Evaluating Deep Learning Architectures for Early Pancreatic Cancer Detection in CT Imaging
Oscar Rodriguez, Aries Socrates, Robert Ace Gonzales
Department of Biology
Faculty Supervisor: Sara El Alaoui
Pancreatic cancer is among the deadliest cancers, largely due to its aggressive growth and late detection. Once detected on scans, it is often too late, with most dying within a couple of years. This project explores the application of machine learning to improve early detection of pancreatic cancer through CT imaging analysis to identify early signs of tumors to increase survival rate. To strengthen our model’s robustness, we combine the CPTAC-PDA dataset of pancreatic cancer patients with the PANCREAS-CT dataset of healthy individuals, allowing us to account for different class imbalances. Our focus is on experimenting with various architectures, most notably on U-Net for segmentation and VGG networks for classification. By refining preprocessing, enhancing feature selection, and systematically evaluating models, we aim to develop a system that detects early indicators of pancreatic cancer and supports earlier diagnosis for better clinical outcomes.