AI for diagnosis of skin lesions

Our Goal

The goal of our research is to develop a diagnostic model to identify pigmented and non-pigmented skin lesions in all ethnic skin types. An AI model for the diagnosis of skin lesions will augment the work of dermatologists in providing enhanced skincare services across various demographics; particularly in remote under-served communities, and others that are under-represented in specialist dermatological support.

Our Motivation

There are relatively few specialist dermatologists providing skincare services globally. This is particularly acute in countries in sub-Saharan Africa. According to the Nigerian Association of Dermatologists (NAD), there are about 100 dermatologists subserving an estimated population of over 190 million (i.e. 0.053 per 100,000) [1]. Ethiopia currently has a doctor shortage with approximately 30 dermatologists in a nation of over 109 million people (i.e. 0.028 per 100,000) [2]. This trend is similar to numbers seen in other countries across sub-Saharan Africa.

The situation with respect to dermatologist density in the United States is better, but remains suboptimal, with an estimated 3.4 dermatologists per 100,000 people. This is below the recommended rate of 4 per 100,000 to provide relevant skincare services [3]. Canada has about 623 dermatologists for a population of over 35 million (1.78 per 100,000) [4]. An appreciable improvement is observed in some countries in Europe, with Greece, Italy and Germany boasting a ratio of 11.4, 8.2, and 6.5 per 100,000 respectively; this however is way above what is seen in other regional neighbours like Spain (2.9), Netherlands (2.7), and the UK (1) per 100,000 [5].

Furthermore, dermatology training and practice is predominantly country/region specific; often tailored towards the provision of knowledge and skill for the diagnosis and management of locally prevalent dermatoses in respective populations. For instance, dermatologists trained to diagnose and care for Caucasian skin perform sub-par in the diagnosis of dermatoses in other ethnic skin types. This difference in care may be attributed to the limited range of exposure to dermatoses in other skin types, lack of experience, inadequate education to account for the cultural and racial differences from a social and medical point of view [6]. Research reports that doctors who do not understand the differences in white and black skin types provide sub-optimal medical care to African-American patients [7,8].

Our research seeks to mitigate the above challenges by providing medical practitioners, in particular, general medical physicians, with a tool for appropriate diagnosis of skin diseases and satisfactory skincare services for patients; and to ensure prompt referral to consultant dermatologists. In addition, by training the model on a diversity of white and black skin colour types, the model can better understand the differences in appearance of skin lesions across different demographic groups and races, with the potential to augment the work of dermatologists when they face diagnostic challenges in racially disparate skin types for which they lack clinical exposure and/or experience.

Who our Work Impacts

Dermatologists: to augment diagnosis of skin diseases for racially different skin types.
Medical professionals: to provide skincare services to patients in regions under-served by a specialist dermatologist.
Dermatological research: to stimulate research in applying computer vision, AI and machine learning methods for dermatological research. This research will release the dataset into the public domain for reproducibility and use by other researchers.