EARLY DETECTION OF PANCREATIC CANCER USING BIOMARKER-DRIVEN MACHINE LEARNING ALGORITHMS

Abstract

pancreatic cancer is one of the more deadly cancers because of the lack of timely symptoms and delayed patient presentation. It is evident from existing methods that conventional screening methods, such as the biomarker CA19-9, cannot perform with the desired level of sensitivity and specificity in identifying cancer at an early stage. This manuscript will explore challenges associated with identifying cancer in its early stages based on a machine learning approach and discuss a biomarker-based machine learning approach for identifying cancer more accurately by integrating various clinical, metabolic, and inflammatory markers. It can be inferred from the experiment conducted on real-time patient data and evaluated based on various biomarkers and the machine learning approach implemented within the framework for making predictions on the identified features for cancer and its predictions during the early stages of cancer. The performance was found better for XGBoost with 90.1% accuracy, 91.4% sensitivity, 88.9% specificity, and 0.94 AUC, where the existing cancer biomarker and its additional biomarkers have shown high significance in improving the overall performance predictions. The proposed approach provides a comprehensive, non-invasive tool for understanding cancer more deeply for healthy individuals and those already affected with cancer.

Keywords