DEEP NEURAL ARCHITECTURE FOR ROBUST AND EXPLAINABLE BREAST CANCER CLASSIFICATION USING HISTOPATHOLOGICAL IMAGING

Abstract

Breast cancer among womens is one of the most common and life-threatening diseases which has been affecting the health of womens worldwide, and an early diagnosis of the breast cancer plays a very important role in improving the treatment effectiveness and the survival rates.The recent advancements in the fields of Artificial Intelligence (AI), Machine Learning(ML), and Deep Learning(DL) have very significantly improved and automated medical diagnosis systems. This research work presents an explainable deep learning (DL) framework for the breast cancer classification by using the Wisconsin Breast Cancer Dataset (WBCD), which consists of the diagnostic features extracted from the digitized images of breast cell nucleus. Initially, the traditional machine learning (ML) algorithms which includes techniques like Logistic Regression(LR), Decision Tree(DT), and Random Forest(RF) were first im plemented to establish the baseline performance for breast cancer classification. A one dimensional(1-D) Convolutional Neural Network (CNN) model was then developed and further it was enhanced using techniques like Batch Normalization (BN), Dropout regu larization, and an attention mechanism to enhance the feature extraction and accuracy of classification . The Wisconsin Breast Cancer Dataset (WBCD) was pre-processed using the feature standardization and then it was divided into the training set and testing set by using an 80:20 ratio. To improve transparency of classification and the interpretability of classification, Ex plainable Artificial Intelligence(XAI) techniques including the Local Interpretable Model Agnostic Explanations(LIME) and the SHapley Additive exPlanations (SHAP) were then integrated into the framework. Local Interpretable Model-Agnostic Explanations(LIME) was utilised to generate the local explanations for the individual predictions, while SHap ley Additive exPlanations (SHAP) provided both, the local and the global feature im portance analysis. The experimental results obtained demonstrate that the proposed Attention-Enhanced Convolutional Neural Network(CNN) model achieves the improved performance of classification while providing reliable and interpretable predictions which are suitable for health-care applications and clinical decision support systems.