorSpectra.

Figure 5. CPS532
Figure 5. CPS532 laser
laser spectrum
spectrum at 20 °C,
at 20 using the “Peak Track” tool from ThorSpectra.

Subsequently, the laser’s operating temperature was varied to 25 ◦ C, 30 ◦ C, 35 ◦ C and

Influence of laser operating temperature on the FWHM of its main
40 ◦ C, with the obtained spectra presented in Appendix A. Figure 6 illustrates the variation
peak
in the FWHM
Figure 5. CPS532aslaser
a function
spectrumof the
at 20laser’s operating
°C, using temperature.
the “Peak Track” tool from ThorSpectra.
1300
1212.4 pm
Influence of laser operating temperature on the FWHM of its main
1100 peak
1300 1013.7 pm
FWHM (pm)

1212.4 pm
900
1100
1013.7 pm
763.5 pm 767.6 pm
FWHM (pm)

700
900 649.8 pm

500
20 763.5
25 pm 30 767.6
35 pm 40
700
Laser operating temperature (ᵒC)
649.8 pm
Figure 6. Influence of laser operating temperature on the FWHM of its main peak.
500
20 25 30 35 40
Table 1. Acquisition parameters used
Laserfor the acquisition
operating temperatureof(ᵒC)
the ethanol spectrum.

Exposure Time(s) Gain (dB) ROI (px) Average Number of Images

Figure6.6.Influence
Figure Influenceof
oflaser
laseroperating
operatingtemperature
temperatureon
onthe
theFWHM
FWHMof
ofits
itsmain
mainpeak.
peak.
31.6 11.5 128 3
TableContrary to expectations,
1. Acquisition the relationship
parameters used between
for the acquisition the
of the CPS532
ethanol laser’s operating tem-
spectrum.
perature and its FWHM is non-linear. The FWHM decreases until 30 ◦ C and increases at
Exposure Time(s) Gain (dB) ROI (px) Average Number of Images
higher temperatures, peaking at 40 ◦ C, the laser’s maximum operating temperature. At
31.6 11.5 128 3
40 ◦ C, the main peak exhibits instability in shape and a significant intensity reduction,
dropping to 0.51. These results indicate that high temperatures negatively affect the laser
Sensors 2025, 25, 659 9 of 22

performance, making operation under such conditions undesirable. At 30 ◦ C, the FWHM of

the main peak reaches its minimum value of 649.8 pm, with an acceptable intensity of 0.82.
Thus, 30 ◦ C is the temperature that minimizes the spectral range and that, consequently,
increases the spectral resolution of Raman spectra.
In a second phase, the spectrum of ethanol (chemical formula: CH3 CH2 OH) was
acquired using the developed Raman system at three different laser operating temperatures:
25 ◦ C, 30 ◦ C and 35 ◦ C (cf. Figure 7). The main objective of this phase was to assess how the
laser’s operating temperature influenced the quality and accuracy of Raman spectra. The
ethanol spectrum was acquired with Spectrum Analyzer software 1.2, with the acquisition
Sensors 2025, 25, x FOR PEER REVIEW
parameters reported in Table 1. The visualization of the spectrum obtained was performed 10 of 24
using SpectraGryph software, version 1.2.

0.037
0.036
0.035
0.034
0.033 Etanol 25ºC
0.032 Etanol 30ºC
0.031 Etanol 35ºC

0.03
0.029
0.028
0.027
0.026
0.025
0.024
0.023
0.022
0.021
Intensity (a.u.)

0.02
0.019
0.018
0.017
0.016
0.015
0.014
0.013
0.012
0.011
0.01
0.009
0.008
0.007
0.006
0.005
0.004
0.003
0.002
0.001
0
400 500 600 700 800 900 1000 1100 1200 1300 1400 1500 1600 1700 1800 1900 2000 2100 2200 2300 2400 2500 2600 2700 2800 2900 3000 3100 3200 3300 3400 3500
Wavenumbers [cm-1 ]

◦ C, 30 ◦ C and 35 ◦ C.
Figure 7. Raman
Figure 7. Raman spectrum
spectrum of
of ethanol
ethanol at
at laser
laser operating
operating temperatures
temperatures of
of25
25 °C, 30 °C and 35 °C.

1. Acquisition
TableThe presence parameters used forethanol
of characteristic the acquisition
peaks ofinthe
theethanol spectrum.
spectrum, although not highly
intense, demonstrates that the system operates with sufficient precision to capture molec-
Exposure Time(s) Gain (dB) ROI (px) Average Number of Images
ular vibration signals from the sample.
31.6 11.5 128 3
When comparing the three ethanol spectra, the spectrum obtained at 35 °C shows the
lowest peak intensities, particularly in zone (4), around 2800 to 3000 cm−1. Between the
Theobtained
spectra presenceat of25 °C and 30 °C,
characteristic ethanol peaksat
the peaks the°C
in25 spectrum, although
exhibit slightly not highly
lower in-
intensity.
tense, demonstrates
Considering that thethat theoptimization
first system operates
phasewith
identified °C as the
sufficient30precision to optimal
capture molecular
operating
vibration signals from the sample.
temperature for enhancing laser performance and given the quality of the ethanol spec-
◦ C shows
trumWhen comparing
acquired the three ethanol
at this temperature, spectra, the
we concluded that 30 °C maximizes
spectrum obtained at the35overall effi-
the lowest peak intensities, particularly in zone (4), around 2800 to 3000 cm − 1 . Between
ciency of the developed Raman system.
the spectra obtained at 25 ◦ C and 30 ◦ C, the peaks at 25 ◦ C exhibit slightly lower intensity.
2.3.2. Optimization
Considering that theoffirst
Spectrum Analyzer
optimization Software
phase identified 30 ◦ C as
Acquisition Parameters
the optimal operating
temperature for enhancing
The software laser performance
used allows andof
for adjustments given the quality
acquisition of the such
settings, ethanol
as spectrum
exposure
acquired at this temperature, we concluded that 30 ◦ C maximizes the overall efficiency of
time and the average number of images. These adjustments are critical for ensuring result
the developed
stability and for Raman system.
obtaining high-quality spectra with a strong signal-to-noise ratio (SNR).
For this optimization, the ethanol Raman spectrum was acquired using different ac-
2.3.2. Optimization of Spectrum Analyzer Software Acquisition Parameters
quisition parameters, with the laser operating at 30 °C. Each parameter was varied three
times,The
as software
reported used allows
in Table 2. for adjustments of acquisition settings, such as exposure
time and the average number of images.
In the first stage, the exposure time,These
whichadjustments
controls theare
timecritical
during forwhich
ensuring result
the Point-
stability and for
Grey camera obtaininglight,
“captures” high-quality spectra with
was evaluated. a strong signal-to-noise
It is important to note that theratio (SNR).
maximum
exposure time allowed by the software is 31.6 s. Additionally, spectra acquired with ex-
posure times shorter than 2.1 s exhibited high noise levels, which made their analysis ex-
haustive and confusing.
The shortest exposure time of 2.1 s resulted in a Raman spectrum with high noise,
which interfered with the ethanol peaks by artificially increasing their intensity. However,
Sensors 2025, 25, 659 10 of 22

For this optimization, the ethanol Raman spectrum was acquired using different
acquisition parameters, with the laser operating at 30 ◦ C. Each parameter was varied three
times, as reported in Table 2.

