Covid-19 Diagnosis Empowered With Deep Semi-supervised Learning Techniques Using X-ray And Ct Images: A Systematic Review

Résumé: Abstract. Background: The early detection and diagnosis of COVID-19 and the accurate separation of non-COVID-19 cases in a quick and cheap manner are among the main challenges in the COVID- 19 pandemic to ensure recovery treatment for patients which will help to save patient’s lives. Deep learning techniques proved themselves to be a novel prediagnostic detection methodology of COVID- 19. In particular, the deep supervised learning has successfully applied to analyse and detect COVID-19 on chest X-ray and CT scan images. However, the performance of such models is heavily dependent on the availability of a large labelled dataset. This is often a limitation because the creation of which is a very expensive and time consuming task, and especially imposes a great challenge for a novel disease. Semi-supervised learning has shown the ability to improve the detection accuracy of supervised models whilst requiring a small labelled datasets. This makes the semi-supervised an interesting alternative of significant practical importance for identifying COVID-19. Material and Methods: The present systematic review was conducted by searching the three databases of PubMed, Web of Science and Science Direct from December 1, 2019, to May 15, 2022, based on a search strategy. A total of 392 articles were extracted and, by applying the inclusion and exclusion criteria, 33 articles were selected as the research population. Result: In this study we reviewed studies which used deep semi-supervised learning methods on chest X-ray images and CT scans for the detection and diagnosis of COVID-19 and compared their performance. According to the findings, deep semi-supervised learning-based models are able to improve the diagnostic accuracy and robustness without exhaustive labeling. Conclusion: The application of deep semi-supervised learning in the field of COVID-19 radiologic image processing facilitates an accurate and reliable diagnosis. The use of deep learning technology and semi-supervised learning paradigm in the detection and diagnosis for COVID-19 reduce false-positive and negative errors, leading to an improved patient care and management.

Publié dans la revue: Journal of Molecular and Pharmaceutical Sciences