Abstract
This narrative review focuses on the artificial intelligence (AI) approaches to the diagnosis and diagnostic support of COVID-19 in a clinical and regulatory setting. Although initial research found high internal accuracy, most had bias, little extra validation and poor real-world applicability. Artificial intelligence strategies include imaging (CT, CXR and ultrasound), clinical data, and multimodal systems based on machine learning, deep learning and new transformer models. Performances are very sensitive to the quality of data, labeling and representativeness. Key challenges include dataset shift, confounding, and fairness. Recent innovations in reporting standards and regulatory frameworks are geared towards enhancing reliability. The external validation, integration with clinical and provisional impact on patient outcomes and healthcare workflows should be prioritized in future work.
1. Introduction
The pandemic accelerated a sudden demand by COVID-19 on rapid, scalable, and dependable diagnosis and triage, especially when the ability to perform molecular tests, turnaround times, and supply chains were limited at the start of the pandemic [1]. Global guidelines have since made nucleic acid amplification testing (NAAT; e.g., RT PCR) and, subsequently, antigen testing the key method to diagnose SARS-CoV-2 infection, with the role of chest imaging being contextual (e.g., to assess complications, severity, or alternate diagnoses) as opposed to a stand-alone screening test [2]. Diagnosis AI approaches to COVID-19 have thus approached two closely similar, yet different tasks: to detect imaging patterns that are consistent with COVID-19 pneumonia (a radiologic phenotype), and predict the presence of SARS-CoV-2 or clinical COVID 19 based on non-imaging inputs (clinical variables, laboratory tests or multimodal combinations) [3,4,5]. Strong internal performance demonstrations were the rule, and a number of high profile original studies were reporting external testing and/or multi-site testing—particularly of chest CT [6]. As an illustration, a big CT pipeline trained on CC-CCII data achieved an internal 0.98 AUROC and accuracy of 0.92 and external Chinese testing showed an AUROC of over 0.97 with some reduction in performance [7]. An independent test set of a multinational CT study with 0.91 accuracy and sensitivity 0.84 and specificity 0.93 demonstrates the attention to cross-site diversity early on [8]. Across modalities, however, the prevailing scientific finding of scientifically methodologically oriented critiques is that large portions of the initial COVID-19 diagnostic AI literature were of high risk of bias, confounding, limited external validation, and weak support of actual world clinical impact, making generalizability a central bottleneck [9,10,11]. Since 2020, reporting and evaluation standards have matured to the extent that there are now AI-specific guidance on imaging studies (CLAIM, updated 2024), prediction model studies (TRIPOD + AI, 2024), risk of bias appraisal (PROBAST + AI, 2025), and AI diagnostic accuracy studies (STARD AI, 2025) [12]. The reality of regulatory and deployment has also changed. In the United States (U.S.), AI in diagnostics is generally subject to software as a medical device (SaMD) requirement, such as lifecycle management and Good Machine Learning Practice, and the U.S. Food and Drug Administration has a public list of AI enabled medical devices [13]. The AI Act came into effect in the EU in August 2024, with applicability by stages, which applies to high-risk clinical AI systems, and software used in medical devices continues to be under MDR/IVDR guidance (including AI relevant interpretations) [14].
This narrative review synthesizes peer-reviewed evidence and authoritative guidance on AI methods used for diagnosis and diagnostic support in COVID-19, emphasizing: diagnostic needs and intended clinical roles; AI method families (ML, deep learning, CNNs, transformers, ensembles); and data types and datasets (CT, CXR, ultrasound, clinical/lab, multimodal), including sample sizes.
