EXPLAINABLE AI FOR BREAST CANCER DETECTION USING GRAD-CAM AND MASK MAMMOGRAM NEURAL NETWORKS

Abstract

Early identification is essential for increasing survival rates of breast cancer, which is still a top cause of death among women worldwide. This research introduces a cutting-edge system that uses deep learning and state-of-the-art image processing to automatically identify breast cancer in mammography pictures. Bilateral filtering is used for image preprocessing in the approach, which also incorporates CNN for classification, MMNN for segmentation, and Grad-CAM for explainability. Microcalcifications and lesions, which are critical for correct diagnosis, are preserved throughout the preprocessing stage by using bilateral filtering to decrease noise. Using a convolutional neural network (CNN), intricate patterns linked to breast cancer may be learnt and used to classification tasks. Clinicians can better understand the model's decision-making process thanks to Grad-CAM, which shows them exactly which parts of the picture were considered. In order to pinpoint tumours, the MMNN is used for ROI segmentation and delineation; subsequently, bounding box detection is employed, yielding invaluable spatial data for further clinical investigation.All of the work is done in a Python environment, making use of well-known modules to display the findings. Automated breast cancer screening and diagnosis might benefit greatly from the suggested framework because to its impressive accuracy in locating and identifying breast tumours. An effective, scalable, and interpretable method for early-stage cancer diagnosis may be achieved by the combination of Python's image processing capabilities with deep learning.