Classification of Metastatic Lymph Node Sections Using Deep Learning

Cancer is an extremely aggressive disease, in which abnormal cells divide uncontrollably and destroy vital body tissue. Metastasis is a form of cancer when cancer cells break away from the main site and propagate to other vital organs. Since lymph nodes are responsible for the transportation of lymphocytes (the main disease-fighting cells), these are specifically central to this phenomenon. Incidentally, the highest cause of death in cancer patients is metastasis. To automate detection of metastasis for better diagnosis, many CNN-based architectures have been employed for improved accuracy. These are, however, a bit tedious to run on local systems due to their computation time and performance. To obtain a similar level of accuracy with much faster and lighter infrastructure, we have proposed the use of FullyInceptionResNet, a model that builds on InceptionResNetV2. We achieved training accuracy, AUC, and precision of 93.4%, 0.97, 95.26% respectively. Future improvements on our work can help reduce overfitting and increase ability to deploy in accordance with realistic datasets.