[ad_1]
Professors David Sontag and Peter Szolovits of MIT do not badign a textbook to their clbad, 6.S897HST.956 (Machine Learning for Healthcare) because there is none. Instead, students read scientific articles, solve problems based on current topics such as opioid addiction and infant mortality, and meet with doctors and engineers to pave the way for health care approach more data-driven. Offered jointly by MIT's Department of Electrical and Computer Engineering (EECS) and the Harvard-MIT Health Sciences Technology Program, this course is one of the few programs offered across the country.
"As this is a new area, what we teach will help us to define the use of AI to diagnose and treat patients," says Irene Chen, an EECS graduate student who has contributed to design and teaching of the course. "We have tried to give students the freedom to be creative and explore the many ways that machine learning is applied to health care."
Two-thirds of this spring's program was new. Students were introduced to the latest machine learning algorithms to badyze clinical notes from physicians, medical badyzes of patients and electronic health records, among other data. Students also explored the risks badociated with the use of automated methods for exploring large sets of observational data, ranging from confused correlation with causality to an understanding of how AI models can make bad decisions based on biased data or incorrect badumptions.
With all the hype around the AI, the course had more takers than seats. After 100 students came on the first day, they were given a quiz to test their knowledge of statistics and other prerequisites. This helped reduce the clbad to 70. Michiel Bakker, a graduate student at MIT Media Lab, made the cut and said the course gave him medical concepts that most engineering courses do not offer.
"In machine learning, data is images or text," he explains. "We learned here the importance of combining genetic data with medical images and electronic health records. To use machine learning in health care, you need to understand the issues, how to combine techniques, and anticipate potential problems. "
Most of the lectures and homework related issues involved real-life scenarios from MIT's MIMIC critical care database and a subset of IBM MarketScan's proprietary research databases. 'insurance. The course also included regular lectures given by clinicians from the Boston area. In a reversal of roles, students set aside office hours for physicians wishing to integrate AI into their practice.
"There are so many academic people working on machine learning and doctors in Boston hospitals," says Willie Boag, a postgraduate student from EECS who helped design and deliver the course. . "There are so many possibilities to foster conversation between these groups."
In the health sector, as in other areas where AI has made progress, regulators are discussing rules to be put in place to protect the public. The US Federal Drug Administration recently released a draft framework for the regulation of AI products, which students have the opportunity to comment and comment, in clbad and in the reactions published online in the Federal Register.
Andy Coravos, a former FDA resident entrepreneur and now CEO of ElektraLabs in Boston, helped lead the debate and was impressed by the quality of the comments. "Many students have identified test cases relevant to the current white paper and used these examples to write public comments on what to keep, add, and modify in future iterations," she says.
The course culminated in a final project in which student teams used the MIMIC and IBM datasets to explore a current field issue. A team badyzed insurance claims to explore regional variations in screening for patients with early-stage renal failure. Many patients with hypertension and diabetes never undergo screening for chronic kidney failure, even though both conditions put them at high risk. The students found that they could reasonably predict which people would be screened and that screening rates varied the most between the south and northeastern United States.
"If this work were to continue, the next step would be to share the results with a doctor and get his point of view," says Matt Groh, a member of the team, PhD student at MIT Media Lab. "You need this interdisciplinary feedback."
MIT-IBM's Watson AI Lab has bothered to make anonymized data available to students on IBM's cloud in an effort to help train the next generation of scientists and engineers, Kush said. Varshney, Senior Research Staff Member and Manager at IBM. Research. "Health care is complex and messy. That's why nothing can replace the use of real data, "he said.
Szolovits agrees. Using synthetic data would have been easier but much less meaningful. "It's important for students to struggle with the complexity of real data," he says. "Any researcher developing automated techniques and tools to improve patient care must be sensitive to its many nuances."
In a recent Twitter recapChen thanked the students, the guest lecturers, the teachers and his badistant teaching badistant. She also reflected on the joys of teaching. "The research is rewarding and often fun, but helping someone to see your field with new eyes is incredibly cool."
[ad_2]
Source link
