Implicit Neural Representations for Multimodal Longitudinal Image Imputation and Interpolation
Ranking
Observed public metrics from 1 member.
Merged summary
TL;DR - A patient-specific conditional implicit neural representation models longitudinal multiparametric MRI as a continuous function of world coordinates, time, and modality, filling in missing sequences and time points. It targets a real clinical pain point: follow-up oncology imaging with missing sequences, heterogeneous protocols, and inconsistent resolutions.
- Continuous coordinate-space formulation enables both spatial and temporal interpolation without resampling to a fixed voxel grid; stochastic modality dropout during training handles incomplete multimodal data.
- Evaluated on longitudinal MRI from paediatric brain tumour patients, with statistically significant gains over linear interpolation for T1CE and FLAIR (p < 0.05) and mean MS-SSIM of 0.95 ± 0.02 for T1CE.
- A self-consistency confidence estimator, derived from cross-modal reconstruction performance at inference time, correlates strongly with true reconstruction quality (Pearson r up to 0.996).
- The confidence signal is framed as a deployment safeguard for heterogeneous clinical settings, though results come from a single paediatric cohort.
Sources (1)
Implicit Neural Representations for Multimodal Longitudinal Image Imputation and Interpolation
TL;DR - A patient-specific conditional implicit neural representation models longitudinal multiparametric MRI as a continuous function of world coordinates, time, and modality, filling in missing sequences and time points. It targets a real clinical pain point: follow-up oncology imaging with missing sequences, heterogeneous protocols, and inconsistent resolutions.
- Continuous coordinate-space formulation enables both spatial and temporal interpolation without resampling to a fixed voxel grid; stochastic modality dropout during training handles incomplete multimodal data.
- Evaluated on longitudinal MRI from paediatric brain tumour patients, with statistically significant gains over linear interpolation for T1CE and FLAIR (p < 0.05) and mean MS-SSIM of 0.95 ± 0.02 for T1CE.
- A self-consistency confidence estimator, derived from cross-modal reconstruction performance at inference time, correlates strongly with true reconstruction quality (Pearson r up to 0.996).
- The confidence signal is framed as a deployment safeguard for heterogeneous clinical settings, though results come from a single paediatric cohort.