2. COVID-19 Diagnostic Context
The diagnostic “gold standard” for SARS-CoV-2 infection has been NAAT (including RT-PCR), with guidance covering specimen collection, NAAT, antigen tests, antibody testing, and quality assurance [15]. This guidance context matters because many AI studies (especially imaging AI) implicitly treat RT-PCR as the reference standard while simultaneously operating in a clinical reality where RT-PCR can be delayed, variably sensitive, or operationally constrained [16]. Chest imaging (CT, radiography) was widely used in clinical management, but major professional guidance cautioned against using CT as a first-line screening test for COVID-19 diagnosis, emphasizing appropriate indications and local epidemiology [17]. The World Health Organization published a rapid advice guide on the use of chest imaging in COVID-19, and the American College of Radiology advised against using CT for screening, recommending that CT be reserved for hospitalized symptomatic patients with specific clinical indications [18,19]. From an AI perspective, this positioning implies that most imaging-based “COVID diagnosis” algorithms are better conceptualized as tools for: (a) detecting a radiologic pattern consistent with COVID-19 pneumonia versus other pneumonias; (b) triage and workflow prioritization; (c) severity quantification; and (d) decision support for clinicians—rather than definitive virologic diagnosis by imaging alone [20]. Peer-reviewed commentaries explicitly warned that CT findings are not pathognomonic and can overlap substantially with other pneumonias, reinforcing the risk of over claiming “diagnosis” from imaging AI.
3. Data Foundations and Dataset Practices
The quality of data, label quality, and representativeness of data are the key factors in determining the level of AI performance in COVID-19 diagnosis. In modalities, there are a number of common data foundations. Chest CT data can exist as a 3D volume, or can be broken down into 2D slices. Some of the most influential CT pipelines were two-stage: lesion/lung segmentation, and classification. The authors provided a dataset of 617,775 CT images of 4154 patients, the manual pixel level annotation of 4695 slices to train the segmentation, and a three class classification task (novel coronavirus pneumonia vs. common pneumonia vs. normal) in a large CC-CCII-based system [7]. The datasets of Chest X-rays (CXRs) are usually 2D and ubiquitous in clinics yet possess a proven confounding risk in publicly available datasets (portable device marks, hospital-site artifacts, divergent acquisition guidelines). A prototype open benchmark (COVIDx) developed in COVID Net consisted of 13,975 CXR images in 13,870 patient cases, although only 266 COVID-19 patient cases at that point, which highlights extreme imbalance of the classes and the use of multisource aggregation [21]. Lung ultrasound (LUS) is attractive to bedside and low-resource environments, yet the data are frequently provided as a video loop with high operator bias. Training is usually on frames that are cut out of clips, which can increase the risks of correlation/leakage unless one splits on the patient or exam level. To investigate the detection of COVID-19 patterns and the severity scoring, a hospital ultrasound dataset was reported with 2908 frames of 450 hospitalized patients [22]. Clinical/laboratory data approaches use routine blood tests, blood gas, and vital signs to diagnose via a screening/classification problem. As an example, one of the feasibility studies involved standard hematochemical measurements of 279 symptomatic ED patients who have undergone RT PCR testing [23].
4. Dataset Summary and Sample Sizes
CC-CCII/China National Center for Bioinformation (CNCB) resources powered multiple CT studies; the CC-CCII dataset is explicitly referenced and used as the backbone for a large CT diagnostic system and was made available via CNCB infrastructure [24]. MosMedData (Russia) is widely used for CT work, reported as a dataset of 1110 chest CT scans in a peer-reviewed dataset paper [25]. Large public CXR aggregations (e.g., COVIDx) were assembled from multiple repositories, enhancing scale but amplifying heterogeneity and confounding [21]. Common preprocessing steps include DICOM handling, de-identification, lung field segmentation (often U-Net variants), intensity normalization (especially for CT), resizing/standardization, and augmentation (rotation, flips, contrast) to mitigate small sample sizes [26]. One of the most well-known radiology data preparation papers stresses that the decisions made during curation, labeling, de-identification, and harmonization have a direct impact on generalizability and downstream bias [27]. COVID-19 diagnostic AI is challenged by label: RT PCR is not a perfect reference (false negative, time effect) and patterns in imaging are similar to other viral/bacterial pneumonias. There were studies that annotated lesions at the pixel level (expensive, but informative), and numerous studies that used study level annotations and retrospective report extraction [28]. In the CC II system, thousands of slices had been annotated pixel level, polygonal to train segmentation models [7]. The methodological critiques highlight that label noise and dataset biases are not side effects, but are rather focal points of threat to validity; numerous early prediction and diagnostic models were judged as highly prone to bias [29,30].
