Early Diagnosis of Pancreatic Cancer using Urinary Biomarkers and Machine Learning

Diagnosing pancreatic ductal adenocarcinoma (PDAC) and chronic pancreatitis (CP) early is crucial for better patient outcomes. However, the lack of specific biomarkers makes this challenging. This study investigated the potential of a panel of four urinary biomarkers (creatinine, LYVE1, REG1B, and TFF1) combined with machine learning (support vector machines and neural networks) for diagnosis. We analyzed a publicly available dataset of 590 urine samples categorized as healthy, benign pancreatic disease, or PDAC. All samples were collected before treatment and age and sex were balanced across groups. Both machine learning models achieved high accuracy (>89%) and sensitivity (>84%) for PDAC diagnosis. For CP diagnosis, neural networks performed slightly better (accuracy: 75.3%, sensitivity: 65.1%) compared to support vector machines (accuracy: 71.0%, sensitivity: 59.0%). These results suggest that both methods hold promise for diagnosing these diseases. The use of urinary biomarkers offers several advantages. Urine collection is non-invasive and patient-friendly, the biomarkers are stable and detectable in early stages of disease, and measurement is relatively inexpensive. Our study has limitations, including a small sample size and a retrospective design. Further research with larger, more diverse populations and prospective studies is needed. Additionally, the potential use of these biomarkers for risk stratification and disease progression monitoring requires investigation. In conclusion, this study demonstrates the potential of a combined approach using urinary biomarkers and machine learning algorithms for diagnosing PDAC and CP. While neural networks showed a slight edge for CP diagnosis, both methods performed well. Further research is needed to validate these findings and explore the broader potential of these biomarkers.