Artificial Intelligence in Healthcare: Early Pancreatic Cancer Detection Using Urinary Biomarkers

Abstract

Pancreatic cancer is one of the deadliest malignancies due to its late-stage diagnosis and lack of effective early detection tools. Existing detection and screening methods currently fail to identify the tumor at its early, more treatable stages, contributing to persistently low survival rates and necessitating alternative approaches. However, in recent times, machine learning (ML), which is a branch of artificial intelligence (AI), has shown immense promise in the field, potentially enhancing early cancer detection by identifying minute and subtle patterns in clinical data. This study explores the application of machine learning and deep learning in the prediction of pancreatic cancer, using notably as input a set of patient urinary and blood biomarkers identified in previous studies as potentially promising for early detection of pancreatic cancer. The goal, after all, of this study is to predict the presence of the disease before it is diagnosed. Four classification models (Neural Network, Decision Tree, Random Forest, and K-Nearest Neighbors) were implemented to analyze the data features, classifying individuals as healthy, having benign hepatobiliary disease, or having pancreatic cancer. To further improve prediction reliability, a Multiplicative Weight Update (MWU) method was applied to dynamically adjust the influence of each model based on their testing performance, finally forming an overall more robust and accurate program. The integration of four distinct classification models, in tandem with the MWU method, distinguishes this research from previous studies and enhances its predictive performance. Given the varying concentrations of biomarkers associated with different pancreatic conditions, the use of multiple diverse models to capture both linear and complex non-linear patterns in the biomarker data was particularly important, something prior studies relying on individual models rarely achieved. As a result, the final prediction accuracy was significantly improved. The results demonstrate high accuracies for most models, with the Decision Tree achieving the highest predictive accuracy of 98.7%. These results highlight the potential of AI-driven diagnostic tools in improving early pancreatic cancer detection.