Skin color in dermatology textbooks: An updated evaluation and analysis.
About: This article is published in Journal of The American Academy of Dermatology.The article was published on 2021-01-01. It has received 98 citations till now. The article focuses on the topics: Textbooks as Topic.
Citations
More filters
Posted Content•
TL;DR: This work shows the need to explicitly account for underspecification in modeling pipelines that are intended for real-world deployment in any domain, and shows that this problem appears in a wide variety of practical ML pipelines.
Abstract: ML models often exhibit unexpectedly poor behavior when they are deployed in real-world domains. We identify underspecification as a key reason for these failures. An ML pipeline is underspecified when it can return many predictors with equivalently strong held-out performance in the training domain. Underspecification is common in modern ML pipelines, such as those based on deep learning. Predictors returned by underspecified pipelines are often treated as equivalent based on their training domain performance, but we show here that such predictors can behave very differently in deployment domains. This ambiguity can lead to instability and poor model behavior in practice, and is a distinct failure mode from previously identified issues arising from structural mismatch between training and deployment domains. We show that this problem appears in a wide variety of practical ML pipelines, using examples from computer vision, medical imaging, natural language processing, clinical risk prediction based on electronic health records, and medical genomics. Our results show the need to explicitly account for underspecification in modeling pipelines that are intended for real-world deployment in any domain.
374 citations
Cites background from "Skin color in dermatology textbooks..."
...In dermatology in particular, differences between the presentation of skin conditions across skin types has been linked to disparities in care (Adelekun et al., 2020)....
[...]
01 Jun 2021
TL;DR: In this paper, the authors trained a deep neural network model to classify 114 skin conditions and found that the model was most accurate on skin types similar to those it was trained on, and evaluated how an algorithmic approach to identifying skin tones, individual typology angle, compares with Fitzpatrick skin type labels annotated by a team of human labelers.
Abstract: How does the accuracy of deep neural network models trained to classify clinical images of skin conditions vary across skin color? While recent studies demonstrate computer vision models can serve as a useful decision support tool in healthcare and provide dermatologist-level classification on a number of specific tasks, darker skin is under-represented in the data. Most publicly available data sets do not include Fitzpatrick skin type labels. We annotate 16,577 clinical images sourced from two dermatology atlases with Fitzpatrick skin type labels and open-source these annotations. Based on these labels, we find that there are significantly more images of light skin types than dark skin types in this dataset. We train a deep neural network model to classify 114 skin conditions and find that the model is most accurate on skin types similar to those it was trained on. In addition, we evaluate how an algorithmic approach to identifying skin tones, individual typology angle, compares with Fitzpatrick skin type labels annotated by a team of human labelers.
65 citations
TL;DR: The Diverse Dermatology Images (DDI) dataset is created—the first publicly available, expertly curated, and pathologically confirmed image dataset with diverse skin tones and identifies important weaknesses and biases in dermatology AI that should be addressed for reliable application to diverse patients and diseases.
Abstract: An estimated 3 billion people lack access to dermatological care globally. Artificial intelligence (AI) may aid in triaging skin diseases and identifying malignancies. However, most AI models have not been assessed on images of diverse skin tones or uncommon diseases. Thus, we created the Diverse Dermatology Images (DDI) dataset—the first publicly available, expertly curated, and pathologically confirmed image dataset with diverse skin tones. We show that state-of-the-art dermatology AI models exhibit substantial limitations on the DDI dataset, particularly on dark skin tones and uncommon diseases. We find that dermatologists, who often label AI datasets, also perform worse on images of dark skin tones and uncommon diseases. Fine-tuning AI models on the DDI images closes the performance gap between light and dark skin tones. These findings identify important weaknesses and biases in dermatology AI that should be addressed for reliable application to diverse patients and diseases.
35 citations
TL;DR: The first-ever prespecified criteria that allow for the assessment of diversity in the dermatologic literature is developed and could be used by journal editors to include at least 15% SOC-relevant articles in each issue.
21 citations
References
More filters
291 citations
TL;DR: Whether the race and skin tone depicted in images in textbooks assigned at top medical schools reflects the diversity of the U.S. population is considered, which suggests that racial inequities are embedded in the curricular edification of physicians and patients.
125 citations
TL;DR: Overall, the coverage of dark skin at national meetings and in photographs in the major dermatology resources is limited and variable and more consistent photographic coverage and textual information describing common and serious skin diseases in people of color should be incorporated into educational resources.
