
Hi, this is Yinheng. I am a postdoc at dep. of radiation oncology, utsw. In 2025, I got my Ph.D. degree in BME from Tsinghua University, supervised by Prof. Shaohua Ma. In 2019, I got my B.E. degree in CogSci from Xiamen University. Along the journey, I had the opportunity to visit Prof. Roberto Horowitz's group at UC Berkeley(2023), Dr. Fangzhou Liao at Careverse Research(2021), Prof. Lu Fang's group at Tsinghua(2019), and Dr. Qiong Yan at SenseTime Research(2018).
My research goal is to develop an interventional autopilot capable of percepting, reasoning, and manipulating during endovascular interventions, to reduce the workload and expertise required of clinicians, ultimately making advanced endovascular care more accessible and affordable. Technically, this spans medical imaging, motion modeling, and policy learning for slender deformable objects. I am currently working along two axis:
- reconstructing 3D vascular structure from real-time 2D X-ray angiography;
- learning control policies for autonomous guidewire navigation.
I use this blog to share the research/tools/thoughts as I work toward them.
News
-
π« 1 poster accepted by 2026 ASTRO Annual Meeting, Boston, MA.
-
π« 1 poster and 1 oral accepted by 2026 AAPM Annual Meeting Vancouver, BC, Canada.
-
π« 1 paper accepted by Nature Machine Intelligence as cover article.
-
πΌ β π» One year visiting at Prof. Horowitzβs lab, UC Berkeley.
-
Join Digital Heart group at careverse.com as a research intern.
-
π§ A simplified 2D implementation of the Incremental Potential Contact (IPC) algorithm β course project for GAMES 201 by Yuanming HU β is available on GitHub.
-
A course project on continuous potential functions in CRF was presented in the Student Oral Presentation section at the TBSI Workshop on Learning Theory.
-
π· A paper on reference-based super-resolution is accepted by IEEE TPAMI.
-
π©Ί Fighting COVID-19 with our system; code & dataset. Read the paper in Patterns Journal, Cell Press.
-
πΌ The PANDA dataset is accepted by CVPR 2020.
-
β½ A paper is accepted by IEEE VR 2020 and IEEE TVCG.


