A Validation Study of the Power and Effectiveness of Machine Learning in Mammogram Interpretation رسالة ماجستير

Abstract

Breast Cancer is widespread across the globe. It’s the primary cause of death in cancer fatalities. According to the Palestinian Ministry of Health (MoH) annual report, it ranked as the third reported death of all reported cancer deaths in the West Bank. Breast cancer has many symptoms like breast pain, discharge, lumps, and masses. It also has risk factors that can be categorized into genetics and lifestyle which helps in developing the cancer. Moreover, mammogram screening is the most common technique to diagnose breast abnormalities, but there is a challenge in the lack of skilled experts able to interpret mammograms accurately. Machine Learning (ML) plays an important role in medical image processing particularly in early detection when the treatment is less expensive and available. This thesis presents two approaches to detecting breast abnormalities based on the Convolutional Neural Network (CNN). The first approach classified images as normal and abnormal, while the second classified them into Breast Imaging-Reporting and Data System scores (BI-RADS). Furthermore, six CNN models were implemented in both approaches, namely VGG16, VGG19, DenseNet121, ResNet50, Xception, and EfficientNetB7. The used dataset is a unique (first-hand) dataset collected from the Palestinian MoH. Based on the results, DenseNet121 outperformed other models in the first approach with 0.83 and 0.85 for testing accuracy and Area Under Curve (AUC) respectively. In contrast, Xception and EfficientNetB7 obtained the best results in the second approach with more than 90% accuracy. As a future work, the outperformed model can be integrated with other patient data like genetic information, medical history, and lifestyle factors to evaluate the risk of developing specific diseases. This would increase the survival rate and enable proactive VI measures. Finally, a larger dataset of mammogram images should be collected to improve results and further generalize the models