Attention-Enhanced Xception Network for Automated Microsatellite Instability Classification in Colorectal Cancer Histopathology

Abstract Background : Microsatellite instability (MSI) is a critical molecular biomarker in colorectal cancer (CRC), guiding prognosis and immunotherapy decisions. However, standard MSI detection methods such as immunohistochemistry (IHC) and PCR remain costly, time-consuming, and inaccessible in many settings. Recent advances in computational pathology have enabled the use of deep learning models to extract clinically meaningful features from routine hematoxylin and eosin (H&E)-stained slides, offering a non-invasive and scalable alternative. Methods : We developed an interpretable deep learning framework integrating the Xception backbone with a Convolutional Block Attention Module (CBAM) to enhance MSI classification performance. The model was trained and tested on a large-scale public dataset containing 194,178 image patches derived from FFPE histology slides of CRC patients in the TCGA cohort. Model performance was evaluated using standard metrics including accuracy, precision, recall, F1-score, and area under the ROC curve (AUC). Grad-CAM was employed to visualize model attention and assess interpretability. Results : Our proposed model achieved superior performance with an AUC of 0.9551, an accuracy of 88.4%, a precision of 90.7%, a recall of 84.1%, and an F1-score of 87.3%, outperforming baseline CNN architectures including DenseNet121, InceptionV3, ResNet50, and others. Grad-CAM analysis revealed that the model consistently focused on histopathological features associated with MSI, such as tumor-infiltrating lymphocytes and glandular disorganization, while suppressing non-diagnostic regions. Conclusion : The attention-enhanced Xception model provides a robust, interpretable, and clinically scalable solution for MSI status prediction from H&E-stained slides. It holds promise as a non-invasive alternative to traditional assays, especially in resource-constrained environments, and supports the growing role of deep learning in precision oncology.