Table 2. Acquisition parameters used to study the impact of exposure time on the quality of the
ethanol Raman spectrum.

Name Exposure Time(s) Gain (dB) Average Number of Images

Ethanol 2.1 s 2.1 3.8 2
Ethanol 10 s 10 3.8 2
Ethanol 31.6 s 31.6 3.8 2

In the first stage, the exposure time, which controls the time during which the Point-
Grey camera “captures” light, was evaluated. It is important to note that the maximum
exposure time allowed by the software is 31.6 s. Additionally, spectra acquired with ex-
posure times shorter than 2.1 s exhibited high noise levels, which made their analysis
exhaustive and confusing.
Sensors 2025, 25, x FOR PEER REVIEW The shortest exposure time of 2.1 s resulted in a Raman spectrum with high11noise, of 24
which interfered with the ethanol peaks by artificially increasing their intensity. However,
this increase was solely due to the added noise, compromising the spectrum analysis. The
noise
to havemay have noise
a high originated
level.from
Basedtheon
instrumentation,
this, we concludeincluding
that anlaser or thermal
exposure timefluctuations
of 31.6 s is
or
optimal for improving Raman spectra quality, especially for urine analysis. See the
electronic noise. Increasing the exposure time to 10 s significantly reduced noise,
Figure 8.
as expected, since a longer acquisition time allowed the camera to integrate the incoming
light,
Table thus increasing
2. Acquisition the signal-to-noise
parameters used to studyratio.
the Further
impact ofincreasing the acquisition
exposure time time
on the quality to
of the
the maximum
ethanol Raman value, 31.6 s, caused subtle differences in noise levels, especially in the range
spectrum.
between 1500 cm−1 and 2800 cm−1 . Although this longer exposure did not substantially
Name Exposure Time(s) Gain (dB) Average Number of Images
improve the ethanol spectrum, it can be beneficial for urine spectra, which tend to have a
Ethanol 2.1 s 2.1 3.8 2
high noise level. Based on this, we conclude that an exposure time of 31.6 s is optimal for
Ethanol 10 s 10 3.8 2
improving Raman spectra quality, especially for urine analysis. See Figure 8.
Ethanol 31.6 s 31.6 3.8 2
0.06

0.055
Ethanol 2.1 s
0.05 Ethanol 10 s
Ethanol 31.6 s

0.045

0.04

0.035
Intensity (a.u.)

0.03

0.025

0.02

0.015

0.01

0.005

0
400 600 800 1000 1200 1400 1600 1800 2000 2200 2400 2600 2800 3000 3200 3400
Wavenumbers [cm-1 ]

Figure 8. Impact
Figure 8. Impact of
of exposure
exposure time
time on
on the
the quality
quality of
of the
the ethanol
ethanol Raman
Raman spectrum.
spectrum.

Then,
Then, the
the impact
impact of
of the
the number
number of of images
images acquired
acquired and
and combined
combined to to generate
generate the
the
final
final spectrum was evaluated; the parameters are reported in Table 3 and the obtained
spectrum was evaluated; the parameters are reported in Table 3 and the obtained
spectra
spectra are
arereported
reportedininFigure
Figure9. 9.
The analysis
The of the
analysis obtained
of the spectra
obtained reveals
spectra that when
reveals only
that when
two images are acquired, the noise is significantly higher, particularly in regions
only two images are acquired, the noise is significantly higher, particularly in regions without
peaks,
withoutcomplicating the analysis
peaks, complicating theofanalysis
more complex
of morespectra,
complexsuch as urine
spectra, spectra.
such Increasing
as urine spectra.
Increasing the number of images to 12, the noise level in these regions decreases consid-
erably while preserving the intensity of ethanol’s characteristic peaks. This provides the
highest SNR ratio, as it minimizes noise while maintaining distinct peaks. Acquiring spec-
tra with a higher number of images resulted in a similar noise level in the regions without
Sensors 2025, 25, 659 11 of 22

the number of images to 12, the noise level in these regions decreases considerably while
preserving the intensity of ethanol’s characteristic peaks. This provides the highest SNR
ratio, as it minimizes noise while maintaining distinct peaks. Acquiring spectra with a
higher number of images resulted in a similar noise level in the regions without peaks.
However, the intensity of the peaks reduced considerably. Therefore, we conclude that
the average number of images that improve the overall quality of the spectrum while
maintaining a high SNR ratio is 12 images.

Table 3. Acquisition parameters used to study the impact of the average number of images on the
quality of the ethanol spectrum.

Sensors 2025, 25, x FOR PEER REVIEW Name Exposure Time(s) Gain (dB) 12 of 24
Average Number of Images
Ethanol 2 images 10 3.8 2
Ethanol 12 images 10 3.8 12
Ethanol
Ethanol24
12images
images 1010 3.8
3.8 24
12
Ethanol 24 images 10 3.8 24
0.065

0.06

Ethanol 2 images
0.055 Ethanol 12 images
Ethanol 24 images
0.05

0.045

0.04
Intensity (a.u)

0.035

0.03

0.025

0.02

0.015

0.01

0.005

0
400 600 800 1000 1200 1400 1600 1800 2000 2200 2400 2600 2800 3000 3200 3400
Wavenumbers [cm-1 ]

Impact of
Figure 9. Impact
Figure of the
the average
average number
number of
of images on the quality of the ethanol Raman spectrum.

Theofgain
2.4. Role parameter,
the Laser’s Powerwhich
in the represents the amplification of the electrical signal gen-
Raman Spectrum
erated by the camera’s sensor after capturing the light, was also evaluated. However,
Maintaining the overall hardware configuration unchanged, we tested our system
the ethanol spectra did not exhibit significant variations in noise or intensity, making it
with a new laser with a higher power, exceeding 80 mW, and a short spectral linewidth
impossible to draw concrete conclusions about its impact.
(nominal FWHM = 0.003 nm), which is manufactured by Frankfurt Laser Company, i.e.,
This study faced limitations, as it was not feasible to explore all possible parameter
model FPYL-532-80T-LN-S. To evaluate its performance compared to the previously used
values. However, the optimal settings to enhance the quality of the ethanol spectrum were
laser, we acquired the ethanol spectrum under similar conditions (cf. Figure 10).
31.6 s as the exposure time and an average of 12 images.
5
4.5
4 2.4. Role of the Laser’s Power in the Raman Spectrum
FPYL-532Sw
3.5
CPS532
3 Maintaining the overall hardware configuration unchanged, we tested our system
2.5
2 with a new laser with a higher power, exceeding 80 mW, and a short spectral linewidth
1.5
1 (nominal FWHM = 0.003 nm), which is manufactured by Frankfurt Laser Company, i.e.,
0.5
0
model FPYL-532-80T-LN-S. To evaluate its performance compared to the previously used
0.04
0.035
laser, we acquired the ethanol spectrum under similar conditions (cf. Figure 10).
Intensity (a.u.)

0.03
0.025
0.02
0.015
0.01
0.005
0
400 600 800 1000 1200 1400 1600 1800 2000 2200 2400 2600 2800 3000 3200 3400
Wavenumbers [cm-1 ]

Figure 10. Comparison of the ethanol spectrum obtained with two different lasers.

