Deep Learning-Driven Multimodal Integration of miRNA and Radiomic for Lung Cancer Diagnosis
Yuanyuan Chen, Dikang Chen, Xiaohui Liu, Hui Jiang, Xuemei Wang
Abstract
Lung cancer remains one of the most common and deadly malignancies worldwide. Current diagnosis and staging primarily rely on biopsy techniques, which fail to comprehensively characterize the molecular profiles and tumor microenvironment. Current studies demonstrate the promising performance (AUC = 82%) of miRNA-based predictive models, but exclusive reliance on miRNA signatures is limited by incomplete capture of tumor heterogeneity. Integrating imaging and genomic data can further enhance model accuracy, with functional nanomaterials serving as core advanced biosensing platforms to bridge miRNA sensing and radiomic fusion. Consequently, integrating imaging and genomic data can further enhance model accuracy. Recent research employing DenseNet architecture for the multimodal fusion of miRNA and radiomic features achieved an AUC of 0.98 with 85.7% sensitivity. This review summarizes advances in miRNA biomarkers, deep learning-driven radiogenomics, and critical roles of functional nanomaterials in biosensing-enabled multimodal integration, along with challenges and future directions for clinical translation.