This Covid-19-classifier is a Deep Learning based image classifier which is able to categorize CT Scans as either COVID-19, PATHOLOGICAL (which groups together MERS, pneumonia and other diseases), or as NORMAL (non-pathological) lungs scans.

Our accuracy is 94.5% (True positive rate of 92%). To the best of our knowledge this in the range of the top-notch Covid-19 lung radiology AI classifiers in the literature. Please see the methodology page for more details how we achieved this high detection rate. We encourage you to play around with it and improve our code. Yes, it is open source! (for the good of humanity).

Note well: as of 2020/4/24 we uploaded our "medium-sized-model" and replaced the old small-model. This new model was trained with 6700 pictures.

The main focus of this work is to push it out as open source quickly so that it can be improved by others. Only collectively (Internet wide) can we beat the pandemic.



Try it out!

Here you can upload an axial lung CT-scan. If you don't have any pictures at hand, feel free to use pictures from our test dataset.


Note: only .jpg or .png files please. DICOM files will be available soon.
Currently, we only accept axial CT-images since they give higher accuracy and have higher sensitivity.


Please note that this software is not (yet) certified for medical diagnosis, nor has this software been used in a clinical study. Certification of medical software is a long and slow process. The urgency of the current Covid-19 crisis however requires immediate action. We will aim for certification, but decided to release the code AS-IS as open source right now, in order to benefit from the global skills of the AI community. Therefore, we CAN NOT claim any responsibility for the functioning or correctness of this software. It is provided AS-IS in the hope that it supports medical practioners, who need to make their own analysis and judgements.
Software, especially AI-based software can only be an additional tool for humans and shall not replace a professional's judgement.

Copyright 2020 (C) by All rights reserved. E-mail.

Made with in Vienna