AI Health

Friday Roundup

The AI Health Friday Roundup highlights the week’s news and publications related to artificial intelligence, data science, public health, and clinical research.

October 28, 2022

In today’s Duke AI Health Friday Roundup: time series classification for sensor data; most US maternal deaths are preventable; confronting stigmatizing language about substance use; digital repository houses wealth of 3-D specimen scans; user evaluations for explainable AI systems; retooling research funding mechanisms; the immunological reverberations of the Black Death pandemic; much more:


A shiny chrome human-shaped robot stands in a museum gallery, looking at framed pictures hanging on the walls. Image created with Stable Diffusion.
Image created with Stable Diffusion
  • Business Insider’s Beatrice Nolan reports on growing concerns from artists whose work is being used – with or without their knowledge or consent – to train AI-powered image generators (such as DALL-E, Midjourney, and Stable Diffusion) to produce stylistically similar images.
  • “The use of AI in diabetes care has already irreversibly shifted my role as an endocrinologist, but I no longer fear that I will become obsolete. AI will not come between me and my patient, but will instead relieve me of the burden of analyzing the data. AI can restore human-to-human connections, allowing me to be a coach, supporter and healer, giving me space to gain a broader perspective into a patient and their life.” A viewpoint article published in Nature Medicine by endocrinologist Aaron B. Neinstein accentuates the positive (possibilities) for health AI.
  • “While high time series classification performance has been achieved in digital clinical, physiological, or biomedical measures, no standard benchmark datasets, modeling methods, or reporting methodology exist. There is no single widely used method for time series model development or feature interpretation, however many different methods have proven successful.” A systematic review by Wang and colleagues published in Sensors examines the array of approaches currently in use for time series classification of data gathered from personal devices and wearable sensors.
  • “After identifying and thoroughly analyzing 85 core papers with human-based XAI evaluations over the past five years, we categorize them along the measured characteristics of explanatory methods, namely trust, understanding, fairness, usability, and human-AI team performance. Our research shows that XAI is spreading more rapidly in certain application domains, such as recommender systems than in others, but that user evaluations are still rather sparse and incorporate hardly any insights from cognitive or social sciences.” A review article by Rong and colleagues, available as a preprint from arXiv, characterizes the current state of user studies in the domain of explainable AI and proposes some guidelines for future investigations.
  • “If AI systems can glean a deeper, more nuanced understanding of reality beyond what’s specified in a training data set, they potentially will be more useful and ultimately will bring AI closer to real-world medical applications. Self-supervised learning is much more scalable than traditional supervised learning because class label annotation is not required.” A brief blog post by Qinmei Xu at Radiology: Artificial Intelligence provides a quick primer on self-supervised learning for medical image analysis.


Close-up black and white photo of a tiny infant’s hands holding on to mother’s finger. Image credit: Liv Bruce/Unsplash
Image credit: Liv Bruce/Unsplash
  • “The data highlights multiple weaknesses in the system of care for new mothers, from obstetricians who are not trained (or paid) to look for signs of mental trouble or addiction, to policies that strip women of health coverage shortly after they give birth….The number one problem, as Sheffield-Abdullah sees it, is that the typical six-week postnatal checkup is way too late.” NPR’s April Dembosky unpacks a recent CDC report that finds the vast majority of maternal deaths in the US are preventable.
  • “Originally envisioned as a way to store 3D scans produced by the lab Boyer worked in as a postdoctoral researcher, MorphoSource is now one of the world’s most important scientific data repositories. In a recent survey asking natural sciences researchers which repositories were most important for their work, it tied up in first place with GenBank, the National Institutes of Health genetic sequence database holding all publicly available DNA sequences. And it reached this status in less than 10 years.” An article by Duke University’s Marie Claire Chelini highlights the work of evolutionary anthropologist Doug Boyer, whose MorphoSource repository provides a digital library of high-resolution, 3-D scans of museum biological and fossil specimens.
  • “In a paper published Wednesday in Nature, Poinar, together with researchers from the University of Chicago and France’s Institut Pasteur, studied DNA samples from individuals at the East Smithfield graveyard as well as people who lived just before and a few decades after the Black Death. The team found that the Black Death not only created selective evolutionary pressure for genes that promote better immune response to Y. pestis but also may have contributed to humans’ current susceptibility to autoimmune diseases like Crohn’s disease. STAT News’ Brittany Trang reports on findings, recently published in Nature, that trace the immunological reverberations of the Black Death pandemic.
  • “Among outpatients with mild to moderate COVID-19, treatment with ivermectin, compared with placebo, did not significantly improve time to recovery. These findings do not support the use of ivermectin in patients with mild to moderate COVID-19.” A research article published last week in JAMA by Naggie and colleagues presents results from the ACTIV-6 randomized clinical trial compared ivermectin vs placebo in patients with mild to moderate COVID.

COMMUNICATION, Health Equity & Policy

Shepherd dog jumping through a hoop at a dog agility competition. Image credit: Andrea Lightfoot/Unsplash
Image credit: Andrea Lightfoot/Unsplash
  • “This competitive model, which is almost universally used, has well-documented flaws. It is excessively wasteful in terms of researchers’ time. If only 20% of grants are funded, every dollar given by a research grant agency represents many valuable hours spent by skilled researchers writing appealing grant proposals, filling forms, and jumping through the hoops of the reviewing process. These are hours which were channeled away from fruitful work to advance science, create new medical therapies, inform policymakers, and generate innovations.” An opinion article published in STAT News by Lionel Page and Adrian Barnett makes the case for a lottery system to replace a research funding paradigm that increasingly absorbs massive amounts of effort and features slender odds for success.
  • “…we identified physical, communication, knowledge, structural, and attitudinal barriers to care for people with disabilities. Physicians reported feeling overwhelmed by the demands of practicing medicine in general and the requirements of the Americans with Disabilities Act of 1990 specifically; in particular, they felt that they were inadequately reimbursed for accommodations. Some physicians reported that because of these concerns, they attempted to discharge people with disabilities from their practices.” A study published in Health Affairs by lagu and colleagues shines a light on physicians’ attitudes and conduct toward patients with disabilities.
  • “…new research has emerged showing that simple word choices can have a big impact on the way health professionals view their patients and, accordingly, the care they receive. And in recent years, a coalition of doctors, recovery advocates, researchers, and even government officials has pushed to swap out stigmatizing terms like “addict” in favor of language that recognizes addiction as a medical condition — and acknowledges those who suffer from it as human beings.” A feature article at STAT News by Lev Facher highlights the emerging recognition of the importance of language in confronting problems related to substance use and overdose.