Artificial Intelligence and COVID-19 Using Chest CT Scan and Chest X-ray Images: Machine Learning and Deep Learning Approaches for Diagnosis and Treatment

Objective: To report an overview and update on Artificial Intelligence (AI) and COVID-19 using chest Computed Tomography (CT) scan and chest X-ray images (CXR). Machine Learning and Deep Learning Approaches for Diagnosis and Treatment were identified. Methods: Several electronic datasets were analyzed. The search covered the years from January 2019 to June 2021. The inclusion criteria were studied evaluating the use of AI methods in COVID-19 disease reporting performance results in terms of accuracy or precision or area under Receiver Operating Characteristic (ROC) curve (AUC). Results: Twenty-two studies met the inclusion criteria: 13 papers were based on AI in CXR and 10 based on AI in CT. The summarized mean value of the accuracy and precision of CXR in COVID-19 disease were 93.7% ± 10.0% of standard deviation (range 68.4–99.9%) and 95.7% ± 7.1% of standard deviation (range 83.0–100.0%), respectively. The summarized mean value of the accuracy and specificity of CT in COVID-19 disease were 89.1% ± 7.3% of standard deviation (range 78.0–99.9%) and 94.5 ± 6.4% of standard deviation (range 86.0–100.0%), respectively. No statistically significant difference in summarized accuracy mean value between CXR and CT was observed using the Chi square test (p value > 0.05). Conclusions: Summarized accuracy of the selected papers is high but there was an important variability; however, less in CT studies compared to CXR studies. Nonetheless, AI approaches could be used in the identification of disease clusters, monitoring of cases, prediction of the future outbreaks, mortality risk, COVID-19 diagnosis, and disease management.


Introduction
In December 2019, a large outbreak of a novel coronavirus infection occurred in Wuhan, Hubei Province, China. The novel coronavirus was named severe acute respiratory syndrome coronavirus 2 (SARS-CoV-2) by the International Committee on Taxonomy of Viruses and led to a dramatic pneumonia outbreak in China [1][2][3]. The disease caused by the virus, named coronavirus disease  by the World Health Organization (WHO), can be spread through human-to-human contact. On January 30, 2020, the WHO declared a global public health emergency against the outbreak of COVID-19 [4][5][6].
The COVID-19 diagnosis is confirmed by the positive results of the nucleic acid amplification test of the respiratory tract or blood specimens using reverse transcription real-time fluorescence polymerase chain reaction (RT-PCR) [7,8]. However, methods like chest X-ray (CXR) and chest Computed Tomography (CT) scan are medical imaging techniques, which are widely used to assess the pneumonia due to COVID-19 [9][10][11][12][13][14][15][16][17][18][19]. The reported sensitivity of CXR for COVID-19 pneumonia is relatively low in the early phase of the disease and in mild cases (69%). Conversely, CT shows greater sensitivity for early pneumonic change, disease progression, and alternative diagnosis; the administration of the intravenous contrast medium, is essential for the diagnosis of pulmonary thromboembolism [20][21][22][23][24][25][26][27][28][29][30][31][32][33][34][35][36][37]. Despite recent advances in diagnostic tools, radiologic imaging alone is not sufficient for the COVID-19 pneumonia diagnosis. Imaging should be associated to clinical and laboratory testing. In addition, the American College of Radiology, so as the Italian Society of Radiology (SIRM) does not recommend chest CT as a screening tool, suggesting this method only for symptomatic patients with specific clinical indications. Bilateral distribution of ground glass opacities (GGO) with or without consolidation in posterior and peripheral lungs was the cardinal hallmark of COVID-19 disease [6,22]. Among COVID-19 patients, it is reasonable to assume that those with a very severe disease could exhibit high risk of venous thromboembolism, including deep vein thrombosis and/or pulmonary embolism. In this scenario, it is opened the question on the use of contrast medium during CT studies [20][21][22].
The mathematical models for COVID-19 pandemic, confirmed by practical evidence in China, in Italy, and in the rest of the world, have shown that the rapid substantial increase in the number of critically ill patients exceeds in the total capacity of Intensive care units (ICUs), even excluding routine critical admissions for trauma, stroke, and other emergencies.
