Longitudinal Imaging and Temporal Deep Learning for Dynamic Risk Prediction in Lung Cancer: Current Progress and...

Lung cancer is a dynamically evolving disease characterized by marked temporal heterogeneity across tumor progression, treatment response, resistance development, and clinical outcomes. Nevertheless, the majority of existing artificial intelligence (AI) systems continue to rely predominantly on single time-point imaging, imposing inherent constraints on their capacity to characterize disease evolution and capture continuously shifting risk states. The growing availability of longitudinal imaging data, combined with ongoing advances in temporal deep learning, is driving a transition in lung cancer AI from static image interpretation toward dynamic risk prediction. In this narrative review, relevant literature was identified through searches of PubMed/MEDLINE, Web of Science Core Collection, and Scopus from database inception to September 2026, using terms related to lung cancer, longitudinal imaging, temporal deep learning, and dynamic risk prediction, with representative studies selected according to their methodological relevance, clinical applications, and implications for clinical translation. Longitudinal imaging tracks disease trajectories over time and offers distinct opportunities to characterize tumor growth dynamics, treatment-induced changes, and evolving recurrence risk. Temporal deep learning, in turn, enables the modelling of complex time-dependent relationships that conventional static approaches are poorly suited to represent. The field has progressed from early dual time-point analyses to long-range temporal modelling, multimodal temporal fusion, and continuous risk updating, with encouraging applications spanning pulmonary nodule management, treatment response assessment, recurrence surveillance, survival prediction, and individualized disease management. Nonetheless, substantial obstacles persist, including the scarcity of large-scale longitudinal datasets, irregular temporal sampling, limited model generalizability, insufficient causal interpretability, and a lack of robust evidence supporting real-world clinical deployment. Addressing these challenges will require coordinated efforts toward standardized longitudinal data infrastructure, continuous-time modelling frameworks, multimodal integration strategies, and closer alignment of AI systems with clinical workflows. Taken together, longitudinal imaging and temporal deep learning are steering lung cancer AI away from static image analysis and toward dynamic disease management, with the potential to support continuous risk assessment and potentially inform adaptive clinical decision-making throughout the full spectrum of lung cancer care.