Doctor Qs Flashcards
Do you charge patients to use it?
No, our platform is free for users with skin issues.
Are you a corporation? Why should I trust you?
We are a business, and our mission and goals is the same as yours - to help patients get from sick to healthy faster, and find the right medical resources.
How do you make money?
We make money by providing patients with information about relevant medical resources, and are paid for advertising those services.
Who are your competitors?
Google Image Search and WebMD, as these sites provide resources to people with skin issues. They do not do visual analysis and search, so the information they provide to patients is very limited. The word “rash” is used to describe hundreds of diseases.
Have you heard of/talked to Visual Dx?
Yes, very familiar. We’ve spoken with Art before and appreciate what he has been building.
How are you different than Visual Dx?
VisualDx it’s basically a textbook and visual atlas in your phone where you can do text-based searching by symptoms and description. We do the visual analysis based on the image, which replicates how a dermatologist evaluates a skin issue in the office.
How do patients use your app?
Patients can access the app in their phone browser at home, quickly and privately. It is completely anonymous, and points them toward medical resources.
What is your accuracy?
The algorithm is 80% top 5 accurate, which means in 4 of 5 cases we identify the right disease in the top 5 visual matches. This is between the accuracy of a primary care physician and the accuracy of an expert dermatologist. We’re working to improve this through gathering more data. Any suggestions for diagnosed image resources?
Why isn’t the accuracy better?
The accuracy relies on diagnosed images that we train the AI on, just like a dermatology resident needs to see a lot of cases before they can differentiate different diseases with high accuracy. Getting more images to improve this is our top priority.
How do you determine the top 5 diseases?
Our computer is first trained on regular images like cats, birds, and squirrels and learns how to recognize different things. Then, we train it specifically on skin diseases, showing it hundreds to thousands of examples of each disease. The computer generates millions of guesses of what differentiates each condition, and wipes out all of the guesses that aren’t good enough, leaving the final model.
What is your sensitivity and specificity?
We have 80% Top 5 accuracy, which is a measure of sensitivity, as we’re evaluating if we identified it in the top 5 matches. In the future we’re adding in two measures of success, making sure the other 4 of 5 make sense on the differential (specificity), and evaluating high accuracy for the most critical (infectious and precancerous or cancerous) conditions.
What if your algorithm is wrong?
We help direct patients to medical resources so they get the care that they need no matter what the results are.
Who is liable if you are wrong?
Legal issues are really about your labeling and claims that ou make. We have clear labeling about how our tool is not intended for primary diagnosis, but is a visual search tool to enable patients to make more informed choices. We’re not trying to replace the doctor evaluation experience, but rather point patients in the right direction, so they see it’s easy to get the care that they need.
Does likely occurance matter? Why is something very unlikely showing up?
We provide patients with the most visually similar skin conditions, whether or not they are likely, and provide them information on likelihood of the disease so they can have this information. We want to make sure people who actually do have rare diseases are informed of this possibility.
How do you validate the images uploaded to the webapp and make sure they are identified correctly?
We have dermatologists helping us to identify them, and multiple strategies for validating those identities. In the future, we plan to do a study of the accuracy comparing our platform to doctors prospectively.
