Page 159 - AI for Good Innovate for Impact
P. 159
AI for Good Innovate for Impact
Use Case 22: EquiDermAI: An Inclusive Deep-Generative
Framework to Overcome Racial & Regional Barriers in Skin Disease
Diagnosis� 4.1-Healthcare
Country: United States
Organization: Indian Institute of Technology, Madras
Contact Person:
Akshat Santhana Gopalan, akshatsg@ outlook .com
Narayanan Madaboosi Srinivasan, narayananms@ smail .iitm .ac .in
1 Use case summary table
Item Detail
Category Healthcare
The problem to be In today’s world, AI and computer vision-based diagnosis of dermato-
addressed logical diseases are playing a key role in the healthcare industry. With a
rapidly increasing number of medical institutions relying on such intelli-
gent models for skin disease diagnosis, it is vital to ensure their accuracy
is maintained across different populations.
However, a pivotal problem arises where a bias in the training data in
terms of the skin colour of the training instances prevents consistent
accuracy across all skin tones. Most dermatological training datasets
are heavily skewed to contain a majority of white skin tones, and models
trained on such datasets perform poorly on diagnosing other darker,
underrepresented skin tones. Black patients with psoriasis receive
advanced treatments less often (8.3% vs. 13.3% for white patients). Mela-
noma survival rates starkly differ: 93% for white adults, but only 71% for
Black individuals (American Cancer Society, 2023). Research on top skin
conditions for people of colour is alarmingly scarce (Takeshita et al). The
digital divide exacerbates these disparities.
Many underdeveloped regions in the world do not have access to power-
ful diagnostic tools. About 32% of the global population was said to
have no internet access in 2024, displaying a clear digital divide and
inaccessibility to such diagnostic tools (International Telecommunication
Union, 2020). The current scenario thus has prominent racial and digital
divisions around the world in the field of dermatological healthcare.
123
