MIT Jameel Clinic - AI & Health

Can we slow the pace of cognitive decline? John Werner for Forbes covers the work of MIT Jameel Clinic - AI & Health PI Randall Davis in human-computer interaction in the early detection of dementia. Along with a team of researchers, Davis devised a stylus that could detect the onset of dementia through inadvertent actions such as eye fixation, blink rate, pupil size, etc. — methods which are significantly less invasive and time-consuming than existing tests. Read more: https://lnkd.in/eee5GD5B #aiforgood #aiforhealth #healthcareinnovation #dementia

Day 1 of the MIT Jameel Clinic - AI & Health High School Summer Bootcamp is off to an exciting start! Some of the alumni from last year's program greeted the new students in the morning, followed by an incredible day filled with talks, lectures, and tutorials from AI researchers, physicians, policy researchers, and a Nobel Prize winner! This year's cohort was double the size of last year's cohort, making it the largest bootcamp we've hosted so far! 📸: Alexander Laiman #aiforgood #aiforhealth #healthcareinnovation #aieducation

The majority of FDA-approved AI models are for medical imaging, but gradual changes in data could cause the performance to deteriorate over time. Cami R. of Psychology Today recently covered a Nature Medicine paper published by MIT Jameel Clinic - AI & Health researchers, highlighting the key findings, which showed that AI models that excel at distinguishing race from medical images demonstrate biased and reduced performance when conducting disease diagnoses from medical images. Article: https://lnkd.in/dcA7cx2J Paper: https://lnkd.in/gEMJKaTT #aiforgood #aiforhealth #machinelearningsolutions #healthcareinnovation

Remember when researchers at MIT and Harvard realized that AI could accurately detect race from X-rays 90% of the time and human experts couldn't? Now, researchers from MIT Jameel Clinic - AI & Health and Emory University School of Medicine (Yuzhe Yang, Haoran Zhang, Judy Gichoya, Dina Katabi and Marzyeh Ghassemi) have found that AI models that are particularly good at detecting race from medical images also show the biggest "fairness gaps" when it comes to correctly diagnosing medical images from patients of difference races or genders. Read more from Anne Trafton in MIT News: https://lnkd.in/e5Gcmu3Y #aiforgood #aiforhealth #healthcareinnovation #aibias

What can Aristotle teach us when it comes to ethics in artificial intelligence? MIT Jameel Clinic - AI & Health PI Fotini Christia recently spoke at the Institute for Ethics in AI at the University of Oxford on the Practitioners' Panel, in which leading figures from the technical, entrepreneurial, and regulatory sides of AI discuss bridging theory and practice. The panel is part of the #LyceumProject, which seeks to apply Aristotelian ethics to regulation of AI, whether it be the personal self-regulation, regulation through industry codes or social convention, or regulation through domestic or international law. Current discourse around the ethics and politics of AI has been dominated by utilitarian and rights-based paradigms, neglecting Aristotelian ethics, which centers human flourishing in ethics and democratic politics. #aiforgood #aiforhealth #aiethics #aisafety #airegulation

Machine learning models are increasingly being trained on biased datasets that contain explicit and/or illegal content, and negative societal biases. This means that the same ML models often perform poorly when analyzing subgroups that are underrepresented in the dataset. To prevent this, Saachi Jain, Kimia Hamidieh, Kristian Georgiev, Andrew Ilyas, MIT Jameel Clinic - AI & Health PI Marzyeh Ghassemi, and Aleksander Madry introduce Data Debiasing with Datamodels (D3M), a data-centric debiasing approach that: (1) Pinpoints a small number of examples that are disproportionately driving the worst-group accuracy and removes them. (2) Achieves competitive debiasing performance that outperforms standard data-based and model-based approaches. (3) Discovers hidden biases. Paper: https://lnkd.in/eM3p96Gm #aiforgood #aiforhealth #aibias #aisafety #aiethics

We're excited to watch the development of all of this incredible cancer research, especially from MIT Jameel Clinic - AI & Health AI faculty lead Regina Barzilay and Tyler Jacks, PhD of the MIT Koch Institute for Integrative Cancer Research. The two MIT researchers are working together to build machine learning models that can predict interactions between T-cell receptors and peptide-MHC complexes — a formidable challenge due to a high degree of genetic diversity, the complex 3D structures of the molecules involved, and scarcity of data. #aiforgood #aiforhealth #healthcareinnovation #cancerresearch

We are delighted to be working with Abdul Latif Jameel Poverty Action Lab (J-PAL) student Kei Chuen (Philip) Ma to study the cost-effectiveness of low-dose CT (LDCT) screening using Sybil, a deep learning model that analyzes a patient's LDCT scan to predict lung cancer risk up to 6 years in advance. The last study done on the cost-effectiveness of LDCT lung cancer screening was published a decade ago (2014) in the NEJM Group. Understanding the cost-effectiveness of medical procedures is a major consideration for policymakers who are responsible for deciding on screening guidelines, practice measures, and insurance coverage and we hope this paper can illuminate how AI models like Sybil can benefit both patients and providers. #aiforgood #aiforhealth #machinelearningsolutions #healthcareinnovation

Ultrafast 3D imaging is critical for visualizing complex and dynamic biological processes like blood flow and observing how signals travel between neurons. But there is often a tradeoff between resolution and speed, as there is a limit to how much data can be processed in such a short period of time. Ruipeng Guo, Qianwan Yang, Andrew S. Chang, Guorong Hu, Joseph Greene, Christopher Gabel, MIT Jameel Clinic - AI & Health PI Sixian You, and Lei Tian introduce EventLFM, which uses deep learning to bypass the limitations of current 3D imaging techniques, leading to improved 3D reconstructions of tissues and organs. Paper: https://lnkd.in/dFq25kGR #aiforgood #aiforhealth #healthcareinnovation #machinelearningsolutions

A sprayable gel that can seal wounds? MIT Jameel Clinic - AI & Health PI Elazer Edelman recently co-authored a paper with Gonzalo Muñoz Taboada, Daniel Dahis, Pere Dosta Pons, Ph.D., and Natalie Artzi in which they developed a hydrogel sealant called GastroShield that can be sprayed through an endoscope. Hydrogels are used for surgical procedures because they are soft and flexible, yet many hydrogels quickly lose adhesion, especially when used for gastrointestinal (GI) procedures. As such, there is no commercially available sealant that is able to effectively prevent post-operative and delayed GI complications (3-7 days after surgery). While GastroShield's main advantage is that can be sprayed on wounds resulting from colon surgery, researchers have found a broad array of applications for GastroShield including treating stomach ulcers and inflammatory conditions like Crohn's. Read more from Anne Trafton in MIT Technology Review: https://lnkd.in/eBS39gHD #healthcareinnovation #medicaldevice #medtech #gisurgeon