How does the AI machine learning work to detect COVID-19 and other health issues?
It is very similar to flashcards and how young children learn to distinguish a cat from a dog and a dog from a pig. Teachers show images of each animal and audibly say what each animal is on the cards. With repetition children begin to recognize the shape and patterns of a cat, dog and pig. iDetect’s machine learning system utilizes a similar teaching method. We feed the neural network eye images and tell it if the patient is COVID-19 positive or negative based on PCR tests. The system puts the positives in one bucket and the negatives in another bucket. As we feed the system more eyes it becomes more, and more accurate. When you submit your eye for a test the AI software looks at the signatures on your eye and compares them to those signatures in the positive and negative buckets until it finds a match. iDetect is currently at 90%+ accuracy with only 400 + COVID positive eyes tested.
How long does it take to get the results of a test?
From the time you submit your image it's just 3 to 6 seconds
iDetect is currently not FDA approved. We are going through the process to get the approval done. As a non-invasive device we hope they will grant us approval soon. Hey, if you work for the FDA please help us out.
Can you test for other diseases?
We believe we can but we would need to utilize the same machine learning format to develop tests for other health issues i.e. diabetes, heart disease, pregnancy, concussions, etc.
How many tests can I take each month?
You can take as many tests as you want. The CDC recommends being tested for COVID-19 once per week but there is no limit.
Currently not for sale in the USA until we get clearance from the FDA. The anticipated pricing will be less than a cup of coffee.
Can you detect the different variants?
In order for us to have data on variants we would have to train the ML (machine learning) on variants and that means the PCR test would need to identify the variant so we could tag the images and then train the ML model.
Have you experienced false positives from other vascular diseases like diabetes?
No machine learning algorithm can explain itself in regards as to why a false positive/negative occurred, which is also the case with PCR testing. From the data we have seen regarding PCR testing false positives are generally caused due to sampling corruption. As of now we have a very low false negative rate and a higher false positive rate simply due to the limited sample set of eye images available to train our ML model.
Do we know if the threshold time for detection is similar to a PCR test?
As of now our threshold would likely be the same as PCR because we are mapping our detection directly to PCR testing.