The differences between the spectra are notable, with the higher-powered laser
 Maintaining the overall hardware configuration unchanged, we tested our system
with a new laser with a higher power, exceeding 80 mW, and a short spectral linewidth
(nominal FWHM = 0.003 nm), which is manufactured by Frankfurt Laser Company, i.e.,
Sensors 2025, 25, 659 12used
of 22
model FPYL-532-80T-LN-S. To evaluate its performance compared to the previously
laser, we acquired the ethanol spectrum under similar conditions (cf. Figure 10).
5
4.5
4
FPYL-532Sw
3.5
CPS532
3
2.5
2
1.5
1
0.5
0
0.04
0.035
Intensity (a.u.)

0.03
0.025
0.02
0.015
0.01
0.005
0
400 600 800 1000 1200 1400 1600 1800 2000 2200 2400 2600 2800 3000 3200 3400
Wavenumbers [cm-1 ]

Figure 10. Comparison

Figure 10. Comparison of
of the
the ethanol
ethanol spectrum
spectrum obtained
obtained with
with two
twodifferent
differentlasers.
lasers.

The
The differences
differencesbetween
betweenthethe
spectra are notable,
spectra with the
are notable, higher-powered
with the higher-poweredlaser provid-
laser
ing a substantial
providing enhancement
a substantial in the intensity
enhancement of the Raman
in the intensity of the peaks,
Ramanlower
peaks,noise
lower and shorter
noise and
acquisition times. Using the CPS-532 laser, the maximum intensity achieved
shorter acquisition times. Using the CPS-532 laser, the maximum intensity achieved was was only 0.035,
which was which
only 0.035, notablywas lower. In contrast,
notably lower. Inthe FPYL-532
contrast, laser allows
the FPYL-532 forallows
laser intensities to reach
for intensities
up to 4, representing a remarkable improvement. This enhancement
to reach up to 4, representing a remarkable improvement. This enhancement directly im-directly improves
the quality
proves the and clarity
quality andofclarity
the obtained data, highlighting
of the obtained the critical
data, highlighting role
the of laser
critical rolepower in
of laser
Raman spectroscopy analyses.
power in Raman spectroscopy analyses.
By
By analyzing
analyzing the
the acquired
acquired spectra,
spectra, we
we may
may conclude
conclude that
that the
the resolution
resolution of of the
the system
system
with CPS-532 is about 15 cm −1 ; a similar value was obtained with FPYL-532, being the
with CPS-532 is about 15 cm ; a similar value was obtained with FPYL-532, being the
−1

main
main advantage
advantage of of the
thehigher
highersignal/noise
signal/noise ratio.
ratio. Further
Further improvements
improvements in in the
the resolution
resolution
could be obtained by taking advantage of the enhanced laser power for reducing the slit
size. However, the FPYL laser is significantly more expensive than the one employed in
this system, making it unsuitable for low-cost approaches, and it has been used only as
a comparison term. Once the laser operation entered a constant temperature condition,
no significant fluctuations in the peak positions were observed, which can be considered
lower than the system resolution.
The spectral resolution of available commercial Raman portable setups ranges from
8 cm−1 for the Wasatch WP532X (532 nm) to 9.7 cm−1 for the Thorlabs Portable Coded-
Aperture Raman Spectrometer RASP2 (682 nm) and 10 cm−1 for the Hamamatsu C15471
(785 nm). All these systems are extremely compact and efficient, but they are available at a
price level that is too high to address the large-scale employment of a POC system.
While a research-level Raman spectrometer can present an impressive spectral resolu-
tion at less than 1 cm−1 over a wide spectral range (see, for example, the Horiba LabRAM
Soleil), when compared with top-class Raman equipment, one major advantage of the
simplicity of the mechanical parts are the soft requirements for precise alignment, which
support the stability we observed over time and the low sensitivity to environmental
conditions, like humidity and temperature. Targeting our application playground, once
the spectral range of interest is defined within a specific clinical context, as assessed by a
large-scale clinical study, the spectral resolution can be improved by tailoring the properties
of the diffraction grating, sacrificing the spectral range.

3. Urine Spectrum Acquisition

To validate the developed system for urine spectra acquisition, its performance was
evaluated in this context. The focus of this evaluation was not to identify the biomarkers
present in the urine, nor to conduct their detailed analysis, but to ensure that the Raman
system is able to acquire urine spectra with precision and consistency. For this purpose,
Sensors 2025, 25, 659 13 of 22

five samples of urine were used. The acquisition parameters used are described in Table 4,
and the spectra are reported in Figure 11.

Table 4. Acquisition parameters used for the acquisition of urine spectra.

Exposure Time(s) Gain (dB) ROI (px) Average Number of Images

Sensors 2025, 25, x FOR PEER REVIEW 31.6 1.5 128 10 14 of 24

0.065

0.06

Urine sample 1
0.055 Urine sample 2
Urine sample 3
0.05 Urine sample 4
Urine sample 5

0.045

0.04
Intensity (a.u.)

0.035

0.03

0.025

0.02

0.015

0.01

0.005

0
400 450 500 550 600 650 700 750 800 850 900 950 1000 1050 1100 1150 1200 1250 1300 1350 1400 1450 1500
Wavenumbers [cm-1 ]

Figure11.
Figure Ramanspectrum
11.Raman spectrumofofthe
thefive
fiveurine
urinesamples.
samples.