5. AI Methods and Evaluation Evidence
This section reviews the dominant AI method families (ML, deep learning, CNNs/transformers, ensembles, multimodal fusion) and the evaluation practices that determine clinical credibility. Classical ML: logistic regression, random forests, gradient boosting and radiomics pipelines (handcrafted texture/shape features). These are typical of tabular lab tests and occasionally of CXR/CT radiomics [31,32]. Brinati et al. (2020) applied ML to regular blood features and compared models based on AUC, sensitivity, specificity, PPV and accuracy [23]. Deep learning (CNNs): 2D CNNs for CXR and slice-based CT; 3D CNNs/volume models for CT. CT systems often incorporate segmentation → classification pipelines to reduce background confounding and to expose lesion burden features [7,33,34]. Transformers: by 2022–2025, transformer architectures and foundation models emerged as players in the general analysis of chest X-rays [35]. Foundation models such as EVA-X adopt vision transformer backbones pre-trained on large unlabeled CXR corpora to learn generalizable representations [36]. Although COVID specific transformer diagnostic papers do exist, the important conceptual point is that training on large COVID era convenience datasets may yield better performance compared to training on large unlabeled CXR corpora [37]. Ensembles: ensembles can either be model ensembles (meaning predictions) or feature level ensembles (combining radionics and deep features). In imaging, a feature level ensemble methodology was reported to have better classification than single feature families [38]. XGBoost and random forests are typical ensembles in clinical data [39]. Multimodal fusion: combines imaging and symptoms, exposure and lab tests. In one of the most notable AI systems, CT of the chest was combined with clinical variables and reported AUC 0.92 in a test group of 279 patients and detected RT PCR-positive patients with normal CT more accurately than radiologist readings in the test group [40]. The rigor of evaluation is diverse. The high credibility protocols consist of: patient level splits (to prevent leakage across correlated images/frames), strict separation of training/validation/test, external validation (different hospital/site), prospective validation or observer studies (radiologist with AI vs. radiologist alone), calibration assessment (especially of triage thresholds), reporting of prevalence dependent metrics (PPV/NPV) and sensitiveness/specificity [41]. In CT work, Zhang et al. (2020) report internal validation and multiple external/prospective cohorts, with explicit internal and external performance metrics [7]. Bai et al. report both internal testing and independent external testing, plus a radiologist augmentation experiment. For diagnostic AI, the primary metrics requested by the user are standard: sensitivity, specificity, AUROC, PPV, NPV, and F1. However, narrative interpretation must emphasize: PPV/NPV dependence on disease prevalence, which changed rapidly across waves and settings; AUROC can mask threshold behavior; clinical deployment usually requires operating points for triage (favoring high NPV to “rule out,” or high sensitivity to avoid missed cases).
