Integrating artificial intelligence and bioimpedance spectroscopy for enhanced pancreatic disease diagnosis.

The purpose of this study is to develop bioimpedance spectroscopy methods for diagnosing pancreatic diseases. We developed a descriptor approach using impedance spectroscopy results, generating four amplitude-phase-frequency responses from four quasi-orthogonal leads. They create the feature spaces required for our hybrid classifier for diagnosing pancreatic diseases, consisting of five macro levels, the first of which is based on probabilistic neural networks, and the remaining four are based on fuzzy inference. We also presented a device structure for generating informative feature spaces. The use of multi-frequency sensing in classifiers based on fully connected neural networks made it possible to develop a clinical decision support system for diagnosing pancreatic diseases. The results were confirmed in groups of patients aged from 25 to 80 years, male and female, at different stages of pancreatic disease. The training and testing samples were formed using various diagnostic methods, including medical history, physical examination, assessment of comorbidities, laboratory tests, ultrasound, laparoscopy, intraoperative revision, and computed tomography. An assessment of the diagnostic quality indicators of a hybrid neural network for three classes of diseases: "pancreatic cancer," "chronic pancreatitis" and "without pancreatic pathology" showed that their maximum value for differentiating these classes of diseases was 89%, the minimum -63%. Experimental software studies and clinical trials of our decision support system demonstrated diagnostic sensitivity and specificity comparable to existing methods, validating the practicality of our methods. This research opens new possibilities for accessible and objective diagnosis of pancreatic diseases, expanding the ability to make intelligent medical decisions.