By analyzing
4. Data Treatmentthe obtained spectra, it is evident that all five samples exhibit consistent
and Validation
−1 −1 −1
peaks around 415 cm and 735 cm . Regarding the 415 cm peak, a limited amount of
To validate that the system runs a correct spectra acquisition process, a typical
information is available in the literature; however, this peak is likely associated with the
method is to compare the results with the production data of a well-known composition.
presence of glycogen in urine, which typically exhibits a Raman peak around 480 cm−1 [17].
So, following−this well-established procedure, we tested our Raman system by comparing
The 735 cm 1 peak is probably related to creatinine, despite its usually reported values
its results with other data published in the literature about simple alcohols, like ethanol
being around 640 cm−1 and 700 cm−1 [9,18–20]. Additionally, a broader and less intense
and methanol. The Raman analysis − of these alcohols was largely developed to address the
peak is observed between 990 cm 1 and 1090 cm−1 , likely associated with urea, which
problem of detecting the so-called “fake wine” − production, which is often accompanied
typically presents Raman peaks around 1000 cm 1 and 1006 cm−1 [9,18,19]. The intensities
by some methanol components, producing toxicity for excessive dosage [21]. The im-
of these peaks varied across the samples, demonstrating the system’s sensitivity to the
portance of the clinical problem associated with wine adulteration justifies the abundance
differing compound concentrations in urine and its consistency in the acquisition of the
of Raman data available. We may conclude that the results obtained for both alcohols
spectra. Overall, the intensity of the peaks remained relatively low, with noise present
agree with the data available in the literature [22], as confirmed below in Figure 12. It is
throughout the spectra. Enhancing peak intensity and reducing noise would require a laser
possible to operate a Gaussian convolution on the raw data to obtain a clean plot of the
with higher spectral resolution and greater power.
spectra, as reporteddemonstrates
This analysis in Figure 13. that
Whilethethe readability
developed of thesystem
Raman plots isisgreatly improved,
both consistent in
the optimal filtering parameters are dependent on the data quality and the
its spectra acquisitions and sensitive to variations in compound concentrations within anumber and
intensity
complexof the peaks,
matrix. Thesesoresults
they can be inferred
reinforce correctlypotential
the system’s only in a to
second
detectphase; otherwise,
changes in urine
the blurring operation directed to remove noise may become associated
composition, essential for its application in future diagnostic analyses. with the risks of
removing relevant details. This is evident in the treatment of the urine samples, which are
much
4. Dataricher in content components
Treatment and Validationand, correspondingly, in the number of peaks, as de-
picted in Figure 14. If the complete composition of a urine sample is not known a priori,
To validate that the system runs a correct spectra acquisition process, a typical method
the filtering operation should be executed only within a clinical framework, having pre-
is to compare the results with the production data of a well-known composition. So,
viously defined which peaks need to be searched. To obtain more information on this
following this well-established procedure, we tested our Raman system by comparing its
topic, a clinical study, with urine samples collected in a hospital emergency department,
results with other data published in the literature about simple alcohols, like ethanol and
is foreseen for the future. If enough urine samples are available with an associated clinical
methanol. The Raman analysis of these alcohols was largely developed to address the
diagnosis, it will be possible to conduct a direct search of the clinically significant peaks.
problem of detecting the so-called “fake wine” production, which is often accompanied by
Within this framework, instead of using Raman spectroscopy for a detailed chemical de-
some methanol components, producing toxicity for excessive dosage [21]. The importance
scription of the urine contents, it will be possible to establish a correlation between a spe-
cific disease risk and the presence and intensity of determinate peaks. This consideration
drives the attention directly to the development of a classification scheme suitable to be
elaborated by a machine learning model. To support this point of view and demonstrate
the feasibility of a machine learning approach for Raman spectra interpretation, in the
Sensors 2025, 25, 659 14 of 22

of the clinical problem associated with wine adulteration justifies the abundance of Raman
data available. We may conclude that the results obtained for both alcohols agree with the
data available in the literature [22], as confirmed below in Figure 12. It is possible to operate
a Gaussian convolution on the raw data to obtain a clean plot of the spectra, as reported
in Figure 13. While the readability of the plots is greatly improved, the optimal filtering
parameters are dependent on the data quality and the number and intensity of the peaks,
so they can be inferred correctly only in a second phase; otherwise, the blurring operation
directed to remove noise may become associated with the risks of removing relevant details.
This is evident in the treatment of the urine samples, which are much richer in content
components and, correspondingly, in the number of peaks, as depicted in Figure 14. If
the complete composition of a urine sample is not known a priori, the filtering operation
should be executed only within a clinical framework, having previously defined which
peaks need to be searched. To obtain more information on this topic, a clinical study, with
urine samples collected in a hospital emergency department, is foreseen for the future. If
enough urine samples are available with an associated clinical diagnosis, it will be possible
to conduct a direct search of the clinically significant peaks. Within this framework, instead
of using Raman spectroscopy for a detailed chemical description of the urine contents, it
will be possible to establish a correlation between a specific disease risk and the presence
and intensity of determinate peaks. This consideration drives the attention directly to the
development of a classification scheme suitable to be elaborated by a machine learning
model. To support this point of view and demonstrate the feasibility of a machine learning
approach for Raman spectra interpretation, in the following section, we report a case study
elaborating on the detection of the methanol percentage in an ethanol solution.
The analysis of urine samples is interesting as a proof of concept, pointing the attention
to an application domain where a portable Raman analysis can contribute to addressing an
important social problem. However, the peak intensities and distributions in the spectra
of urine samples can present a large variability depending on the specific condition of the
donors, so they cannot be used to objectively infer the performance level of the prototype.
So, as a comparison with other data reported in the literature for methanol detection using
portable Raman systems, Table 5 reports the level of detection (LOD) and some technical
specifications of the systems. Both the CPS532 and FPYL532 versions are included in the
table. As a reference of values, it is worth recalling that drinks made from “industrial
methylated spirits” [5% (v/v) methanol:95% (v/v) ethanol] can cause severe and even fatal
illness, and the current EU general limit for naturally occurring methanol is 10 g methanol/L
ethanol (which equates to 0.4% (v/v) methanol at 40% alcohol) [23]. Regarding the low-
power version with the CPS532 laser, the LOD is higher than the EU recommendation,
as the 2840 cm−1 peak remains hidden by the noise for a concentration lower than 4.76%
of methanol. Moving to the higher-power FPYL532 (leaving all the other components
unchanged), the S/N is much improved, and the LOD falls to a value below 0.25%. It
should be noted that the 80 mW of laser power is still well below the standard hundreds of
mW normally used in infrared models, making the management of the safety requirements
for a portable system easier.

Table 5. Level of detection (LOD) of methanol in a solution with ethanol.

Model LOD (%) LOD (g/L) Laser Wavelength (nm) Laser Power (mW) More Significant Peak (cm−1 )
CPS532 4.76 37.67 532 5 2840
FPYL532 0.25 1.97 532 80 2840
Ref. [24] 0.23–0.39 830 400 1030
Ref. [25] 0.025 1064 450 1023
Ref. [22] 20 532 40 2840
Sensors 2025, 25, 659
Sensors 2025, 25, x FOR PEER REVIEW 15 of 2415 of 22
Sensors 2025, 25, x FOR PEER REVIEW 15 of 24

0.04

2839 0.03031
0.038
0.04

2839 0.03031
0.036
0.038

2947 0.02565
0.034
0.036

2947 0.02565
0.032
0.034
0.03
0.032
0.028
0.03
0.026
0.028
0.024
0.026

1039 0.01319
0.022
0.024

1039 0.01319
0.02
0.022
0.018
0.02

1471 0.006812
0.016
0.018

1471 0.006812
0.014
0.016

1122 0.003587
0.012
0.014

2716 0.001745
2151 0.001601

2591 0.001574
1122 0.003587
0.01
0.012

2716 0.001745
2151 0.001601

2591 0.001574
0.008
0.01
0.006
0.008

2928 0.03612
0.004
0.006

2928 0.03612
0.002
0.004
0.04
0.002
0.038
0.04
0.036
0.038
0.034
0.036
0.032
0.034
0.03
0.032
885.6 0.01916

0.028
0.03
885.6 0.01916

0.026
0.028
Intensity (a.u.)

0.024
0.026
0.022
Intensity (a.u.)

