Research Article
Research Article
Impact Analysis of Machine Learning Techniques for COVID-19 Diagnosis: A Critical
Review
Anshul a, and Raju Kumar b A
Research Scholar, University Institute of Computing, Chandigarh University, Gharuan, Mohali, Punjab, India
Assistant Professor, Department of Computer Applications, Chandigarh Group of Colleges, Mohali, Punjab,
bAssociate Professor, Chandigarh University, Gharuan, Mohali, Punjab, India
Article History: Received: 11 January 2021; Accepted: 27 February 2021; Published online: 5 April 2021 _____________________________________________________________________________________________________
Abstract: COVID-19 virus belongs to the severe acute respiratory syndrome (SARS) family raised a situation of health
emergency in almost all the countries of the world. Numerous machine learning and deep learning based techniques are used to diagnose COVID positive patients using different image modalities like CT SCAN, X-RAY, or CBX, etc. This paper provides the works done in COVID-19 diagnosis, the role of ML and DL based methods to solve this problem, and presents limitations with respect to COVID-19 diagnosis.
Keywords: Machine Learning, Deep Learning, Artificial Intelligence, COVID-19.
___________________________________________________________________________
1. Introduction
Corona disease was declared a pandemic in March 2020. It was first recognized in China in December 2019. This disease is extending across the globe, causing gigantic societal commotion and stretching our ability to deliver efficient healthcare[1]. In the world, till 16 Jan 2021 total confirmed cases were 94,248,741 in which total deaths were 2,016,116 [2] and the count is increasing day by day. The hasty spread of this virus throughout the world is caused by the scarcity of knowledge about this virus along with a lack of antiviral therapies. Early infection symptoms include cough, fever, and other respiratory issues. Cough is playing a major role in transmitting the coronavirus from human to human [3]. Table 1 shows the most affected countries due to corona.
Table 1. Top 10 Country Wise Data for CORONA -19[4]
Country Total Confirmed Cases Total Death
USA 2,43,06,043 4,05,261 India 1,05,58,710 1,52,311 Brazil 84,56,705 2,09,350 Russia 35,68,209 65,566 UK 33,57,361 88,590 France 28,94,347 70,142 Turkey 23,80,665 23,832 Italy 23,68,733 81,800 Spain 22,52,164 53,314 Germany 2,038,645 1,657,900
As the infection escalates, the patient has to face astringent respiratory syndromes, kidney failure, and lung issues in some cases. Some patients start losing their blood cells and had an uncontrollable physique temperature. The situation is getting worse as till time there is no effective medicine is there to completely control this problem. The only solution to this problem is early detection and after that quarantine, the infected person immediately to avoid the further spreading of Corona. Fig. 1. Shows the graphical representation of the top 10 countries which are
affected by corona. This graph is representing data in terms of total corona-positive cases and deaths till 16 Jan 2021.
Fig. 1: Graphical representation of Country-wise Corona Death Analysis[4]
From the last few decades, in the healthcare sector, artificial intelligence has been globally stated as a promising approach to aid in disease detection and helping the doctors in diagnosis properly and in comparatively less time [5]. Machine learning has demonstrated remarkable results in the field of medical image data scrutiny. To diagnose COVID patients early, both machine and deep learning are showing excellent outcomes with the support of CT SCAN, X-RAY, and some other image modalities. This paper is working on the analysis of the impact of the machine and deep learning on COVID-19, current research in this field, their results, and limitations in their studies. 2. Related work
In retaliation to the pandemic, researchers have scurried to develop models using artificial intelligence (AI) methods specially machine learning (ML) & deep learning (DL) methods to support doctors. These methods have the potential to require advantage of the huge amount of multi-modal data collected from patients and can be used in successful diagnosis, transform detection, and triage of corona-positive patients.
