Health Discernment System
Health Discernment System
Health Discernment System
01/19/2024
Introduction
• The significant growth of medical images and techniques requires comprehensive and exhaustive
efforts from a medical professional who is susceptible to human error and the result can
also vary widely among various experts.
• ML and AI techniques have been used for medical image processing and image recognition.
Deep learning-based algorithms have demonstrated remarkable outcomes in the medical field
with works like computer-aided diagnosis, speech recognition and so on.
• In this project, we have used the above stated idea behind disease detection, to develop a system
using convolutional neural network (CNN) that will help in detection of a particular disease.
• The system has been made user-friendly with the help of GUI, so that it can be used not only by
the medical professionals but also by the population at large.
01/19/2024
About the Diseases
Malaria
• Malaria is a deadly, infectious mosquito-borne disease caused by Plasmodium parasites.
These parasites are transmitted by the bites of infected female Anopheles mosquitoes.
• A pre-trained convolutional neural network has been used as feature extractors toward
improved malaria parasite detection in thin blood smear images.
• An automated process has been put forward that recognizes both the tasks of malaria
parasite segmentation as well as classification.
01/19/2024
Contd..
Pneumonia
• Pneumonia is an infection in one or both lungs. Bacteria, viruses, and fungi causes it.
The infection causes inflammation in the air sacs in the lungs, which are called alveoli.
• This work mainly proposes a convolutional neural network (CNN) model to group
and identify the occurrence of pneumonia disease from a given assortment of chest X-
ray images.
• Based on the model, disease detection from X-ray images was investigated,
classifications on chest X-ray image views were performed.
01/19/2024
Contd..
Breast Cancer
• Breast cancer is one of the main causes of cancer death worldwide. After skin cancer,
breast cancer is the most common cancer diagnosed in women but men can get breast
cancer too.
• In this model, classification of breast cancer histology images have been done by
extracting features through a convolutional neural network (CNN) and then classifying
them using a fully connected network.
• This model has been tested for binary classification of the diseases with competitive
experimental results.
01/19/2024
Contd..
Skin Cancer
• Skin cancer is an abnormal growth of skin cells. Most skin cancers are caused by
exposure to ultraviolet (UV) light. When the skin is not protected, UV rays from sunlight
or tanning beds can damage and alter skin's DNA that leads to the cancer.
• Deep learning model has been built to classify and identify the binary diagnostic group
of melanocytic images obtained through dermoscopy.
•Based on the model, disease detection through dermal cell images has been investigated,
and classifications on dermal cell images have been performed.
01/19/2024
Literature Survey
Table 1. Comparison of different literatures
01/19/2024
Motivation
• Disease detection plays a very important role in the process of diagnosis. Therefore,
the motivation lies in accurate classification and detection of the diseases based on
medical images.
• The main aim is to minimize the chances of error that might happen due to the doctor's
misjudgement.
• Developing a system that will not only help in detecting the diseases efficiently but
will also save the time and effort of the medical practitioners.
• This will also save the patients from running to the doctor to get their medical reports
verified.
01/19/2024
Software Requirements Specification
01/19/2024
Contd..
• The user also has to have the medical test images in JPEG, JPG or PNG format for detection
purpose.
01/19/2024
Requirement Analysis
Specific Requirements
There are two interfaces namely – User Interface and Software Interface.
• User Interface
The user interface will be implemented using any desktop running on Windows OS. This interface
will be very user friendly so that people from different strata can use it to detect their disease
without any difficulty by just uploading their medical test image.
• Software Interface
A software interface running on Windows OS. It should have Python compiler(3.6).
01/19/2024
Contd..
BEHAVIOUR REQUIREMENTS
01/19/2024
Contd..
PERFORMANCE REQUIREMENTS
• The evaluation performance will depend on some software and hardware components.
The security of the predicting interface depends directly on the security of the workstation.
01/19/2024
Methodology and System Design
Image Processing
• Image processing can be defined as the technical analysis of an image by using
complex algorithms.
• The purpose of early image processing was to improve the quality of the image. Its
use has been increasing exponentially in the last decades.
01/19/2024
Contd..
• Another benefit of CNNs is that they are easier to train and have fewer parameters than
fully connected networks with the same number of hidden units.
01/19/2024
Contd..
• The design pattern of MVC decouples these major components to allow efficient
reuse of code and also allows parallel development of each of the component.
• This architecture is mostly used for desktop graphical user interfaces (GUIs) and is
popular in designing web applications.
01/19/2024
Contd..
The Model
• A Model is the principal central component of an application.
• The Model used in this work comprises of 4 various sub-models which represent each disease.
• The dataset is then divided into the train set and test set of 22046 and 5512 images respectively.
• The custom CNN model used had 3 Convolutional layers and one fully-connected layer. ‘relu’
and ‘softmax’ activation functions are applied to the hidden layers and the output layer
respectively.
01/19/2024
Contd..
• In the pneumonia model, a dataset comprising of chest x-rays has been taken. The images are
of 224x224 pixels.
• The dataset is then divided into the train set and test set of 22046 and 5512 images
respectively.
• CNN model having 10 convolutional layers and one fully-connected layer has been used.
• For activation purpose, ‘relu’ and ‘sigmoid’ functions are used in the hidden and output layer
respectively.
