STABLE & UNSTABLE

SIGNAL
PROCESSING IN
IMAGING SCIENCE
AND INFORMATICS
INTRODUCTION
◦ In imaging science and informatics, signal processing
plays a crucial role in transforming raw data into usable
images and extracting meaningful information from those
images.
◦ Stability is an important criterion in this domain because it
affects the quality, reliability, and interpretation of the data.
This report explores the concepts of stable and unstable
signal processing, examining their implications in the
context of imaging systems and their applications in
scientific research, medical diagnostics, and informatics.
Stable Signal Processing

◦ Stable signal processing refers to methods that provide

consistent and reliable results even when subject to small
perturbations in input data or system parameters. A
system is considered stable if it consistently delivers
output that accurately reflects the input without
unpredictable variations. In imaging science, stability is
crucial for applications requiring precision, such as
medical imaging, satellite imaging, and microscopy.
Key Characteristics of Stable
Signal Processing:

◦ Predictability: Small changes in input data lead to small,

predictable changes in the output.
◦ Noise Resistance: Stable systems are less sensitive to
noise, ensuring that minor distortions do not significantly
affect the final image.
◦ Consistency: Results are consistent across different runs
of the same process, leading to reproducibility.
◦ Convergence: Many stable processes rely on iterative
algorithms, where each step converges toward an optimal
solution without divergent or erratic behavior.
MEDICAL IMAGING

◦ In applications like MRI or CT scans, stable signal

processing is necessary for obtaining high-resolution
images that clinicians can rely on for diagnosis.
Consistency in imaging data ensures accurate
interpretation and long-term monitoring of consistency in
imaging.
UNSTABLE SIGNAL
PROCESSING
◦ Unstable signal processing refers to methods where small
changes in input or system parameters can lead to
significant and often unpredictable variations in output.
These instabilities can arise from noise, insufficient
regularization, or non-linear system dynamics. Unstable
processes are often undesirable in critical applications,
but they can be useful in research or exploratory domains
where sensitivity to small changes may reveal hidden
features of data.
Key Characteristics of
Unstable Signal Processing:

◦ Sensitivity: Even minimal changes in the input data can

result in disproportionately large changes in the output.
◦ Noise Amplification: Noise in the input signal may be
greatly magnified, leading to errors and artifacts in the
resulting image.
◦ Inconsistency: Results may vary sig intly with each run,
making it difficult to reproduce outcomes.
TRANSFORMATION IN
MEDICAL IMAGING
◦ it transforms the image from one form to another form, and
this process also transfers the image style. This approach was
utilised in medical imaging to increase the number of medical
images in the dataset. Deep learning neural network
technology automates medical image analysis.
Digital Transformation - What
it means for radiology

