AN AI-DRIVEN MULTI-DISEASE FRAMEWORK FOR EARLY DETECTION OF PANCREATIC CANCER AND NEURODEGENERATIVE DISORDERS USING SYNTHETIC EHR DATA

Abstract

Pancreatic Cancer and Neurodegenerative Disease Diagnosis Using Artificial Intelligence Approach. Early diagnosis of pancreatic cancer and neurodegenerative disease is a major challenge in contemporary medicine because of their complexity and asymptomatic nature during the initial stages of development. PCN_Early was designed as an artificial intelligence framework based on synthetic electronic health records that could help detect patients prone to developing these conditions before the onset of any clinical symptoms. The system employs the XGBoost classifier (Extreme Gradient Boosting) that evaluates the following factors: age, body mass index, HbA1c, AST, ALT, LDL, APOE-e4 gene, memory score, sleep hours, vitamin B12, depression score, and family medical history. Besides, five features were added for determining the risk patterns associated with diseases. An accuracy of 81% was obtained, while the ROC-AUC value was 0.8936. FastAPI was used for deploying the backend, which was connected to the Streamlit user interface. This study confirmed the great promise of using XGBoost AI models on synthetic electronic health records to enhance the early diagnosis process and provide preventive healthcare services. Building on this predictive capability, the system also includes a conversational assistant called PCN-AI, which allows clinicians to ask questions in plain language and receive answers drawn from a dedicated medical knowledge base covering pancreatic cancer, neurodegenerative disorders, and related clinical topics. This assistant runs on the Groq Llama3-8B model and uses a retrieval-augmented approach to ensure its responses stay grounded in domain-relevant information rather than general knowledge