Gastric Cancer Detection in Endoscopic Images using Deep Learning

Gastric cancer remains a leading cause of cancer-related mortality worldwide, with early and accurate diagnosis being crucial for improving patient survival. This study presents an automated deep learning framework for gastric disease classification using endoscopic images. The system employs a dual-backbone hybrid model combining EfficientNet-B0 and ResNet-50, leveraging complementary spatial and hierarchical feature extraction to achieve precise multi-class classification of eight gastric conditions, including polyps, ulcerative colitis, esophagitis, and normal gastric regions. Input images are pre-processed, normalized, and converted into PyTorch tensors, ensuring consistent and noise-resilient data representation. To enhance interpretability and clinical trust, the framework integrates a Grad-CAM module, generating visual heatmaps that highlight discriminative regions of pathology within the gastric mucosa. Extensive experiments demonstrate the effectiveness of the hybrid model, achieving 94.2% accuracy, 93.8% precision, 92.9% recall, and 93.3% F1-score for multi-class classification, and 96.1% accuracy for binary classification (normal vs. abnormal). Comparative analysis shows that the hybrid approach outperforms single-backbone models by improving convergence, reducing misclassification, and capturing subtle pathological features. The proposed framework provides good predictive accuracy, interpretability through Grad- CAM, and practical applicability for computer-aided gastric disease diagnosis, making it a reliable tool for supporting clinical decision-making in endoscopic image analysis.