Modified ReLU in Deep Learning Models and Explainable AI Techniques for Accurate and Interpretable Breast Cancer Subtype Classification
Wahyu Nugroho, Catur Supriyanto, Guruh Fajar Shidik, Pujiono Pujiono · 2024
Breast cancer is a serious condition that presents a considerable risk to life if not identified in its initial stages. Numerous approaches for identifying breast cancer remain to be performed conventionally through medical examination, which entails limits in accuracy and reliability. Recent technological ad-vancements have led several researchers to create diverse deep-learning methodologies for automated breast cancer diagnosis, particularly convolutional neural network (CNN) models. The main objective of this study is to make the CNN model work better by modifying the rectified linear units (ReLU) activation function to distinguish eight subtypes of breast cancer. Based on the experiment results, the DenseNet201 model with Less-N egativeReLU modified activation function obtained the best performance with accuracy, precision, recall, and F1-score of 96.04 %$(\alpha=0.091)$, “ 96.45 %$(\alpha=0.03)$, 96.04 %$(\alpha=0.09)$, and 96.14%$(\alpha=0.03)$, respectively. The findings demonstrate that the proposed method effectively enhances and optimizes the CNN model performance. This study also employed explainable AI (XAI) techniques to improve the understanding of model prediction results, consequently enhancing clinician's confidence in its implementation. By improving the interpretability of AI-driven predictions, this study aims to support the practical adoption of these models in clinical settings, ultimately contributing to better diagnostic processes and outcomes in breast cancer care.