0.024
0.02
0.022
0.018
1453 0.008087
0.02
1053 0.007144

0.016
0.018
1453 0.008087
1053 0.007144

0.014

2714 0.003992
0.016
1279 0.003764

0.012
0.014

2714 0.003992
1279 0.003764

0.01
0.012
0.008
0.01
0.006
0.008
0.004
0.006
0.002
0.004
0.002
800 850 900 950 1000 1050 1100 1150 1200 1250 1300 1350 1400 1450 1500 1550 1600 1650 1700 1750 1800 1850 1900 1950 2000 2050 2100 2150 2200 2250 2300 2350 2400 2450 2500 2550 2600 2650 2700 2750 2800 2850 2900 2950 3000 3050 3100
Wavenumbers (cm-1)
800 850 900 950 1000 1050 1100 1150 1200 1250 1300 1350 1400 1450 1500 1550 1600 1650 1700 1750 1800 1850 1900 1950 2000 2050 2100 2150 2200 2250 2300 2350 2400 2450 2500 2550 2600 2650 2700 2750 2800 2850 2900 2950 3000 3050 3100
Wavenumbers (cm-1)

Figure Raman
12.12.
Figure Raman spectrum
spectrumofofmethanol
methanol (blue line) and
(blue line) andethanol
ethanol(orange
(orange line).
line). TheThe relevant
relevant peaks, as
peaks,
Figure 12. Raman spectrum of methanol (blue line) and ethanol (orange line). The relevant peaks,
described in theinliterature,
as described are are
the literature, highlighted.
highlighted.
as described in the literature, are highlighted.

2947 0.
0.03

2947 0.
0.03
0.028
0.028
0.026
0.026
0.024
0.024
0.022
1039 0.01321

0.0220.02
1039 0.01321

0.02
0.018
1470 0.008736

0.018
0.016
1470 0.008736

0.016
0.014
0.014
0.012
1124 0.004259

0.0120.01
1124 0.004259

0.01
0.008

2928 0.03599
0.008
0.006

2928 0.03599
0.006
0.004
0.004
0.002
0.002
0.0360
0
0.034
0.036
0.032
0.034
0.0320.03

2878 0.02144
0.028
885.6 0.01914

0.03

2878 0.02144
0.026
0.028
885.6 0.01914

0.024
0.026
0.022
0.024
Intensity (a.u.)

2974 0.01203
0.0220.02
Intensity (a.u.)

2974 0.01203
0.018
0.02
1453 0.008078
1053 0.007139

0.016
0.018
1453 0.008078
1098 0.005646

0.014
1053 0.007139

0.016

2714 0.003981
1279 0.003762
1098 0.005646

0.012
0.014 2714 0.003981
1279 0.003762

0.0120.01
0.008
0.01
Sensors 2025, 25, x FOR PEER REVIEW 0.006
0.008
0.004
16 of 24
0.006
0.002
0.004
0.002 0
0 800 850 900 950 1000 1050 1100 1150 1200 1250 1300 1350 1400 1450 1500 1550 1600 1650 1700 1750 1800 1850 1900 1950 2000 2050 2100 2150 2200 2250 2300 2350 2400 2450 2500 2550 2600 2650 2700 2750 2800 2850 2900 2950 3000 3050 3100
Wavenumbers (cm-1)
800 850 900 950 1000 1050 1100 1150 1200 1250 1300 1350 1400 1450 1500 1550 1600 1650 1700 1750 1800 1850 1900 1950 2000 2050 2100 2150 2200 2250 2300 2350 2400 2450 2500 2550 2600 2650 2700 2750 2800 2850 2900 2950 3000 3050 3100
Wavenumbers (cm-1)

It should
Figure 13. be noted
Raman that theof80methanol
spectrum mW of laser
(bluepower
line) andis still well(orange
ethanol belowline).
the standard
The data hundreds
are filtered
Figure
Figure13.13.Raman
Raman spectrum
spectrum ofofmethanol
methanol (blue
(blue line)line)
andand ethanol
ethanol (orange
(orange line). line). Theare
The data data are filtered
filtered
of
withmW normally
a Gaussian used in infrared
convolution algorithmmodels,
for noisemaking
removal. the management of the safety require-
with a Gaussian convolution algorithm for noise
with a Gaussian convolution algorithm for noise removal. removal.
ments for a portable system easier.
The analysis of urine samples is interesting as a proof of concept, pointing the atten-
The analysis of urine samples is interesting as a proof of concept, pointing the atten-
0.0115 tion to an application domain where a portable Raman analysis can contribute to address-
0.011
tion to an application domain where a portable Raman analysis can contribute to address-
0.0105 ing an important social problem. However, the peak intensities and distributions in the Urine sample 5

0.01
ing an important social problem. However, the peak intensities and distributions in the Gaussian deconvolution of: Urine sample 5

spectra of urine samples can present a large variability depending on the specific condi-
0.0095
spectra of urine samples can present a large variability depending on the specific condi-
0.009
tion of the donors, so they cannot be used to objectively infer the performance level of the
739.7 0.006694
0.006696

0.0085 tion of the donors, so they cannot be used to objectively infer the performance level of the
0.008 prototype. So, as a comparison with other data reported in the literature for methanol
0.0075 prototype. So, as a comparison with other data reported in the literature for methanol
0.007 detection using portable Raman systems, Table 5 reports the level of detection (LOD) and
0.0065
detection using portable Raman systems, Table 5 reports the level of detection (LOD) and
Absorbance

some technical specifications of the systems. Both the CPS532 and FPYL532 versions are
0.006
some technical specifications of the systems. Both the CPS532 and FPYL532 versions are
0.003446

0.0055
1042 0.003412

included in the table. As a reference of values, it is worth recalling that drinks made from
1124 0.003123
0.003124

0.005
included in the table. As a reference of values, it is worth recalling that drinks made from
“industrial methylated spirits” [5% (v/v) methanol:95% (v/v) ethanol] can cause severe and
1041

3545 0.002508
0.002512

0.0045
1220 0.002262

0.002128

0.002141

“industrial methylated spirits” [5% (v/v) methanol:95% (v/v) ethanol] can cause severe and
2717 0.002128
2149 0.002123
1359 0.002066

0.002011
1474 0.002012
921.7 0.001935

0.001915

0.004
2624 0.001912
1696 0.001884

0.001778
3139 0.001773
2033 0.001751
0.001752
1571 0.001674

0.001613

even fatal illness, and the current EU general limit for naturally occurring methanol is 10
1867 0.001609

0.001585

3314 0.001598
2370 0.001578

3386 0.001523
0.00153
0.001455
2991 0.001453

0.0035
2440 0.001408

2909 0.001378
0.001379
0.00141
1957 0.001259

even fatal illness, and the current EU general limit for naturally occurring methanol is 10
2149

0.001218
2290 0.001222
0.00125

0.003
3139

g methanol/L ethanol (which equates to 0.4% (v/v) methanol at 40% alcohol) [23]. Regard-
1867

3387
2440

0.0025
1957

0.002
g methanol/L ethanol (which equates to 0.4% (v/v) methanol at 40% alcohol) [23]. Regard-
ing the low-power version with the CPS532 laser, the LOD is higher than the EU recom-
0.0015

0.001
ing the low-power version with the CPS532 laser, the LOD is higher than the EU recom-
mendation, as the 2840 cm−1 peak remains hidden by the noise for a concentration lower
0.0005
mendation, as the 2840 cm−1 peak remains hidden by the noise for a concentration lower
0
than 4.76% of methanol. Moving to the higher-power FPYL532 (leaving all the other com-
400 500 600 700 800 900
than 4.76% of methanol. Moving to the higher-power FPYL532 (leaving all the other com-
1000 1100 1200 1300 1400 1500 1600 1700 1800 1900 2000 2100
Wavenumbers [cm-1 ]
2200 2300 2400 2500 2600 2700 2800 2900 3000 3100 3200 3300 3400 3500

ponents unchanged), the S/N is much improved, and the LOD falls to a value below 0.25%.
ponents unchanged), the S/N is much improved, and the LOD falls to a value below 0.25%.
Figure 14.14.
Figure Raman
Ramanspectrum
spectrum of one
one urine
urinesample
sample and
and its its corresponding
corresponding filtered
filtered plot, obtained
plot, obtained by a by a
Gaussian
Gaussian convolution
convolutionalgorithm.
algorithm.

