Implicit Neural Representations for Multimodal Longitudinal Image Imputation and Interpolation
2026-08-03 • Computer Vision and Pattern Recognition
Computer Vision and Pattern Recognition
AI summaryⓘ
The authors developed a special AI model that can fill in missing pieces and smooth out differences in repeated MRI scans taken over time from cancer patients. Their model treats the images as continuous data that changes smoothly over space, time, and scan type, allowing it to predict images even when some scanning information is missing. They tested this on brain tumor MRI scans and found it works better than simple interpolation methods. Additionally, the model can estimate how confident it is about its predictions. This means the approach could be useful in real-world medical settings where scan data is often incomplete or inconsistent.
Longitudinal MRIMultiparametric MRIImplicit Neural RepresentationSpatial InterpolationTemporal InterpolationModality DropoutCross-Modal ReconstructionPaediatric Brain TumorMS-SSIMConfidence Estimation
Authors
Sina Wendrich, Lukas Förner, Zoe Reinke, Kartikay Tehlan, Ansgar Berlis, Michael Frühwald, Matthias Wagner, Thomas Wendler
Abstract
Longitudinal multiparametric MRI is central to follow-up imaging in oncology, yet real-world clinical data are characterised by missing sequences, heterogeneous acquisition protocols, and varying spatial resolutions across time points. We propose a patient-specific conditional implicit neural representation (INR) that models multimodal longitudinal MRI as a continuous function of world coordinates, time, and modality conditioning. The model is trained with stochastic modality dropout to handle incomplete data, and its continuous coordinate-space formulation enables both spatial and temporal interpolation without resampling to a fixed voxel grid. A self-consistency-based confidence estimator is derived from cross-modal reconstruction performance at inference time. We evaluate the framework on longitudinal MRI from paediatric brain tumour patients, demonstrating statistically significant improvements over linear interpolation for T1CE and FLAIR (p < 0.05), with mean MS-SSIM of 0.95 $\pm$ 0.02 for T1CE. Predicted confidence correlates strongly with true reconstruction quality (Pearson r up to 0.996), suggesting reliable deployment potential in heterogeneous clinical settings.