A Deep Learning Approach For Efficient Breast Cancer Diagnosis Using Hybrid CNN-BILSTM with Soft Attention Mechanism

Abstract

In recent years, the rising prevalence of breast cancer among women has underscored the critical importance of timely and accurate disease prediction. The ability to predict breast cancer efficiently empowers healthcare professionals with a robust decision-making system, enabling optimal treatment strategies. This study presents a sophisticated diagnostic framework, integrating various techniques from preprocessing to classification to enhance the precision and effectiveness of breast cancer detection. The proposed methodology begins with preprocessing using a Butterworth filter to enhance the quality of input data. The preprocessed output is segmented utilizing Cascaded Fuzzy C-Means segmentation. Subsequently, extracting features through the (GLCM), capturing intricate patterns and texture information crucial for discriminating between benign and malignant tissues. The classification step employs a Hybrid CNN-BiLSTM architecture augmented by a Soft Attention Mechanism. This hybrid model is designed to effectively leverage spatial hierarchies and sequential dependencies in medical images, allowing for a more comprehensive analysis of complex forms related to breast cancer. Soft Attention Mechanism enhances the interpretability of the model by assigning varying weights to different regions of the input data, emphasizing salient features crucial for exact analysis. The attention mechanism donates to the model’s ability to focus on relevant information, improving both sensitivity and specificity in breast cancer classification. The proposed methodology is validated through extensive experiments, demonstrating its superior performance compared to traditional approaches. The results highlight the possibility of this deep learning-based strategy valuable enabling healthcare professionals, facilitating precise and timely decision-making in breast cancer treatment optimization.