Abstract
<title>Abstract</title> <p> <bold>Objective</bold> To develop a fully automated artificial intelligence (AI) model for analyzing body composition at the third lumbar vertebra (L3) level and to evaluate its clinical utility in predicting short-term outcomes and long-term prognosis in patients undergoing pancreatectomy. <bold>Methods</bold> Patients diagnosed with pancreatic or periampullary diseases between July 2017 and July 2024 were retrospectively enrolled. An end-to-end AI pipeline was constructed to automatically localize the L3 vertebral level and segment skeletal muscle (SM), subcutaneous adipose tissue (SAT), and visceral adipose tissue (VAT). Body composition parameters, including sarcopenia, myosteatosis, and visceral obesity, were derived from segmentation results. The associations between these parameters and short-term postoperative complications, as well as long-term survival in pancreatic cancer patients, were systematically analyzed. <bold>Results</bold> The AI model demonstrated exceptional segmentation performance, achieving mean Dice similarity coefficients of 0.9715, 0.9538, and 0.9392 for SM, SAT, and VAT, respectively. Clinically, visceral obesity was significantly associated with prolonged operative duration, increased blood loss, and higher rates of clinically relevant postoperative pancreatic fistula (CR-POPF). Both sarcopenia and myosteatosis were identified as robust predictors of major complications and lower rates of textbook outcomes. In terms of long-term prognosis, multivariate analysis confirmed that sarcopenia and myosteatosis were independent predictors for both recurrence-free survival (RFS) and overall survival (OS). <bold>Conclusion</bold> The fully automated AI model provides an efficient and reproducible tool for body composition analysis. Sarcopenia and myosteatosis serve as critical indicators for surgical risk stratification and long-term prognostication, offering valuable insights for personalized clinical decision-making in pancreatic surgery. </p>