Table 2.

A quick run-through of the related methods.

Reference	Category	Real-life deployment locations (NA indicates not available)	Important points	Future scopes
Chen et al. [40]	Supervised approach	Renmin Hospital, Wuhan University, China	• It is a deep learning-based approach that is used to determine the presence of COVID-19 by analyzing high-resolution chest CT scans. The focal area is selected by exploiting the UNet++ framework. • Experimental results show that this approach can achieve 95.24% accuracy that is quite remarkable. • Apart from this, this work also achieves 100% sensitivity, and 93.55% specificity that shows the effectiveness of this work.	• Some variants of the UNet++ model can be applied to test the performance. • This approach can be applied and evaluated to different types of images.
Wang et al. [41]	Supervised approach	NA	• This work exploits the advantages of the modified transfer learning to analyze CT scan images and bring out some useful features. • This approach can work with an accuracy of 79.3%, a specificity of 83.00%, and a sensitivity of 67.00%.	• Some hybrid transfer learning approaches can be evaluated and compared with this approach. • Different set of images can be used to check the performance of the proposed approach. • Attempts can be made to improve the accuracy and other parameters.
Butt et al. [42]	Supervised approach	NA	• A three-dimensional CNN-based segmentation method is developed for effective segmentation of the infected areas of the chest using the CT images. • This work uses multiple CNNs for classification purposes. • The probability of the infection is computed with the help of the Bayesian function. • This approach can work with 98.2% sensitivity and 92.2% specificity.	• Attempts can be made to improve the feature representation. • Various image datasets can be explored. • Different functions can be tested that replace the Bayesian function.
Xu et al. [43]	Supervised approach	NA	• Two 3D CNN-based classification methods are hybridized and a novel classification approach is presented. • This approach is applied and tested to classify three different types of lung infections namely COVID-19, Influenza, and not any specific infection category. • It can perform with an overall accuracy of 86.7%.	• Attempts can be made to improve the accuracy of the system. • This approach can be further enhanced to mine fine features from the biomedical images to effectively classify different other classes. • This approach can be designed to solve multi-class classification problem and can be tested for different types of images and also for different classes of images.
Jin et al. [44]	Supervised approach	16 different hospitals in China	• A Transfer learning-supported framework is designed that can effectively analyze CT images. • This approach exploits the ResNet-50 as the underlying model. • The segmentation work is performed with the help of the 3D UNet++ model. • This approach can achieve a sensitivity of 97.4% and a specificity of 92.2%.	• Same approach can be exploited and evaluated after incorporating various other standard models. • Some other models can also be taken under consideration to perform the segmentation task. • This approach can also be extended to work on different datasets.
Wang et al. [45]	Weakly-supervised approach	NA	• This is a lung lesion detection approach that analyzes the chest CT chest using a weakly supervised model. • This approach uses a UNet architecture as a backbone segmentation model. • The possibility of the COVID-19 infection is analyzed with a 3D deep neural architecture.	• The proposed approach can also be used to develop a new segmentation model by incorporating different architectures and segmentation backbone. • This problem can be extended to solve the multi-class classification problem. • This approach some modifications of this approach can be applied to the biomedical images of different modalities.
Mohammed et al. [46]	Weakly-supervised approach	NA	• It is weakly-supervised that can perform the segmentation job with the help of the segmentation mask. • This method is known as ResNext+ and it can extract spatial features using spatial and channel attention. • This method proves to be efficient enough and can achieve a precision rate 81.9% and F1 score value 81.4%.	• This problem can also be extended to the multiclass classification problem. • This approach can also be applied to some benchmark image segmentation datasets.
Laradji et al. [47]	Weakly-supervised approach	NA	• This work is a feebly regulated methodology that utilizes a few focuses to check the tainted regions that can viably discover the depictions naturally. • This plan is known as point checking and it ends up being sufficiently productive to be applied, in actuality, applications.	• The point checking method may incorporate gradient information efficiently. • This approach can be further hybridized with some metaheuristic approaches.
Laradji et al. [48]	Weakly-supervised	NA	• This is a feebly regulated methodology that can mark the CT pictures successfully and in a timebound style. • This work is essentially a functioning learning approach. • By investing 7% annotation effort, this methodology can perform with 90% productivity contrasted with the totally clarified dataset.	• This approach can be further extended to reduce the existing annotation overheads. • Heterogeneous datasets may be closely investigated to apply this approach in interdisciplinary applications.
Gozes et al. [49]	Supervised approach	NA	• This methodology can robotize the investigation interaction of the chest CT pictures utilizing a 2D profound CNN. • The Resnet-50 model is utilized as the foundation of the characterization model while the U-net engineering manages the division issue. • This technique can work with 98.2% sensitivity and 92.2% specificity.	• This approach can be further investigated and may be deployed in real-life robotic disease investigation. • The foundation model can be further updated to enhance the quality of the investigation. • The effectiveness of this approach can be further investigated on some standard image datasets.
Chakraborty et al. [50]	Unsupervised approach	NA	• A morphological recreation activity-based chest CT picture division approach is proposed in this work. • The edge content-based organized network approach is utilized to decide the ideal organizing components. • The trial results show that this methodology beats many cutting-edge approaches as far as some standard assessment measurements. • On average this approach achieves 307.1888625 MSE value, 23.7246505 PSNR value, and 0.831718459 SSIM value.	• This approach can be further extended to compare the obtained results with some other standard approaches like dice coefficient, accuracy etc. so that the performance of this approach can be assessed with respect to some manually delineated images. • The edge content-based network can be further modified to incorporate point-wise region information.
Han et al. [51]	Semi-Supervised approach	NA	• This methodology is a semi-supervised methodology that analyzes the COVID-19 infection from the chest CT examine pictures. • This methodology utilizes both marked and unlabeled information to improve exactness. • This methodology can successfully separate between normal pneumonia and COVID-19 contamination with 97.32% accuracy, 0.9971 sensitivity, 0.9598 specificity, and 0.9326 PPV. • The comparative outcomes prove its superiority.	• This approach can be applied to different modalities of the biomedical images. • Multi-class classification problem can also be addressed. • Dependency on the marked label can be reduced