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.
November 15, 2024
In this week’s Duke AI Health Friday Roundup: perils of predictive algorithms; soil fungus grooves to white noise; a new angle on Alzheimer disease; dishonest data rampant in online surveys; dispensing with tattooing for radiation therapy; using NLP to extract adverse events in postmarket scenarios; teaching convolutional neural nets to recognize shapes; first-in-human stem cell trials for repairing corneas; going beyond paywalls in ensuring accessibility; much more:
AI, STATISTICS & DATA SCIENCE
- “To the best of our knowledge, our experiment is the first of its type that revealed shape to be the least prioritized feature learned by CNNs. Learning shape information requires the absence of alternative correlated features. To obtain a shape-learning CNN, strong regularization or augmentation would be needed in practice. Furthermore, a sufficiently large receptive field size relative to object size is vital for a segmentation network to learn shape. We also concluded that shape learning is indeed a useful property of a network as it helps with out-of-domain generalization.” A research article by Zhang and Mazurowski, available as a preprint from arXiv, examines how a machine learning application known as a convolutional neural network (CNN) can be prompted to learn to recognize shapes.
- “…automation has also privileged utilitarianism, as it is much more amenable to calculation. Non-utilitarian considerations resist quantification…. In contrast, the veneer of data-driven decision making, even though it hides many normative choices, allows decision makers to reach consensus and to deploy algorithms without endless debate.” In a recent post at the AI Snake Oil blog, Arvind Narayanan, Angelina Wang, Sayash Kapoor, and Solon Barocas drill down on some of the fraught consequences of using predictive algorithms to guide clinical decision-making.
- “…the AI boom of the last 12 years was made possible by three visionaries who pursued unorthodox ideas in the face of widespread criticism. One was Geoffrey Hinton, a University of Toronto computer scientist who spent decades promoting neural networks despite near-universal skepticism. The second was Jensen Huang, the CEO of Nvidia, who recognized early that GPUs could be useful for more than just graphics…The third was Fei-Fei Li. She created an image dataset that seemed ludicrously large to most of her colleagues. But it turned out to be essential for demonstrating the potential of neural networks trained on GPUs.” Ars Technica’s Timothy B. Lee profiles AI scientists Geoffrey Hinton, Jensen Huang, and Fei-Fei Li and examines how their work set the stage for a surge of progress in the development of artificial intelligence applications.
- “The promise of quantum computers lies in their potential to carry out certain calculations much faster than conventional computers. Realizing this promise will require much larger quantum processors than we have today. The biggest devices have just crossed the thousand-qubit mark, but achieving an undeniable advantage over classical computers will likely require tens of thousands, if not millions….But for many quantum algorithms with more obvious commercial applications, like searching databases, solving optimization problems, or powering AI, the speed advantage is more modest.” In an article for MIT Technology Review, Edd Gent considers the emerging argument over whether advances in AI will render further quantum computing capabilities moot.
- “The purpose of this study was to show that AEs that were extracted from text reflect known occurrence frequencies of AEs using EMRs. Although several studies have used NLP to extract AEs from medical text… to the authors’ knowledge, no studies have evaluated extracted AEs as time-to-event outcomes. We found that AEs were significantly detected in all 12 analyses in this study, suggesting that AEs extracted by NLP may be useful for PMS [postmarket surveillance]. A research article published in npj Digital Medicine by Kawazoe and colleagues presents findings from a study that assessed the capacity of natural language processing (NLP) to extract information about adverse events associated with the use of cancer drugs from patient electronic medical records.
BASIC SCIENCE, CLINICAL RESEARCH & PUBLIC HEALTH
- “Fungal soil microbes may get a boost of energy from white noise, according to new research that found the microbes exposed to a particular sound frequency in the lab grew faster. Scientists say they hope the findings…could lead to sonic techniques that spur the growth of microbes that play critical supportive roles in plant microbiomes, helping rejuvenate stressed ecosystems.” Science’s Katherine Bourzac reports on recent research showing that some strains of fungus grow more rapidly when exposed to certain kinds of white noise.
