Automatic Breast Cancer Cell Classification using deep Convolutional neural Networks

Automated cell classification in cancer biology is an active and challenging task for computer vision and machine learning algorithms. In this Thesis, we first compiled a vast data set composed of JIMT-1 human breast cancer cell line images, with and without therapeutic drug treatment. We then train a Convolutional Neural Network architecture to perform classification using per-cell labels obtained from fluorescence microscopy images associated with each brightfield image. The study revealed that our classification model achieves 65% accuracy in breast cancer cells under chemotherapeutic drug treatment with doxorubicin and paclitaxel. Furthermore, it reached 70% accuracy when classifying breast cancer cells without drug treatment. Our results highlight the potential of machine learning and image analysis algorithms to build new diagnosis tools.