Problem
Research questions and friction points this paper is trying to address.
longitudinal MRI
missing data
multimodal imaging
image imputation
temporal interpolation
Implicit Neural Representations for Multimodal Longitudinal Image Imputation and Interpolation
🤖 AI Summary
This work addresses key challenges in clinical longitudinal multiparametric MRI—namely, missing sequences, heterogeneous acquisition protocols, and inconsistent spatial resolution across timepoints—by proposing a patient-specific conditional implicit neural representation. The method models longitudinal multimodal MRI as a continuous function of world coordinates, time, and imaging modality. Through a modality–time joint conditioning mechanism, stochastic modality dropout during training, and a self-consistency–based confidence estimation scheme, it enables robust spatiotemporal interpolation and reconstruction quality assessment without requiring image resampling. Evaluated on pediatric brain tumor data, the approach achieves a mean MS-SSIM of 0.95 ± 0.02 for T1CE, significantly outperforming linear interpolation (p < 0.05), and demonstrates a strong correlation between predicted confidence and actual reconstruction quality, with Pearson’s r reaching up to 0.996.
📝 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.
Innovation
Methods, ideas, or system contributions that make the work stand out.
Implicit Neural Representation
Longitudinal MRI
Multimodal Imputation
Continuous Coordinate Interpolation
Self-consistency Confidence Estimation
🔎 Similar Papers
No similar papers found.
