Using generative AI to investigate medical imagery models and datasets.

BACKGROUND: AI models have shown promise in performing many medical imaging tasks. However, our ability to explain what signals these models have learned is severely lacking. Explanations are needed in order to increase the trust of doctors in AI-based models, especially in domains where AI prediction capabilities surpass those of humans. Moreover, such explanations could enable novel scientific discovery by uncovering signals in the data that aren't yet known to experts. METHODS: In this paper, we present a workflow for generating hypotheses to understand which visual signals in images are correlated with a classification model's predictions for a given task. This approach leverages an automatic visual explanation algorithm followed by interdisciplinary expert review. We propose the following 4 steps: (i) Train a classifier to perform a given task to assess whether the imagery indeed contains signals relevant to the task; (ii) Train a StyleGAN-based image generator with an architecture that enables guidance by the classifier ("StylEx"); (iii) Automatically detect, extract, and visualize the top visual attributes that the classifier is sensitive towards. For visualization, we independently modify each of these attributes to generate counterfactual visualizations for a set of images (i.e., what the image would look like with the attribute increased or decreased); (iv) Formulate hypotheses for the underlying mechanisms, to stimulate future research. Specifically, present the discovered attributes and corresponding counterfactual visualizations to an interdisciplinary panel of experts so that hypotheses can account for social and structural determinants of health (e.g., whether the attributes correspond to known patho-physiological or socio-cultural phenomena, or could be novel discoveries). FINDINGS: To demonstrate the broad applicability of our approach, we present results on eight prediction tasks across three medical imaging modalities-retinal fundus photographs, external eye photographs, and chest radiographs. We showcase examples where many of the automatically-learned attributes clearly capture clinically known features (e.g., types of cataract, enlarged heart), and demonstrate automatically-learned confounders that arise from factors beyond physiological mechanisms (e.g., chest X-ray underexposure is correlated with the classifier predicting abnormality, and eye makeup is correlated with the classifier predicting low hemoglobin levels). We further show that our method reveals a number of physiologically plausible, previously-unknown attributes based on the literature (e.g., differences in the fundus associated with self-reported sex, which were previously unknown). INTERPRETATION: Our approach enables hypotheses generation via attribute visualizations and has the potential to enable researchers to better understand, improve their assessment, and extract new knowledge from AI-based models, as well as debug and design better datasets. Though not designed to infer causality, importantly, we highlight that attributes generated by our framework can capture phenomena beyond physiology or pathophysiology, reflecting the real world nature of healthcare delivery and socio-cultural factors, and hence interdisciplinary perspectives are critical in these investigations. Finally, we will release code to help researchers train their own StylEx models and analyze their predictive tasks of interest, and use the methodology presented in this paper for responsible interpretation of the revealed attributes. FUNDING: Google.

Assistive AI in Lung Cancer Screening: A Retrospective Multinational Study in the United States and Japan.

"Just Accepted" papers have undergone full peer review and have been accepted for publication in Radiology: Artificial Intelligence. This article will undergo copyediting, layout, and proof review before it is published in its final version. Please note that during production of the final copyedited article, errors may be discovered which could affect the content. Purpose To evaluate the impact of an artificial intelligence (AI) assistant for lung cancer screening (LCS) on multinational clinical workflows. Materials and Methods An AI assistant for LCS was evaluated on two retrospective randomized multireader multicase studies, where 627 (141 cancer positive) low-dose chest CT cases were each read twice (with and without AI assistance) by experienced thoracic radiologists (6 US-based or 6 Japan-based), resulting in a total of 7,524 interpretations. Positive cases were defined as those within two years before a pathology-confirmed lung cancer diagnosis. Negative cases were defined as those without any subsequent cancer diagnosis for at least two years and were enriched for a spectrum of diverse nodules. The studies measured the readers' level of suspicion (LoS, on a 0-100 scale), country-specific screening system scoring categories, and management recommendations. Evaluation metrics included the area under the receiver operating characteristic curve (AUC) for LoS and sensitivity and specificity of recall recommendations. Results With AI assistance, the radiologists' AUC increased by 0.023 (0.70 to 0.72, P = .02) for the US study and by 0.023 (0.93 to 0.96, P = .18) for the Japan study. Scoring system specificity for actionable findings increased 5.5% (57%-63%, P < .001) for the US study and 6.7% (23%-30%, P < .001) for the Japan study. There was no evidence of a difference in corresponding sensitivity between unassisted and AI-assisted reads for the US (67.3%-67.5%, P = .88) and Japan (98%-100%, P > .99) studies. Corresponding standalone AI AUC system performance was 0.75 95% CI [0.70-0.81] and 0.88 95%CI [0.78-0.97] for the US and Japan-based datasets, respectively. Conclusion The concurrent AI interface improved LCS specificity in both US and Japan-based reader studies, meriting further study in additional international screening environments. ©RSNA, 2024.