6. Generalizability, Interpretability, Bias, and Fairness
The decrease in performance when comparing internal and external datasets is a common theme and no minor issue: it is the initial factor of whether a COVID diagnostic model can be used in clinics outside of the place of development. An AI model in CT had both high internal test accuracy (96%), and lower performance on independent testing (87%), and the authors openly discuss cross institution differences and demographic imbalances as potential confounders [20]. Internal AUROC 0.98 and external AUROC 0.97 were reported in the large CC-CCII CC II CT system, and external specificity decreases occurred in one area [7]. Dataset shift COVID-19—this is a moving target in at least three senses that apply to AI: epidemiologic shift (prevalence, testing policy, admitted treatment criterion), clinical shift (vaccination status, variants, treatment protocols), and data acquisition shift (portable vs. fixed radiography use, scanner protocols under surge conditions). Opponents of imaging AI point out that models may unwittingly discover shortcuts in correlation with site or acquisition and not disease status [11]. This is especially harmful when the training data is a combination of hospitals or countries that have varying protocols (a common problem in early COVID imaging data). A meta-analysis of COVID 19 predictive models found that most of them were poorly reported and at significant risk of bias, with little confidence in published performance [29]. Pervasive pit-falls listed in methodological analyses of imaging AI included non-representative sampling, leakage in labels, uncontrolled confounding, inconsistent evaluation, and absence of prospective evidence [10]. There are fairness issues as the outcomes of COVID and access to testing/imaging vary based on age, comorbidity, socioeconomic status, and geography. Models that do not explicitly encode demographic attributes can learn proxies (site, device, care pathway) which cause differences in error rates. Contemporary AI reporting standards now explicitly require information on dataset practices be disclosed and algorithmic bias and fairness be considered (STARD AI, 2025) and risk of bias tools have been revised to include AI particular traps (PROBAST + AI, 2025). Saliency/attention maps (e.g., Grad CAM) and feature attribution techniques (e.g., SHAP with tabular models) dominate explainability in COVID diagnostic AI. Grad CAM style analyses are frequently used in imaging CT work (Bai et al. (2020) explicitly includes Grad CAM in its list of abbreviations and interpretability workflow) [20]. COVID Net emphasized explainability as a method to audit the dependence of network on clinically relevant regions [21]. Interpretability must be taken as some supportive and not confirmatory evidence: saliency methods are not always stable, can point to non-causal areas and do not imply causal inferences. Based on this, explainability is best justified combined with: rigorous external validation, artifact control (masking text markers, device labels) and clinician review studies one can determine whether the explanations are consistent with known pathophysiology [42]. Major scientific and governance milestones in COVID diagnostic AI include early CT-based diagnostic systems and augmentation studies (2020), rapid expansion of open CXR benchmarks with explicit calls for explainability (2020), systematic critiques of bias and generalization failures (2020–2021), growth of prospective/operational screening tools using routine clinical data (2021–2022), and maturation of reporting and appraisal standards plus emergence of foundation models (2024–2025) [43].
7. Conclusions
AI has demonstrated a high potential in aiding the diagnosis of COVID-19, especially regarding imaging analysis, triage, and multimodal prediction systems. Nevertheless, it should be complementary to the already known diagnostic methods like NAAT and antigen testing instead of substituting them. The initial period of the pandemic showed significant methodological flaws, such as bias, a lack of generalizability, and overdependence on small datasets. The recent advances in evaluation criteria, regulatory controls, and the development of models have enhanced the predictability and precision of the AI systems. However, such issues as dataset drift, equity, and practical deployment remain critical. Future studies should not rely on retrospective accuracy research but, instead, on prospective, clinically integrated research that provides evidence of real patient care, workflow efficiency, and health system outcomes improvements. Safe and effective implementation of the change requires robust external validation, ethical governance and ongoing monitoring.
Funding
This research received no external funding.
Institutional Review Board Statement
Not applicable.
Informed Consent Statement
Not applicable.
Data Availability Statement
No new data were created or analyzed in this study. Data sharing is not applicable to this article.
Conflicts of Interest
The author declares no conflicts of interest.
Abbreviations
The following abbreviations are used in this manuscript: AI—Artificial Intelligence; CNCB—China National Center for Bioinformation; CXR—Chest X-ray; LUS—Lung ultrasound; SaMD—software as a medical device.
