Two-Steps Approach for Breast Cancer Detection and Classification Using Convolutional Neural Networks
Mohammad Shubeitah, Ahmad Hasasneh, Shadi Albarqouni · International Journal on Engineering Applications (IREA) · 2024
This work presents a two-step deep learning framework to streamline breast cancer diagnosis using unannotated mammogram images. The first step employs a U-net model to segment mass abnormalities, followed by a reprocessing technique to refine and magnify the segmented masks. The second step utilizes a VGG16 model in order to classify the extracted regions as benign or malignant. Applied to a dataset from Al-Mutlaa Hospital in Palestine, the model has achieved a mass classification accuracy of 91% and has demonstrated robustness with an average Intersection over Union score of 0.70 on a diverse public dataset. The results validate the potential of deep learning models to improve breast cancer detection and diagnostic accuracy, especially in under-resourced regions.