Evaluate performance of the PreciseDx AI-digital LDT Breast assay to predict breast cancer recurrence using clinical data and H&E images from study patients

Principal Investigator

Name
Michael Donovan

Degrees
Ph.D., M.D.

Institution
University of Miami

Position Title
Vice Chair, Professor

Email
mjdonovan54@gmail.com

About this CDAS Project

Study
PLCO (Learn more about this study)

Project ID
PLCOI-741

Initial CDAS Request Approval
Mar 9, 2021

Title
Evaluate performance of the PreciseDx AI-digital LDT Breast assay to predict breast cancer recurrence using clinical data and H&E images from study patients

Summary
Breast cancer patients have benefited from assorted genomic risk assessment tools such as OncotypeDx, Mammaprint and EndoPredict (2-6) to predict recurrence; however, all patients prior to any DNA / RNA mutation strategy are required to have an accurate grade. (NCCN, 2020; Clinical Practice Guidelines in Oncology, Breast) Although most groups applying artificial intelligence in digital pathology have focused on pathologist workflow and efficiency, we set out to enhance the overall science of cancer (prognostic) phenotyping at diagnosis through the development of AI-digital-laboratory developed tests (AI-D-LDT’s). The objective was to develop an automated cancer grading platform integrated with machine learning to provide accurate risk classification. Unlike all other approaches, we only require the standard Hematoxylin and Eosin stained image of the tumor as the source material along with clinical features / outcome data to generate a risk score and probability for disease progression. This is most important for breast cancer (projected incidence of 284,200 new cases in 2021) (8) but also relevant for the majority of solid malignancies diagnosed today including prostate, lung, colon, liver, head and neck, neuroendocrine, and skin. The defined process as outlined is readily accessible to all healthcare systems, and meets the unmet needs associated with our current demand for telemedicine in a pandemic and the emerging worldwide shortage of pathologists.

Aims

Primary endpoint: AUC PreciseDx Breast > AUC Clinical feature only model and > AUC imaging features alone for predicting breast cancer recurrence.

Secondary endpoint(s): Absolute risk in the high and low risk groups as identified by PPV and NPV, respectively is > PPV and NPV in the clinical feature only model.

Collaborators

members of the PreciseDx team which consists of image analysis engineers, statisticians and clinical personnel.