AHA Scientific Sessions 2026 Abstract · Nov 2, 2026
Machine Learning with MRI Adiposity Profiling Predicts Cardiac Structural and Hemodynamic Reverse Remodeling Post-Bariatric Surgery
By CHANG YL, HSUAN CF, Lin J
Background: Bariatric surgery induces profound weight loss and cardiovascular risk reduction, including the reversal of obesity-related cardiac remodeling. However, predicting the exact magnitude of individual cardiovascular recovery remains challenging. This study aimed to evaluate the structural and hemodynamic changes 12 months post-bariatric surgery and utilize a machine learning approach to identify baseline predictors for weight loss, blood pressure (BP) reduction, and left ventricular (LV) mass regression. Methods: We prospectively enrolled 40 severe obesity patients (age 40 ± 10 years, 22 males, baseline BMI 42.5 ± 7.1 kg/m ² ) scheduled for bariatric surgery. Patients underwent comprehensive clinical, laboratory, transthoracic echocardiography, and abdominal MRI (for liver fat fraction [LFF], visceral, and subcutaneous adipose tissue [SAT]) at baseline and 12 months postoperatively. To predict 12-month changes in body weight (ΔBW), mean BP (ΔMeanBP), and LV mass index (ΔLVMI), we employed an elastic net regression algorithm with k-fold cross-validation to rigorously filter highly correlated baseline variables. Results: At 12 months, patients demonstrated massive reductions in weight (–43.0 ± 15.0 kg), mean BP (–23 ± 16 mmHg), and LVMI (–11.8 ± 6.4 g/m 2.7 ) (all p < 0.001). The prevalence of normal LV geometry dramatically improved from 32.5% to 70.0%. The elastic net algorithm successfully modeled postoperative outcomes. Prediction of ΔBW (R-squared = 0.77) was overwhelmingly driven by baseline weight severity, with MRI adiposity adding no significant predictive value. Conversely, MRI fat profiling was critical for predicting cardiovascular recovery. ΔMeanBP (R-squared = 0.66) was independently predicted by baseline BP and baseline SAT, where higher baseline SAT drove larger BP drops. ΔLVMI (R-squared = 0.62) was primarily driven by baseline LVMI and baseline LFF, with higher liver fat predicting greater cardiac mass regression. Conclusions: Bariatric surgery yields substantial improvements in LV geometry and hemodynamics at one year. Machine learning modeling reveals that while standard anthropometrics are sufficient to predict total weight loss, direct MRI fat profiling provides significant, independent predictive value for estimating cardiovascular reverse remodeling. This individualized approach offers a clinical framework to guide patient expectations and tailor specific goals for both weight reduction and cardiovascular improvement.