Booth et al. [6] proposed an SVM using parameters of serum laboratory for predicting patient expiration status. They used 398 patients of which 43 deceased and 355 non-deceased for death expectation up to 48 hours before the patient's death. The objective of the paper was to find out the serum biomarkers in corona effected people at the utmost risk of mortality. Hu et al. [7] proposed DL method for COVID-19 classification and detection. This paper also worked on the localize lesions on CAP CT scans and COVID from the image-level description. The proposed framework showed astronomical accurateness, precision, and AUC for COVID classification and encouraging qualitative visualization for the detection of the lesion. Amyar et al. [8] proposed a multitasking DL model to recognize corona patients and fragment COVID-19 lesions from chest CT images. For this, the paper proposed an architecture composed of an encoder and two decoders for segmentation reconstruction and an MLP to do classification. Dansana et al. [9] proposed a DL algorithm using CT SCAN and X-Ray image modalities. This paper used CNN for twofold classification pneumonia-based conversion of DT, VGG-19, Inception_V2 model on different image modalities. This paper showed that deep learning have improved results as than machine learning. Ouyang et al. [10] developed a twofold sampling attention network for COVID -19 detection from CAP using CT image modalities. They used 3D CNN to concentrate on the lung's infected area.
Yasar and Ceylan [11] worked on the early recognition of COVID-19 using the 23- layer CNN and also giving the comparison of texture analysis, ML and DL methods. To do classification, they used CNN and also equated the results with the K-NN and SVM. Alexnet and Mobilenetv2 CNN architectures are used for training and testing purposes. The results were computed by increasing the image count by 5,10,20 times using data augmentation methods. Ahmad et al. [12] proposed a model using a supervised ML algorithm to recognize the features that can help in predicting the COVID-19 diagnosis with astronomical accuracy. This paper encompassed the features like gender of patient,age , observation of fever,travel history, and some clinical minutiae likem cough severity and lung infection incidence. Paper showed that the XGBoost algorithm utmost accurateness (more than 85%) to forecast and select the descriptions to identify COVID-19 patients for each age group.
Pathak et al. [13] used the CNN model to study the x-ray for covid-19 patients and non-corona patients. They cleaned the images and applied data augmentation. They compared the Inception of V3, Xception, and ResNeXt models in terms of precision. Wang et al. [14] compared the feature extraction and DL methods to work with CT image features that can help in COVID-19 prediction and uncover their importance for COVI-19 pneumonia from DL and radionics framework. To unearth deep learning and radionics featured a double-directional adversarial network-based framework and PyRadiomics package was expended. Kuchana et al. [15] worked on two segmentation tasks the first Prediction of lung spaces (and second COVID-19 irregularities from chest CT scans. Shankar and Perumal [16] proposed a DL model titled FM-HCF-DLF for COVID -19 using X-Ray. In this, there
are three processes: preprocessing based on Gaussian filtering, feature extraction done by FM, classification, and CNN bases Inception v3 techniques. Afshar et al. [17] used a capsule network aimed at COVID-19 detection using CT scans and X-ray.
Yan et al. [18] suggested a formula to forecast covid effected people at extreme risk, permitting them to be prioritized and possibly diminishing the mortality rate. Wang et. al [19] proposed an efficient Simulation Model (RSM), they apply GRU alongside a bidirectional and attentional mechanism to categorize 6 clinically significant respiratory patterns. Narin et al. [20] proposed five pre-trained CNN-based models for the detection of coronavirus pneumonia infected people with the help of X-ray radiographs. Ozturk et al. [21] suggested a model used for truthful diagnostics for binary and multi-class classification. Garg et al. [22] compared 40 CNN models and resulted that EfficientNet-25 is showing the best results with 99% inaccuracy, sensitivity, F1 score, and specificity. Islam et al. [23] proposed a CNN+ LSTM technique to diagnose COVID-19 using X-ray. Panahi et al. [24] proposed FCOD which is a DL model based on inception architecture. Usama et al. [25] used ResNet-50 on the publicly available dataset by the University of Montreal and NIH. Table 2 shows the comparison between ML and DL methods on Corona diagnosis.
Table 2. Comparative Analysis of DL and ML Methods on Corona Diagnosis
Sr. No.
Paper Reference
Methodology Dataset Results Limitation/ Future Work
1. [6] SVM 398 patients The SVM model resulted
Sensitivity - 91%, Specificity- 91% AUC- 0.93
1.Dataset is unbalanced and institutional biased.
