Deep Unlearning of Breast Cancer Histopathological Images for Enhanced Responsibility in Classification
Laman Aliyeva, Nihat Abdullayev, Saida Zarbiyeva, Ismayil Huseynov, Umid Suleymanov · 2024
This study is focused on the applicability of machine unlearning techniques towards histopathological images for breast cancer classification with the usage of the DenseNet model. The core focus area of the research is to mitigate data privacy problems and to comply with jurisdictions regarding accepting the request for data removal. This is achieved through an influence-based unlearning technique that adjusts the model's parameters to minimize the impact of targeted data points. The dataset used for this study is the BreaKHis histopathological images dataset. Additionally, after applied techniques, accuracy, F1-score, unlearning time, and confidence distribution are used for evaluation of the method. This study is among the first to use deep unlearning-based approaches in the domain of classification of medical images, thus, making AI in healthcare more responsible and safe.