The Winner | SmartTest

It optimises the distribution of tests among patients with different infection risk, so we can test 2.5x more people with the same amount of tests by grouping multiple samples.

Project description:

One of the breakthrough ideas to deal with this problem is to mix multiple samples from different patients and to use one test on them. It is possible to put mixed samples from 10 people into one test. When the test is positive, the diagnosticians test each case separately to find the infected patient(s). But if the test is negative, this means that all patients are healthy and there was only 1 test instead of 10! Nevertheless, it is not always optimal to mix 10 samples, as this can easily lead to a decrease in performance. That approach was already described by Israeli and German scientists with good results.

Our application is designed for laboratories. Firstly, the diagnostician fills the questionnaire in for each patient. With a simple checkbox list, he marks if the patient has various symptoms if he was abroad or had contact with someone infected. Then, the app calculates the probability of infection for each case. Initially, the weights are set programmatically, but eventually, we will use Machine Learning solutions to cross-reference positive cases with the questionnaire answers to provide even better results.

The result of our application is the best distribution of samples into buckets, minimizing the required number of tests for the whole batch. Using our approach laboratories can increase the throughput to 250% or more, while using the existing equipment, only by changing testing procedures.

Team name: Covictory
Project name: SmartTest

Team members: Maciej Wróbel, Łukasz Strzałka, Tomasz Zmarz, Robert Migas-Mazur