Back

COVID-19 Detection From Chest Radiographs Using Machine Learning and Convolutional Neural Networks

Li, A. C.; Lee, D. T.; Misquitta, K. K.; Uno, K.; Wald, S.

2020-09-02 health informatics
10.1101/2020.08.31.20175828 medRxiv
Show abstract

Accurate and efficient diagnosis of potential COVID-19 patients is vital in the fight against the current pandemic. However, even the gold-standard COVID-19 test--reverse transcription polymerase chain reaction--suffers from a high false negative rate and a turnaround time of up to one week, preventing the infected from accessing the timely care they require, and impeding efforts to isolate positive cases. To address these shortcomings, this study develops a machine learning model based on the DenseNet-201 deep convolutional neural network, that can classify COVID-19 from chest radiographs in less than one minute and far more accurately than conventional tests (F1-score: 0.96; precision: 0.95; recall: 0.98). It uses a significantly larger dataset and more control classes than previously published models, demonstrating the promise of a machine learning approach for accurate and efficient COVID-19 screening. A live web application of the trained model can be accessed at https://cov2d19-classifier.herokuapp.com/.

Matching journals

The top 6 journals account for 50% of the predicted probability mass.

50% of probability mass above

"Similar papers" are the closest papers from that journal in the model's embedding space. They show what the match is built on, but the ranking comes mostly from a classifier over the whole training set, not from these examples alone.