|
No | Applications | Type of data | Challenges | AI methods | Sources |
|
1 | COVID-19 early detection using radiology images. Typically CXR and CT images | CXR images | Limited availability of annotated medical images and medical image classification remains the biggest challenge in medical diagnosis. | DeTraC deep convolution neural network | [14] |
2 | CXR images | Finding optimal parameters for the SVM classifier can be seen as a challenge. Finding optimal parameters for the SVM classifier can be seen as a challenge. Finding optimal values for the relief algorithm can be seen as another limitation of the study | COVIDetectioNet | [92] |
3 | CT images | Redundant data such as interferential vessels can be misdiagnosed as pathology. Radiologists have difficulty differentiating COVID-19 and other atypical and viral pneumonia diseases, which are the same in CT imagery and have similar symptoms. | AlexNet, VGG-16, VGG-19, SqueezeNet, GoogleNet, MobileNet-V2, ResNet-18, ResNet-50, ResNet-101, Xception | [93] |
4 | CXR images | Due to the sudden existence and infectious nature of COVID-19, systematic collection of the extensive data set for CNN training is formidable. Biomarkers found in the CXR images can be misleading. | Patch-based convolutional network | [68] |
5 | CXR images | The research is dealing with images taken directly from patients with severe COVID-19 or some form of pneumonia. However, in the real world, more people are unaffected by pneumonia. The limited number of data available provides a limitation to provide feasible results. | Multiclass classification and hierarchical classification, using texture descriptors and also pretrained CNN model | [94] |
6 | CXR images | Insufficient pulmonary diseases data limit us to conduct verification techniques. | Localise the areas in CXR symptomatic of the COVID-19 presence | [95] |
7 | CT images | Shortage of radiology image labelled “data” | Segmentation deep network (Inf-net) | [96] |
|