DEVELOPING A BREAST CANCER DISEASE DETECTION MODEL USING CNN APPROACH

Abstract

Cells that divide uncontrollably and spread into the surrounding tissues are what cause cancer. DNA changes are a cause of cancer. The majority of DNA alterations that lead to cancer occur in regions of DNA known as genes. One of the cancer diseases that is commonly recognized from a variety of angles as being quite diverse is breast cancer. It is among the main causes of death for females between the ages of 20 and 59 worldwide. According to the World Health Organization's (WHO) 2020 cancer country profile report, breast cancer has the highest age-standardized mortality rate of 22.9 per 100,000 people in Ethiopia, making it the most common cancer there. Early detection and care help patients receive adequate treatment and, as a result, reduce the risk of breast cancer morbidity. According to research, most experienced physicians can diagnose cancer with 79 percent accuracy, while machine learning techniques can achieve 91 percent accuracy. The main aim of this study is to develop a model that can assist a physician in detecting breast cancer and classifying it. Mammogram images were collected from the Korea hospital repository and used for developing a deep learning model. A pre-trained model such as VGG16, Inception, and SDDmobilenet are used as transfer learning for fine tuning. Also, there was a CNN model built from scratch with learning rate, batch size, and epoch and optimizer parameter optimization technique. The model built on InceptionV3 score the highest accuracy of 88% on training. The developed models have the capability of categorizing breast cancer. But the data is not sufficiently available for some classes. To solve the problem the researcher applied augmentation to overcome the problem of overfitting. Therefore collecting a large amount of data for all classes and developing a more reliable classification model is the future work of this thesis.