Research

Current research

My current work focuses on cervical spine imaging, interpretable vertebral-fracture assessment, and medical anomaly detection.

Cover artwork depicting cervical MRI analysis and human-AI collaboration

Published

Cervical Spine MRI Analysis

Interpretable cervical MRI analysis spanning anatomical segmentation, clinical measurements, and spinal-cord T2 hyperintensity detection.

Clinical motivation: Cervical spondylosis assessment requires consistent interpretation of multiple anatomical structures, compression patterns, and signal abnormalities.
Technical focus: Pathology-guided segmentation, automated imaging indicators, uncertainty-aware anomaly detection, and external validation.
Related publications: 2026 · PublishedReconstructing Isotropic 3D Cervical MRI from Anisotropic Clinical Scans via Anatomy-Style Decoupled INRsPaper Code
2026 · PublishedPathology-Guided AI System for Accurate Segmentation and Diagnosis of Cervical SpondylosisPaper Code
2026 · PublishedU2AD: Uncertainty-based unsupervised anomaly detection framework for detecting T2 hyperintensity in MRI spinal cordPaper Code
2026 · Under ReviewArtificial intelligence for degenerative cervical myelopathy imaging: a systematic review across tasks and modalities

Project details Paper Code

HealthiVert-GAN artwork showing pseudo-healthy vertebra synthesis and height-loss measurement

Published

Interpretable Vertebral Fracture Assessment

Pseudo-healthy vertebral synthesis and relative height-loss measurement for quantitative CT-based compression-fracture assessment.

Clinical motivation: Compression-fracture grading is difficult when no pre-fracture reference is available and deformity varies across the vertebral body.
Technical focus: Anatomy-informed image synthesis, quantitative height-loss mapping, and interpretable grading.
Related publications: 2025 · PublishedHealthiVert-GAN: A Novel Framework of Pseudo-Healthy Vertebral Image Synthesis for Interpretable Compression Fracture GradingPaper Code

Project details Paper Code

UniTransAD Figure 1 comparing reconstruction, cyclic-translation, cross-translation, and unified anomaly-detection paradigms

Published

Generalizable Medical Anomaly Detection

Translation-, reconstruction-, and uncertainty-based methods for detecting abnormalities across brain and spinal MRI.

Clinical motivation: Rare and heterogeneous lesions make exhaustive annotation difficult, while acquisition changes can produce false anomaly signals.
Technical focus: Cross-sequence translation, content-style disentanglement, cyclic inference, dual-level evidence, and uncertainty-guided learning.
Related publications: 2026 · PublishedUniTransAD: Unified Translation Framework for Anomaly Detection in Brain MRIPaper Code
2026 · PublishedU2AD: Uncertainty-based unsupervised anomaly detection framework for detecting T2 hyperintensity in MRI spinal cordPaper Code
2025 · PublishedMultiTransAD: Cross-Sequence Translation-Driven Anomaly Detection in Multi-sequence Brain MRIPaper Code

Project details Paper Code

Next steps

From images to quantitative assessment

I am extending these projects toward multimodal cervical imaging, isotropic 3D reconstruction, structured quantitative analysis, and uncertainty-aware review.