Framework for automatic segmentation of breast cancer using lightweight convolutional neural networks

Given a public database of medical images, testing and comparing different neural network algorithms using that database without a region of interest is a challenging task. This work aims to use the CBIS-DDSM (Curated Breast Imaging Subset of the DDSM - Database for Screening Mammography) and InBreast database, to test a lightweight model of convolutional neural networks (CNN) to segment the masses of tumors. The proposed model is a reduced version of the "U-Net" architecture to be lighter and faster, this semi-automatic tool could be attached to CAD systems to help health professionals in real-time decision-making. The proposed method-available in Google Colaboratory format for easy replication and modifications-works faster than the default "U-Net" without harming the results. This work is a reproducible tool and does not achieve state of art results that uses other heavy methods, but should be enhanced since it is open source. Results showed that the model can predict tumors masks of both Medial Lateral Oblique (MLO) and Craniocaudal (CC) cases. We created a premise using the region of interest to define positive results and with that premise, the model achieved a mean dice coefficient of 0.60 and a mean accuracy of 0.40. With test CPU hardware the model can predict 32 images per second, with dedicated GPU the model can predict 237 images per second.