We will ask you to provide differential diagnoses of skin conditions based on their appearances in images. Some images contain sensitive, graphic content that could be disturbing to some viewers. On some but not all images, you can zoom into the details by moving your cursor over the image. Please make sure to spell your differential diagnosis based on the auto-complete selection, so we can correctly keep a record of your differential accuracy.
After you share your differential diagnosis, we will show you an AI model's prediction of the leading diagnosis and ask you to update your differential diagnosis and confidence based on the AI's prediction. If your top diagnosis is the same as the AI model's prediction, we will skip this step.
Please note that the AI model is not perfectly accurate. As such, you should use your experience as a trained clinician to verify the model's predictions and determine whether changing your submission is necessary.
Diagnosing diagnosis is an MIT research project examining the science and art of visual diagnosis. This project is driven by recent research that highlights underrepresentation of skin types in dermatology. We are curious about what makes an image of a skin condition easy or hard to diagnose? Is it just as easy to diagnose skin conditions across skin types? Or, are certain skin types more difficult to diagnose for certain types of conditions? Does this depend on a physician's training? How does clinical experience relate to diagnostic accuracy in this visual task? Can computer vision models support physicians to improve diagnostic accuracy in general and in rare skin types and skin conditions specifically? Alternatively, could second opinions and crowd-sourced diagnoses improve diagnoses and patient outcomes downstream?
For more information, email dermatology-diagnosis@mit.edu.