Deep learning (DL) and machine learning (ML) are branches of AI that focus on producing systems that can learn from examples and improve without being explicitly programmed. ML is the study of computer algorithms that can improve automatically through experience and using data. Machine learning algorithms build a model based on sample data, known as "training data", to make predictions or decisions without being explicitly programmed to do so. Machine learning algorithms are used in a wide variety of applications, such as in medicine, email filtering, speech recognition, and computer vision, where it is difficult or unfeasible to develop conventional algorithms to perform the needed tasks. A subset of machine learning is closely related to computational statistics, which focus on making predictions using computers; but not all machine learning is statistical learning. The study of mathematical optimization delivers methods, theory and application domains to the field of machine learning. Data mining is a related field of study, focusing on exploratory data analysis through unsupervised learning.
Deep learning is a class of machine learning algorithms that uses multiple layers to progressively extract higher-level features from the raw input. For example, in image processing, lower layers may identify edges, while higher layers may identify the concepts relevant to a human such as digits or letters or faces. Deep-learning architectures such as deep neural networks, deep belief networks, deep reinforcement learning, recurrent neural networks and convolutional neural networks have been applied to fields including computer vision, speech recognition, natural language processing, machine translation, bioinformatics, drug design, medical image analysis, material inspection and board game programs, where they have produced results comparable to and in some cases surpassing human expert performance. DL and ML have been applied successfully in many fields, including health care and medical informatics. One important research direction leverages DL and ML to understand and fight COVID-19. Numerous lines of research have been initiated for the application and development of COVID-19-related DL and ML algorithms.
Several review articles have been published on the use of artificial intelligence approaches in COVID-19 research. Agbehadji et al. [54], summarized how big data and AI models can be used for case detection and contact tracing of COVID-19. Bullock et al. [55], discussed how AI is used to evaluate the challenges of COVID-19 at different scales, including molecular, medical, and epidemiological applications. Naud e [56] highlighted the actual and potential applications of AI in fighting COVID-19. Wu et al. [57], surveyed the application of big data technology for preventing and managing COVID-19 in China.
Alballa et al. [58], review the recent ML algorithms in this field and focus on their potential in two main applications: diagnosis of COVID-19 and prediction of mortality risk and severity, using simple clinical and laboratory data; they analyze the main features that were found to be the most relevant to these applications.
Our aim is to report an overview and update on AI-based methods application in COVID-19 disease using radiological images including CXR and CT focus on their potential in two main applications: diagnosis of COVID-19 and prediction of mortality risk and severity.
This narrative review is the result of an autonomous study without protocol and a registration number.
The search covered the years from January 2019 to June 2021. The reference lists of the found papers were analyzed for papers not indexed in the electronic databases. All titles and abstracts were analyzed. The inclusion criteria were studied evaluating the use of AI methods in COVID-19 disease reporting performance results in terms of accuracy or precision or area under Receiver Operating Characteristic (ROC) curve (AUC). Articles published in the English language were included. Exclusion criteria were different topics, unavailability of full text, and not sufficient data.

Statistical Analysis
The summarized accuracy, precision or specificity were calculated in terms of mean, standard deviation value and range. The Chi square test was used to assess differences statistically significant between CXR and CT results. p value < 0.05 was considered significant for all tests.

Results
We identified 84 potentially relevant references through electronic searches. We identified 15 references through scanning reference lists of the identified paper that we added to the 84 references previously selected (total number of articles was 99). We then excluded 51 clearly irrelevant articles through screening titles and reading abstracts. We excluded 25 articles for the reasons listed in the exclusion criteria. A total of 23 article met the inclusion criteria. A diagram of included and excluded studies was summarized in the study flow diagram (Figure 1). Table 1 reports the classification problem, the classification approach and the performance results of the selected papers.
Thirteen papers using CXR and AI approaches in the COVID-19 disease were identified. The summarized mean value of the accuracy and precision of CRX in COVID-19 disease were 93.7% ± 10.0% of standard deviation (range 68.4-99.9%) and 95.7% ± 7.1% of standard deviation (range 83.0-100.0%), respectively.
Ten papers using chest CT and AI approaches in the COVID-19 disease were found. The summarized mean value of the accuracy and specificity of CT in COVID-19 disease were 89.1% ± 7.3% of standard deviation (range 78.0-99.9%) and 94.5% ± 6.4% of standard deviation (range 86.0-100.0%), respectively.
