Explainable AI-enabled hybrid deep learning architecture for breast cancer detection.

Introduction: Breast cancer stands is a leading prevalent and potential fatal infection affecting women worldwide, posing the requirement of a reliable and interpretable diagnostic system. The Deep Learning (DL) methods highly contribute towards medical imagery analysis but due to the black-box nature, its clinical adoption is limited due to lack of interpretability. Methods: This proposed work introduces a hybrid Deep Learning (DL) framework for that integrates three distinct convolutional neural network (CNN) pre-trained architectures: DENSENET121, Xception and VGG16. The proposed fusion strategy enhances feature representation and classification performance through model integration. To address the DL's black-box nature and promote clinical acceptance, the proposed framework incorporates an explainable artificial intelligence (XAI) component utilizing GradCAM++. Results: Experimental evaluation on benchmark breast cancer datasets demonstrates improved classification accuracy by approximately 13\% compared to individual models, demonstrating high performance of the fusion method with an accuracy of 97\%. Discussion: The use of fused DL model enhances the performance of the classification system offering higher accuracy and robust feature extraction. With the introduction of XAI, the cancer classification system presents interpretable results making it applicable in clinical contexts. GRADCAM++ method highlights the multiple lesions with finer edges from the ultrasound images that leads towards the model's predictions, offering transparency and aiding medical professionals in diagnostic validation.