Scribd
Upload
35 views3 pages

Bacteria Classification Using Image Processing and Deep Learning

Uploaded by

190401013
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd
35 views3 pages

Bacteria Classification Using Image Processing and Deep Learning

Uploaded by

190401013
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd
You are on page 1/ 3

Bacteria Classification using Image Processing and Deep learning

Treesukon Treebupachatsakul1, Suvit Poomrittigul2


1
Department of Biomedical Engineering, King Mongkut's Institute of Technology
Ladkrabang, Thailand
2
Department of Software Engineering and Information System, Pathumwan Institute of
Technology, Thailand
E-mail: [email protected], 2 [email protected]
1

Abstract The purpose of this research is to study possibility


to use image classification and deep learning method
An automizing process for bacteria recognition for classify bacteria species by distinguish two basic
becomes attractive to reduce the analyzing time and shapes of long-rod and spherical. For this report, we
increase the accuracy of diagnostic process. This researched to find the methodology to use deep
research study possibility to use image classification learning to classify the standard resolution bacteria
and deep learning method for classify genera of image for further research in mobile application or
bacteria. We propose the implementation method of tablet use case.
bacteria recognition system using Python programing
and the Keras API with TensorFlow Machine 2. Method
Learning framework. The implementation results
have confirmed that bacteria images from microscope 2.1 Bacteria cultivation and photomicrography
are able to recognize the genus of bacterium. The Two species of bacteria were used in this research,
experimental results compare the deep learning Staphylococcus aureus TISTR 746 and Lactobacillus
methodology for accuracy in bacteria recognition delbrueckii TISTR 1339. S. aureus TISTR 476 was
standard resolution image use case. Proposed method cultivated in nutrient agar (NA) (0.5% peptone, 0.3%
can be applied the high-resolution datasets till yeast extract, 0.5% NaCl, 1.5% agar) for 24 h at 30°C.
standard resolution datasets for prediction bacteria While L. delbrueckii TISTR 1339 was cultivated in
type. However, this first study is limited to only two MRS agar (Himedia) for 24 h at 37°C. The cultivated
genera of bacteria. S. aureus and L. delbrueckii were stained by Gramm’s
method. The digital images of gram stained bacteria
Keywords: Bacteria recognition, Image processing, were taken with Optika B-292 Biological Microscope
Deep learning, Image classification equipped with an Optikam B3, Italy. They were
evaluated using a 100 times objective under oil-
1. Introduction immersion. We prepare own dataset Digital images of
both bacteria.
The recognition of various genera and species of
bacteria is useful diagnostic process. The classic 2.2 Deep Learning with Python Programming
techniques of bacteria recognition mainly use Deep learning is a type of Machine Learning using
biological techniques such as gram staining and deep layered structure Artificial Neural Network
identify from contained specific metabolism of (ANN). In this paper, we propose deep learning
bacteria. This classical recognition of bacteria species methodology for bacteria images classification using
is based on comparative analysis of the obtained Python programing included Keras API with
samples with referential ones (American Bank ATCC; TensorFlow Machine Learning framework. This
https://www.lgcstandards-atcc.org). Consequently, paper implemented the LeNet Convolutional Neural
these methods are time-consuming process and Network (CNN) architecture using Python [3].
require an expert knowledge and experience of worker.
Therefore, automatizing the process of bacteria 2.3 Training and Test Datasets
recognition is attractive to reduce the analyzing time Two type of S. aureus dataset and L. delbruekii,
and increase the accuracy of worker [1, 2]. The most each has more than 400 sample images. Each dataset
important features of bacteria that can be recognized has been separated the training datasets and test
on the images is the color of gram stained and the datasets by percentage of 80 and 20, consequently.
shape of its cell.

1
2.4 Research Methodology

The aim of this study was to develop a system that


can immediately and automatically identify and
classify bacteria image in standard resolution for
mobile and tablet use cases. To first observation, we
try to use and prepare own datasets of bacteria in
standard resolution for studying possibility to using
deep learning for classify the bacteria images with
practical LeNet CNN method.

(a) Staphylococcus Correct Prediction with


3. Experimental results probability 85.82%
Experimental results have shown that deep
learning methodology can use to classification high-
resolution bacteria images by existing research [4]
and standard resolution bacteria images which prepare
by authors as shown in Figure 1 and Figure 2

(b) Staphylococcus Correct Prediction with


probability 82.21%

Figure 1 Training Loss and Accuracy existing


research datasets

(c) Lactobacillus Correct Prediction with probability


100%

Figure 2 Training Loss and Accuracy own datasets


(d) Lactobacillus Correct Prediction with probability
Figures 1 and 2, the results confirmed that the
99.83%
standard resolution bacteria images which prepare by
Figure 3 Experimental Results of bacteria detection
authors could be used for training by supervised
output and the percentage of possibility bacteria
learning method with LeNET method.
prediction

2
Figure 3 (a) (b) (c) (d) show examples of bacteria 5. Conclusion
detection output and the output can detect type of
bacteria and inform the percentage of possibility The experimental results have confirmed that two
bacteria prediction. species of bacteria in different cell shape,
Figures 1-3, the results have confirmed that the Staphylococcus aureus (spherical or round shaped)
training method Training accuracy and validation and Lactobacillus delbrueckii (long-rod shaped) are
accuracy have reached more than 75 percent by more able to automatically predict using machine learning
training times of 4 Epoch both of high-resolution and by image classification and deep learning method.
standard resolution bacteria image. The experimental results compare results using
existing research dataset and own prepared dataset for
4. Comparison of Experimental Results accuracy in bacteria prediction use case.
and Discussion Training accuracy and validation accuracy has
reached more than 75 percent by more training times
of 4 Epochs both of high-resolution and standard
Moreover, the trend of improvement for better resolution bacteria image.
accuracy could be investigated by implementing more Accordingly, this research also tries to modify
numbers of Epoch and activation function variables LeNET methodology and observe the trend to make
choosing. more training with greater number of Epochs. The
results have confirmed that standard resolution
bacteria images datasets could be improve more
accuracy in future. It can be more apply another CNN
methodology for comparison such as ResNET,
AlexNET and etc.
However, this first study is limited to only two
species of bacteria in different cell shape. We will
propose more than two genera of bacteria and improve
the accuracy for using with mobile and tablet use
cases in future.

5. Acknowledgement

Figure 4 Training Loss and Accuracy in 10 Epochs This research also was financial supported by
with Sigmoid Activation Function KMITL Research and Innovation Service (KRIS).

References
[1] K. Lim, S. Hyun Park, et al., Cell image processing
methods for automatic cell pattern recognition and
morphological analysis of mesenchymal stem cells –
An algorithm for cell classification and adaptive
brightness correction, J. of Biosystems Eng, 38(1), pp.
55-632013, February 2013.
[2] L. Shamir, J. D. Delaney, et al., Pattern recognition
software and techniques for biological image analysis,
Figure 5 Training Loss and Accuracy in 10 Epochs Plos computational biology, 6(11), November 2010.
with Softmax Activation Function [3] Y. LeCun, L. Bottou, Y. Bengio, and P. Haffner.
Gradient-based learning applied to document
The results of Figures 4 and 5 confirmed that when recognition. Proceedings of the IEEE, November
applying more numbers of epochs, Training Loss and 1998.
Accuracy would be better. Training Loss and [4] B. Zielinski, A. Plichta, et al., Deep learning
Accuracy in 10 Epochs with Softmax Activation approach to bacterial colony classification, Plos one,
Function methodology is better than Sigmoid 12(9), September 2017.
Activation Function, which reached 96% of validation
accuracy.

You might also like