Table 3.
Comparison between AI diagnosis algorithms based on X-ray for COVID-19 patients.
| Ref. | Year | Method | Task | Dataset | Evaluation Measures | ||
|---|---|---|---|---|---|---|---|
| ACC (%) | P (%) | SN (%) | |||||
| [52] | July 2020 | Multi-image augmented Deep learning | Using both X-ray and CT images to provide binary classification model | 100 cases of COVID-19 and non-COVID-19 | 99.4 for X-ray, 95.3 for CT scans |
95.98 | 94.78 |
| [53] | April 2020 | VGG16, VGG19, ResNet, DenseNet, and InceptionV3 | Evaluate the performance of CNN architecture and transfer learning in the COVID-19 classification process | 1427 X-ray images include (224 COVID-19 + cases, 700 pf pneumonia, and 503 normal cases) | 96.78 | 98.65 | 96.46 |
| [54] | November 2020 | Using SVM (Support Vector Machine), CNN (Conventional Neural Networks), ResNet50, InceptionResNetV2, Xception, VGGNet16 |
Examine the health status of the patient’s lung based on CT scan and X-ray | 5857 Chest X-rays and 767 Chest CTs for COVID-19 positive cases | (84 for X-ray, 75 for CT scan) |
- | - |
| [55] | September 2020 | Machine learning techniques | Multiclass classification | 350 images from confirmed cases, 220 images from suspected cases, and 130 images from normal cases | 67.5 | - | - |
| [56] | May 2020 | Using encoder and decoder for segmentation, then use multilayer perceptron for image classification | Multitask model that includes three main steps: (1) image classification; (2) lesion segmentation; and (3) image reconstruction | 1044 divided as (449 patients with COVID-19, 100 normal cases, 98 patients with lung cancer, and 398 with different pathology kinds | 78 | - | - |
| [57] | April 2020 | COVID-net model: CNN model that trained first on ImageNet dataset then trained in COVIDx dataset | Analyzing patient data, predicting patient risk and hospitalization duration | 13,975 images with many X-ray positive cases from various countries) | 92.4 | 88.3 | - |
| [50] | May 2020 | Detecting features of X-ray image using CNN model then fed into SVM to make COVID-19 classification | Binary classification | Total of 50 images (25 for COVID-19 + 25 for pneumonia) | 95.33 | 95.33 | - |
| [58] | April 2020 | COVID-Xnet model that builds on CNN models such as VGG19 and google MobileNet | Binary classification | Total of 50 images (25 for COVID-19 + 25 for non-COVID-19) | 90 | ||
| [24] | May 2020 | Using a darknet model for classification, YOLO for real-time object detection | Developed binary classification model that differentiates COVID-19 cases from healthy cases | 1125 X-ray images (500 health cases, 125 COVID-19 positive cases, and 500 from pneumonia cases | 98.02 | 95.13 | 95.3 |
| [59] | October 2020 | Deep learning and transfer learning models (ResNet50, inception V3, etc.) | COVID-19 diagnosis using X-ray images | 100 X-ray images (50 COVID-19, 50 non-COVID-19) extracted form Dr. Chohen GitHub repository | 98 | ||
| [60] | March 2020 | Supervised pre-trained based 2D model called DeCOVNET | Diagnostic tool for COVID-19 detection using 3D images | 499 CT images aggregated from 13 December 2019, to 23 January 2020, used for the training process. 131 CT images aggregated from 24 January to 6 February, were used for the testing process |
90.01 | 90.65 | 91.21 |
| [61] | February 2020 | DL model based on relation extraction | Using 3D images to fast diagnose COVID-19 from pneumonia | CT scans images from 88 patients with positive COVID-19, 101 images from patients infected with bacteria pneumonia, and 86 images of healthy cases. | 94.21 | 96.32 | 94.0 |
| [62] | July 2020 | Anomaly detection algorithm with efficient Net | Multiclass classification based on anomaly detection technology | Model firstly trained on 5977 images of viral pneumonia (no COVID-19) cases and 37,393 healthy cases. Then testing on the X-COVID dataset that include106 COVID-19 cases | 72.77 | 71.30 | - |
| [63] | June 2020 | Using different pre-trained models (ResNet, AlexNet, SGDM- SqueezNet) | Using image augmentation in enhancing COVID-19 classification | 423 X-rays of COVID-19 cases, 1485 X-rays of viral pneumonia cases, and 1579 of normal cases | 98.2 | 96.7 | 98.2 |
| [64] | June 2020 | Feature optimization technique with Deep CNN model, known as COVXNet | COVID-19 detection | Viral, normal, and bacterial dataset available at (https://github.com/Perceptron21/CovXNet) (Last access date: 10 February 2021) | 98.1 | 98.5 | 98.9 |
| [65] | May 2020 | Data augmentation and DL classification models | COVID-19 detection | A set of 5232 anterior–posterior (AP) images of children with ages from 1 to 5. It includes 1583 normal cases, 2780 bacterial pneumonia, and 1493 CXRs with COVID-19 |
99.25 | - | - |