References
- Chen, L.; Meng, Q.H. Advancing Laboratory Diagnostics for Future Pandemics: Challenges and Innovations. Pathogens 2025, 14, 1135. [Google Scholar] [CrossRef] [Scilit] [PubMed]
- Zhu, Y.; Zhang, M.; Jie, Z.; Tao, S. Nucleic acid testing of SARS-CoV-2: A review of current methods, challenges, and prospects. Front. Microbiol. 2022, 13, 1074289. [Google Scholar] [CrossRef] [Scilit] [PubMed]
- Chapman, S.R.; Willner, L.; Abouafech, A.; Roberti, C.; Willner, C. Diagnostic Performance of Artificial Intelligence in Detecting COVID-19 Pneumonia on Chest Imaging. Cureus 2026, 18, e101775. [Google Scholar] [CrossRef] [Scilit] [PubMed]
- Caliman Sturdza, O.A.; Filip, F.; Terteliu Baitan, M.; Dimian, M. Deep Learning Network Selection and Optimized Information Fusion for Enhanced COVID-19 Detection: A Literature Review. Diagnostics 2025, 15, 1830. [Google Scholar] [CrossRef] [Scilit] [PubMed]
- Dharmik, A. COVID-19 Pneumonia Diagnosis Using Medical Images: Deep Learning-Based Transfer Learning Approach. JMIRx Med. 2025, 6, e75015. [Google Scholar] [CrossRef] [Scilit] [PubMed]
- Topff, L.; Groot Lipman, K.B.W.; Guffens, F.; Wittenberg, R.; Bartels-Rutten, A.; van Veenendaal, G.; Hess, M.; Lamerigts, K.; Wakkie, J.; Ranschaert, E.; et al. Is the generalizability of a developed artificial intelligence algorithm for COVID-19 on chest CT sufficient for clinical use? Results from the International Consortium for COVID-19 Imaging AI (ICOVAI). Eur. Radiol. 2023, 33, 4249–4258. [Google Scholar] [CrossRef] [Scilit] [PubMed]
- Zhang, K.; Liu, X.; Shen, J.; Li, Z.; Sang, Y.; Wu, X.; Zha, Y.; Liang, W.; Wang, C.; Wang, K.; et al. Clinically Applicable AI System for Accurate Diagnosis, Quantitative Measurements, and Prognosis of COVID-19 Pneumonia Using Computed Tomography. Cell 2020, 181, 1423–1433.e11, Erratum in Cell 2020, 181, 1360. [Google Scholar] [CrossRef] [Scilit] [PubMed]
- Harmon, S.A.; Sanford, T.H.; Xu, S.; Turkbey, E.B.; Roth, H.; Xu, Z.; Yang, D.; Myronenko, A.; Anderson, V.; Amalou, A.; et al. Artificial Intelligence for the Detection of COVID-19 Pneumonia on Chest CT Using Multinational Datasets. Nat. Commun. 2020, 11, 4080. [Google Scholar] [CrossRef] [Scilit] [PubMed]
- Aruleba, R.T.; Adekiya, T.A.; Ayawei, N.; Obaido, G.; Aruleba, K.; Mienye, I.D.; Aruleba, I.; Ogbuokiri, B. COVID-19 Diagnosis: A Review of Rapid Antigen, RT-PCR and Artificial Intelligence Methods. Bioengineering 2022, 9, 153. [Google Scholar] [CrossRef] [Scilit] [PubMed]
- Roberts, M.; Driggs, D.; Thorpe, M.; Gilbey, J.; Yeung, M.; Ursprung, S.; Aviles-Rivero, A.I.; Etmann, C.; McCague, C.; Beer, L.; et al. Common Pitfalls and Recommendations for Using Machine Learning to Detect and Prognosticate for COVID-19 Using Chest Radiographs and CT Scans. Nat. Mach. Intell. 2021, 3, 199–217. [Google Scholar] [CrossRef] [Scilit]
- DeGrave, A.J.; Janizek, J.D.; Lee, S.I. AI for radiographic COVID-19 detection selects shortcuts over signal. Nat. Mach. Intell. 2021, 3, 610–619. [Google Scholar] [CrossRef] [Scilit]
- Park, S.H.; Suh, C.H. Reporting Guidelines for Artificial Intelligence Studies in Healthcare (for Both Conventional and Large Language Models): What’s New in 2024. Korean J. Radiol. 2024, 25, 687–690. [Google Scholar] [CrossRef] [Scilit] [PubMed]
- U.S. Food and Drug Administration (FDA). Artificial Intelligence in Software as a Medical Device; FDA: Silver Spring, MD, USA, 2025. Available online: https://www.fda.gov/medical-devices/software-medical-device-samd/artificial-intelligence-software-medical-device (accessed on 5 April 2026).
