Spatial Perception & Registration

Algorithmic foundations for spatial correspondence and deep learning-based medical image registration.

Spatial perception is the foundation of spatial intelligence: before an AI system can understand or reason about anatomy, it must first estimate how structures correspond and deform across images and time.

My research develops the algorithmic foundations for spatial correspondence and alignment, including transformation regularization, similarity learning, and uncertainty estimation. These components form the infrastructure for building reliable and generalizable spatial models in medical imaging.

Transformation Regularization

I study how to represent and regularize deformation fields so that they are physically plausible, diffeomorphic, and generalizable across datasets and anatomies.

GradICON (Tian et al., 2023) achieves state-of-the-art diffeomorphic registration via gradient inverse consistency regularization.

NePhi (Tian et al., 2024) represents deformation fields as neural implicit functions for memory-efficient high-resolution registration.

CARL (Greer et al., 2025) introduces an equivariance framework for image registration.

Similarity Measure

I study how to measure anatomical similarity across images, especially across modalities where intensity values are not directly comparable.

SAMConvex (Li et al., 2023) and SAME++ (Tian et al., 2023) leverage self-supervised anatomical embeddings for fast and accurate cross modality registration.

Uncertainty Estimation

I study how to quantify the uncertainty of spatial correspondence, which is critical for clinical decision-making and downstream tasks.

I propose test-time uncertainty estimation (Tian et al., 2026) via transformation equivariance.

References

2026

  1. arXiv
    UE_teaser.png
    Test-time Uncertainty Estimation for Medical Image Registration via Transformation Equivariance
    Lin Tian, Xiaoling Hu, and Juan Eugenio Iglesias
    arXiv preprint arXiv:2509.23355, 2026
    arXiv

2025

  1. CVPR
    CARL: A Framework for Equivariant Image Registration
    Hastings Greer, Lin Tian, François-Xavier Vialard, and 3 more authors
    In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, 2025
    arXiv

2024

  1. ECCV
    NePhi_teaser.png
    NePhi: Neural Deformation Fields for Approximately Diffeomorphic Medical Image Registration
    Lin Tian, Hastings Greer, Raúl San José Estépar, and 2 more authors
    arXiv preprint arXiv:2309.07322, 2024
    ECCV 2024
    arXiv

2023

  1. CVPR
    GradICON_teaser.png
    GradICON: Approximate Diffeomorphisms via Gradient Inverse Consistency
    Lin Tian, Hastings Greer, François-Xavier Vialard, and 6 more authors
    In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, 2023
    arXiv Code
  2. MICCAI
    SAMConvex_teaser.png
    SAMConvex: Fast Discrete Optimization for CT Registration Using Self-supervised Anatomical Embedding and Correlation Pyramid
    Zi Li, Lin Tian, Tony CW Mok, and 8 more authors
    In Medical Image Computing and Computer Assisted Intervention – MICCAI 2023, 2023
    arXiv Code
  3. arXiv
    SAME_teaser.png
    SAME++: A Self-supervised Anatomical eMbeddings Enhanced medical image registration framework using stable sampling and regularized transformation
    Lin Tian, Zi Li, Fengze Liu, and 7 more authors
    arXiv preprint arXiv:2311.14986, 2023
    arXiv Code