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.

September 30, 2022

In today’s Duke AI Health Friday Roundup: transformer neural networks mimic the human hippocampus; NIH undertakes to ID function for every human gene; FDA releases new guidance for health AI; “nanorattles” shine a light on cancer detection; the impact of elite universities on hiring for US faculty; light pollution gets worse across much of Europe; association between type 1 diabetes and COVID infections in kids; much more:


Abstract black and white drawing of tree branches forming the shape of a human brain. Image credit: Gordon Johnson/Pixabay
Image credit: Gordon Johnson/Pixabay
  • “For years, neuroscientists have harnessed many types of neural networks — the engines that power most deep learning applications — to model the firing of neurons in the brain. In recent work, researchers have shown that the hippocampus, a structure of the brain critical to memory, is basically a special kind of neural net, known as a transformer, in disguise.” Quanta’s Stephen Ornes describes recent work that reveals similarities in the human hippocampus and a type of neural net known as a transformer.
  • A research article by Chaudhari and colleagues, published in Radiology: Artificial Intelligence, describes the use of bidirectional encoder representations from transformers (BERT) model to detect errors arising from speech-recognition tools in radiology reports and to offer suggestions for correcting them.
  • “Deep learning is an artificial-intelligence (AI) technique that relies on many-layered artificial neural networks inspired by how neurons interconnect in the brain. Based as they are on black-box neural networks, the algorithms have their limitations. Those include a dependence on massive data sets to teach the network how to identify features of interest, and a sometimes inscrutable way of generating results. But a fast-growing array of open-source and web-based tools is making it easier than ever to get started.” A splendid feature article by Sandeep Ravindran at Nature walks readers through a handful of examples illustrating transformative advances in AI-enabled image analysis.
  • “Most prokaryotes are not available as pure cultures and therefore ineligible for naming under the rules and recommendations of the International Code of Nomenclature of Prokaryotes (ICNP). Here we summarize the development of the SeqCode, a code of nomenclature under which genome sequences serve as nomenclatural types. This code enables valid publication of names of prokaryotes based upon isolate genome, metagenome-assembled genome or single-amplified genome sequences.” An article by Hedlund and colleagues, published in Nature Microbiology, describes a new approach to developing nomenclature for prokaryotes, one based on genomic sequences.


Photograph of a mouse gene microarray with grids of illuminated dots that reflect patterns of gene expression. Image credit: National Cancer Institute
Image credit: National Cancer Institute
  • “The National Institutes of Health is launching a program to better understand the function of every human gene and generate a catalog of the molecular and cellular consequences of inactivating each gene. The Molecular Phenotypes of Null Alleles in Cells (MorPhiC) program, managed by the National Human Genome Research Institute, aims to systematically investigate the function of each gene through multiple phases that will each build upon the work of the previous.” The National Institutes of Health has announced that it’s launching a 5-year program to catalog the function of every human gene.
  • “When a laser shines on the nanorattles, it travels through the extremely thin outer shell and hits the Raman reporters within, causing them to emit light of their own. Because of how close the surfaces of the gold core and the outer gold/silver shell are together, the laser also excites groups of electrons on the metallic structures, called plasmons. These groups of electrons create an extremely powerful electromagnetic field due to the plasmons’ interaction of the metallic core-shell architecture, a process called plasmonic coupling, which amplifies the light emitted by the Raman reporters millions of times over.” An article by Duke Pratt School of Engineering’s Ken Kingery focuses on an effort to create “nanorattles,” a light-amplifying nanoparticle that be used to test for malignancies without performing a biopsy.
  • “Here, using the unique opportunity provided by the DSLR images taken by astronauts aboard the ISS, we have shown that across the past decade, these spectral changes have been widespread across Europe. They have been spatially very uneven, reflecting historical variation in the kinds of lighting systems that have been used and the diversity of national and regional lighting policies and approaches, but the net effect has nonetheless been a pronounced whitening of the artificial light that is eroding natural nighttime cycles across the continent.” A new report by Sánchez De Miguel and colleagues, published in Science Advances, tallies the effects and trajectory of light pollution across Europe.
  • “In this study, new T1D diagnoses were more likely to occur among pediatric patients with prior COVID-19 than among those with other respiratory infections (or with other encounters with health systems). Respiratory infections have previously been associated with onset of T1D…but this risk was even higher among those with COVID-19 in our study, raising concern for long-term, post–COVID-19 autoimmune complications among youths.” A research paper by Kendall and colleagues, just published in JAMA Network Open, reports on an observational study that found an association between COVID infection and the development of Type 1 diabetes in children.

COMMUNICATION, Health Equity & Policy

Photograph from above and behind a group of seated students at commencement, with the students’ mortarboard hats obscuring their features. Image credit: Joshua Hoehne/Unsplash
Image credit: Joshua Hoehne/Unsplash
  • “Our analyses show universal inequalities in which a small minority of universities supply a large majority of faculty across fields, exacerbated by patterns of attrition and reflecting steep hierarchies of prestige. We identify markedly higher attrition rates among faculty trained outside the United States or employed by their doctoral university. Our results indicate that gains in women’s representation over this decade result from demographic turnover and earlier changes made to hiring, and are unlikely to lead to long-term gender parity in most fields.” An analysis published in Nature by Wapman and colleagues examines the effects of elite institutions on the hiring and retention of faculty at US universities.
  • “The Food and Drug Administration on Tuesday published a list of artificial intelligence tools that should be regulated as medical devices, in some cases appearing to expand its oversight of previously unregulated software products….In a new final guidance for industry, the agency specified that tools designed to warn caregivers of sepsis, a life-threatening complication of infection, should come under regulatory review. STAT News’ Casey Ross reports on a new guidance for industry issued by FDA that offers a look at what kinds of AI applications might be subject to clearance review by the agency.
  • “Our peer-to-peer, multi-modal training prepares student volunteers to become community resource navigators. Student, eager for meaningful clinical experiences, are an untapped resource that can help patients with their social needs.” An article published in Frontiers in Public Health by a group of researchers from Duke University describes a program designed to train student volunteers to act as navigators for community-based health services.