Introduction
A previous study, which found that focal parenchymal atrophy of the pancreas is an indirect sign of small pancreatic cancer (PC), documented the changes in the pancreas until cancer onset.［1-3］ However, the pancreas atrophies and changes with age, so issues remain regarding the criteria for determining pancreatic atrophy as a sign of PC. The application of artificial intelligence (AI) technology to quantify changes in the pancreas over time and predict the onset of PC and high-risk groups for PC from objective indicators has the potential to facilitate early detection of PC.
Aims & Methods
Based on the rate of change in the pancreatic volume on computed tomography (CT) using AI technology, the present study predicted PC onset. Forty patients in whom PC was histopathologically diagnosed at Komagome Hospital between 2015 and 2023 and who had a history of three or more follow-up CT examinations over five years before PC onset (the PC group) were analyzed. Forty patients who underwent CT periodically for malignant skin disease but had no pancreatic disease were enrolled into a non-PC group. Patients with pancreatic carcinoma in situ (pancreatic intraepithelial neoplasm), an intraductal papillary mucinous neoplasm or chronic pancreatitis with multiple calcifications were excluded. Changes in the pancreatic volume in the groups over time were then compared using CT images taken one year or earlier before PC onset. The pancreatic volume was calculated using SYNAPSE VINCENT cloud (Fuji Film, Japan), an AI image application program. Age and pancreas volume were combined to create a scatter diagram, and the rate of pancreatic volume change (the temporal pancreatic volume change coefficient) was calculated from the regression line and compared.
Results
The lesion site in PC group included the head (n=17), body (n=7), and tail (n=16). The median maximum tumor diameter was 31 mm (interquartile range: 27-37). A comparison of the two groups found a cStage classification of IB, IIA, IIB, III, and IV in eight, ten, three, eight, and 11 patients, respectively. The median age was 72.5 and 72.5 years in the respective group (P=0.182). The male: female ratio was 23:17 and 24:16 (P=0.820), the median observation period was 83.5 and 107 months (P<0.05); the mean number of examinations was 10.7 and 18; the total number of CT examinations was 430 and 745; the mean total pancreatic volume was 74.4 and 78.15 ml (P<0.01 ); and the mean pancreatic volume change coefficient was -0.141 and -0.036 (P=0.01), respectively. The rate of decrease in the pancreatic volume was faster in the PC group with increasing age, the average being 8.46 ml over five years in the PC group compared to 2.16 ml in the non-PC group. The cutoff value for the coefficient of pancreatic volume change over time was set at -0.125. The sensitivity was 50%, the specificity was 90%, the positive predictive value was 83%, and the AUC value was 0.67.
Conclusion
In comparison with the age-related changes in the pancreas, the rate of decrease in the pancreatic volume was faster in the PC group, suggesting that PC onset may be predicted using images obtained one year or earlier before PC onset. These findings further suggested that identifying patients with a high risk of PC may be possible by creating a pancreatic volume histogram using AI technology based on age at the time of CT, intervals between past examinations, and the rate of decrease in the pancreatic volume.
References
［１］Nakahodo J, Kikuyama M, Nojiri S, et al. Pancreatology. 2020 Dec;20(8):1689-1697.
［２］Nakahodo J, Kikuyama M, Fukumura Y, et al. Pancreatology. 2022 Dec;22(8):1148-1158
［３］Kikuyama M, Nakahodo J, Honda G, et al. Pancreatology. 2023 Jun;23(4):420-428.
