Bio

Lin Tian is a Research Fellow at Harvard Medical School and Massachusetts General Hospital, working with Dr. Juan Eugenio Iglesias and Dr. Matthew Rosen. She received her PhD in Computer Science from the University of North Carolina at Chapel Hill under the supervision of Prof. Marc Niethammer, and has held research scientist internships at ByteDance AI, Alibaba DAMO, and Google X. Her research spans medical imaging, machine learning, and data science, with a focus on 3D spatial alignment across images and modalities for precise, generalizable, and trustworthy healthcare AI. She serves as an Area Chair for MICCAI 2025, is a recipient of the MICCAI Travel Award and the CVPR Doctoral Consortium NSF Travel Award, and coauthored work recognized with Best Oral Presentation at the MICCAI WBIR.

Spatial Intelligence for Healthcare: Precise, Generalizable, and Trustworthy

Spatial intelligence, the ability to understand and align complex structures, is central to many challenges in healthcare. From image-based analysis in biomedicine and bioscience, to morphology-driven diagnosis, to surgical navigation, and to large-scale, population-level studies of disease, progress depends on accurate, robust, and generalizable spatial learning. Yet, while humans reason effortlessly about continuous, high-dimensional spatial changes, current machine learning systems still struggle to generalize across anatomies, modalities, and scales. My research aims to close this gap by building generalizable and data-efficient AI systems for spatial correspondence, alignment, and reasoning.

To this end, I have developed algorithms that advance the foundations of medical image registration. My work on GradICON and uniGradICON introduced novel regularization strategies and produced the first registration foundation model, enabling out-of-the-box alignment across modalities and anatomies, and released as an open-source Python package with over 6,000 downloads. I further explored physics-grounded approaches, such as a differentiable radiograph generator for CT/X-ray alignment, and simulation-based training engines, such as domain-randomized synthesis for reconstructing 3D brain anatomy from dissection photographs. These methods address a fundamental barrier in healthcare AI: the lack of ground-truth, high-dimensional spatial correspondences.

Looking forward, my research will center on physics-grounded simulation for data-efficient spatial learning, new methods for spatial correspondence, alignment, and reasoning across modalities and scales, and trustworthy AI frameworks that quantify uncertainty and ensure reliability in deployment. Together, these directions aim to establish robust and interpretable systems that generalize across domains, adapt to imperfect data, and deliver actionable spatial insights, driving advances in biomedical discovery and clinical care.