In a large-scale digital experiment on dermatology diagnosis, we found that specialists and generalists achieved diagnostic accuracy of 38% and 19%, respectively. With decision support from a fair deep learning system, the diagnostic accuracy of physicians improved by more than 33%, but the gap in accuracy of generalists widened across skin tones.
This is a preview of subscription content, log in via an institution to check access.
References
Rajpurkar, P., Chen, E., Banerjee, O. & Topol, E. AI in health and medicine. Nat. Med. 28, 31–38 (2022). This review article covers advances in medical image analysis, problem formulation in human–AI collaboration, and common challenges, such as data scarcity and racial bias.
Liu, Y. et al. A deep learning system for differential diagnosis of skin disease. Nat. Med. 26, 900–908 (2020). This paper demonstrates the potential of AI assistance in supporting general practitioners and nurse practitioners in diagnosing common skin diseases.
Daneshjou, R. et al. Disparities in dermatology AI performance on a diverse curated clinical image set. Sci. Adv. 8, eabq6147 (2022). This paper reports that state-of-the-art dermatology AI models are less accurate on dark skin tones than on light skin tones.
Almaatouq, A. et al. Beyond playing 20 questions with nature: Integrative experiment design in the social and behavioral science. Behav. Brain. Sci. https://doi.org/10.1017/S0140525X22002874 (2022). This paper proposes an integrative experimental design whereby researchers map the design space of possible experiments and test these experiments together to promote commensurability in behavioral science.
Groh, M. et al. Evaluating deep neural networks trained on clinical images in dermatology with the Fitzpatrick 17k dataset. Proc. 2021 IEEE/CVF Conference on Computer Vision and Pattern Recognition Workshops (CVPRW) 1820–1828 (2021). This paper presents a large dataset of clinical images annotated with the Fitzpatrick skin type scale and demonstrates that deep learning classifiers are most accurate on skin tones similar to those it was trained on.
Additional information
Publisher’s note Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.
This is a summary of: Groh, M. et al. Deep learning-aided decision support for diagnosis of skin disease across skin tones. Nat. Med. https://doi.org/10.1038/s41591-023-02728-3 (2024).
Rights and permissions
About this article
Cite this article
Physician–machine partnerships boost diagnostic accuracy, but bias persists. Nat Med 30, 356–357 (2024). https://doi.org/10.1038/s41591-023-02733-6
Published:
Issue Date:
DOI: https://doi.org/10.1038/s41591-023-02733-6
- Springer Nature America, Inc.