Table 4 Characteristics of studies that developed machine learning-based algorithms and tools for COVID-19 diagnosis or estimation of disease severity based solely on chest imaging (n = 65).

Characteristic	n	%
Purpose^a
Image segmentation only	2	3
Detection of COVID-19	56	86
Estimation of COVID-19 severity	9	14
Type of chest imaging^a
CT	35	54
X-ray	32	49
Data sources
Private medical imaging data	23	35
Publicly available data	42	65
Machine learning models^a,b
Convolutional neural network	58	89
Support vector machine	9	14
Other^c	11	17
Proprietary	2	3
Publication status^d
Peer-reviewed	7	11
Preprint	58	89

COVID-19 coronavirus disease 2019, CT computed tomography.
^aTotals add to more than 65 and percentages add to more than 100; some studies cover multiple categories.
^bSix studies also covered the Prognosis of Illness and Response to Treatment use case; in these cases, some of the models may have been applied for prediction of disease severity, response to treatment, or death.
^cAdaptive boosting (3); autoencoder (2); decision tree (6); explainable deep neural network (1); fuzzy neural network (1); gradient boosted trees (2); k-nearest neighbors (3); multi-layer perceptron (4); random forest (5); unspecified neural network (1).
^dAt the time of database searches (May 4, 2020).

Quick links

Search