Applying Artificial Intelligence Technologies to Detect Pancreatic Ductal Adenocarcinoma Using Routine Laboratory...

Objective This study aimed to investigate whether artificial intelligence could identify pancreatic ductal adenocarcinoma (PDAC) in patients aged <70 years using only routine laboratory tests (RLTs) applicable to regular health checkups. Methods The RLTs comprised 27 items, including blood and urine tests and the body mass index. Data were analyzed using a self-organizing map (SOM) and a Bayesian regularized neural network (BRNN). Two independent studies were conducted: Study-I compared patients with PDAC (n=46) with healthy individuals (n=72), and Study-II compared patients with PDAC (n=35) with outpatients (n=89) who were negative for PDAC via imaging. Results In Study-I, the SOM clearly separated the PDAC cluster from the healthy cluster. The BRNN analysis achieved a high diagnostic performance: for males, the sensitivity was 96.2% and specificity was 100%, for females, the sensitivity was 100% and specificity was 96.6%. In Study-II, when the study subjects were projected onto the SOM map created in Study-I, the sensitivity and specificity were 90.5% and 65.9%, respectively, for males, and 85.7% and 75.0% for females. In the BRNN analysis for Study-II, the diagnostic performance showed a sensitivity of 81.0% and specificity of 84.6% for males and 73.6% and 95.6% for females. Conclusion Although this was a small-scale pilot study, both SOM and BRNN demonstrated the potential to identify PDAC patients aged <70 years using RLTs.