Authors
Ahmed Ossama Ibrahim, Marwan Hashem, Mostafa Ahmed, Omar Ahmed Salah
Supervised by: Dr. Alaa Hamdy, Eng. Verina Saber
Publishing Date
January 3, 2022
Abstract
This research aims to develop an intelligent system that can detect and classify various liver diseases using image analysis and machine learning methods. This study will employ Ultrasound (US) imaging, which has been identified as one of the most common and affordable methods of medical imaging in Egypt [1] to identify liver diseases in a non-invasive manner. However, because to the low price, the images may be of poor quality or lack detail. As a result, we will use a feature-based technique to correctly categorise the US photographs, and after completing significant research, we will select the most efficient classifier (such as KNN, ANN, or SVM) to use in the feature extraction phase.
1.1 Purpose of this document
The objective of this Software Requirements Specification document is the laying out of requirements for the ’Automatic Classification of Diffuse Liver Diseases Cirrhosis & Hepatosteatosis using Ultrasound Images’ application. The paper will be useful to future project users as well as anyone involved in the project’s development and maintenance.
1.2 Scope of this document
This document is aimed at the software’s future users, which are doctors and other medical personnel working in this field. As well as the patients who will receive the application’s diagnostic and findings. It also targets any developer who will be involved in the system’s future development or maintenance.
1.3 System Overview
The system overview as shown in Figure 8, will begin with splitting the image dataset into a training and testing set from which Regions of interest ROIs will be extracted from while ensuring that no image has an ROI in both training and testing sets. Multiple features including Gray Level Co-occurrence Matrix (GLCM), Gray Level Run-Length Matrix (GLRLM), and First Order Statistical features will be extracted from the ROIs, from which the most prominent features will be selected using future selection algorithms, the selected features will be used to train on three binary classifiers. The binary classifiers will be classifying individual ROIs, the prediction for each ROI of an image will be inserted into a majority function to decide on that image’s classification for each binary classifier, finally the result of the majority functions will be inserted into another majority function to decide that image’s final classification over the three classes, if no two classifiers can agree on a classification the system will abstain from classifying the image.
1.4 System Scope
Our proposed system uses an ultrasound image to detect the three types of liver diseases (cirrhosis, steatosis, and normal). Our classification system will detect the abnormality faster than a traditional diag- nosis.
The system will:
• Classify whether the liver ultrasound is normal or abnormal.
• Classify the type of abnormality (cirrhosis or steatosis).
• Classify the stage of the abnormality (mild,moderate and severe).
• Generate a patient report.
• Abstain from giving a classification if the model thinks the result is inaccurate.