Detection of Metastatic Cancer on Lymph Node Sections with Deep Convolutional Neural Network
Yikun Han, Wentao Lin, Mo Zhou · 2021 3rd International Conference on Machine Learning, Big Data and Business Intelligence (MLBDBI) · 2021
Cancer detection is designed to diagnose the presence of cancerous tissue from a medical scan. Detecting cancer as early as possible is very important, which provides a better chance of saving patients' lives. Generally, plenty of resources and time are required to train a high skill in identifying medical images for cancer detection. However, machine learning models can be used to help doctors to make a more precise and quick diagnosis. Typically, most previous models focus on detecting a single type of cancer which only requires a model to learn limited features. In this paper, we will focus on detecting metastatic cancer based on machine learning models, which has many more features and requires a more generalized model. Therefore, we propose a machine learning model to identify metastatic tissue in histopathologic scans of lymph node sections. A deep convolutional neural network model is used with carefully tuned hyperparameters to predict the image's label on the PatchCamelyon benchmark dataset. Besides, a ResNet model and a simple convolutional neural network model are trained. Our experiment results on the PatchCamelyon benchmark dataset show that our model performs much better than baselines. On this basis, deep convolutional neural networks are a proper method for cancer image classification. These results shed light on how we can help doctors detect metastatic cancer earlier with our deep convolutional neural network model.