Deep Learning-Based Integrated Multi-Omics and Radiomics Model for Diagnosis of Kidney Renal Clear Cell Carcinoma

Kidney renal clear cell carcinoma (KIRC) is a highly aggressive malignancy that accounts for 7080% of renal cell carcinomas. Current diagnostic approaches for KIRC face several limitations that compromise early detection capabilities. Conventional imaging modalities frequently fail to identify early-stage lesions due to their subtle radiographic characteristics. Molecular diagnostic methods, though promising, mostly rely on unimodal gene data, making it harder for further improvements in accuracy, particularly in the case of early-stage prediction. To address these issues, this paper proposed an attention-based deep learning framework with one-dimensional convolutional neural network (genomics) and three-dimensional residual network (imaging) branches. After highly relevant genes were identified through feature engineering methods, they were fused with representations extracted from CT scans using a 3D Residual Network. The framework then employed a dynamic cross-modal attention mechanism to generate a fused vector, which then goes to a fully connected layer to make final decisions. A series of experiments demonstrate that the integrated model achieves the highest prediction performance in all four clinical stages.