Deep Learning for Pathology Image Analysis

Principal Investigator

Name
Lily Peng

Degrees
MD, PhD

Institution
Google

Position Title
Product Manager

Email
lhpeng@google.com

About this CDAS Project

Study
PLCO (Learn more about this study)

Project ID
PLCO-257

Initial CDAS Request Approval
May 16, 2017

Title
Deep Learning for Pathology Image Analysis

Summary
Deep learning is a family of machine learning techniques that have been applied successfully to a variety of image classification tasks (LeCun, Bengio, and Hinton 2015). It has also been applied to medical imaging with very promising results (Esteva et al. 2017; Gulshan et al. 2016; Wang et al. 2016), including in pathology images. In this project, we propose the application of deep learning to digitized images of tissues from the PLCO trial to build algorithms that (1) automatically classify the type and grade of cancer as well as regions of interest; (2) predict protein expression or mutation status directly from the H&E stained samples; and (3) directly predict outcomes from the images (e.g. survival, recurrence).

Aims

This project aims to use deep learning to
(1) automatically detect regions of interest and classify the type and grade of cancer from digitized histopathological images from PLCO tissue samples
(2) Predict protein expression or mutation status directly from the H&E stained samples, and
(3) Directly predict outcomes from images and clinical data (e.g. survival, recurrence).

Collaborators

Martin Stumpe, Google Research