An intentional approach to managing bias in general purpose embedding models.

Advances in machine learning for health care have brought concerns about bias from the research community; specifically, the introduction, perpetuation, or exacerbation of care disparities. Reinforcing these concerns is the finding that medical images often reveal signals about sensitive attributes in ways that are hard to pinpoint by both algorithms and people. This finding raises a question about how to best design general purpose pretrained embeddings (GPPEs, defined as embeddings meant to support a broad array of use cases) for building downstream models that are free from particular types of bias. The downstream model should be carefully evaluated for bias, and audited and improved as appropriate. However, in our view, well intentioned attempts to prevent the upstream components-GPPEs-from learning sensitive attributes can have unintended consequences on the downstream models. Despite producing a veneer of technical neutrality, the resultant end-to-end system might still be biased or poorly performing. We present reasons, by building on previously published data, to support the reasoning that GPPEs should ideally contain as much information as the original data contain, and highlight the perils of trying to remove sensitive attributes from a GPPE. We also emphasise that downstream prediction models trained for specific tasks and settings, whether developed using GPPEs or not, should be carefully designed and evaluated to avoid bias that makes models vulnerable to issues such as distributional shift. These evaluations should be done by a diverse team, including social scientists, on a diverse cohort representing the full breadth of the patient population for which the final model is intended.

Clinicopathologic features of ecthyma gangrenosum: a single integrated health system cohort.

Ecthyma gangrenosum is an uncommon cutaneous eruption that can initially present with painless macules, which rapidly evolve into necrotic ulcers. This study sought to characterize clinicopathologic features of ecthyma gangrenosum from a single integrated health system. Our cohort consisted of 82 individuals diagnosed with ecthyma gangrenosum. Lesions were most commonly found in the lower extremities (55%) and the truncal region (20%). A wide variety of fungal and bacterial etiologies were found among our cohort. The majority of patients with EG were immunocompromised (79%) and 38% of patients also experienced sepsis. The mortality rate seen in our cohort was approximately 34%. No statistical differences in mortality outcome due to EG related complications were seen between pathogen etiology, and distribution or location of lesions. Patients who were septic or immunocompromised died more frequently than non-septic or immunocompetent patients, suggesting poorer prognosis.

Intralesional cidofovir for the management of refractory cutaneous verrucae: a review of applications and opportunities.

Viral warts - manifestations of cutaneous infection by human papilloma virus - can be a significant physical and emotional burden for patients when common treatments fail, particularly for individuals who are immunocompromised or with multiple lesions. Cidofovir, an antiviral agent typically used for the treatment of cytomegalovirus infection, has emerged as an alternative treatment option for viral warts when administered topically or intralesionally. In this review, we highlight the scientific rationale, published evidence, and practical clinical uses of intralesional cidofovir for the management of cutaneous warts as well as ongoing questions requiring further research and exploration of this emerging therapy for refractory verrucae.

A Qualitative Study Exploring Motivators and Barriers to COVID-19 Vaccine Uptake among Adults in South Africa and Zimbabwe.

