Effect of pancreatic parenchymal fibrosis on long-term outcomes in pancreatic cancer patients measured using AI-guided MRI image.

Abstract

660 Background: Pancreatic parenchymal fibrosis has been reported as a risk factor for surgical complications and a prognostic indicator in pancreatic cancer patients. The significance of assessing pancreatic fibrosis is widely recognized. However, preoperative prediction of pancreatic fibrosis is difficult and has depended on intraoperative findings and postoperative histopathological examination. In this study, we analyzed MRI images of pancreatic cancer patients using an AI engine designed for surgical simulation and commonly used for pancreatic analysis. We evaluated the engine's ability to assess overall pancreatic fibrosis and its correlation with long-term prognosis. Methods: We analyzed 262 cases of resected pancreatic cancer for which semi-automatic extraction of the pancreatic parenchyma was possible using AI with a Dice coefficient of 0.8 or higher. We then examined the factors associated with pancreatic fibrosis using multivariate analysis. We defined pancreatic fibrosis using Kloppel's fibrosis score based on intraoperative findings or histopathological examination. We used preoperative MRI images and calculated the RSID value from the ADC value and fat-suppressed images (in-phase and out-of-phase images) based on phase difference. We performed a multivariate analysis to examine prognostic factors in patients with resected pancreatic cancer, including ADC and RSID values. Results: In 262 pancreatic cancer patients, the presence or absence of pancreatic fibrosis was associated with lesion location (head/body and tail: 122/35 vs 38/67), pancreatic thickness (12 vs 16 mm), pancreatic duct diameter (5 vs 2 mm), pancreatic volume (115 vs 165 mL), and ADC value (1.13±1.25 mm 2 /s vs 1.64±1.98 mm 2 /s), and relative signal intensity difference (RSID) value (16.9 ± 20.6 % vs. 10.5 ± 18.8 %). Multivariate analysis identified head lesions (odds ratio [OR], 4.125; p = 0.015) and high RSID (OR, 5.586; p = 0.001) as independent predictors of pancreatic fibrosis. Multivariate analysis identified elevated CA19-9 (hazard ratio [HR], 2.11), lymph node metastasis (HR, 5.25), elevated RSID (HR, 3.01), and absence of postoperative adjuvant therapy (HR, 4.23) as poor prognostic factors. Conclusions: AI-based MRI image analysis is useful for evaluating pancreatic fibrosis. This efficacy is useful for surgical planning and determining treatment strategies for drug therapy in pancreatic cancer patients.