A Dual-Branch Deep Learning Framework Combining Xception and ResNet for Accurate Lung and Colon Cancer Detection

Lung and colon cancers are among the leading causes of cancer-related deaths worldwide. Early detection significantly enhances survival rates, but traditional diagnostic methods, which rely on manual analysis of histopathological images, are labor-intensive, error-prone, and inconsistent. While deep learning has shown promise in automating medical image analysis, existing models often struggle with issues like overfitting, poor generalization, and class imbalance, which limit their clinical effectiveness. This research proposes a hybrid deep learning model combining the strengths of Xception and ResNet architectures. The Xception model excels at feature extraction, while ResNet’s residual connections improve training stability and address issues like gradient vanishing. The model is trained using a large dataset of lung and colon cancer images, with a dynamic learning rate and backpropagation via Stochastic Gradient Descent (SGD) momentum, optimizing performance. Advanced data augmentation techniques, such as rotations and flips, further enhance model generalization. The proposed hybrid model achieves an impressive accuracy of 98.96%, demonstrating high effectiveness in distinguishing between different cancer classes. The model outperforms traditional diagnostic methods, offering a robust and reliable tool for automated cancer detection with significant clinical potential.