Table 5. Level of detection (LOD) of methanol in a solution with ethanol.

Model LOD (%) LOD (g/L) Laser Wavelength (nm) Laser Power (mW) More Significant Peak (cm−1)
CPS532 4.76 37.67 532 5 2840
FPYL532 0.25 1.97 532 80 2840
Ref. [24] 0.23–0.39 830 400 1030
Ref. [25] 0.025 1064 450 1023
Sensors 2025, 25, 659 16 of 22

5. Development of the Supervised Learning Model

5.1. Data Acquisition
By developing a learning model, we aim to evaluate its capability to differentiate and
classify Raman spectra. If successful, the model could contribute to making the diagnosis
of kidney diseases simpler, faster and more effective.
To train the model, Raman spectra of six methanol–ethanol solutions with different
compositions were acquired. The decision to use methanol–ethanol solutions was based on
the fact that they generate simpler Raman spectra. This characteristic facilitated the initial
development and optimization of the algorithm, as it allowed us to evaluate the model’s
ability to classify straightforward Raman signals before progressing, in the future, to more
complex biological samples.
The methanol–ethanol spectra were acquired using the developed Raman system. To
ensure dataset diversity, the acquisition parameters (exposure time, gain and ROI) of the
Sensors 2025, 25, x FOR PEER REVIEW 17 of 24
Spectrum Analyzer software were varied between each acquisition. A total of 618 spectra
(cf. Table 6) were acquired and divided into two subsets: 80% (495 spectra) for model
training and 20% (123 spectra) for model testing. Each spectrum was converted into a
differences among the six solutions, exemplifying the variability in the dataset used. The
2D image with a resolution of 1000 × 2000 pixels, enabling the model to extract spectral
acquisition parameters used for the shown alcohol spectra are listed in Table 7, and the
features for correct classification.
data are reported in Figure 15.
Table 6. Data used for the supervised learning model.
Table 6. Data used for the supervised learning model.
Solutions Number of Spectra Obtained Total No. of Data
Solutions
100% Ethanol Number106 of Spectra Obtained Total No. of Data
90% Ethanol
100%10% Methanol
Ethanol 103 106
80% Ethanol 20% Methanol 103
90%50%Ethanol 10%
Ethanol 50% Methanol
Methanol 102 103
618
80%20%Ethanol 20%
Ethanol 80% Methanol
Methanol 102 103
100% Methanol 102 618
50% Ethanol 50% Methanol 102
20% Ethanol 80% Methanol 102
Next, we Methanol
100% present a graph showcasing an example102 spectrum for each alcohol solution
used in training the supervised learning model. This visualization highlights the spectral
differences among parameters
Table 7. Acquisition the six solutions,
used forexemplifying
the acquisitionthe variability
of the in the dataset
example alcohol spectra. used. The
acquisition parameters used for the shown alcohol spectra are listed in Table 7, and the
Exposure Time(s) Gain (dB) ROI (px) Average Number of Images
data are reported in Figure 15.
10 3.8 128 12
0.05
0.04
0.03
0.02
0.01 100% Ethanol
0.05 90% Ethanol 10% Methanol
0.04 80% Ethanol 20% Methanol
0.03 50% Ethanol 50% Methanol
0.02 20% Ethanol 80% Methanol
0.01 100% Methanol
0.05
0.04
0.03
0.02
0.01
0.05
Intensity (a.u)

0.04
0.03
0.02
0.01
0.05
0.04
0.03
0.02
0.01
0.05
0.04
0.03
0.02
0.01

400 600 800 1000 1200 1400 1600 1800 2000 2200 2400 2600 2800 3000 3200 3400
Wavenumbers [cm-1 ]

Figure
Figure 15. Example Raman
15. Example Raman spectra
spectra of
of the
the alcohol
alcohol solutions
solutions used.
used.

5.2. Supervised Learning Model Development

The learning model was developed using MATLAB Version 2023, and it is a Convo-
lutional Neural Network (CNN) tailored to classify Raman spectra.
As previously mentioned, the input data consists of 2D images of the Raman spectra
from alcohol solutions, all standardized to the same dimensions and represented with
Sensors 2025, 25, 659 17 of 22

Table 7. Acquisition parameters used for the acquisition of the example alcohol spectra.

Exposure Time(s) Gain (dB) ROI (px) Average Number of Images

10 3.8 128 12

5.2. Supervised Learning Model Development

The learning model was developed using MATLAB Version 2023, and it is a Convolu-
tional Neural Network (CNN) tailored to classify Raman spectra.
As previously mentioned, the input data consists of 2D images of the Raman spectra
from alcohol solutions, all standardized to the same dimensions and represented with
three color channels (RGB). The network architecture, as depicted in Figure 16, includes
two hidden layers following the input layer. Each hidden layer comprises a convolutional
layer, a batch normalization layer, a ReLu activation function and a pooling layer. The
Sensors 2025, 25, x FOR PEER REVIEW 18 layer,
output layers, essential for the final spectra classification, consist of a fully connected of 24
followed by a SoftMax activation function and a classification layer.

Figure 16. Supervised learning model architecture.

architecture.

5.3. Supervised Learning

5.3. Supervised Learning Model
Model Optimization
Optimization
Hyperparameters
Hyperparameters significantlyinfluence
significantly influence a model’s
a model’sperformance, particularly
performance, its train-
particularly its
ing speed and accuracy. These hyperparameters can optimize critical
training speed and accuracy. These hyperparameters can optimize critical aspects of a aspects of a model,
making it more it
model, making robust
moreand efficient.
robust and efficient.
The
The MaxEpoch hyperparameter defines
MaxEpoch hyperparameter the maximum
defines the maximum number
number of of training
training iterations,
iterations,
i.e., epochs. Higher epoch values allow a model to learn more effectively
i.e., epochs. Higher epoch values allow a model to learn more effectively but increase but increase
the
the risk of overfitting and extend the training time. During an epoch,
risk of overfitting and extend the training time. During an epoch, a model processes the a model pro-
cesses the entire
entire training training
set, dividingset,itdividing it into mini-batches,
into mini-batches, another key another key hyperparameter.
hyperparameter. The Mini-
The Mini-batch size impacts training efficiency, as larger groups can
batch size impacts training efficiency, as larger groups can result in a longer training result in a longer
time.
training time.
The InitialLearnRate hyperparameter defines the rate at which a model adjusts its
weightsThe during
InitialLearnRate hyperparameter
training. This rate directly defines thethe
influences ratespeed
at which
and a model adjusts
effectiveness its
of the
weights during training. This rate directly influences the speed and
training process. An improperly adjusted rate can prevent a model from converging and effectiveness of the
learning from the training data.
The model was optimized by varying the mentioned hyperparameters, and their im-
pact was evaluated based on the training time, accuracy and precision of the model (cf.
Table 8). Despite its simplicity, this empirical approach allows for the identification of the
Sensors 2025, 25, 659 18 of 22

training process. An improperly adjusted rate can prevent a model from converging and
learning from the training data.
The model was optimized by varying the mentioned hyperparameters, and their
impact was evaluated based on the training time, accuracy and precision of the model (cf.
Table 8). Despite its simplicity, this empirical approach allows for the identification of the
most effective hyperparameters for our dataset and model architecture.

