A Comprehensive Deep Learning Framework for Breast Cancer Detection and Classification Using Multiple Convolutional Neural Network Architectures

Abstract

Improving rates of survival of patient for breast cancer (BC) requires early identification and precise categorization. Using the DDSM mammography dataset , this work The goal is to enhance the detection and classification of breast cancer through the application of deep learning techniques. The suggested framework uses a method based on C.N.N. 55,890 pre-processed mammography pictures, divide into train and test sets, make up the dataset. Both positive and negative instances were included in the data, which was further divided into benign masses, benign calcifications, malignant masses, and benign calcifications. The images were scaled to 299x299 resolution. Accuracy, sensitivity, specificity, and other pertinent measures were used to train and assess four models. The classification accuracies of C.N.N, VGG16, ResNet-50, DenseNet121, and Efficient Net were 90.83%, 87.00%, 91.35%, 92.40%, and 94.97%, respectively. Efficient Net achieved the highest performance, demonstrating superior generalizability across diverse imaging modalities and demographic variations. The proposed frameworks, supported by pre-trained models, demonstrates significant potential for improving Early recognition and identification of breast cancer as well. The integration of ethical considerations, interpretability, and a focus on clinical impact ensures its relevance for real-world applications.