CERSI Collaborators: Kim RM Blenman, PhD, MS (PI); Joseph Ross, MD, MHS (CERSI PI); Jessica Ritchie, MPH (PM)
FDA Collaborators: Brandon D. Gallas, PhD; Nicholas Petrick, PhD; Wen Si, PhD; Katherine Elfer, PhD, MPH
Project Start: October 14, 2022
Regulatory Science Challenge
Many U.S. patients are seen in low resource community care centers and practices located close to their homes. There have been significant advances in technology, such as the creation of digital imaging microscopes and artificial intelligence software, to improve the health of patients. But these expensive platforms are not offered as possible options for most low resource environments. Not having access to these tools leads to further health inequities.
This project aims to improve general knowledge around these new tools by expanding data with real-world evidence from community healthcare providers. The data collected will be used to develop affordable tools and best practices for low resource environments.
This project will focus on the common practice of using a panel of experts to score the density of immune cells in digital pathology images of hematoxylin and eosin (H&E) slides that will be used as a reference standard for software developers who will be creating tools for digital images for use in clinical care. Including community healthcare providers in this work will reveal the challenges and limitations arising in low resource environments. In this project we will summarize best practices and create user-friendly tools to expand understanding of how artificial intelligence models are developed and can be used in clinical practice.
Project Description and Goals
This project will focus on developing resources for studies that are used to assess the performance of artificial intelligence software tools. This includes designing hardware that can be controlled virtually through the web, software, and statistical methods that can be used in all environments, especially low resource. This study will aim to determine the validity of artificial intelligence software tools.
Researchers will create a virtual bridge between existing software on the precisionFDA website for immune cell density scoring and a physical microscope for remote control of the microscope. Accompanying software tools for controlling a local microscope by a local computer and connecting the local computer to special web-based digital pathology software, will be created. The online software will be modified to accept the virtual bridge and to control the microscope from the precisionFDA website. These tools are currently being used independently of each other for viewing and collecting data from pathologists. After researchers integrate these tools together as one unit, they will develop worklists that specify the tasks that need to be completed for each slide. Following this, task-specific workflows with the previous worklists will be created.
A panel of experts will also be used to score immune cell density in these images. This data will be used to train pathologists to the task so they may participate in studies to assess performance of related artificial intelligence software. These evaluations will help in the development of tools that can be used to assess performance of a diverse population of pathologists using artificial intelligence software.