- European Commission; DG Health. Artificial Intelligence in Healthcare; European Commission: Brussels, Belgium, 2025; Available online: https://health.ec.europa.eu/ehealth-digital-health-and-care/artificial-intelligence-healthcare_en (accessed on 5 April 2026).
- Eftekhari, A.; Alipour, M.; Chodari, L.; Dizaj, S.M.; Ardalan, M.; Samiei, M.; Sharifi, S.; Vahed, S.Z.; Huseynova, I.; Khalilov, R.; et al. A Comprehensive Review of Detection Methods for SARS-CoV-2. Microorganisms 2021, 9, 232. [Google Scholar] [CrossRef] [Scilit] [PubMed]
- Han, G.R.; Goncharov, A.; Eryilmaz, M.; Ye, S.; Palanisamy, B.; Ghosh, R.; Lisi, F.; Rogers, E.; Guzman, D.; Yigci, D.; et al. Machine learning in point-of-care testing: Innovations, challenges, and opportunities. Nat. Commun. 2025, 16, 3165. [Google Scholar] [CrossRef] [Scilit] [PubMed]
- Inui, S.; Gonoi, W.; Kurokawa, R.; Nakai, Y.; Watanabe, Y.; Sakurai, K.; Ishida, M.; Fujikawa, A.; Abe, O. The role of chest imaging in the diagnosis, management, and monitoring of coronavirus disease 2019 (COVID-19). Insights Imaging 2021, 12, 155. [Google Scholar] [CrossRef] [Scilit] [PubMed]
- World Health Organization (WHO). Use of Chest Imaging in COVID-19: A Rapid Advice Guide; WHO: Geneva, Switzerland, 2020; Available online: https://www.who.int/publications/i/item/use-of-chest-imaging-in-covid-19 (accessed on 5 April 2026).
- American College of Radiology (ACR). ACR Recommendations for the Use of Chest Radiography and Computed Tomography (CT) for Suspected COVID-19 Infection. 11 March 2020. Available online: https://www.acr.org/Advocacy/Position-Statements/Recommendations-for-Chest-Radiography-and-CT-for-Suspected-COVID19-Infection (accessed on 5 April 2026).
- Bai, H.X.; Hsieh, B.; Xiong, Z.; Halsey, K.; Choi, J.W.; Tran, T.M.L.; Pan, I.; Shi, L.-B.; Wang, D.-C.; Mei, J.; et al. Performance of Radiologists in Differentiating COVID-19 from Non-COVID-19 Viral Pneumonia at Chest CT. Radiology 2020, 296, E46–E54. [Google Scholar] [CrossRef] [Scilit] [PubMed]
- Wang, L.; Lin, Z.Q.; Wong, A. COVID-Net: A Tailored Deep Convolutional Neural Network Design for Detection of COVID-19 Cases from Chest X-Ray Images. Sci. Rep. 2020, 10, 19549. [Google Scholar] [CrossRef] [Scilit] [PubMed]
- La Salvia, M.; Secco, G.; Torti, E.; Florimbi, G.; Guido, L.; Lago, P.; Salinaro, F.; Perlini, S.; Leporati, F. Deep learning and lung ultrasound for COVID-19 pneumonia detection and severity classification. Comput. Biol. Med. 2021, 136, 104742. [Google Scholar] [CrossRef] [Scilit] [PubMed]
- Brinati, D.; Campagner, A.; Ferrari, D.; Locatelli, M.; Banfi, G.; Cabitza, F. Detection of COVID-19 Infection from Routine Blood Exams with Machine Learning: A Feasibility Study. J. Med. Syst. 2020, 44, 135. [Google Scholar] [CrossRef] [Scilit] [PubMed]