◦ Radiology departments are faced with a multitude of

challenges today like staff shortages, increasing amounts of
data as well as the need for faster and precise diagnosis and
innovative solutions. These demands need to be met effectively
and efficiently. This is where digital solutions can play an
increasingly vital part in transforming radiology. Using digital
technologies to turn large amounts of data into insights will
support more precise diagnosis, targeted treatment, and
greater patient satisfaction.
PRE MEDICAL
IMAGE PROCESSING
◦ Pre-medical image processing refers to the set of
techniques and steps applied to raw medical imaging data
before the main analysis or reconstruction takes place.
This stage is crucial because it helps to prepare, clean,
and enhance the raw data for better image quality and
more accurate diagnosis or reconstruction.
I. IMAGE RECONSTRUCTION
◦ Image reconstruction refers to the process of generating or restoring an image from incomplete,
degraded, or abstract representations.
◦ In radiography, image reconstruction plays a crucial role, especially in advanced imaging
modalities such as computed tomography (CT)
3 TYPES OF
RECONSTRUCTION
ALGORITHM
◦ Filtered Back Projection (FBP): This is the traditional and most widely
used method.
It works by applying a mathematical filter to the raw data and then
back-projecting it to reconstruct an image. However, it requires a
higher radiation dose and may result in artifacts or noise in low-dose
scans.
◦ Iterative Reconstruction (IR): This method iteratively refines the image
by comparing it to the measured projection data, improving image
quality and reducing noise. It allows for lower radiation doses while
maintaining high image quality.
◦ Model-Based Iterative Reconstruction (MBIR): This is a more
advanced form of iterative reconstruction that incorporates physical
models of the imaging system (e.g., detector geometry, noise, etc.) to
further enhance image quality and reduce artifacts.
APPLICATION OF IMAGE
RECONSTRUCTION IN
RADIOLOGY
◦ Low-Dose CT Scanning: Reducing the radiation dose is a priority in
medical imaging to minimize risk to patients. Advanced reconstruction
techniques, particularly iterative and deep learning-based methods,
help achieve high-quality images at much lower doses.
◦ Artifact Reduction: Metal implants, patient movement, and other
factors can cause artifacts in CT images. Iterative and Al-based
reconstruction algorithms help reduce these artifacts, making the
images clearer and more reliable for diagnosis.
◦ Improved Detail in Soft Tissue Imaging: Radiography traditionally
struggles with soft tissue contrast, but advanced reconstruction
algorithms improve the visualization of soft tissues, making
radiography more useful in diverse clinical situations.
 ◦ In radiology pre-processing refers to the
technique of isolating the region of interest
(ROl) from the surrounding areas that are
not clinically relevant, such as air, table
II. surfaces, or irrelevant tissues in the
scanned image. This process enhances
BACKGROUND the visibility of specific anatomical
REMOVAL structures (e.g., bones, organs, or lesions)
and facilitates more accurate analysis,
segmentation, and diagnosis. It is
particularly important in modalities like CT,
MRI, X-rays, and ultrasound.
EXAMPLE WORKFLOW
◦ Input Image: A raw radiological image with noise, background structures, and the region of
interest.
◦ Thresholding: Apply intensity thresholding to distinguish the air or irrelevant structures (e.g., the
surrounding environment, imaging table) from the body part being imaged.
◦ Segmentation: Use edge detection or region-growing algorithms to refine the separation of the
region of interest.
◦ Morphological Operations: Apply dilation or erosion to clean up small areas of noise or
incomplete background removal.
◦ Output: A clean image with the background removed, highlighting the region of interest for
further analysis or diagnosis.
APPLICATIONS OF
BACKGROUND REMOVAL
◦ Bone Structure Isolation: In X-rays and CT scans, background
removal is commonly used to isolate the bone structures, making it
easier to identify fractures, deformities, or lesions.
◦ Tumor Segmentation: In oncology imaging, removing the background
helps focus on the tumor and surrounding tissues for more accurate
measurement and treatment planning.
◦ Organ Segmentation: Removing the background is crucial when
segmenting organs like the liver, lungs, or brain in CT or MRI scans
for better anatomical analysis.
◦ Vessel Analysis: In angiography, removing the background helps
visualize blood vessels more clearly, improving the detection of
blockages, aneurysms, or malformations.
III. NOISE REMOVAL
◦ In radiography refers to the process of reducing unwanted
or extraneous signals (noise) from radiographic images to
enhance their quality and make important anatomical
features clearer for diagnosis. In radiographic images
such as X-rays, CT scans, or MRI, noise can degrade
image clarity, making it difficult for radiologists to interpret
the images accurately. Various techniques are employed
to filter or remove noise while preserving important
diagnostic details.
TYPES OF NOISE IN
RADIOGRAPHY
◦ Quantum Noise: This is caused by the statistical variation in the
number of X-ray photons reaching the detector. It tends to occur when
a low dose of radiation is used, leading to a grainy appearance in the
image.
◦ Electronic Noise: This comes from the electronic components of the
imaging system, including the detectors and amplifiers, and can
contribute to random variations in the image signal.
◦ Shot Noise: This occurs due to the random nature of the interaction
between X-rays and the body, leading to fluctuations in the image
signal.
◦ Scatter Noise: Scatter radiation refers to X-rays that deviate from their
original path after interacting with tissues, leading to noise in the
image. It can lower contrast and blur the edges of structures.
IMPORTANCE OF
NOISE REMOVAL
◦ Improves Image Quality: Reducing noise enhances the
visibility of fine details, making the image clearer and
easier to interpret.
◦ Enhances Diagnostic Accuracy: Lower noise levels allow
radiologists to make more accurate diagnoses, especially
in identifying subtle abnormalities like small tumors or
fractures.
◦ Supports Lower Radiation Doses: Noise removal allows
the use of lower radiation doses without compromising
image quality, which is beneficial for patient safety.
APPLICATION OF
NOISE REMOVAL
◦ Low-Dose CT Imaging: Noise reduction techniques, particularly iterative
reconstruction and deep learning-based methods, are essential in low-dose
CT scans to maintain image quality while reducing radiation exposure.
◦ MRI: MRI scans, particularly when performed quickly or at lower field
strengths, often contain noise that can obscure important features.
Advanced denoising techniques like wavelet-based or bilateral filtering help
enhance image clarity without losing important details.
◦ X-rays: In digital X-ray systems, noise due to photon variations and
electronic interference can degrade image quality, especially in low-dose
applications.
Smoothing and Fourier-based techniques are frequently used to clean up
noisy X-rays.
◦ Ultrasound: Ultrasound images can contain speckle noise, which results
from the interaction of sound waves with tissues. Techniques like speckle
reduction filters and anisotropic diffusion are used to clean up ultrasound
images while maintaining edge information.
 ◦ Image compression attempts to reduce or eliminate
the presence of these redundancies to minimize the
storage size or transmission time requirements for a
given image. Depending on the methods used,
coding and spatiotemporal redundancy reduction
can be either reversible or irreversible, while psy-
chovisual redundancy reduction is always
irreversible.

