Science Buddies: "Ask an Expert"

I'm looking to attend and hopefully place at ISEF and JSHS this year. My project involves using a smartphone application I built to take a picture of a patient in order to diagnose a condition of the body that can cause permanent blindness. It's a major health problem in underdeveloped areas of the world (primarily Africa), and has been completely eliminated in most developed countries (including the United States). (I don't want to go into too much detail here for privacy reasons).

The issue that I am encountering now is that it's not possible to test the accuracy of the application in diagnosing this condition on real patients because there are none within the United States. What I have instead chosen to do is to test the success rate of the app in diagnosing images of the condition on infected patients that I found on the websites of the World Health Organization and other major health organizations. In theory, there should be very little difference between the success rate of testing on these images on the Internet compared to testing on actual patients, because my algorithm makes its diagnoses based on the shapes and colors in the image, which will remain constant regardless of whether it's an image on the Internet or whether's it an actual patient.

Is this a major flaw in my project? If so, what can I do to combat it? If it's not, what can I do to ensure judges that the accuracy of the app should be very similar even when testing on patients?

Thanks in advance!

Moderator notes: I've moved this post to the Life sciences where the appropriate experts can see your question and help you. I've also deleted the duplicate post. Please follow the posting guidelines and do not post the same question in more than one forum. That makes it hard for the appropriate expert to notice your question. Thank you.

Thank you for your patience in waiting for a reply!

You ask several questions, but I only feel qualified to try to address just part of your question:
"what can I do to ensure judges that the accuracy of the app should be very similar even when testing on patients?"

First, even if you did have access to real patients, I think that you will need to consider several aspects of the image testing:

(1) the ability of your imaging technique to detect (or miss) real cases, that is, how often does your technique MISS real cases. This is usually referred to as "false negatives." How many images from unique patients do you expect to have access to?

(2) the ability of your imaging technique to distinguish "normal" images from "positives". This is usually referred to as "false positives". How many images from unique persons without the infection do you expect to have access to?

(3) the effect of image quality on the ability of your technique in the above 2 situations.

So, the judges will be looking at the number of "test" cases (#1 above) that you use and, secondly, the number of "control" cases (#2) above.

You might want to review the project guidelines: https://www.sciencebuddies.org/science- ... ience-fair and this: https://www.sciencebuddies.org/science- ... mpetitions, looking specifically at these 2 sections:

a. Experimental Design for Advanced Science Projects

b. Increasing the Ability of an Experiment to Measure an Effect

Please do not hesitate to come back with more questions!