Deep learning to estimate biological age from clinical routine abdominal MRI and investigate its association with major adverse cardiovascular events and cancer death

• Applicant: Dr. Matthias Jung
• Subject Area: Radiology
• Term: since 2023
• Project identifier: Deutsche Forschungsgemeinschaft (DFG) - Project number 518480401

Objective: This study assesses whether a deep learning framework that estimates biological age based on 3D body composition (BC) segmentation masks derived from MR imaging (BC-age framework) can predict future major adverse cardiovascular events (MACE) and mortality beyond chronological age in a large patient cohort. Background: Chronological age is a cornerstone of medical decision-making but an imperfect measure of aging. Measures of biological age, which reflect an individual's overall health status, could provide a more nuanced understanding of aging, and potentially improve personalized medical decision-making. In this project, we apply our BC-age framework to a large clinical patient cohort that underwent abdominal MRI for cancer staging or workup of an incidental finding (e.g. kidney cyst) and investigate its value in predicting MACE (including myocardial infarction, stroke, and heart failure) and mortality in a clinical setting. Preliminary work: We developed the BC-age framework in two steps using data from 30,389 asymptomatic individuals (20-75 years; 44.2% female) from the German National Cohort (NAKO): First, a model was trained to segment 3D BC compartments as subcutaneous (SAT), visceral (VAT), intramuscular adipose tissue (IMAT) and skeletal muscle (SM) from MRI. Then, a second model was trained that takes the 3D BC segmentation masks as an input and outputs an age estimate in years. For downstream analyses, we calculated MRI-Age acceleration, defined as an age-specific z-score of the output of the model in Step 2 (positive MRI-Age acceleration is biologically older, negative is younger). We validated this framework in an independent testing set of 36,317 individuals from the general population using the UK Biobank (65.1±7.8 years, 51.7% female; 2.7% [969/36,317] MACE; median follow-up 4.8 years): cumulative incidence curves showed a higher incidence of MACE in individuals with positive vs. negative MRI-Age acceleration (Log-rank p<0.0001). Cox regression revealed an independent association between MRI-Age acceleration and MACE (aHR:1.09, 95% CI [1.01-1.18], p=0.02) after adjustment for chronological age, sex, BMI, race, body composition volumes, diabetes, hypertension, cancer, alcohol consumption, and smoking status. Methods: Our BC-age framework will be validated for clinical translation using more than 30,000 routine abdominal MRIs of Mass General Brigham patients (20-90 years; 54.9% female) that will be subdivided into two different datasets: 1) cancer patients and 2) patients with an incidental finding. The BC-age framework’s biological age estimate will be tested to predict MACE (both cohorts), all-cause, and cancer mortality (cancer patients) beyond chronological age and traditional clinical risk factors. Relevance: If successful, we will have validated a new way to opportunistically measure biological age using clinical routine MRI to help assess risk and guide patient care.

• DFG Programme: WBP Fellowship
• International Connection: USA
• Host: Professor Dr. Michael Lu