Experimental Evaluation of Pancreatic Cancer Identification based on CT Images by using Intelligent Deep Learning Procedure

• Published in: 2025 International Conference on Frontier Technologies and Solutions (ICFTS) pp. 1 - 9
• Main Authors: R, Ramachandran, Shathik, J. Anvar, Sivakumar, J., Manothini, P., Asha, G. S. Jackulin, Venkatanaresh, M.
• Format: Conference Proceeding
• Language: English
• Published: IEEE 27.03.2025
• Subjects: Accuracy; AUC-ROC; Biomedical imaging; Classification; CNN; Computed tomography; CT scan; Data augmentation; Data models; Deep learning; Explainability; Grad-CAM; Medical Imaging; Noise reduction; Pancreatic cancer; Performance evaluation; Robustness
• DOI: 10.1109/ICFTS62006.2025.11031801

Pancreatic cancer is one of the most aggressive malignancies, necessitating early and accurate detection for effective treatment. This study proposes an intelligent deep learning framework for pancreatic cancer identification using Computed Tomography (CT) images. The methodology involves advanced preprocessing techniques, including noise reduction, contrast enhancement, and segmentation, followed by a deep learning-based classification approach. The proposed model integrates a CNN backbone with an attention mechanism, enhancing its ability to focus on critical pancreatic regions. Additionally, data augmentation strategies such as rotation, flipping, and brightness adjustments improve the model's generalization ability. Performance evaluation using 10-fold cross-validation demonstrates that the proposed model achieves 95.8% accuracy, 94.5% precision, 96.7% recall, and an F1-score of 95.6%, outperforming baseline models, including ResNet-50 (90.3% accuracy) and EfficientNet-B0 (93.6% accuracy). Explainability techniques such as Grad-CAM visualization provide interpretability, highlighting the regions influencing predictions. Comparative analysis confirms the model's superior capability in distinguishing pancreatic cancer cases from healthy subjects. The impact of data augmentation is also assessed, showing significant improvements in model robustness. The high specificity (95%) and AUC-ROC (97.3%) further validate the effectiveness of the proposed framework in medical imaging applications. This research underscores the potential of deep learning in early pancreatic cancer detection, contributing to improved diagnostic accuracy. Future work will focus on multi-modal data integration, including clinical and histopathological information, to further enhance model reliability and deployment in real-world clinical settings.