Sara Noor Eldin, Jana Khaled Hamdy, Ganna Tamer Adnan, Maysoon Hossam, Dr Ammar Mohammed, Eng Noha Elmasry, and Eng Menna Gamil
Cancer is the second-highest cause of death as reported by the World Health Organization. A correct diagnosis of Breast Cancer ensures that appropriate treatment plans and procedures are provided to patients. Due to the disparity in the pathologists’ skills, manual histopathology examinations conducted by them are complex, time-intensive, and might be vulnerable to misinterpretations. Using Deep Learning algorithms, the key concept of our proposed system is to diagnose breast cancer types from microscopy biopsy images. The proposed system will detect whether the abnormal lesions presented in these images are benign or malignant and if it is malignant, it will be stated if it is an In-Situ Carcinoma or Invasive Carcinoma. We seek to reduce the time results take to reach patients who are in both physical and emotional distress, as well as reducing the possibility of misdiagnosis that might lead to more severe complications in patients’ lives.
1.1 Purpose of this document
The purpose of this Software Requirements Specification document, is to present the requirements of ’Breast Cancer Diagnosis using Microscopy Biopsy Images’ software. The document will act as an aid to the future users of the project, as well as anyone who is concerned with the future development and maintenance of the project.
1.2 Scope of this document
This document targets the future users of this software, which are the pathologists or any of the medical staff concerned with this domain. As well as the patients who will receive their diagnosis and reports from the application. It also targets any developer that will take part in the future development or maintenance of the system.
1.3 System Overview
The system overflow will begin with the phase of preprocessing the input images, datasets of histopathological images will be used as inputs. First, the images will go through some enhancements as smoothing filters to find the region of interest as well as transformations to enhance the image quality. Then, the enhanced images will go through the next phase. The next phase will be the training phase, the enhanced images of the dataset will be split into training data and testing data, and several trials will be done to decide the distribution of the training and testing data. The one resulting in the highest accuracy will be chosen. The training data will enter the deep learning model which is a CNN classifier, and several CNN architectures will be used to compare between them to obtain the classifier that will achieve the highest accuracy. The classifier will study the different types of tumors and differentiate between them. The testing data will then pass through the model to predict the accuracy of the model. Lastly, the detecting phase also starts after the preprocessing phase, the image will go through the evaluated model which is obtained from the training model, the percentages of prediction of which type of tumor will be outputted, and finally, a report will be created with the results as the pathologist can view it or send it to the patient.
1.4 System Scope
Our proposed system is designed to detect the type of breast cancer(Benign, In-Situ Carcinoma, Invasive Carcinoma). The system will detect breast cancer faster than normal pathology procedures. It aims to cut down the chances of misdiagnosis and reduce the time they spend until they receive the results to start the right treatment path.
The system will:
• Detect if the breast biopsy image is normal or abnormal.
• Detect the breast cancer type if it is abnormal.
• Generate a report with the patients’ data and results.
• Does not detect the breast cancer stage.