The model used by this application is a fine-tuned EfficientNet-Lite4 Model provided by TensorFlow Hub making it fast to run on mobile devices. I have trained this model for a total of 12789 images from the following datasets linked below achieving a testing accuracy of 97.32%.
For more information you can look at the raw code from '/Machine Learning Files/'.
Here is the confusion matrix that came from the evaluation of the model on the testing dataset.
0 - Covid-19 Detected
The model detects covid-19 infection
1 - Covid-19 Not Detected
The model detects no signs of covid-19 infection
2 - Non-Informative
The model detects an image that cannot be used to make a conclusion.
The A.I. model used to run this application hasn't been tested on 'real world data' but only on open anonymized datasets online. I haven't tested the model rigorously to guarantee it's safety.