A study by MIT researchers has found that doctors are less accurate when diagnosing skin diseases in patients with darker skin, based on images only. The study, which included over 1,000 dermatologists and general practitioners, found that dermatologists accurately diagnosed about 38% of the images they saw, but only 34% of those that showed darker skin. General practitioners showed a similar decrease in accuracy with darker skin.
The researchers also discovered that the use of artificial intelligence (AI) algorithms could improve a doctor’s diagnostic accuracy, although this improvement was greater when diagnosing patients with lighter skin. One reason for this discrepancy could be that dermatology textbooks and training materials predominantly feature lighter skin tones, thereby limiting doctors’ experience and knowledge of skin diseases on darker skin tones.
This study is the first to provide empirical evidence of physician diagnostic disparities based on skin tone. The lead author, Matt Groh, argues that the findings should lead to changes in dermatology education policies. For the study, the researchers compiled 364 images from various sources, representing 46 skin diseases across different skin shades. The participants, who were all doctors, were shown 10 of these images and were asked to provide their top three diagnoses for each image. The results indicated a significant drop in diagnostic accuracy when doctors were faced with darker skin images.
In addition to this, after assessing how doctors performed independently, the researchers also used an AI algorithm that they developed to help with the diagnosis. This algorithm, which had an accuracy rate of 47%, improved the diagnostic accuracy of dermatologists (up to 60%) and general practitioners (up to 47%). However, while the AI aid enhanced the diagnostic accuracy for both groups, it had a larger impact on images of lighter skin than darker skin.
The study indicates the importance of integrating more patient images with darker skin in medical training and materials, and the potential of AI in improving diagnostic accuracy. It provides valuable insights that could guide future improvements in medical training and the development of AI aids in dermatology.