2. The study did not perform on another clinical factor-like patient diagnosis.
2. [7] CNN 60 3D CT lung scans Results with No contrast adjustments are Accuracy -81.0% Precision- 83.2% Sensitivity- 80.2% Specificity (%)-82.6% AUC- 0.816
1.Proposed model is not enough in discrimination of CAP from COVID-19.
2.Possible existence of non-infection slices in between the CAP and COVID -19 subject images may introduce noise in training.
3. [8] 2D UNET +
CNN
1044 images For segmentation: Dice coefficient > 0.78 For Classification: an area under the ROC curve > 93%
the larger database should be used for better performance
4. [9] CNN 360 images (295 COVID-19 patients +16 SARS + 18 streptococcus) validation accuracy (91%)
The data set used is very small
5. [10] 3DCNN For training-validation stage:
2186 CT scans For the testing stage: 2796 CT scans. AUC: 0.944, Accuracy: 87.5%, Sensitivity: 86.9%, specificity : 90.1%, F1-score :82.0%.
1.Consistency on tracking the evolution of Covid-19 throughout treatment is not tested.
2.Accuracy in a small Covid infected area is not satisfactory.
6. [11] CNN 1.396 CT images For 2- Fold Cross-
Validation The highest mean sensitivity. : 0.91, Specificity: 0.9891, Accuracy: 0.9473, F-1 Score: 0.9058, AUC : 0.9888 10 - Fold Cross- Validation The highest mean sensitivity. : 0.91 Specificity : 0. 9404 Accuracy : 0. 9901 F-1 Score : 0. 9284 AUC : 0. 9903
1.More complex CNN can be used. 2.The results can be enhanced using the realization of pipeline methods
7. [12] XGBoost 6,512 patients Accuracy:88% A large dataset can improve
prediction accuracy.
8. [13] CNN 6432 chest x-ray scans Accuracy- 97.97% Overfitting problems can occur. To handle this more complex and larger dataset can be used. 9. [14] Bi-directional adversarial network-based framework & 266 patients Sensitivity: >73% Specificity: >75%
1.This paper did not examine other features examined by a framework
other than the BigBiGAN
PyRadiomics package
2.Enhancement in interpretability can be done using Image encoding processes.
10. [15] CNN+Unet GitHub and Kaggle F1- Score: 97.31% Mean IoU: 84.6%
Improvement in 3D reconstructed volumes can be done for anomaly detection purposes. 11. [16] CNN 27 healthy people+ 220 COVID-19+ 11 SARS+ 15 Pneumocystis class.
Results in percentage are Sensitivity (93.61), Specificity (94.56), Precision (94.85), Accuracy (94.08), F score (93.2 ) Kappa value(93.5)
Multi classified data set can be used
12. [17] Capsule Network
X-Ray Data set Accuracy : 95.7%, Sensitivity: 90% Specificity: 95.8% AUC -0.97
For future diagnosis, overlap with other lung infections can be considered.
13. [18] XGBoost 404 infected patients more than 90% accuracy The Deep Learning model can improve the results
14. [20] ResNet50 341 Chest X- Ray for Covid patient Accuracy in different dataset are: Dataset-1: 96.1%, Dataset- 2: 99.5% Dataset-3.: 99.7%
The larger data set can be used
15. [21] Convolutional Neural Network DarkCovidNet Architecture 127 (Female (43) + Male(82)) Accuracy Binary classes: 98.08% Multi-classes: 87.02%
The larger data set can be used
16. [23] CNN+LSTM 4575 X-ray images Accuracy -99.4 AUC - 99.9 Specificity 99.2 Sensitivity 99.3 F1-score- 98.9
1.Larger data set can be used. 2.Different angles of X-ray images should be used.
3. Multiple diseases in the images should be classified.
17. [24] FCOD 940 X-ray Accuracy- 96%
AUC-0.95% F1-Score- 96%
The larger data set can be used to check the performance.
18. [25] ResNet-50 89 corona infected +93 healthy people+96 infected with another disease
Accuracy- 98% F1-Score- 98%
A more complex dataset can be used.
