Table 2.

List of related methods of COVID-19 detection from CXR images with pros/ cons.

Author	Year	Methods with advantages	Drawbacks
Apostolopoulos & Mpesiana [30]	2020	MobileNet (v2): designed by the concept of residual network for COVID detection	Suffer in less sensitivity and specificity values for Inception, Xception; datasets were imbalanced
Ucar & Korkmaz [31]	2020	SqueezeNet: it was a pre-trained model for COVID detection and achieve good accuracy in augmented data in less computational time.	Achieved poor result (76.3%) in raw data used only 76 COVID CXR images.
Loey et al. [32]	2020	GAN and deep transfer learning solve the overfitting problem used only 307 original images.	Sometimes validation accuracy was higher than the test accuracy due to highly augmented datasets.
Keles et al. [33]	2021	COV19-CNNet and COV19-ResNet: hybrid deep learning model using CNN and ResNet architecture for the detection of COVID-19 from CXR and achieved satisfactory result	They used small number (910) of CXR images in training and model suffers in false positive rate.
Banerjee et al. [34]	2021	COVID-19 detection from audio dataset using residual neural network architecture	Dataset was created with the sound of COVID-19 coughs and non-COVID-19 coughs; accuracy was not acceptable at all.