Development and validation of a novel chronic pancreatitis pathological grade based on artificial intelligence

Effective chronic pancreatitis (CP) treatment requires accurate severity evaluation, but no histopathology grading system exists. This study aimed to develop and validate a novel CP pathological grade (Histopathology-derived CPpG) using quantified pathological and radiological characteristics through deep learning.

Patients with pathologically/clinically confirmed CP or recurrent acute pancreatitis were retrospectively enrolled (2011-2023). Whole-slide CP images were automatically segmented and quantified via DeeplabV3+, followed by latent class analysis to develop Histopathology-derived CPpG. A deep learning radiomics score (DLRS) was created to predict Histopathology-derived CPpG using preoperative CT scans of patients with pathologically confirmed CP. CT-predicted CPpG was then validated in an independent group of patients with clinically confirmed CP and recurrent acute pancreatitis.

The study included 2054 patients with CP and recurrent acute pancreatitis, with 181 cases of pathologically confirmed CP. Histopathology-derived CPpG I had a higher proportion of acini, acinus-to-stroma ratio, acinus-to-islet ratio, islet-to-stroma ratio, and (acinus + islet)-to-stroma ratio, and a lower proportion of stroma and lymphocytes compared to CPpG II. The DLRS demonstrated high performance in the validation (AUC, 0.84; 95 % CI: 0.75-0.92) and test (AUC, 0.76; 95 % CI: 0.65-0.87) sets. In a large-scale clinical validation, CT-predicted grades were significantly associated with endocrine and exocrine function, as well as prognosis (P < .05).

This study developed a novel pathological classification, Histopathology-derived CPpG, which accurately reflects disease severity. Additionally, the non-invasive DLRS shows great potential for dynamically monitoring CP severity and evaluating pancreatic endocrine and exocrine function, as well as prognosis.