- “The findings suggest that Alzheimer’s treatments are most likely to help early in the disease, and that one strategy might be to protect vulnerable inhibitory neurons…The results also show how scientists’ understanding of Alzheimer’s is being changed by new tools and techniques that can reveal detailed information about millions of individual brain cells.” NPR’s Jon Hamilton reports on new research suggesting that damage from Alzheimer disease may proceed in distinct phases – slowly at first, and then undergoing rapid acceleration.
- “For cancer patients who require radiation therapy, permanent tattoos have long been necessary to help doctors accurately target and deliver treatment. While seemingly insignificant, these permanent spots can remind many patients of their cancer journey every time they see themselves in the mirror. It’s time to put ink and needles in the rear view, because radiotherapy tattoos do more than mar the skin; they can mar the survivorship mindset.” An opinion article at STAT News by William Chun-Ying Chen and Louis Potters examines an emotionally charged medical practice – tattooing cancer patients to guide targeting in radiation therapy – and asks whether the practice is still necessary or desirable.
- “At 52 weeks, secondary measures of efficacy showed that the disease stage had improved, corrected distance visual acuity was enhanced, and corneal opacification had diminished in all treated eyes. Corneal epithelial defects, subjective symptoms, quality-of-life questionnaire scores and corneal neovascularisation mostly improved or were unchanged.” A paper published last week in Lancet by Soma and colleagues describes a first-in-human attempt to use stem cell therapy to restore damaged corneas.
COMMUNICATION, Health Equity & Policy
- “Despite some naysayers, the progress toward a more open ecosystem has been undeniable. However, one significant gap in this march toward openness is the comparative lack of attention to the accessibility of content. Content that is inaccessible is no more open to those who need supportive reading functionality than content that is behind a paywall. Access without accessibility doesn’t serve the goal of providing federally funded resources to the public.” In an essay for Scholarly Kitchen, Todd A. Carpenter shines a light on barriers to accessibility in scholarly publishing that go beyond the mere fact of paywalls (or not).
- “…many researchers undertake online survey research without knowledge of the prevalence and likelihood of experiencing survey questionnaire fraud nor familiarity with measures used to identify fraud once it has occurred. This research note is based on the experience of researchers across four sites who implemented an online survey of families’ experiences with COVID-19 in the U.S. that was subject to substantial fraud. By the end of data collection, over 70% of responses were flagged as fraudulent with duplicate IP addresses and concurrent start/end times representing the most common indicators of fraud observed.” An article published in the International Journal of Social Research Methodology by Nur and colleagues describes ongoing efforts to ameliorate rampant fraud in online survey questionnaires.
- “Few of us in the medical or public health field want or have the time to read long articles. Concise, well-written articles are a joy to read and can have the greatest impact. Lincoln’s Gettysburg address was only 268 words long; the same day, Edward Everett delivered a speech that was 13,607 words in length. We all know which one had the greatest impact and is most remembered. Editors can help authors by pointing out what can be cut, what can be moved to an online supplement, and what can be said in fewer words.” In a viewpoint article for JAMA Pediatrics, Editor in Chief Frederick P. Rivara shares some publishing insights from more than two decades of helming the journal.
- “Our retrospective study of adult, intermediate-acuity ED patients found that African-American women were less likely to be admitted than White men at all 3 EDs regardless of subgroupings by comorbidity score, age, or insurance status (with the only exception being Medicare patients at ED A). Moreover, African-American women trended toward the lowest likelihood of admission compared to all other race-sex groups. In addition, we observed that African-American men also had lower likelihood of admissions compared to White men consistently across EDs within certain subpopulations.” In a research article published in Health Services Insights, Lin and colleagues present findings from a retrospective study of emergency department admissions that found racial and gender differences in rates of admission across 3 emergency departments in the US Southeast.
- “Cardiology care before delivery is intended to lessen that risk, but a study of nearly 30,000 insured patients’ health records found that serious cardiovascular events were more common in Black patients than white patients in the first year after delivery despite receiving this care. This pre-delivery cardiology care — something only 1 in 9 patients with preeclampsia receive — was linked to a 69% lower risk for white patients but not for Black patients.” An article by STAT News’ Elizabeth Cooney describes a trio of recent research projects being presented at this month’s American Heart Association meeting that underscore the persistent presence of disparities in health outcomes – sometimes in spite of efforts to mitigate them.