IV. IMAGE ◦ If the reduction is reversible, then the compression

is said to be lossless since no information is lost by
COMPRESSION the process, and the original image can be
reconstructed exactly. Compression algorithms
which result in irreversible redundancy reduction
are said to be lossy, and the reconstructed image
after lossy compression is only an approximation of
the original image. In general, lossy compression
algorithms achieve higher compression levels than
lossless algorithms.
BASIC COMPRESSION METHODS
Huffman Coding
◦ Huffman coding (Huffman, 1952) is one of the most common
methods of variable-length encoding used to reduce the coding
redundancy. As we have seen, representing every pixel of an
image with a constant number of bits results in a suboptimal
bitrate. Huffman coding is a simple algorithm that optimizes the
bit rate by representing the pixel values that appear most times in
an image with the shortest codes (symbols that require fewer
bits), and using progressively longer codes for the pixel values
that appear fewer times in the image.
Pixel-Difference Encoding
◦ the values of adjacent pixels in an image (or the-same pixel in
adjacent frames in time-sequence images) do not vary
considerably; consequently, each pixel does not provide much
more information than its neighbor. Therefore, a simple method of
reducing spatial redundancy is to store only the differences
between pixels which are either adjacent by location(for two-
dimensional [2D] and 3D images) or by time (time-sequence
images).
Arithmetic Coding
◦ As opposed to Huffman coding which replaces each pixel value in an image
with a special code on a one-to-one basis, another method of variable-length
encoding, arithmetic coding, replaces a set of pixel values with one code
(Abramson, 1963).
◦ As can be seen, for this image, arithmetic coding improved slightly on the
results of Huffman coding, and achieved almost the theoretical maximum
compression ratio predicted by the image’s entropy.
JPEG Compression
◦ In JPEG compression algorithm, an international standard, is
oneof the most commonly used image compression methods
today,and is therefore an algorithm that is worth studying in
detail.
◦ TheJPEG compression standard actually includes three different
com-pression algorithms: (1) the standard algorithm, (2) the
progressiveJPEG algorithm for higher quality or compression
ratios, and (3) alossless algorithm (JPEG-LS).