Images and Algorithms Task Force
In July of 2018, PDS launched an Images & Algorithms taskforce, inviting experts from FDA, academia, and industry. This taskforce is co-chaired by Larry Schwartz, MD, a Chief Radiologist from Columbia University and George Demetri, MD, a Senior Oncologist from Dana-Farber, and managed by Asba (AT) Tasneem, PhD from PDS.
On November 11, 2019, this taskforce convened an FDA-PDS Symposium VIII focused on ‘Artificial Intelligence in Tumor Imaging’. This symposium was co-hosted by Stanford University School of Medicine and attended by more than 125 experts. The experts emphasized the urgency to take immediate, practical steps to move the field forward with an initial focus in data sharing and developing machine-learning algorithms using imaging data from clinical trials. Among the distinguished speakers were: Lloyd Minor (Stanford), Robert Califf (Duke/Verily), Larry Schwartz (Columbia), George Demetri (Dana-Farber), and Sean Khozin (now at Janssen Pharmaceuticals).
Interested in learning more about Images and Algorithms program? Please reach out to the program director.
Task Force Value Statement
Members of the PDS Images and Algorithms Task Force represent a joint effort of radiological, oncological, machine-learning, regulatory, and statistical experts from academia, industry, and government to develop a series of AI algorithms that can be used for image recognition and categorial measurement and classification of tumor dynamics in response to anticancer therapy. The primary objectives of this effort are:
- Development of reliable algorithms to improve the efficiency and accuracy of independent human radiologic review assessments of tumor dynamics according to standard categorical RECIST in registrational clinical trials; and
- Development of a tumor burden index based on holistic algorithmic assessment and classification of imaging data.
This effort will yield the first curated and aggregated groups of imaging data sets which will be shared via Project Data Sphere’s open-access platform.
A machine-learning algorithm for a specific tumor type will initially be developed through academic collaborations; the objective is to then further develop and adapt this algorithm for additional tumor types and additional tumor metrics as they evolve.
The hypothesis is that these data sets and the development of a related deep-learning algorithm will reduce the time and cost and improve the performance for imaging in clinical trials, shortening the time from discovery to implementation, improving the accuracy of the reviews, and, ultimately, improving patient lives.