- Wu, X.; Chen, C.; Zhong, M.; Wang, J.; Shi, J. COVID-AL: The diagnosis of COVID-19 with deep active learning. Med. Image Anal. 2021, 68, 101913. [Google Scholar] [CrossRef] [Scilit] [PubMed]
- Morozov, S.P.; Andreychenko, A.E.; Blokhin, I.A.; Gelezhe, P.B.; Gonchar, A.P.; Nikolaev, A.E.; Pavlov, N.A.; Chernina, V.Y.; Gombolevskiy, V.A.; Aлeкcaндpoвич, Г.B. MosMedData: Data set of 1110 chest CT scans performed during the COVID-19 epidemic. Digit. Diagn. 2020, 1, 49–59. [Google Scholar] [CrossRef] [Scilit]
- Musanga, V.; Chibaya, C.; Viriri, S. Using domain adaptation and transfer learning techniques to enhance performance across multiple datasets in COVID-19 detection. Front. Comput. Sci. 2026, 7, 1639421. [Google Scholar] [CrossRef] [Scilit]
- Willemink, M.J.; Koszek, W.A.; Hardell, C.; Wu, J.; Fleischmann, D.; Harvey, H.; Folio, L.R.; Summers, R.M.; Rubin, D.L.; Lungren, M.P. Preparing Medical Imaging Data for Machine Learning. Radiology 2020, 295, 4–15. [Google Scholar] [CrossRef] [Scilit] [PubMed]
- Jaiswal, A.; Fervers, P.; Meng, F.; Zhang, H.; Móré, D.; Giannakis, A.; Wailzer, J.; Bucher, A.M.; Maintz, D.; Kottlors, J.; et al. Performance of AI Approaches for COVID-19 Diagnosis Using Chest CT Scans: The Impact of Architecture and Dataset. Rofo 2026, 198, 185–198. [Google Scholar] [CrossRef] [Scilit] [PubMed]
- Wynants, L.; Van Calster, B.; Collins, G.S.; Riley, R.D.; Heinze, G.; Schuit, E.; Bonten, M.M.J.; Dahly, D.L.; Damen, J.A.; Debray, T.P.A.; et al. Prediction models for diagnosis and prognosis of COVID-19: Systematic review and critical appraisal. BMJ 2020, 369, m1328, Erratum in BMJ 2021, 372, n236. [Google Scholar] [CrossRef] [Scilit] [PubMed]
- Albalawi, U.; Mustafa, M. Current Artificial Intelligence (AI) Techniques, Challenges, and Approaches in Controlling and Fighting COVID-19: A Review. Int. J. Environ. Res. Public Health 2022, 19, 5901. [Google Scholar] [CrossRef] [Scilit] [PubMed]
- Chen, R.; Zhang, H.; Huang, X.; Han, H.; Jian, J. CT Radiomics-based machine learning approach for the invasiveness of pulmonary ground-glass nodules prediction. Eur. J. Radiol. Open 2025, 15, 100680. [Google Scholar] [CrossRef] [Scilit] [PubMed]
- Basha, S.M.; de Albuquerque, V.H.C.; Chelloug, S.A.; Elaziz, M.A.; Mohisin, S.H.; Pathan, S.P. Robust Machine Learning Technique to Classify COVID-19 Using Fusion of Texture and Vesselness of X-Ray Images. CMES 2024, 138, 1981–2004. [Google Scholar] [CrossRef] [Scilit]
- Musa, A.; Prasad, R.; Onwualu, P.; Hernandez, M. A Systematic Review of Cross-Population Shifts in Medical Imaging Analysis with Deep Learning. Big Data Cogn. Comput. 2026, 10, 76. [Google Scholar] [CrossRef] [Scilit]
- Ahmad, I.S.; Dai, J.; Xie, Y.; Liang, X. Deep learning models for CT image classification: A comprehensive literature review. Quant. Imaging Med. Surg. 2025, 15, 962–1011. [Google Scholar] [CrossRef] [Scilit] [PubMed]
