Stage-Specific Prognostic Modeling for Pancreatic Cancer Using Convolutional Neural Networks
Jagendra Singh, Malleswari Akurati, Harishchander Anandaram, Chinnala Balakrishna, Ramy Riad Al–Fatlawy, Sakil Ahmad Ansari · 2024
This study focuses on the stage-specific prognosis of pancreatic cancer using advanced Convolutional Neural Networks (CNN). The study aims to evaluate three different CNN architectures (VGG 16, VGG 19 and ResNet 50) in identifying four stages for each of the advanced 3400 medical images. The dataset has been carefully preprocessed by resizing, normalization and data augmentation. This ensures high-quality input for model training and testing. Models are then tested according to the accuracy, precision, recall and F1 score of CNN models. Results show that ResNet 50 has the highest accuracy, with an accuracy sizeable at 97.88%. Precision, recall, F1 score values are all high. VGG 19 is next with an accuracy of 94.50%, followed by VGG 16. VGG 16 is effective but this minor false predictions leads to an accuracy of only 91.20%. Confusion matrices for every model further showcase their predictive power, with ResNet 50 clocking in at the lowest misclassification across stages. These results highlight ResNet 50's strong potential for accurate staging of pancreatic cancer; it appears to be a boon in precision diagnostics. If these models can be successfully applied in medical imaging, early detection and treatment planning for pancreatic cancer patients will be greatly improved. Practically every form of cancer can be said to have been ruled out or at least controlled to some extent so that survival rates increase with time: this represents progress towards personalized medicine.