The use of artificial intelligence could be useful in diagnosis but it must be trained using these pictures and so researchers have developed a new data set for dark skin tones.
Demand is high, the lack of dermatologists acute: The availability of dermatology specialists is even worse than the WHO recommended minimum in many countries in Africa; the ratio is approximately one specialist for every one million people.
There are suggestions that the presence of such specialists or rather the lack of them, is much more apparent in rural Africa, where up to 87% of children are affected by skin diseases that are not treated.
The PASSION project (abbreviation for Pediatric AI Skin Support in Outreach Nations) has been set up with the aim of remedying this problem: Working with the Malawian, Madagascan, and Guinean dermatologists, the team of the University of Basel, headed by Professor Alexander Navarini, laid a basis for the application of artificial intelligence in dermatological diagnostics in these countries.
They are showcasing the project during the MICCAI 2024 conference, which stands for International Conference on Medical Image Computing and Computer Assisted Intervention in Marrakesh.
Absence of pictures with black skin complexion For instance, if AI is to identify eczema from a photo it needs to firstly learn what eczema is by analysing hundreds of photos.
However, the current images are mainly of skin disorders in light skinned people and these have been used as evidence by clinics in Europe and the U.S. The scarcity of medical professionals in a number of countries in Africa also implies that there are not many pictures of skin issues in people of colour available.
It also means that if an AI program is trained on the photos of light skin tones, such programs may be considerably less effective at detecting the changes in darker skin types.
These images were captured by dermatologists across the Malagasy Republic, Malawi, and Republic of Guinea from 2020 to 2023 with patient permitting permission.
The images were annotated to contain the necessary diagnosis as well as information on age, gender and the body part of the corpus from which the image was taken before it was incorporated in the anonymized database.
We are at the moment implementing it gradually as part of a validation study in Madagascar. We plan to release the new diagnostic tool with scientific oversight once diagnostic accuracy is above 80 percent,” said Philippe Gottfrois, a doctoral student in Navarini’s research group and lead author of the paper.
In the further stages, the creators intend to increase the database and use more representative image material, including diseases such as neglected tropical skin diseases. They expect that AI will help to reduce a significant shortage of dermatological services in these areas.
Reference: University of Basel. AI-supported dermatology for darker skin tones, thanks to new data setÂ