- Lee, D. The Application of Foundation Models in Radiology: Bridging Images, Reports, and Beyond. J. Korean Soc. Radiol. 2025, 86, 919–937. [Google Scholar] [CrossRef] [Scilit] [PubMed]
- Yao, J.; Wang, X.; Song, Y.; Zhao, H.; Ma, J.; Chen, Y.; Liu, W.; Wang, B. EVA-X: A foundation model for general chest x-ray analysis with self-supervised learning. npj Digit. Med. 2025, 8, 678. [Google Scholar] [CrossRef] [Scilit] [PubMed]
- Mehboob, F.; Rauf, A.; Jiang, R.; Saudagar, A.K.J.; Malik, K.M.; Khan, M.B.; Hasnat, M.H.A.; AlTameem, A.; AlKhathami, M. Towards robust diagnosis of COVID-19 using vision self-attention transformer. Sci. Rep. 2022, 12, 8922. [Google Scholar] [CrossRef] [Scilit] [PubMed]
- Ho, T.K.K.; Gwak, J. Feature-level ensemble approach for COVID-19 detection using chest X-ray images. PLoS ONE 2022, 17, e0268430. [Google Scholar] [CrossRef] [Scilit] [PubMed]
- Imani, M.; Beikmohammadi, A.; Arabnia, H.R. Comprehensive Analysis of Random Forest and XGBoost Performance with SMOTE, ADASYN, and GNUS Under Varying Imbalance Levels. Technologies 2025, 13, 88. [Google Scholar] [CrossRef] [Scilit]
- Mei, X.; Lee, H.C.; Diao, K.-Y.; Huang, M.; Lin, B.; Liu, C.; Xie, Z.; Ma, Y.; Robson, P.M.; Chung, M.; et al. Artificial Intelligence–Enabled Rapid Diagnosis of Patients with COVID-19. Nat. Med. 2020, 26, 1224–1228. [Google Scholar] [CrossRef] [Scilit] [PubMed]
- Fazilati, F.; Rajabi, M.Z.; Alihosseini, N.; Farsani, M.E.; Sandid, S.H.; Zamani, S.; Farahani, M.A.; Biriaei, F.; Sadeghipour, F.; Mirshamsi, M.T.; et al. TRIAGE: Trustworthy Reporting and Assessment for Clinical Gain and Effectiveness of AI Models. Diagnostics 2026, 16, 666. [Google Scholar] [CrossRef] [Scilit] [PubMed]
- Obaido, G.; Mienye, I.D.; Aruleba, K.; Chukwu, C.W.; Esenogho, E.; Modisane, C. A Systematic Review of Contrastive Learning in Medical AI: Foundations, Biomedical Modalities, and Future Directions. Bioengineering 2026, 13, 176. [Google Scholar] [CrossRef] [Scilit] [PubMed]
- Huang, S.; Yang, J.; Fong, S.; Zhao, Q. Artificial intelligence in the diagnosis of COVID-19: Challenges and perspectives. Int. J. Biol. Sci. 2021, 17, 1581–1587. [Google Scholar] [CrossRef] [Scilit] [PubMed]
Disclaimer/Publisher’s Note: The statements, opinions and data contained in all publications are solely those of the individual author(s) and contributor(s) and not of MDPI and/or the editor(s). MDPI and/or the editor(s) disclaim responsibility for any injury to people or property resulting from any ideas, methods, instructions or products referred to in the content. |
© 2026 by the author. Licensee MDPI, Basel, Switzerland. This article is an open access article distributed under the terms and conditions of the Creative Commons Attribution (CC BY) license.