Annotating data at scale is time consuming especially for medical data, and the scarcity of labeled data can limit the effectiveness of supervised learning. We are developing self-supervised learning approaches, in addition to multimodal learning and transfer learning approaches, that allow for high model performance with small amounts of labeled data.
While AI technologies for clinical decision making are expected to be used in conjunction with clinical decision makers in practical scenarios, this setup is largely ignored in the design of these technologies. We are working on investigating how to optimize human-AI collaboration in the context of clinical workflows and deployment settings.
The validation of academically and commercially developed AI models has relied on private, non-standardized datasets that do not allow for a transparent assessment of their performance in the real world. We are developing open benchmarks to help the community transparently measure advancements in generalizability of algorithms to new geographies, patient populations, and clinical settings.
We tackle important problems in medicine that require artificial intelligence solutions.
We work in close collaboration with clinicians worldwide, and often partner with hospitals and
We work on problems across clinical domains including radiology, emergency medicine, and
cardiology, and develop computer vision and natural language processing solutions.
Please contact us at
hms.harvard.edu if any of the open positions or
calls for collaboration fit you.
We are looking for Harvard affiliates with a background in any one of AI, medicine, or web development to join our team. We expect that you will be able to commit 10-20 hours per week to research and be able to commit at least 6 months of research with the lab. We also have open post-doctoral positions for those with a strong demonstrated background in AI and medical image interpretation.
We cultivate collaborations with academic institutions, hospitals and industry. We are currently interested in partnering with industry to deploy medical imaging AI models. We are also seeking partnerships with hospitals to create large multi-institutional imaging datasets.