No statistically significant difference in summarized accuracy mean value between CXR and CT was observed using the Chi square test (p value > 0.05).    CT CT There were 406 clearer COVID-19-positive lung CT images. The marked areas in the mask images are 0-"ground glass opacity," 1-"consolidations," 2-"lungs other," 3-"background.

Discussions
Artificial intelligence approaches have been used to predict the outbreak, to diagnose the disease, to analyze CXR and CT scan images, and more recently to predict mortality or progression risk to severe respiratory failure. This evidence clearly indicates the need for the most rapid and accurate diagnostic and stratification of patients with COVID-19, with technologies and expertise easily accessible from all nodes of the healthcare system with responsibility of diagnosis of COVID-19 and management of patients (either in the health structures or at home) [59].
Chest radiographs are first-line investigations in many countries. Researchers could examine not only the initial imaging findings and extent of respiratory involvement, but also how radiographic progression in serial studies correlates with patients' clinical outcome [60][61][62]66]. CT examination has been used extensively worldwide to evaluate the grade and the extension of the viral pneumonia by COVID-19 and in the followup, which are also based on AI algorithms [67][68][69][70]. Several radiological organizations do not recommend CT as primary diagnostic/screening tool for COVID-19 [71][72][73][74] or have excluded CT findings from its diagnostic criteria [75]. Radiologists focus on main CT findings (GGO, consolidation, reticulation/thickened interlobular septa, nodules), and lesion distribution (left, right or bilateral lungs) [76][77][78][79][80].
AI methods seek to exploit mainly for characterizing COVID-19 pneumonia CT patterns, for monitoring patients in clinical settings and for estimating efficacy of treatment. Based on the data derived from clinical parameters, AI may provide critical data for resource allocation and decision-making by prioritizing the need of ventilators and respiratory supports in the Intensive Care Unit [81][82][83]. AI was used for the COVID-19 disease detection and quantification from CXR and CT images [63,[81][82][83][84][85][86][87][88]. AI can also be used for predicting the chances of recovery or mortality in COVID-19 and to provide daily updates, storage and trend analysis and charting the course of treatment.
At the same time, radiomics approaches can be usefully implemented, focusing on segmentation techniques of the lung parenchyma based on region growing techniques and on other radiomics COVID-19 specific features and their use with machine learning such as Support Vector Machines (SVMs) or Random Forests [86][87][88].

Application on Chest X-Ray Images
In the study of Sethy et al. [83], the deep learning methodology is reported for detection of a coronavirus infected patient by CXR. The suggested classification model, Resnet50 plus Support Vector Machine (SVM), achieved accuracy and false positive rate of 95.38% and 95.52% respectively for detecting COVID-19.
Jiao et al. [89], using the CXR as input to an EfficientNet deep neural network combined with clinical data, assessed the ability to predict COVID-19 disease severity (critical or noncritical). They reported that when CXR was added to clinical data for severity prediction, the area under the receiver operating characteristic curve (ROC-AUC) increased from 0.821 to 0.846 on internal testing and from 0.731 to 0.792 on external testing; when deep-learning features were added to clinical data for progression prediction, the concordance index (C-index) increased from 0.769 to 0.805 on internal testing and from 0.707 to 0.752 on external testing; when image and clinical data were combined C-index increase from 0.805 to 0.781 on internal testing and from 0.752 to 0.715 on and internal testing.
Al-Waisy et al. [90], proposed COVID-CheXNet system that is made by combining the results generated from two different deep learning models (e.g., ResNet34 and HRNet) on CXR: two predicted probability scores are computed, and the highest probability score is used to assign the input image to one of two classes for detecting COVID-19. The proposed COVID-CheXNet system reached to diagnose the COVID-19 patients with a detection accuracy rate of 99.99%, a sensitivity of 99.98%, a specificity of 100% and a precision of 100%. Cases used in this study come from different databases: 200 X-ray images with confirmed COVID-19 infections come by Cohen's GitHub database [91]; 200 COVID-19 CXRs gathered from three different repositories: Radiopaedia dataset [92], Italian Society of Medical and Interventional Radiology (SIRM) [93] and Radiological Society of North America (RSNA) [94]; 400 normal CXR by Kaggle's CXR dataset [95].
