DermAI

A "DermAI" dashboard featuring a lesion image input, patient data, and a detailed differential diagnosis report.
A picture of 2 moles over time and the AI analysis of it
A table showing a table of secure refferals

Inspiration

In a clinical setting, misdiagnoses of skin conditions occur frequently and can sometimes leads to late diagnoses for malignant diseases. According to a study conducted from January 2015 to December 2016, out of the 114 diseases recorded from 769 patients, 55 of these diseases were misdiagnosed (Onsoi et al. 2019). With the rise of implementing AI in a clinical setting, we were inspired to create a tool that specifically medical professionals could use to assist them in their diagnosis process as well as shine a light upon possibilities or characteristics of the symptoms they have missed. Overall, we wanted to create a tool that reduces the number of misdiagnoses for a variety of skin diseases, and provide patients with the necessary care they need before the condition intensifies.

References

Onsoi W, Chaiyarit J, Techasatian L. 2019. Common misdiagnoses and prevalence of dermatological disorders at a pediatric tertiary care center. Journal of International Medical Research. 48(2):030006051987349. doi:https://doi.org/10.1177/0300060519873490.

What it does

DermAI's main function is to analyze a photo of the skin area of interest, and analyzes the image as well as basic patient information to generate possible diagnoses of the patient's physical symptoms. The AI tool then suggests possible treatments for the diagnoses, the intensity level of the condition and makes a list of referrals that the doctor can provide to the patient if necessary.

How we built it

We built this application using python, react and javascript. We did research on possible datasets to use (we chose HAM10000), and we use Gemini AI throughout the building process as well as implemented its features throughout our project.

Accomplishments that we're proud of

We are very proud of our project as a whole, especially its accuracy in determining skin conditions.

What we learned

From this experience building our project, we learned about the various types of skin lesions especially the ones presented in HAM10000. In addition, we learned how to integrate Gemini AI throughout our entire project.