Unsupervised Machine Learning of Intrapancreatic Fat Deposition Reveals Distinct Clinical Phenotypes: A CT Study.
Authors
Affiliations (1)
Affiliations (1)
- Department of Gastroenterology, The Affiliated Hospital of Qingdao University, Qingdao, Shandong, People's Republic of China.
Abstract
Intrapancreatic fat deposition (IPFD) is an ectopic fat phenotype associated with metabolic abnormalities. However, the clinical heterogeneity of IPFD in relation to generalized adiposity remains insufficiently characterized. This study aimed to explore clinical profiles associated with IPFD using unsupervised clustering and characterize their metabolic features. In this retrospective, single-center cross-sectional study, 558 adults undergoing non-contrast abdominal computed tomography were included. K-means clustering was performed using age, body mass index (BMI), pancreas-to-spleen attenuation difference, fasting plasma glucose (FPG), and triglyceride (TG) levels. Clinical characteristics, metabolic comorbidities, and circulating biomarkers were compared across the derived exploratory clusters. Three exploratory clinical phenotypes were identified: an obesity-dominant cluster (n=140), an FPD-like cluster characterized by greater IPFD (n=182), and a reference cluster (n=236). The FPD-like cluster showed the lowest pancreas-to-spleen attenuation difference, older age, and lower BMI than the obesity-dominant cluster. In contrast, the obesity-dominant cluster showed stronger associations with metabolic dysfunction-associated steatotic liver disease and higher uric acid levels. Free fatty acid concentrations were highest in the FPD-like cluster. Differences in glycemic and lipid-related characteristics should be interpreted cautiously because FPG and TG were included as clustering variables. In this exploratory cross-sectional analysis, a cluster characterized by greater IPFD, older age, and lower BMI than the obesity-dominant cluster was associated with less favorable metabolic characteristics. These findings suggest that IPFD may provide complementary information beyond BMI-based assessment; however, the identified clusters should be considered exploratory clinical profiles rather than validated biological subtypes. External validation in independent populations and longitudinal studies are required to confirm their reproducibility and clinical relevance.