Deep Learning Architectural Performance Evaluation on Breast Cancer MRI and Mammography Image Datasets

Abstract

Since breast cancer continues to be the top cause of death among women all around, it emphasizes the need of having good early diagnosis instruments. Advances in deep learning have given convolutional neural networks (CNNs) great promise in medical imaging uses. On two separate image datasets comprising MRI and mammography images, this work assesses and contrasts the performance of four deep learning architectures: CNN, ResNet50, InceptionV3, and EfficientNet B0. Accuracy, precision, recall, and F1-score among other measures were used to evaluate the models. Particularly in terms of recall and F1-score, EfficientNetB0 and ResNet50 showed better performance, thereby demonstrating their resilience in spotting positive breast cancer cases. The results highlight the need of effective network designs and transfer learning in improving diagnosis accuracy in imaging of breast cancer.