news

Sep 21, 2026 Happy to know that we’ll start our Automated PET Report Generation with Visual Grounding project under the support from NIH (1R01EB038426-01).
Sep 12, 2026 OpenSkill: Open-World Self-Evolution for LLM Agents is accepted by EMNLP 2026. This work asks whether an agent can evolve given nothing but a task prompt, building both its skills and its own verification from documentation and repositories.
Sep 12, 2026 Happy to know that we’ll start our UniBrain: An End-to-End Unified Toolkit for Comprehensive Neuroimaging Analysis project under the support from NIH (1R01EB038264-01A1).
Aug 09, 2026 LiveMedBench: A Contamination-Limited Medical Benchmark for LLMs is accepted by KDD 2026.
Jul 26, 2026 Honored to serve as the Associate Editor of IEEE Transactions on Medical Imaging.
Jul 26, 2026 Agentic AI Enhances Physician Trust in Clinical Decision Making is accepted by AMIA. This work finds that physicians trusted the system’s clinical reasoning in 89.57% of treatment-planning cases.
Jul 15, 2026 Interpretable agentic AI system with localized reasoning for radiology is accepted by npj Digital Medicine.
Jul 09, 2026 Evaluating GPT-4V for Multi-modal Medical Applications: A Comparative Study Using Open-source Datasets is accepted by IEEE Transactions on Radiation and Plasma Medical Sciences.
May 26, 2026 Released the NeuroClaw technical report (arXiv:2604.24696). NeuroClaw is a closed-loop agentic AI for executable and reproducible neuroimaging research, and the harness behind our AI4Neuroscience effort.
May 20, 2026 Our intern Sophia Liu won first place at an event organised by AMD and 36Kr, for a public-interest project building educational AI agents for teachers in rural mountain schools. Congratulations, Sophia!
May 14, 2026 Just posted Metis AI: The Overlooked Middle Zone Between AI-Native and World-Movers onto arXiv (2605.14407). The dominant framing places the boundary of AI capability between digital and physical tasks. I argue the more consequential boundary runs within digital work itself: a wide class of tasks that are computationally accessible yet resist automation because they are institutionally, socially, and normatively entangled. The paper proposes five structural characteristics that define this zone, and argues the right design response is centaur architectures rather than better automation.
Apr 26, 2026 Our fMRI foundation model NeuroSTORM is now online at Nature Biomedical Engineering. NeuroSTORM is pre-trained on 28.65 million fMRI frames — over 9,000 hours from more than 50,000 subjects, covering all major public fMRI datasets. We validated it across five downstream tasks and two clinical datasets, where it consistently outperformed existing methods. Source code and model weights are released.
Apr 20, 2026 Our high-school intern Emmett Chen, a junior at Noble and Greenough School, won second place at the 2026 Massachusetts Science and Engineering Fair for his work at CAMCA on 3D hand pose estimation from limited 2D views. Congratulations, Emmett!
Mar 14, 2026 Beyond Adapter Retrieval: Latent Geometry-Preserving Composition via Sparse Task Projection is accepted by AAAI 2026.
Sep 20, 2025 Our student received a mentorship award for work on agentic AI for clinical workflows.
Aug 22, 2025 Our paper was selected as an Editor’s Choice in Medical Physics.
Jun 19, 2025 Our collaborative work MedReason with UCSC won 3rd place in the Hugging Face Reasoning Dataset Competition. Congratulations to the team!
Jun 17, 2025 Three papers accepted at MICCAI 2025.
May 01, 2025 Our fairness-in-segmentation paper was accepted as a Spotlight Poster at ICML 2025.
Apr 02, 2025 Released MedReason, a medical reasoning dataset with 33K examples.
Jan 22, 2025 Two papers accepted at ICLR 2025.
Dec 19, 2024 Large Language Models and Large Multimodal Models in Medical Imaging: A Primer for Physicians is accepted by the Journal of Nuclear Medicine as a CME article. This work introduces physicians and researchers in nuclear medicine to the fundamental principles and applications of these models.
Nov 14, 2024 Automated Detection and Management of Valvular Heart Disease Using Machine Learning is published as a patent (US-20240379239-A1). This work was a collaboration between our team and GE Healthcare.
Nov 12, 2024 Two podium abstracts are accepted by AMIA, including Cross-Modal Retrieval for Alzheimer’s Disease Diagnosis Using CLIP-Enhanced Dual Deep Hashing.
Jul 24, 2024 Robot Control via Natural Instructions Empowered by Large Language Model is accepted by Discovering the Frontiers of Human-Robot Interaction (Springer Nature).