Abstract: Patients with dark skin can present with morphologic variants, subtle disease presentations, and disease manifestations requiring unique management and therapies. With African Americans, Asians, and Hispanic Americans becoming a significant portion of the population, dermatologists must be able to diagnose and manage skin conditions in people of color. In this study, core dermatology educational sources were examined to determine if they provide dermatologists and trainees with the knowledge base necessary to diagnose and treat skin disease in the ethnic patient. Overall, the coverage of dark skin at national meetings and in photographs in the major dermatology resources is limited and variable. More consistent photographic coverage and textual information describing common and serious skin diseases in people of color should be incorporated into educational resources.
109 citations
TL;DR: Imbalance in the depiction of skin of colour in teaching images can have deeper and broader effects than simply missed educational opportunities, and can affect the integrity of the field.
Abstract: Healthcare disparities are regrettably familiar: black people are 50% more likely to die from heart attacks or stroke than white people; residents of rural areas of the U.S.A. have higher prevalence of chronic obstructive pulmonary disease and Mexican-American adults with hypertension are less likely to have controlled blood pressure. An especially troubling healthcare disparity for our specialty is dermatologists’ suboptimal familiarity with diagnosing skin disease in skin of colour. In one study, 47% of dermatologists felt that their training was inadequate to diagnose skin disease in skin of colour. One of the authors (J.C.L.), witnessed this situation leading to suboptimal care – a patient with a rash that was subsequently diagnosed as toxic epidermal necrolysis waited in the emergency room for several hours because the ‘characteristic’ redness that dermatologists seek to make the diagnosis can be subtle in skin of colour. Unfamiliarity with darker skin may have contributed to the delay in diagnosis and treatment for this patient. This example of care disparity is arresting but not rare. Many dermatologists can recall similar situations where visual diagnosis was debated or delayed, until a biopsy revealed a common disorder that presented in a way that was not ‘classic’ because of the patient’s darker skin. Why might dermatologists feel ill-equipped to diagnose certain skin diseases in persons of colour? Past studies have documented that teaching images disproportionately depict white skin. To learn whether this disparity still exists, we trained three independent image reviewers who categorized photos from two common textbooks and a frequently used teaching set. Of the 5026 images we reviewed, the proportion of images depicting skin of colour was estimated to be 22–32% in textbooks and 21–38% in the teaching set. However, for images of sexually transmitted infections (STIs), the proportion of skin of colour varied from 47% to 58%, compared with 28% for images of infections that were not STIs. Thus, the depiction of skin of colour in our teaching images remains imbalanced. Similarly, in one survey, only 25 4% of dermatology trainees and 19 5% of programme directors of dermatology residencies approved by the Accreditation Council for Graduate Medical Education reported having lectures by an expert that were specific to skin of colour. These discrepancies, as noted by others, can affect the quality of our care by failing to expose physicians in our specialty to a diverse range of clinical appearances. We are not teaching (and possibly not learning) skin of colour. However, we believe that imbalance in the depiction of skin of colour in teaching images can have deeper and broader effects than simply missed educational opportunities. Firstly, suboptimal comfort with diagnosing and caring for patients of colour can affect the physician–patient relationship. Within communities of colour, there is a legacy of mistrust in the medical system, which we strengthen when we are unsure. If we are not confident in our diagnostic abilities in skin of colour, we may hesitate when faced with common diagnoses. Uncertainty in clinical diagnosis can contribute to disparities in powerful but subtle ways: if there is a dearth of taught material, then stereotypes may dominate. In this way, our lack of diversity in images and over-representation of certain diseases can contribute to disparities. Secondly, the lack of visual representation of all types of skin in our common teaching texts can affect our trainees. For trainees of colour it can be conflicting to work in a system in which their race is overlooked or presented in a distorted way. This conflict could contribute to decreased job satisfaction and burnout, precisely at a time when leaders are calling for more diversity in our field. Finally, this imbalance in conventional dermatology images can affect the integrity of our field. The content of our teaching materials reflects what we value. Without balanced representation of skin of colour in our teaching, our specialty has a narrowed scope of practice and limited impact. An important step forward is naming this disparity and its consequences as a more serious issue than a lost educational opportunity. Most diagnoses should have representative photos in a broad spectrum of skin colours. We should consider over-representation of skin of colour, as we often lack the opportunity to see actual patients, making teaching images more important. There are full textbooks and digital resources (https://www.visualdx.com/visualdx/7/) with images of skin of colour, so these photos already exist. We should involve more experts in skin of colour in the authorship of commonly used textbooks. To expand our databases of existing photos, we should also actively photograph common dermatoses in skin of colour. By 2060, 47% of Americans will have nonwhite skin. The visual images from which we teach and learn must reflect this
85 citations