Consistent with radiology and pathology, dermatology is another specialty where a lot of the diagnosis comes down to a visual inspection of the skin and doing pattern recognition. This is good fit for AI to assist. Assist is the right term since the diagnosis will need to take other factors into consideration such as patient symptoms, other exam findings, labs, etc. However, given that the skin’s appearance is a big part of the diagnosis, AI can provide a strong assist to the dermatologist. There is significant activity in this area involving google and startup companies focused in developing solutions.
In some initial studies, algorithm outperformed the dermatologists, in both skin lesion images and dermascopic digital images. Esteva et al used a CNN on 129,000 of dermatological lesions to classify whether the lesion was a benign seborrheic keratosis versus a keratinocyte carcinoma or a benign nevus versus a malignant melanoma. This group found that their CNN performed about as well as a panel of 21 board-certified dermatologists. Most skin lesions are diagnosed by PCPs with significantly less expertise so use of algorithm can represent significant improvement in accuracy and detection.
One of the most interesting developments in this area has been the release of google’s tool to assist people in getting an initial idea of what their skin issue may be. Google’s new tool helps identify skin conditions by analyzing images uploaded by people into google’s portal. To use this AI-powered dermatology assist tool, which is a web-based application, users upload three well-lit images of the skin, hair or nail concern from different angles. The tool then asks a series of questions about the user’s skin type, how long they’ve had the issue and other symptoms that help the tool narrow down the possibilities. The AI model analyzes this information and draws from its knowledge of 288 conditions to provide a list of possible matching conditions.
For each matching condition, the tool will show dermatologist-reviewed information and answers to commonly asked questions, along with similar matching images from the web. Users can either save their results, delete them or donate them to Google’s research efforts. Two billion people worldwide suffer from dermatologic issues, but there’s a global shortage of specialists. While many people’s first step involves going to a Google search bar, it can be difficult to describe what you’re seeing on your skin through words alone. It took three years of machine learning research and product development to build the AI-powered dermatologist assistant.
To date, Google has published several peer-reviewed papers that validate the AI model and more are in the works, according to a blog post. In a study published in Nature in 2021, the AI system was shown to be as good as a dermatologist at identifying 26 skin conditions, and more accurate than the primary care physicians and nurses in the study. A recent paper published in JAMA Network Open demonstrated that Google Health’s AI tool may help clinicians diagnose skin conditions more accurately in primary care practices, where most skin diseases are initially evaluated.