Table 8. Analysis of how changing hyperparameters impacts model evaluation.

Hyperparameters of Training Model Performance

MaxEpoch InitialLearningRate MiniBatchSize Accuracy (%) Precision (%) Training Time
20 0.001 64 96.75 96.79 24 min 04 s
20 0.001 32 98.37 98.71 10 min 52 s
20 0.001 16 100 100 05 min 31 s
30 0.001 16 99.19 99.21 08 min 20 s
10 0.001 16 99.19 99.21 02 min 45 s
20 0.1 16 17.07 16.67 05 min 28 s
20 0.01 16 87.81 87.78 05 min 30 s

Analyzing the results, the hyperparameter with the greatest influence on the model’s
accuracy and precision is InitialLearningRate. When this parameter was increased from
0.001 to 0.01, the model’s performance declined. With a rate of 0.1, its accuracy dropped
significantly to 17.07% and precision to 16.67%, as excessively high learning rates can cause
large weight adjustments, preventing the model from adequately capturing data patterns.
Thus, the optimal learning rate for the developed model is 0.001.
Adjusting the Mini-batch size especially influenced the training time. With 16 batches,
the model achieved 100% accuracy and precision, within approximately 5 min of training.
These values reflect the limitations imposed by the small dataset size used, which likely
contributed to overfitting.
The MaxEpoch hyperparameter did not behave as anticipated. Increasing the number
of epochs from 20 to 30 resulted in a slight decrease in performance, indicating that the
model could be overfitting to the training data. On the other hand, with the decrease in
epoch cycles from 20 to 10, the model’s accuracy and precision stayed equal, though it
minimized training time to just about 2 min.
In summary, the hyperparameter combination that maximizes model performance,
while maintaining efficient training times and avoiding overfitting, is MaxEpoch of 10,
InitialLearningRate of 0.001, and MiniBatchSize of 16.
These results highlight the potential of the developed learning model for Raman
spectrum classification, evidencing its capability to identify patterns in these spectra.

6. Conclusions and Future Perspectives

This study demonstrated the potential of a simplified and portable Raman spec-
troscopy system for acquiring the spectra of liquid and complex samples, including urine.
The application of this system could simplify and enhance the accessibility of diagnosing
diseases outside clinical contexts. Future work will focus on the identification and analysis
of specific biomarkers associated with kidney diseases using the developed system.
The system’s main challenge was the high noise levels and the low intensity of the
characteristic peaks in the spectrum obtained. Overcoming this issue would require a
higher-power laser since the one used in this study had a relatively low power of 4.5 mW.
Additionally, urine’s fluorescence spectrum indicates a low fluorescence intensity when
Sensors 2025, 25, 659 19 of 22

irradiated with a red laser with 635 nm, suggesting the potential development of a Raman
system using a high-power laser with this emission wavelength.
The assembly, alignment and calibration of the Raman system were successfully
achieved at this stage. In a future phase, the use of more sophisticated optical elements
could be explored to amplify Raman scattering. Similarly, the acquisition parameters were
optimized using ethanol spectra; however, their direct impact on urine spectra should be
explored in subsequent phases.
During the optimization phases, the system exhibited high stability, as evidenced
by minimal spectral variations across repeated measurements over time. Despite higher
noise levels in some instances, the system successfully acquired the ethanol spectrum
with its characteristic Raman peaks, demonstrating its robustness and reliability under
challenging conditions.
The analysis of the five urine spectra with the developed system demonstrates
consistency and sensibility to variations in compound concentrations in urine. Future
work will involve expanding the dataset to include urine samples from healthy and un-
healthy individuals, as well as validating the system’s performance in detecting clinically
relevant biomarkers.
The implementation of a supervised learning model for spectra classification proved
to be an effective tool, capable of recognizing patterns in Raman spectra and achieving accu-
rate classification. The application of these kinds of models can facilitate the differentiation
between healthy and pathological spectra. The results obtained with methanol–ethanol
solutions demonstrated the model’s ability to classify Raman spectra. These results serve
as a foundation for future work, where the model will be trained with urine spectra to
identify kidney disease biomarkers.
Although a direct comparison with commercial Raman instruments was not performed
in this study, the developed system is significantly more affordable than the systems
available on the market, having an estimated cost below EUR 5000. Furthermore, its design
allows for portability, highlighting its potential for point-of-care applications.
Despite the limitations imposed by instrumentation and budget, this work paves the
way for the development of a Raman spectroscopy system that, combined with machine
learning, creates an accessible, portable and effective tool for non-invasive diagnosis,
particularly for kidney disease detection.

Author Contributions: Conceptualization, C.D. and A.F.; methodology, C.D. and M.F.; software, C.D.
and A.F.; validation, A.F. and S.A.P.; investigation, C.D.; data curation, C.D.; resources, M.F. and J.F.;
writing—original draft preparation, C.D.; writing—review and editing, A.F.; supervision, A.F., M.F.
and S.A.P.; project administration, A.F.; funding acquisition, A.F. All authors have read and agreed to
the published version of the manuscript.

Funding: This research was supported by Fundação para a Ciência e Tecnologia (FCT), Center of
Technology and Systems (CTS) UIDB/00066/2020/UIDP/00066/2020, by project IPL/IDI&CA2023/
LUMINA_ISEL by Instituto Politécnico de Lisboa (IPL/2023/IDI&CA), and by Sapienza Visiting
Professor Programme 2022.

Institutional Review Board Statement: Not applicable.

Informed Consent Statement: Informed consent was obtained from all subjects involved in this study.

Data Availability Statement: The data used in this project as well as a version of the algorithm are
available upon request.

Acknowledgments: The authors would like to thank Caterina Serafinelli for preparing the alcohol
solutions used to train the supervised learning model. The collaboration of Paolo Di Giamberardino,
Sensors 2025, 25, 659 20 of 22

under the Sapienza Visiting Professor Programme 2022, for designing the Matlab algorithm is
also acknowledged.

Conflicts of Interest: The authors declare no conflicts of interest.

Sensors 2025, 25, x FOR PEER REVIEW 21 of 24
Appendix A
Sensors 2025, 25, x FOR PEER REVIEW 21 of 24

Figure
Figure A1.A1. CPS532laser
CPS532 laserspectrum
spectrum with
withan
anoperating
operatingtemperature of°C.
of 25
temperature 25 ◦ C.
Figure A1. CPS532 laser spectrum with an operating temperature of 25 °C.

Figure A2. CPS532 laser spectrum with an operating temperature of 30 °C.

Figure
Figure A2.A2. CPS532laser
CPS532 laserspectrum
spectrum with
withan
anoperating
operatingtemperature of°C.
of 30
temperature 30 ◦ C.
Sensors 2025, 25,
Sensors 25, x FOR PEER REVIEW 22 of 24
2025, 25, 659
Sensors 2025, x FOR PEER REVIEW 2221of
of 24
22

CPS532 laser spectrum ◦°C.