Ozturc at al. [97] proposed a model for automatic COVID-19 detection using raw CXR images in order to perform the binary classification COVID-19 versus no findings and multiclass classification COVID-19 versus no findings. Their model produced a classification accuracy of 98.08% for binary classes and 87.02% for multi-class cases. The DarkNet model was used in the study as a classifier implementing 17 convolutional layers and introducing different filtering on each layer.
Du et al. [98], applied machine learning (ML) to the task of detection of SARS-CoV-2 infection using basic laboratory markers. Moreover, they tested ML accuracy adding at laboratory markers the radiologist interpretations of chest radiographs. When they used the combination of laboratory markers and radiologist interpretations, the sensitivity of ML was over 90% while keeping moderate specificity.
Dey et al. [99], proposed a classifier ensemble technique, utilizing Choquet fuzzy integral. It classifies CXR images in common pneumonia, confirmed COVID-19, and healthy lungs. They utilized the pre-trained convolutional neural network models to extract features and classify the CXR images using two dense layers and one softmax layer. The proposed method provides 99.00%, 99.00%, 99.00%, and 99.02% average recall, precision, F-score, and accuracy, respectively.
Alruwaili et al. [100], proposed an enhanced Inception-ResNetV2 deep learning model that can diagnose chest X-ray scans with high accuracy of 99.83%. Besides, a Grad-CAM algorithm is used to enhance the visualization of the infected regions of the lungs in CXR images.
Bukhari et al. [101], employed ResNet50 for COVID-19 detection using CXR images. They tried to differentiate four types of classes, which are healthy normal, bacterial pneumonia, viral pneumonia, and COVID-19 cases. They achieved an average accuracy of 98.18% and a F1-score of 98.19%.
Khan et al. [102], proposed a model named CoroNet to identify COVID-19 in x-ray and CT scans utilizing a pretrained Xception convolution network. For the four classes (viral pneumonia, COVID-19, bacterial pneumonia, and normal), the first experiment attained an accuracy of 89.6%, while for three classes (normal, COVID-19, and pneumonia) obtained a total accuracy of 95.0%.
A COVIDX-Net model to help radiologists in identifying and diagnosing COVID-19 in CXR images was developed by Hemdan et al. [103]. They compared seven performances of seven pretrained deep learning networks; they are the InceptionV3, MobileNetV2, VGG19, DenseNet201, Inception-ResNetV2, ResNetV2, and Xception model. Based on their experiments, the VGG19 model achieved the highest accuracy of 90%.
Sethy and Behera [83], introduced a hybrid approach that utilizes deep learning for feature extraction and support vector machine (SVM) for detecting patients contaminated with COVID-19 by using CXR images. Using the pretrained 13 distinct Convolutional Neural Network models, the SVM provided the best results on the deep features of the ResNet50 model achieving accuracy of 95.38% for detecting COVID-19 (ignoring SARS, MERS and ARDS).
Ouchicha et al. [104], proposed a model named CVDNet to diagnose the COVID-19 cases. This model employed local and global features of CXR by using two parallel layers with various kernel sizes reaching an average accuracy of 97.20% for detecting COVID-19 cases.

Application on Chest CT images
Gozes et al. [81], used deep learning models to explore AI CT image analysis tools in the detection, quantification, and tracking of coronavirus. A total of 106 COVID-19 chest CT scans (50 labeled by a radiologist, and other 56 by RT-PCR test) and 99 normal ones were used to find potential COVID-19 thoracic CT features and to evaluate disease progression over time, generating a quantitative score. Utilizing the deep-learning image analysis system developed, they achieved classification results for COVID -19 versus  Ko et al. [105], investigated a simple 2D deep learning framework, and named the fasttrack COVID-19 classification network (FCONet), in order to diagnose COVID-19 pneumonia based on a single chest CT image. FCONet was developed by transfer learning using one of four state-of-the-art pretrained deep learning models (VGG16, ResNet-50, Inception-v3, or Xception) as a backbone. Among the four pretrained models of FCONet, ResNet-50 showed excellent diagnostic performance (sensitivity 99.58%, specificity 100.00%, and accuracy 99.87%) and outperformed the other three pretrained models in the testing data set. In the additional external testing data set using low-quality CT images, the detection accuracy of the ResNet-50 model was the highest (96.97%).