3. Classification of energy efficiecy technique
III. IMPACT OF MACHINE LEARNING IN COVID DIAGNOSIS
Machine learning & deep learning models are showing promising fallouts in COVID diagnosis. Many applications of machine learning are utilized ranging from tracking, containing, and treating the virus infection [26]. The main problem of this virus is that it travels from human to human very quickly, so early detection of this virus can save many human lives. Machine learning diagnostic implements are particularly precise in detecting the virus appearance by means of radiological images & can be used as a tool for assessment. In hospitals and other health care systems during this pandemic machine learning tools are a better alternative as radiologists are very expensive and not possible available everywhere. These tools use different image modalities like CT SCAN, X-RAY, CBX, etc. but most of the researchers prefer to work with CT Scan and X-Ray. Machine learning tools can also help the doctors briskly screen the patients who are hospitalized due to some other reasons than COVID-19.
AI has performed an important role in medical imaging analytic during this pandemic. ML and DL models can comfort the community to numerous extents like early warning and alerts, prognosis and diagnosis of disease, tracking, and prediction, treatment and cures, social controls, etc[27]. A Canadian AI model BlueDot ascertained its mettle and informed the user about the outbreak before WHO. Similarly, another AI model Healthmap also predicted the COVID-19 outbreak. ML models are also contributing to the discovery of drugs for Covid-19. Researchers are using AI models that can speed up the process of searching for treatment and COVID-19 vaccination. Google's Deep Mind a deep learning tool predicted the protein's structure of COVID-19 which is very helpful for COVID-19 vaccine, detection [28]. These models are also working in the area of contact tracing using the mobile app. Table 3 shows some mobile app that is used by various countries.
5. Discussion and conclusion
The influence of the pandemic has driven researchers to create a tremendous amount of exploration pointed toward understanding, checking, and containing the disease; in any case, it stays hazy regardless of whether the research that has been delivered to date adequately addresses existing information gaps. The disease can be confirmed by using the RT-PCR test but the detection using RT-PCR is a prolonged task. ML and DL methods are providing automatic recognition of COVID-19 which can be very beneficial in scrutinizing quantifying and classifying contact-free subjective interaction [30]. For fast and accurate results, a DL technique is also established for the ancestry of graphical features of COVID-19 from CT-SCAN and X-Ray image modalities. CT Scanning is a little advance than X-Ray and can provide more accurate results of body part tissues and organs with machine learning but they are costly as compared to X-Ray [31].
X-Ray is cheap as compared to CT scans but may lead to wrong results. Although AI methods always show promising results in the healthcare sector yet many challenges are yet to be solved. According to a study, both ML and DL are giving a robust performance in COVID-19 detection but deep learning shows better results than machine learning. Convolution neural network is giving the best results till the time in this field but the problem is deep learning-based models require very huge and labeled data for training propose. ResNet is showing the best results till time with an accuracy of 98% [28]. To solve the problems of the data set, data augmentation can be done but the results may be affected due to augmentation. Despite early detection machine learning is also playing an effective role in finding drug discovery, severity, tracking the improvement in COVID treatment [32]. AI algorithms for COVID-19 are developed with an awfully broad range of applications, data collection procedures, and performance assessment metrics [33]. Perhaps, as a result, none are currently able to be deployed clinically. Possible reasons can be:
(i) the bias in small data sets and the variability of enormous internationally sourced data sets; (ii) the poor integration of multi-stream data, particularly imaging data;[34]
(iii) the requirement for clinicians and data analysts to figure side-by-side to confirm the developed AI algorithms are clinically relevant and implementable into routine clinical care.[35]
Without any second opinion, machine learning and deep learning are giving promising results to solve this problem but it is very difficult to fight with COVID-19 due to its unknown biological origin and mysterious behavior. It is said that fever, cold, cough, and breathing problems are the main symptoms of CORONA but in some cases, there were no such types of symptoms found but still, they were infected by this disease. Some precautionary measures like wearing face covers, social distancing, seclusion, sanitization, and quarantine can condense the risks of extending pandemics. The future confronts for AI correlated to the diagnosis of COVID-19 is testing & training various ML & DL architectures with a large and variety of image datasets infested by distinctive hues of pneumonia. A multi- classification technique is obligatory with statistics for more defined COVID-19 diagnosis.
