Prognosis prediction of muscle invasive bladder cancer based on whole slide images

Muscle invasive bladder cancer is a type of bladder cancer with poor prognosis, and is of great clinical significance to predict prognosis survival through various factors. For the evaluation on prognosis, in addition to clinical information, more and more studies have demonstrated the importance of tumor microenvironment. As the whole slide images can provide detailed features at tissue and cell level, they are suitable for the evaluation of tumor microenvironment, and their image features themselves also could contribute to prognosis evaluation. Therefore, this paper proposes a semi-quantitative evaluation method of tumor microenvironment based on deep learning using whole slide images, and establishes a 3-year survival prediction model based on machine learning using a fusion of whole slide images features, tumor microenvironment features and clinical features. The final model achieves 86.00% of accuracy, 0.95 of AUC, 0.8353 of specificity and 0.8923 of sensitivity on muscle invasive bladder cancer from TCGA database, which confirms the effectiveness of whole slide images and these three types of features for prognosis survival prediction.