PatchDetect: Breast Cancer Detection combining Unet-ResNet-50 and Patch Embedding LSTM

Abstract

This study presents a novel framework for breast cancer detection, combining patch embedding, feature extraction using a pre-trained Convolutional Neural Network (CNN) model (ResNet50), Long Short-Term Memory (LSTM) networks for image sequence analysis, and Fully Connected Layers for final classification. The model's performance was optimized using various hyperparameters, achieving an accuracy of 94%, recall of 93%, precision of 92%, and F-measure of 92% while maintaining a minimal error rate of 6%. The findings emphasize the importance of integrating pre-trained CNNs with sequential analysis via LSTMs for feature-rich and temporal data like mammographic patches. The study also highlights the impact of parameter tuning on classification performance, paving the way for more accurate, automated, and non-invasive breast cancer diagnostic tools.