AI Facilitates Diagnosis of Skin Conditions By Primary Care Providers Interview with:

Yun Liu, PhD Google Health Palo Alto, California

Dr. Yun Liu

Yun Liu, PhD
Google Health
Palo Alto, California What is the background for this study? Would you describe the system?  Does it use dermatoscopic images?

Response: Dermatologic conditions are extremely common and a leading cause of morbidity worldwide. Due to limited access to dermatologists, patients often first seek help from non-specialists. However, non-specialists have been reported to have lower diagnostic accuracies compared to dermatologists, which may impact the quality of care.

In this study, we built upon prior work published in Nature Medicine, where we developed a computer algorithm (a deep learning system, DLS) to interpret de-identified clinical images of skin conditions and associated medical history (such as whether the patient reported a history of psoriasis). These clinical images are taken using consumer-grade hardware such as point-and-shoot cameras and tablets, which we felt was a more accessible and widely-available device compared to dermatoscopes.

Given such images of the skin condition as input, the DLS outputs a differential diagnosis, which is a rank-ordered list of potential matching skin conditions. In this paper, we worked with user experience researchers to create an artificial intelligence (AI) tool based on this DLS. The tool was designed to provide clinicians with additional information per skin condition prediction, such as textual descriptions, similar-appearing conditions, and the typical clinical workup for the condition.

We then conducted a randomized study where 40 clinicians (20 primary care physicians, 20 nurse practitioners) reviewed over 1,000 cases — with half the cases with the AI-based assistive tool, and half the cases without. For each case, the reference diagnosis was based on a panel of 3 dermatologists. What are the main findings?

Response: Based on our prespecified primary analyses, we found that with AI assistance, both PCPs and NPs significantly improved their diagnostic ability (defined as agreement with the diagnosis from the dermatologist panel), by over 20% on a relative basis.

We also had multiple prespecified secondary analysis, with a quick overview being that the PCPs’ and NPs’ accuracy for biopsy-derived diagnoses increased; rates of desire for biopsies and referrals decreased — all these at a modest cost of an additional 5-7 seconds of median review time per case. Finally, the AI assistance was associated with higher clinician performance across skin types, ranging from pale skin that does not tan to brown skin that rarely burns. What should readers take away from your report?

Response: AI-based assistive tools can help non-specialist clinicians interpret skin conditions more accurately. What recommendations do you have for future research as a result of this work?

Response: Our study was conducted on retrospective cases from a teledermatology service. Prospective studies in primary care settings are likely warranted to better understand the impact on a potentially different patient population. Is there anything else you would like to add?

Response: We are excited about the potential of this technology to help improve the access and quality dermatologic care, and look forward to additional research in this area.

In terms of disclosures, most authors are employees of Google LLC, own Alphabet stock, and are inventors on patents in various stages.


Jain A, Way D, Gupta V, et al. Development and Assessment of an Artificial Intelligence–Based Tool for Skin Condition Diagnosis by Primary Care Physicians and Nurse Practitioners in Teledermatology Practices. JAMA Netw Open. 2021;4(4):e217249. doi:10.1001/jamanetworkopen.2021.7249 



The information on is provided for educational purposes only, and is in no way intended to diagnose, cure, or treat any medical or other condition. Always seek the advice of your physician or other qualified health and ask your doctor any questions you may have regarding a medical condition. In addition to all other limitations and disclaimers in this agreement, service provider and its third party providers disclaim any liability or loss in connection with the content provided on this website.


Last Updated on April 28, 2021 by Marie Benz MD FAAD