Skin Disease is Common
Skin diseases are very common and affect people of all ages, so it’s no wonder that Artificial Intelligence (AI) for dermatology offers so much potential. In the UK, around one quarter of people who visit their GP do so about a skin-related problem. And outside doctors’ surgeries, over 10 billion Google searches are performed each year in relation to skin, hair and nail conditions.
Worldwide, skin cancer is the most common form of cancer. In fact, one in every five Americans will develop skin cancer by the age of seventy, and every day over 9,500 Americans are diagnosed with skin cancer.
AI Uses in Dermatology
The British Association of Dermatologists (B.A.D.) reports that there is a sense of urgency to implement AI in dermatology. AI can be used within skin care to identify possible skin conditions, differentiate between benign and malignant skin lesions (particularly to diagnose melanoma), and to monitor inflammatory skin conditions.
Efficacious AI solutions could significantly help dermatologists in their decision-making processes and diagnoses. AI could also be used to assist non-dermatology experts that regularly see patients with skin care problems (such as general practitioners and nurses).
Within the broader AI industry, there are very well known instances of racial bias. It is therefore absolutely critical to ensure that imaging datasets for dermatological AI include appropriate representation of different skin colours.
Skin cancer is more common among people with lighter skin, but datasets used for skin cancer AI must be racially diverse. Data should also include areas where people of color are disproportionately affected (e.g. palms and soles).
Widespread adoption of smartphones with cameras offers great opportunities for AI in dermatology. In a few seconds, people can upload photos of skin problems to mobile apps. Smartphone apps for dermatological AI may help to identify conditions, inform people about what steps to take, connect people to dermatologists, and may even increase access to medical advice among those with limited medical care.
An example is Google’s AI “Derm Assist,” designed to identify common skin problems. It provides users with a shortlist of possible conditions. However, Derm Assist is not a diagnostic tool. It only guides users towards information that can help them decide their next step.
The downside is that such apps could also provide inaccurate information. It’s imperative therefore that datasets sufficiently represent different skin conditions as well as different demographics (especially ethnicity and age).
The British Association of Dermatologists (B.A.D.) has questioned the effectiveness of currently available apps. A key reason being that some machine learning algorithms are developed with limited data. B.A.D. stresses the need for clinical images presenting a much wider range of skin disorders. What’s more, B.A.D. states that:
“We believe it is essential for AI app developers to develop algorithms within a tightly integrated ecosystem including computer scientists, clinicians and patient organizations and to be aware of the relevant regulatory and ethics governance frameworks for the usage of their products.“
The American Academy of Dermatology (A.A.D.) also highlights the use of clinical images (taken within optimal settings) versus the use of images taken with smartphones, to develop datasets for AI solutions.
gliff.ai’s software enables users to embed knowledge into high-quality, auditable datasets which can be used for building world-changing and trustworthy AI in skin care. Built to the highest standards for medical AI development, gliff.ai is open-source, end-to-end encrypted, and easy to use.
Regarding dermatological AI development, gliff.ai helps users to:
- Curate datasets with 1000s of images – large enough to ensure different skin conditions and demographics are represented
- Label and annotate images so that dermatologists can embed their specialist knowledge into datasets
- Collaborate so that data scientists, skin clinicians and other collaborators can come together in a single platform.
Furthermore, in 2021 gliff.ai developed a tool which enables users to crowdsource images. This involves a web-based app to which photos can be uploaded and added directly to a dataset within gliff.ai’s software. This tool offers AI developers in dermatology an efficient and cost-effective way to source images of skin lesions. Currently this is not included in the gliff.ai subscription, but anyone who is interested in knowing more can contact us directly.