OC.03.10 AUTOMATED CARDIOVASCULAR RISK PREDICTION IN MASLD PATIENTS USING AI, BY CLINICAL AND CT IMAGING

**Background and Aim:** Accurate assessment of cardiovascular risk is essential for individuals who appear healthy. Metabolic-dysfunction associated steatotic liver disease (MASLD), formerly known as NAFLD, affects 38% of the global population. In patients with MASLD, cardiovascular diseases (CVD) are the leading cause of mortality, suggesting that MASLD should be considered an independent risk factor in developing new cardiovascular risk assessment scores. This study aims to develop a predictive model for cardiovascular risk assessment in MASLD patients by applying artificial intelligence (AI) techniques to non-contrast Cardio-CT scans and clinical data. **Material and Methods:** A retrospective study analyzed clinical and imaging data from 174 patients who underwent Cardio-CT scans at two different Hospitals. Based on Coronary Artery Calcium (CAC) scores and Hounsfield Units (HU), 50% of the patients were affected by metabolic-dysfunction associated steatotic liver disease (MASLD) and cardiovascular disease (CVD), while the remaining patients served as healthy controls. Ninety-six patients were used for training the AI model, and 78 patients formed the internal validation cohort to obtain performance metrics. A U-Net convolutional neural network was employed to segment the liver parenchyma, and a Gray-Level Co-occurrence Matrix (GLCM) was applied to extract radiomic features to evaluate levels of steatosis. These relevant features were combined with clinical data and used as input for a multilayer perceptron neural network to perform binary classification of CAC. **Results:** Our trained algorithm automatically determines the severity of CAC based on liver steatosis and patient clinical data, thereby assessing the level of cardiovascular risk (CVR). Notably, the most significant features influencing the model were dyslipidemia, diabetes, and age. When testing images and clinical data from both centers, the most effective model was the Stacking Model, which achieved an Area Under the Curve (AUC) of 80%. **Conclusions:** Our AI model estimates cardiovascular risk in MASLD patients undergoing abdominal CT scans by integrating radiomic and clinical data. This approach could also be useful for cirrhotic patients undergoing hepatocellular carcinoma (HCC) screening or awaiting orthotopic liver transplantation (OLT).