01/19/2024
Contd..
• For this, a dataset has been taken of cell images of microanatomy that contains RGB images
of 50x50 pixels.
• To the dataset, a CNN model has been applied of 6 convolutional layers and one fully-
connected layer.
• To the output layer and the hidden layers, ‘sigmoid’ and ‘relu’ activation functions are applied
respectively.
01/19/2024
Contd..
• The dataset comprising of RGB images of skin samples has been taken. The images were of
224x224 pixels.
• The dataset is then divided into the train set and test set of 2637 and 660 images respectively.
• A CNN model of 4 convolutional layers and one fully connected layer has been developed.
• For activation, ‘softmax’ and ‘relu’ are applied to the output layer and hidden layers
respectively.
01/19/2024
Contd..
01/19/2024
Contd..
The View
• It helps UI / UX people to operate in parallel with the people at the back-end of the user
interface.
The Controller
• It obtains the user's interaction with the View, transmits them on to the Model that
then processes the input information for output production.
01/19/2024
Contd..
• There are mainly three files in the project: main.py, detector and upload which contains
the GUI, Detector and Controller components, respectively
• We can alter and run the GUI component separately without compromising the
functionality of the other components.
• A single Controller component works for all the Views and Detectors.
01/19/2024
Results
Malaria 95.21%
T02 Pneumonia Normal Normal
Detection Detected Detected Pneumonia 90.06%
Breast Cancer 86.88%
T03 Breast Benign Tissue Benign
Cancer Detected Tissue Skin Cancer 84.6%
Detection Detected
01/19/2024
Preview of the Application
Interface 1 Interface 2
• On opening the application, the opening page is • This page opens on selection of a particular
displayed. disease.
• It consists of 4 buttons (one for each disease). • It consists of – UPLOAD button, DETECT button,
• The user clicks on the disease they want to detect. INPUT image view and, OUTPUT text view.
01/19/2024
Contd..
• This page opens when user clicks the • The selected image is displayed on screen in
UPLOAD button. the INPUT image view.
• This opens a file where the user selects the
image to be detected.
01/19/2024
Contd..
01/19/2024
Conclusion
• A “Health Discernment System” has been proposed for medical image classification
that will work in real-life scenarios.
• Different sub-models pertaining to the four diseases (malaria, pneumonia, breast cancer,
and skin cancer) have been designed using convolutional neural network (CNN) and
they have all been tested separately.
• A GUI based web application has been designed using Tkinter so that the system can
take real-life image inputs for detection.
01/19/2024
Future Scope
• Major attention will be given to increase the accuracy such that our proposed system
can be used to detect a large number of chronic and critical diseases.
• When these enhancements are done, the system can be integrated with an android
application to make it more convenient and easily portable. This will allow people
from all strata to use it effectively even if they do not have a personal computer.
01/19/2024
References
[1] Bibin, Dhanya, Madhu S. Nair, and P. Punitha. "Malaria parasite detection from peripheral blood smear images using deep belief networks." IEEE Access
5 (2017): 9099- 9108.
[2] Mohanty, Itishree, P. A. Pattanaik, and Tripti Swarnkar. "Automatic Detection of Malaria Parasites Using Unsupervised Techniques." In International
Conference on ISMAC in Computational Vision and Bio-Engineering, pp. 41-49. Springer, Cham, 2018.
[3] T. I. Mohammad, A. A. Md, T. M. Ahmed, and A. Khalid,“Abnormality detection and localization in chest x-rays usingdeep convolutional neural
networks,” 2017, http://arxiv.org/abs/1705.09850.
[4] W. Xiaosong, P. Yifan, L. Le, L. Zhiyong, B. Mohammadhadi,and M. S. Ronald, “Chest X-ray8: Hospital-scale chest X-raydatabase and benchmarks on
weakly-supervised classificationand localization of common thorax diseases,” 2017, http://arxiv.org/abs/1705.02315
[5] F.A. Spanhol, L.S. Oliveira, C. Petitjean, L. HeutteA dataset for breast cancer histopathological image classification IEEE Trans Biomed
Eng, 63 (7) (2016), pp. 1455-1462, 10.1109/TBME.2015.2496264 https://doi.org/10.1109/TBME.2015.2496264
[6] F.A. Spanhol, L.S. Oliveira, C. Petitjean, L. HeutteBreast cancer histopathological image classification using convolutional neural networks 2016
international joint conference on neural networks, IJCNN 2016, vancouver, BC, Canada, july 24-29, 2016 (2016),
pp. 2560-2567, 10.1109/IJCNN.2016.7727519 https://doi.org/10.1109/IJCNN.2016.7727519
[7] L. Yu, H. Chen, Q. Dou, J. Qin, and P.-A. J. I. t. o. m. i. Heng, "Automated melanoma recognition in dermoscopy images via very deep residual networks,"
vol. 36, no. 4, pp. 994-1004, 2017.
[8] U.-O. Dorj, K.-K. Lee, J.-Y. Choi, M. J. M. T. Lee, and Applications, "The skin cancer classification using deep convolutional neural network," pp. 1-16,
2018.
[9] www.kaggle.com - The world's largest data science community
[10] www.tensorflow.org - open-source machine-learning platform
01/19/2024
Thank You!
01/19/2024