Underrepresented Tones: Addressing Skin Bias in Medical Imaging for Eczema, Psoriasis, and Melanoma Detection Using CNNs

Medical imaging with convolutional neural networks (CNNs) has the potential to improve early disease detection. However, the success of CNNs relies heavily on the quality and diversity of the training data. Our study focused on training CNNs using datasets for conditions like eczema, psoriasis, and melanoma across various skin tones. We found that Black skin data was underrepresented. This lack of data led to lower accuracy when analyzing African skin samples, highlighting a clear bias in medical imaging. Many believe that just using more data can solve these problems. But our results, along with studies by Romero Lopez et al. (2017) [11] and Yu et al. (2017) [13], show that having a diverse and high-quality dataset is more important. In summary, there’s a need to include data from all skin tones to ensure fairness in medical imaging. This isn’t just a technical problem, but a matter of equal healthcare for all. We hope for a future where medical AI is inclusive and represents everyone fairly.