Aim 1: Develop informatics methods for digital pathology images to support the pathologist based on PLCO image data. We will develop methods that computationally interpret WSI to support the pathologist. We will focus on the following tasks: predicting stage and histology, and predicting clinical outcome. We will point the pathologist to areas of the images that are most important for each task and we will provide visualizations of similar patients and their treatment outcomes in the context of WSI data and multi-scale data.
Aim 2: Validate predictions using publicly available digital pathology images using TCGA data. We will validate our methods on this cohort and determine how the models can generalize to new data.
Aim 3: pancancer vs. single cancer. We will develop a meta-learning approach which is a form of transfer learning to attempt to develop a model that can easily generalize to new cancer sites with fewer training examples. We will compare this approach with single cancer training between PLCO and TCGA data sets.

The work will be done within the Gevaert lab at Stanford university.