References
“World Corona Dashboard,” 2021. [Online]. Available: https://www.worldometers.info/coronavirus/. [Accessed: 16-Jan-2021].
A. Ahmad, S. Garhwal, S. K. Ray, G. Kumar, S. J. Malebary, and O. M. Barukab, “The Number of Confirmed Cases of Covid-19 by using Machine Learning: Methods and Challenges,” Arch. Comput. METHODS Eng. A. Amyar, R. Modzelewski, and S. Ruan, “MULTI-TASK DEEP LEARNING BASED CT IMAGING ANALYSIS FOR COVID-19 : CLASSIFICATION AND SEGMENTATION General Electric Healthcare ,
Buc , France LITIS - EA4108 - University of Rouen , Rouen , France Nuclear Medicine Department , Henri Becquerel Center , Rouen,” pp. 2–8, 2020.
A. Bansal, R. P. Padappayil, C. Garg, A. Singal, M. Gupta, and A. Klein, “Utility of Artificial Intelligence Amidst the COVID 19 Pandemic: A Review,” J. Med. Syst., vol. 44, no. 9, Aug. 2020.
A. Garg, S. Salehi, M. La Rocca, R. Garner, and D. Duncan, “Efficient and Visualizable Convolutional Neural Networks for COVID-19 Classification Using Chest CT,” pp. 1–35, 2020.
A. H. Panahi, A. Rafiei, and A. Rezaee, “FCOD: Fast COVID-19 Detector based on deep learning techniques,” Informatics Med. Unlocked, vol. 22, p. 100506, 2021.
A. L. Booth, E. Abels, and P. McCaffrey, “Development of a prognostic model for mortality in COVID-19 infection using machine learning,” Mod. Pathol., 2020.
A. Narin, C. Kaya, and Z. Pamuk, “Automatic Detection of Coronavirus Disease (COVID-19) Using X-ray Images and Deep Convolutional Neural Networks,” arXiv Prepr. arXiv2003.10849., 2020.
A. S. Albahri et al., “Role of biological Data Mining and Machine Learning Techniques in Detecting and Diagnosing the Novel Coronavirus (COVID-19): A Systematic Review,” J. Med. Syst., vol. 44, no. 7, May 2020.
American Journal of Sociology, “Machine learning for COVID-19 detection and prognostication using chest radiographs and CT scans: a systematic methodological review,” J. Chem. Inf. Model., vol. 53, no. 9, pp. 1689– 1699, 2019.
D. Dansana et al., “Early diagnosis of COVID-19-affected patients based on X-ray and computed tomography images using deep learning algorithm,” Soft Comput., vol. 0123456789, 2020.
D. Javor, H. Kaplan, A. Kaplan, S. B. Puchner, C. Krestan, and P. Baltzer, “Deep learning analysis provides accurate COVID-19 diagnosis on chest computed tomography,” Eur. J. Radiol., vol. 133, no. October, p. 109402, 2020.
H. Wang et al., “Decoding COVID-19 pneumonia : comparison of deep learning and radiomics CT image signatures,” 2020.
H. Yasar and M. Ceylan, “A novel comparative study for detection of Covid-19 on CT lung images using texture analysis , machine learning , and deep learning methods,” 2020.
J. B. Long and J. M. Ehrenfeld, “The Role of Augmented Intelligence (AI) in Detecting and Preventing the Spread of Novel Coronavirus,” J. Med. Syst., vol. 44, no. 3, pp. 2019–2020, 2020.
J. Rasheed et al., “A survey on artificial intelligence approaches in supporting frontline workers and decision makers for the COVID-19 pandemic,” CHAOS SOLITONS & FRACTALS, vol. 141, Dec. 2020.
K. Shankar and E. Perumal, “A novel hand-crafted with deep learning features based fusion model for COVID-19 diagnosis and classification using chest X-ray images,” Complex Intell. Syst., no. March, 2020.
L. Wynants et al., “Prediction models for diagnosis and prognosis of covid-19 infection: systematic review and critical appraisal,” BMJ-BRITISH Med. J., vol. 369, Apr. 2020.
