Table 1.

Recently published AI approaches being undertaken to support clinical decision-making processes in pneumonia.

Reference	Kermany et al. [13] Cell, 2018	Stephen et al. [14] Journal of Healthcare Engineering, 2019	Heckerling et al. [15] Clinical Applications, 2003	Hwang et al. [16] JAMA Network Open, 2019
Main Goal	Detect pneumonia and distinguish viral and bacterial etiology	To handle pneumonia classification	Predict the presence of pneumonia among patients with acute respiratory complaints	Make a deep learning–based algorithm for major thoracic diseases; Comparison with physicians and external validation
Applied Method	Neural network	Neural network and augmentation methods to artificially increase the size and quality of the dataset	Neural networks	Deep learning—neural networks
N°	5232 chest X-ray for training phase and 624 images for test phase	5856 X-ray images—3722 training set and 2134 to the validation set---	1023 patients–training cohort of 907 and a testing cohort of 116	54,221 X-ray with normal finding—41140 with abnormal findings
Results	Detect pneumonia = accuracy of 92.8% Distinguish viral vs bacterial = accuracy of 90.7%	Training accuracy = 0.9531 validation accuracy of 0.9373	Training cohort = sensitivity of 0.842 specificity of 0.593 testing cohort = sensitivity of 0.829 specificity of 0.547	Image-wise classification: in-house = AUROC of 0.965 and external validation = AUROC of 0.979 Lesion-wise localization: in-house = AUAFROC of 0.916 and external validation = AUAFROC of 0.972 -Comparison with physician: DLAD = AUROC 0.983 was higher versus 3 observer groups (p < 0.005)

Abbreviations: AUROC: area under the receiver operating characteristic curve; AUAFROC: area under the alternative free-response receiver operating characteristic curve; DLAD: Deep learning–based automatic detection algorithms.