Comparative Analysis of Transfer Learning and Customized Deep Convolutional Neural Networks for Breast Cancer Detection and Classification
Rakibul Islam, Mahbubur Rahman, Md Sipon Miah, Md. Khairul Islam, Md Shahin Ali, Mingbo Niu · 2023
Breast cancer is a highly lethal form of cancer that affects the cells in the breasts and is second only to lung cancer in terms of its impact on women's health. It is a prevalent type of cancer that affects women globally. Detecting breast cancer at an early stage is crucial to reducing mortality rates. However, the current manual diagnosis process for breast cancer is time-consuming, and there is a shortage of qualified personnel. Additionally, human diagnosis can be prone to errors. Therefore, an automated system is required to address these challenges. Breast cancer detection using image processing techniques has become more popular in recent years. The objective of this study was to examine and evaluate the ability of a Deep Convolutional Neural Network (DCNN) to detect and classify cases of Breast Cancer. Besides this, a customized model is developed. Preprocessing involves normalizing the incoming images, extracting features that aid in accurate classification, and then augmenting the data with more images to increase classification accuracy. To evaluate the performance, the customized models performance is compared with the transfer learning models and an Ensemble model. Our proposed model gives the best accuracy of 84.38% with compared to other transfer learning models and an Ensemble Model.