Computed tomography-based fully automated artificial intelligence model to predict extrapancreatic perineural invasion in pancreatic ductal adenocarcinoma

作者：Jieyu Yu, Chengwei Chen, Mingzhi Lu, Xu Fang, Jing Li, Mengmeng Zhu, Na Li, Xiaohan Yuan, Yaxing Han, Li Wang, Jianping Lu, Chengwei Shao, Yun Bian · 发表于：International Journal of Surgery · 年份：2024 · DOI：10.1097/js9.0000000000001604 · 被引用次数：12 · 研究领域：Pancreatic and Hepatic Oncology Research、Cholangiocarcinoma and Gallbladder Cancer Studies、Lung Cancer Diagnosis and Treatment

BACKGROUND: Extrapancreatic perineural invasion (EPNI) increases the risk of postoperative recurrence in pancreatic ductal adenocarcinoma (PDAC). This study aimed to develop and validate a computed tomography (CT)-based, fully automated preoperative artificial intelligence (AI) model to predict EPNI in patients with PDAC. METHODS: The authors retrospectively enrolled 1065 patients from two Shanghai hospitals between June 2014 and April 2023. Patients were split into training (n=497), internal validation (n=212), internal test (n=180), and external test (n=176) sets. The AI model used perivascular space and tumor contact for EPNI detection. The authors evaluated the AI model's performance based on its discrimination. Kaplan-Meier curves, log-rank tests, and Cox regression were used for survival analysis. RESULTS: The AI model demonstrated superior diagnostic performance for EPNI with 1-pixel expansion. The area under the curve in the training, validation, internal test, and external test sets were 0.87, 0.88, 0.82, and 0.83, respectively. The log-rank test revealed a significantly longer survival in the AI-predicted EPNI-negative group than the AI-predicted EPNI-positive group in the training, validation, and internal test sets (P<0.05). Moreover, the AI model exhibited exceptional prognostic stratification in early PDAC and improved assessment of neoadjuvant therapy's effectiveness. CONCLUSION: The AI model presents a robust modality for EPNI diagnosis, risk stratification, a...