A3. CPS532 laser spectrum with
Figure A3.
Figure with an
an operating
operating temperature
temperature of
of 35 C.
35 °C.

Figure A4. CPS532 laser spectrum with an operating temperature of 40 °C.

Figure A4. CPS532 laser spectrum with an operating temperature of 40 ◦°C.
C.
Sensors 2025, 25, 659 22 of 22

References
1. James, M.T.; Bhatt, M.; Pannu, N.; Tonelli, M. Long-term outcomes of acute kidney injury and strategies for improved care.
Nat. Rev. Nephrol. 2020, 16, 193–205. [CrossRef] [PubMed]
2. Kellum, J.A.; Romagnani, P.; Ashuntantang, G.; Ronco, C.; Zarbock, A.; Anders, H.-J. Acute kidney injury. Nat. Rev. Dis. Prim.
2021, 7, 52. [CrossRef]
3. Coca, S.G.; Parikh, C.R. Urinary Biomarkers for Acute Kidney Injury. Clin. J. Am. Soc. Nephrol. 2008, 3, 481–490. [CrossRef]
4. Turgut, F.; Awad, A.S.; Abdel-Rahman, E.M. Acute Kidney Injury: Medical Causes and Pathogenesis. J. Clin. Med. 2023,
12, 375. [CrossRef] [PubMed]
5. Waikar, S.S.; Betensky, R.A.; Emerson, S.C.; Bonventre, J.V. Imperfect Gold Standards for Kidney Injury Biomarker Evaluation.
J. Am. Soc. Nephrol. 2012, 23, 13–21. [CrossRef] [PubMed]
6. Soto, K.; Devarajan, P. Acute Kidney Injury Biomarkers: From bench to clinical use. Port. J. Nephrol. Hypertens. 2016, 30, 166–169.
7. Sharma, M.; Tsai, C.-L.; Li, Y.-C.; Lee, C.-C.; Hsieh, Y.-L.; Chang, C.-H.; Chen, S.-W.; Chang, L.-B. Utilizing Raman spectroscopy
for urinalysis to diagnose acute kidney injury stages in cardiac surgery patients. Ren. Fail. 2024, 46, 2375741. [CrossRef]
8. Kuhar, N.; Sil, S.; Verma, T.; Umapathy, S. Challenges in application of Raman spectroscopy to biology and materials. RSC Adv.
2018, 8, 25888–25908. [CrossRef] [PubMed]
9. Jeng, M.-J.; Sharma, M.; Lee, C.-C.; Lu, Y.-S.; Tsai, C.-L.; Chang, C.-H.; Chen, S.-W.; Lin, R.-M.; Chang, L.-B. Raman Spectral
Characterization of Urine for Rapid Diagnosis of Acute Kidney Injury. J. Clin. Med. 2022, 11, 4829. [CrossRef] [PubMed]
10. Smith, E.; Dent, G. Modern Raman Spectroscopy—A Practical Approach, 2nd ed.; Wiley: Hoboken, NJ, USA, 2019.
11. Rüdiger, P. Field Guide to Lasers; SPIE: Bellingham, WA, USA, 2007; Volume FG12.
12. Ferraro, J.R.; Nakamoto, K.; Brown, C.W. Introductory Raman Spectroscopy; Academic Press: Cambridge, MA, USA, 2003.
13. Starter Edition Assembly—OpenRAMAN. Available online: https://www.open-raman.org/build/starter-edition/assembly/
(accessed on 1 July 2024).
14. Fantoni, A.; Fernandes, M.; Fidalgo, J.; Soto, M.K.; Pereira, S.A.; Moreno, A.; Domingos, C.; Correia, A.R.; Sousa, H.; Serafinelli,
C.; et al. The LUMINA setup for a light-based urine monitoring and analysis. In Proceedings of the SPIE, Biophotonics in
Point-of-Care III, Strasbourg, France, 7–12 April 2024; p. 26. [CrossRef]
15. Hoffmann, G.G. Infrared and Raman Spectroscopy: Principles and Applications; Walter de Gruyter: Berlin, Germany, 2023.
16. Spectrum Analyzer Suite. Available online: https://www.open-raman.org/build/software/ (accessed on 23 September 2024).
17. Stone, N.; Kendall, C.; Smith, J.; Crow, P.; Barr, H. Raman spectroscopy for identification of epithelial cancers. Faraday Discuss.
2003, 126, 141–157. [CrossRef] [PubMed]
18. Bispo, J.A.M.; Vieira, E.E.d.S.; Silveira, L.; Fernandes, A.B. Correlating the amount of urea, creatinine, and glucose in urine from
patients with diabetes mellitus and hypertension with the risk of developing renal lesions by means of Raman spectroscopy and
principal component analysis. J. Biomed. Opt. 2013, 18, 87004. [CrossRef] [PubMed]
19. Chen, C.; Yang, L.; Zhao, J.; Yuan, Y.; Tang, J.; Yang, H.; Yan, Z.; Wang, H.; Lv, X. Urine Raman spectroscopy for rapid and
inexpensive diagnosis of chronic renal failure (CRF) using multiple classification algorithms. Optik 2019, 203, 164043. [CrossRef]
20. Dou, X.; Yamaguchi, Y.; Yamamoto, H.; Doi, S.; Ozaki, Y. Quantitative analysis of metabolites in urine by anti-Stokes Raman
spectroscopy. Biospectroscopy 1998, 3, 113–120. [CrossRef]
21. Nekoukar, Z.; Zakariaei, Z.; Taghizadeh, F.; Musavi, F.; Banimostafavi, E.S.; Sharifpour, A.; Ghuchi, N.E.; Fakhar, M.; Tabaripour,
R.; Safanavaei, S. Methanol poisoning as a new world challenge: A review. Ann. Med. Surg. 2021, 66, 102445. [CrossRef]
22. Emin, A.; Hushur, A.; Mamtimin, T. Raman study of mixed solutions of methanol and ethanol. AIP Adv. 2020, 10, 065330. [CrossRef]
23. Paine, A.J.; Dayan, A.D. Defining a tolerable concentration of methanol in alcoholic drinks. Hum. Exp. Toxicol. 2001,
20, 563–568. [CrossRef] [PubMed]
24. Ellis, D.I.; Muhamadali, H.; Xu, Y.; Eccles, R.; Goodall, I.; Goodacre, R. Rapid through-container detection of fake spirits and
methanol quantification with handheld Raman spectroscopy. Analyst 2018, 144, 324–330. [CrossRef] [PubMed]
25. Ellis, D.I.; Eccles, R.; Xu, Y.; Griffen, J.; Muhamadali, H.; Matousek, P.; Goodall, I.; Goodacre, R. Through-container, extremely
low concentration detection of multiple chemical markers of counterfeit alcohol using a handheld SORS device. Sci. Rep. 2017,
7, 12082. [CrossRef] [PubMed]

Disclaimer/Publisher’s Note: The statements, opinions and data contained in all publications are solely those of the individual
author(s) and contributor(s) and not of MDPI and/or the editor(s). MDPI and/or the editor(s) disclaim responsibility for any injury to
people or property resulting from any ideas, methods, instructions or products referred to in the content.