Longitudinal Medical Image Analysis

The field studies sequences of medical scans to measure biological change, distinguish disease progression or treatment response from imaging variability, and improve diagnosis and prognosis. Common approaches include image registration and lesion tracking, temporal CNNs and transformers, graph and self-supervised models, and multimodal fusion with clinical data; generative methods also forecast future images, sometimes with uncertainty estimates. Applications range from brain disease and cancer to breast screening, retinal disease, and comparative radiology reporting, with evaluation spanning progression prediction, segmentation, registration, and multi-visit reasoning benchmarks. Sparse and irregular follow-ups, protocol differences, and limited external validation remain challenges, as does showing that predictions capture patient-specific change rather than population averages.