Zhang et al. [64], used artificial intelligence technology proposing a COVSeg-NET model that can segment GGO lesions in COVID-19 chest CT images. The COVSeg-NET model is based on the fully convolutional neural network model structure, which mainly includes convolutional layer, nonlinear unit activation function, maximum pooling layer, batch normalization layer, merge layer, flattening layer, sigmoid layer, and so forth. The results showed a sensitivity and specificity of the COVSeg-NET model of 0.447 and 0.996 respectively.
Song et al. [107], developed a deep learning network, which is called DeepPneumonia, to diagnose COVID-19 cases analyzing CT scans. Their proposed system was built on the ResNet50 using transfer learning technology. It could localize the essential lesion characteristics, especially GGO. Their system achieved an average AUC of 0.99 and sensitivity score of 93%. Besides, it reached an average AUC of 0.95 and sensitivity of 96% for bacterial pneumonia-infected cases.
Wang et al. [65], developed an artificial intelligence system in a time-to-event analysis framework to integrate chest CT and clinical data for risk prediction of future deterioration to critical illness in patients with COVID-19. The artificial intelligence system achieved a C-index of 0.80 for predicting individual COVID-19 patients as having critical illness, and successfully stratified the patients into high-risk and low-risk groups with distinct progression risks (p < 0.0001).
Xu et al. [108], proposed a fully automated COVID-19 diagnosis based on a 3D deep learning network-using chest CT scans. Their proposed system consists of four basic stages, which are pre-processing, candidate region segmentation, classification for each candidate region, and overall infection probability. The experimental results of this study showed that the summarized accuracy rate was 86.7%.

Critical Considerations and Conclusions
In addition, if the summarized accuracy of the selected papers is high, there was an important variability. The accuracy and applicability of AI approaches in COVID-19 from CXR or chest CTs have questioned, based on concerns of the radiologists' association, and given the impact of selection bias reported in first published results. Moreover, the limitation of this methodology is that if the patient is in a critical situation and unable to attend for CXR or CT scanning.
The analyzed papers showed the great potential of AI in COVID-19 pandemic by helping complex decision-making. However, most of the analyzed papers were experimental, and the produced models have not been deployed in real-world clinical setting. Those reported are impeded by several limitations. The available data sets may suffer from selection bias. The prognosis studies mostly encompass inpatients, who are usually sicker, whereas the diagnosis studies typically involve patients who already exhibit symptoms fitting with COVID-19. More data are needed on asymptomatic individuals and those with mild symptoms, who might not visit the hospital. Moreover, most of the studies reviewed employed imbalanced data sets, that is, those where many records in the training data set represent the negative class, and the positive class is under-represented. Thus, the reported performance of various AI algorithms applied in this context may have been affected by polarization of the context: a pandemic scenario. A high accuracy value in such cases could be attributed to the ability of the model to accurately identify negative samples and erroneously exclude all the positive COVID-19 cases. More effort is required to handle imbalanced data sets prior to the application of AI to COVID-19. The predictive performance of the models might also differ when using representative data that incorporates the targeted population, which merits further investigation.
Moreover, although AI is a promising tool in precision medicine, many factors such as low signal-to-noise ratio and complex data integration have challenged its efficacy. Both CXR and CT showed a high accuracy to detect pneumonia by COVID-19 and to predict the disease evolution, but which CXR is the first examination in this context and thus more data is available, CT is more capable to investigate extension and critical issues of the disease. However, CT images represent a difficult classification task due to the relatively large number of variable objects, specifically the imaged areas outside the lungs that are irrelevant to the diagnosis of pneumonia. Notably, the assessed features of the CT images were from patients with severe lung lesions at later stages of disease development. A larger number of databases to associate this with the disease progress and all pathologic stages of COVID-19 are necessary to optimize the diagnostic system.
In conclusion, AI approaches could be used in the identification of disease clusters, monitoring of cases, prediction of the future outbreaks, mortality risk, diagnosis of COVID-19, disease management by resource allocation, facilitating training, record maintenance and pattern recognition for studying the disease trend.