L. Yan et al., “A machine learning-based model for survival prediction in patients with severe COVID-19 infection,” medRxiv, 2020.
M. Ahamad et al., “A Machine Learning Model to Identify Early Stage Symptoms of SARS-Cov-2 Infected Patients,” Expert Syst. Appl., p. 113661, 2020.
M. Allam et al., “COVID-19 Diagnostics, Tools, and Prevention,” DIAGNOSTICS, vol. 10, no. 6, Jun. 2020. M. Ilyas, H. Rehman, and A. Nait-ali, “Detection of Covid-19 From chest x-ray images using artificial intelligence: an early review,” arXiv, pp. 1–8, 2020.
M. Kuchana, A. Srivastava, R. Das, and J. Mathew, “AI aiding in diagnosing , tracking recovery of COVID-19 using deep learning on Chest CT scans,” 2020.
M. Z. Islam, M. M. Islam, and A. Asraf, “A combined deep CNN-LSTM network for the detection of novel coronavirus (COVID-19) using X-ray images,” Informatics Med. Unlocked, vol. 20, no. August, p. 100412, 2020.
P. Afshar, S. Heidarian, F. Naderkhani, A. Oikonomou, K. N. Plataniotis, and A. Mohammadi, “COVID-CAPS: A capsule network-based framework for identification of COVID-19 cases from X-ray images,” Pattern Recognit. Lett., vol. 138, pp. 638–643, 2020.
R. Chaudhari, L. W. Fong, Z. Tan, B. Huang, and S. Zhang, “An up-to-date overview of computational polypharmacology in modern drug discovery,” Expert Opin. Drug Discov., vol. 15, no. 9, pp. 1025–1044, Sep. 2020.
S. Hu, Y. Gao, Z. Niu, Y. Jiang, and L. A. O. Li, “Weakly Supervised Deep Learning for COVID-19 Infection Detection and Classification from CT Images,” vol. 8, no. April, 2020.
S. Lalmuanawma, J. Hussain, and L. Chhakchhuak, “Applications of machine learning and artificial intelligence for Covid-19 (SARS-CoV-2) pandemic: A review,” CHAOS SOLITONS & FRACTALS, vol. 139, Oct. 2020.
S. Manigandan, M. T. Wu, V. K. Ponnusamy, V. B. Raghavendra, A. Pugazhendhi, and K. Brindhadevi, “A systematic review on recent trends in transmission, diagnosis, prevention and imaging features of COVID-19,” Process Biochem., vol. 98, no. August, pp. 233–240, 2020.
S. Usama, S. Safwan, K. Bukhari, A. Syed, and S. Sajid, “The evaluation of convolutional neural network (CNN) for the assessment of chest X-ray of COVID-19 patients,” Ann. Clin. Anal. Med., vol. 11, no. 6, 2020. T. Ozturk, M. Talo, E. A. Yildirim, U. B. Baloglu, O. Yildirim, and U. Rajendra Acharya, “Automated detection of COVID-19 cases using deep neural networks with X-ray images,” Comput. Biol. Med., vol. 121, no. April, p. 103792, 2020.
Worldometer, “COVID-19 CORONAVIRUS PANDEMIC,” 2021. [Online]. Available:
https://www.worldometers.info/coronavirus. [Accessed: 17-Jan-2021].
X. Ouyang et al., “Dual-Sampling Attention Network for Diagnosis of COVID-19 from Community Acquired Pneumonia,” IEEE Trans. Med. Imaging, vol. 39, no. 8, pp. 2595–2605, 2020.
Y. Pathak, P. K. Shukla, A. Tiwari, S. Stalin, S. Singh, and P. K. Shukla, “Deep Transfer Learning Based Classification Model for COVID-19 Disease,” IRBM, vol. 1, pp. 1–6, 2020.
Y. Wang, M. Hu, Q. Li, X.-P. Zhang, G. Zhai, and N. Yao, “Abnormal respiratory patterns classifier may contribute to large-scale screening of people infected with COVID-19 in an accurate and unobtrusive manner,” 2020.
