Enhanced Diagnostic Precision for Lung Cancer: A Hybrid Deep Learning Model Combining Xception and EfficientNetB3

Accurate detection of lung and colon cancer is critical for improving patient outcomes. This study proposes a hybrid deep learning framework combining EfficientNetB3 and Xception architectures to classify histopathological images. Using the LC25000 dataset, comprising 25,000 images across five classes, the model employs preprocessing steps such as resizing, normalization, and augmentation. Features extracted via Xception are refined using EfficientNetB3 and concatenated for classification. The framework achieved a 98.60% test accuracy, with sensitivity and specificity values of 0.9979 and 1.0000 for colon cancer, and 0.9898 and 0.9856 for lung cancer. F1-scores of 0.9990 (colon) and 0.9660 (lung) further demonstrate its reliability. An AUC-ROC of 0.9993 underscores its superior performance in minimizing false outcomes. The results highlight the model's potential for clinical integration, offering an efficient, scalable, and accurate tool for automated cancer detection.