Using artificial intelligence on dermatology conditions in Uganda: a case for diversity in training data sets for machine learning

Abstract

Background: In pursuit of applying universal non-biased Artificial Intelligence (AI) in healthcare, it is essential that data from different geographies are represented. Objective: To assess the diagnostic performance of an AI-powered dermatological algorithm called Skin Image Search on Fitzpatrick 6 skin type (dark skin) dermatological conditions. Methods: 123 dermatological images selected from a total of 173 images were retrospectively extracted from the electronic database of a Ugandan telehealth company, The Medical Concierge Group (TMCG) after getting their consent. Details of age, gender, and dermatological clinical diagnosis were analysed using R on R studio software to assess the diagnostic accuracy of the AI app along with disease diagnosis and body part. Predictability levels of the AI app were graded on a scale of 0 to 5, where 0-no prediction was made and 1-5 demonstrated a reduction incorrect diagnosis prediction rate of the AI. Results: 76 (62%) of the dermatological images were from females and 47 (38%) from males. Overall diagnostic accuracy of the AI app on black dermatological conditions was low at 17% (21 out of 123 predictable images) compared to 69.9% performance on Caucasian skin type as reported from the training results. There were varying predictability levels correctness i.e., 1-8.9%, 2-2.4%, 3-2.4%, 4-1.6%, 5-1.6% with performance along individual diagnosis highest with dermatitis (80%). Conclusion: There is need for diversity of image datasets used to train dermatology algorithms for AI applications to increase accuracy across skin types and geographies.