Risk Stratification of Pancreatic Ductal Adenocarcinoma in New-Onset Diabetes Using Artificial Intelligence Analysis of Retinal Images
Abstract
This project is about automated risk stratification of pancreatic ductal adenocarcinoma (PDAC), also called pancreatic cancer, in diabetic patients which will enhance early detection of PDAC. The project falls under FY21 PCARP Focus Area Early detection research for pancreatic cancer, including studies of individuals with pre-diabetes and diabetes and/or those in underserved ethnic and minority communities. Early detection of the PDAC is complicated as PDAC remains asymptomatic in the initial stages. Resultantly, over 80% of PDAC diagnosis occurs at a late stage when treatment is highly ineffective. Risk stratification of PDAC can improve early detection as follow-up screening of high-risk individuals on regular basis can efficiently assist in detecting cancer even before the symptoms appear or become obvious. Subjects diagnosed with new-onset diabetes (NOD) at age = 50 are usually at higher risk of developing PDAC within ~36 months of their NOD diagnosis compared to the general population, giving a possible opportunity to predict PDAC. Studies also show that many complications in the retina of NOD patients precede PDAC. These complications include changes in the retinal vasculature due to its strong biological connection with pancreatic disorders. Quantifying these changes can efficiently assist in identifying a manageable subpopulation of NOD patients, showing a high risk of PDAC, to be screened and subsequently allow early detection of pancreatic cancer. The primary hypothesis for the study is that retinal images of patients (age = 50), diagnosed with NOD no more than three years ago, present indicative features of existing or future incidence of PDAC. Such features can be discovered and utilized for risk prediction of PDAC using Artificial Intelligence (AI). The objective of the project is to discover, interpret, and quantify such predictive features using AI analysis of retinal images of the NOD population and subsequently develop an improved risk stratification model for PDAC by integrating retina-based indicators with conventional risk factors of PDAC. This study includes developing an AI risk prediction model that will perform extensive analysis of retinal images of NOD patients (a) who developed PDAC, and (b) who did not develop PDAC, within 3 years of their NOD diagnosis. This analysis will help to identify retinal image features that are significantly different in the two groups and are potentially indicative of existing or future incidence of PDAC. The model will then utilize such features and perform risk quantification of PDAC and stratify individuals into high-risk and low risk for PDAC categories by classifying their retinal images. Two institutes including the Cedars-Sinai Medical Center and the Veterans Affairs Greater Los Angeles Healthcare System (VAGLAHS) will collaborate and perform this study. The study will use two sets of retinal images from two groups (NOD, NOD+PDAC), each containing 258 images, for analysis and prediction modeling. A team of AI imaging scientists, gastrointestinal experts, and trained ophthalmologists from both centers will contribute to this study. Extensive AI analysis of retinal images from two groups will be performed, and the proposed model will be rigorously trained/validated to generate reproducible results. The identification of new-onset diabetic individuals with a high risk to develop PDAC will allow regular targeted screening which will help to detect PDAC at an early stage. Identifying these individuals will give the opportunity to perform anti-cancer therapies to decelerate or even prevent the development of PDAC. This will also reduce the associated treatment cost, make the disease more manageable, and improve the quality of life of diabetic patients. In addition, this unique model will offer a completely non-invasive and cost-efficient mechanism to predict PDAC without requiring diabetic patients to undergo any additional or intensive
Document Details
- Document Type
- DoD Grant Award
- Publication Date
- Dec 28, 2022
- Source ID
- W81XWH2210737
Entities
People
- Touseef Ahmad Qureshi
Organizations
- Cedars-Sinai Medical Center
- United States Army