Self-supervised isotropic reconstruction for abnormality detection in anisotropic MRI
Jui-Yo Hsu1, Pin-Hsun Lian2, Tzu-Yi Chuang3
1Program for Precision Health and Intelligent Medicine, Graduate School of Advanced Technology, National Taiwan University, Taipei, Taiwan; Department of Orthopedic Surgery, National Taiwan University Hospital, Taipei, Taiwan; Department of Orthopedic Surgery, College of Medicine, National Taiwan University, Taipei, Taiwan.
Abstract:
Accelerating musculoskeletal magnetic resonance imaging (MRI) while preserving diagnostic detail remains challenging because acquiring fully‑isotropic ground‑truth volumes is clinically costly. In routine practice, anisotropic scans with reduced through-plane resolution degrade multiplanar visualization and slice-by-slice review in reformatted planes, obscure subtle abnormalities spanning only a few slices, and limit automated three-dimensional (3D) analyses that assume comparable spatial resolution across axes. We present a two‑stage, fully self‑supervised pipeline that learns directly from anisotropic scans-obviating any paired high‑resolution data-and converts highly anisotropic (8:1) turbo‑spin‑echo volumes into isotropic images and 3D abnormality maps. Unlike prior self-supervised super-resolution methods, Stage 1 uses a single forward multi-view generative adversarial network (GAN) with patch-based contrastive and adversarial objectives rather than a backward/cycle-consistency approach. Stage 2 leverages an anatomy-conditioned denoising-diffusion model for healthy counterfactual generation, yielding voxel-wise lesion maps without external annotations. On 2225 Osteoarthritis Initiative knee scans from five different imaging centres, the framework reduced Fréchet inception distance from 407.4 → 254.4 (coronal) and 429.9 → 266.9 (axial), achieved the best Kernel Inception Distance (KID) / Learned Perceptual Image Patch Similarity (LPIPS) scores among competing unsupervised methods, and was preferred in 65-67% of blinded orthopedist comparisons. Crucially, isotropic enhancement propagated to downstream tasks: femur-tibia segmentation F1 scores increased and previously confluent bone‑marrow lesions were separated into discrete entities, enabling precise volumetric quantification. Robustness experiments demonstrated consistent gains across five imaging centers, synthetic noise/contrast perturbations, and transfer of the resolution-enhancement module to two additional MRI protocols, supporting robustness across sites and acquisition protocols. By eliminating the need for ground‑truth isotropic images while surpassing state‑of‑the‑art unsupervised super‑resolution in both perceptual quality and clinical utility, our method may facilitate retrospective cohort studies and prospective scan-time reduction in heterogeneous knee MRI settings, with preliminary transferability to additional protocols.
Insights
This study introduces a self-supervised AI pipeline to create high-resolution, isotropic knee MRI scans from standard anisotropic data. This method improves visualization and analysis of subtle abnormalities without needing costly ground-truth images.
Area of Science:
- Medical Imaging
- Artificial Intelligence
- Biomedical Engineering
Background:
- Musculoskeletal magnetic resonance imaging (MRI) acceleration is limited by the challenge of acquiring fully-isotropic, high-resolution volumes, which is clinically costly.
- Anisotropic scans, common in practice, have reduced through-plane resolution, degrading multiplanar visualization, obscuring subtle abnormalities, and hindering 3D analyses.
Purpose of the Study:
- To develop and validate a two-stage, self-supervised deep learning pipeline for converting anisotropic knee MRI scans into isotropic volumes and 3D abnormality maps.
- To obviate the need for paired high-resolution ground-truth data in MRI super-resolution.
Main Methods:
- A two-stage, fully self-supervised pipeline was developed: Stage 1 uses a multi-view generative adversarial network (GAN) for super-resolution, and Stage 2 employs an anatomy-conditioned denoising-diffusion model for generating abnormality maps.
- The pipeline learns directly from anisotropic turbo-spin-echo volumes, converting 8:1 anisotropic data into isotropic images without requiring paired high-resolution datasets.
Main Results:
- The framework significantly improved image quality metrics (Fréchet inception distance) and achieved superior Kernel Inception Distance (KID) / Learned Perceptual Image Patch Similarity (LPIPS) scores compared to other unsupervised methods.
- Orthopedists preferred the enhanced images in 65-67% of blinded comparisons, and the isotropic enhancement improved downstream tasks like femur-tibia segmentation and bone-marrow lesion quantification.
- Robustness was demonstrated across five imaging centers, various perturbations, and transferability to two additional MRI protocols.
Conclusions:
- The proposed self-supervised pipeline effectively generates isotropic knee MRI images and abnormality maps from anisotropic scans, surpassing current unsupervised super-resolution methods in perceptual quality and clinical utility.
- This approach eliminates the need for ground-truth isotropic images, facilitating retrospective studies and prospective scan-time reduction in diverse knee MRI settings.