While vaccines are a well-established method of controlling the spread of infectious diseases, vaccine hesitancy jeopardizes curbing the spread of COVID-19. Through the Vaccine Information Network (VIN), this study explored barriers and motivators to COVID-19 vaccine uptake. We conducted 18 focus group discussions with male and female community members, stratified by country, age group, and-for Zimbabwe only-by HIV status. Participants' median age across both countries was 40 years (interquartile range of 22-40), and most (65.9%) were female. We conceptualized the key themes within the World Health Organization's Strategic Advisory Group of Experts on Immunization (SAGE) 3C (convenience, confidence, complacency) vaccine hesitancy model. Barriers to vaccine uptake-lack of convenience, low confidence, and high complacency-included inaccessibility of vaccines and vaccination sites, vaccine safety and development concerns, and disbelief in COVID-19's existence. Motivators to vaccine uptake-convenience, confidence, and low complacency-included accessibility of vaccination sites, user-friendly registration processes, trust in governments and vaccines, fear of dying from COVID-19, and knowing someone who had died from or become infected with COVID-19. Overall, vaccine hesitancy in South Africa and Zimbabwe was influenced by inconvenience, a lack of confidence, and high complacency around COVID-19 vaccines.

Patient Perspectives on Clinic Note Transparency Within Dermatology.

This survey study assesses dermatology patient experiences with viewing online medical records and seeks to identify areas for improvement.

Generalized pustular psoriasis: Quality and readability of online health information.

Generalized pustular psoriasis (GPP) is a multisystem disease with potentially life-threatening adverse effects. As patients increasingly seek health information online, and as the landscape for GPP changes, the quality of online health information (OHI) becomes progressively more important. This paper is the first of its kind to examine the quality, comprehensiveness and readability of online health information for GPP. Similar to pre-existing studies evaluating OHI, this paper examines 5 key search terms for GPP- 3 medical and 2 laymen. For each search term, the results were evaluated based on HONcode accreditation, an enhanced DISCERN analysis and a number of readability indices. Of the 500 websites evaluated, 84 (16.8%) were HONcode-accredited. Mean DISCERN scores of all websites were 74.9% and 38.6% for website reliability and treatment sections, respectively, demonstrating key gaps in comprehensiveness and reliability of GPP-specific OHI. Additionally, only 4/100 websites (4%) analysed for readability were written at the NIH-recommended sixth-grade level. Academic websites were significantly more difficult to read than governmental websites. This further exacerbates the patient information gap, particularly for patients with low health literacy, who may already be at higher risk of not receiving timely medical care.

Investigating Attitudes, Motivations and Key Influencers for COVID-19 Vaccination Uptake among Late Adopters in Urban Zimbabwe.

The rapid development of vaccines in response to the COVID-19 pandemic has provided an effective tool for the management of COVID-19. However, in many African countries there has been a poor uptake of COVID-19 vaccines with only 32.5% first vaccine dose coverage compared to the WHO global target of 70%. As vaccine access improves, one of the important drivers of low uptake has been vaccine hesitancy, driven by levels of confidence, convenience, and complacency. Between 4 January-11 February 2022, we conducted a survey of vaccine late adopters to assess factors that influenced adults in Harare, Zimbabwe to present for their first COVID-19 vaccine dose almost 12 months after the vaccination program began. Of the 1016 adults enrolled, 50% were female and 12.4% had HIV co-infection. Binary logistic regression models were developed to understand factors associated with vaccine confidence. Women were more likely to have negative views about the COVID-19 vaccine compared to men (OR 1.51 (95%CI 1.16, 1.97, p = 0.002). Older adults (≥40 years) compared with youth (18-25 years) were more likely to have 'major concerns' about vaccines. When asked about their concerns, 602 (59.3%) considered immediate side effects as a major concern and 520 (52.1%) were concerned about long-term health effects. People living with HIV (PLWH) were more likely to perceive vaccines as safe (OR 1.71 (95%CI: 1.07, 2.74, p = 0.025) and effective (1.68 (95%CI: 1.07, 2.64, p = 0.026). Internet users were less likely to perceive vaccines as safe (OR 0.72 (95% CI: 0.55, 0.95, p = 0.021) compared to non-Internet users; and social media was a more likely source of information for youth and those with higher education. Family members were the primary key influencers for 560 (55.2%) participants. The most important reason for receiving the COVID-19 vaccine for 715 (70.4%) participants was the protection of individual health. Improving vaccine coverage will need targeted communication strategies that address negative perceptions of vaccines and associated safety and effectiveness concerns. Leveraging normative behavior as a social motivator for vaccination will be important, as close social networks are key influences of vaccination.