Empowering Medical Imaging with Artificial Intelligence - A Review of Machine Learning Approaches for COVID-19 Detection and Segmentation

Aug 6, 2026

Upload by Finn

Since 2019, the rapid spread of COVID-19 and emerging variants have increased the need for reliable diagnostic support. X-ray and computed tomography (CT) imaging play a central role in assessing infection, while artificial intelligence offers scalable assistance to clinicians. The review focuses on machine learning methods that support COVID-19 diagnosis by detecting and segmenting lesions on radiographic and tomographic images. It examines approaches for classification, detection, and lung-related representations, highlighting how models can reduce manual workload and improve decision consistency despite lesion variability.

LanguageEnglish

CategoryResearch & Report

FormatPDF

Pages16

Reading Time40 min

AccessFree to download

Introduction

COVID-19 impact and diagnostic needs

Imaging modalities: X-ray vs CT

Disease progression patterns in CT

Machine Learning for COVID-19 Imaging

Learning paradigms: supervised, unsupervised, semi-supervised

Deep learning and reinforcement learning

Review Focus: Detection, Classification, Segmentation

Lesion detection on radiographic and CT images

Segmentation of infection regions

Clinical relevance in hospital workflows

Empowering Medical Imaging with Artificial Intelligence: A Review of Machine Learning Approaches for the Detection, and Segmentation of COVID-19 Using Radiographic and Tomographic Images Sayed Amir Mousavi Mobarakeh, Kamran Kazemi, Ardalan Aarabi, Habibollah Danyali Abstract — Since 2019, the global dissemination of the Coronavirus and its novel strains has resulted in a surge of new infections. The use of X-ray and computed tomography (CT) imaging techniques is critical in diagnosing and managing COVID-19. Incorporating artificial intelligence (AI) into the field of medical imaging is a powerful combination that can provide valuable support to healthcare professionals.This paper focuses on the methodological approach of using machine learning (ML) to enhance medical imaging for COVID-19 diagnosis.For example, deep learning can accurately distinguish lesions from other parts of the lung without human intervention in a matter of minutes.Moreover, ML can enhance performance efficiency by assisting radiologists in making more precise clinical decisions, such as detecting and distinguishing Covid-19 from different respiratory infections and segmenting infections in CT and X-ray images, even when the lesions have varying sizes and shapes.This article critically assesses machine learning methodologies utilized for the segmentation, classification, and detection of Covid-19 within CT and X-ray images, which are commonly employed tools in clinical and hospital settings to represent the lung in various aspects and extensive detail.Thereis a widespread expectation that this technology will continue to hold a central position within the healthcare sector, driving further progress in the management of the pandemic. Index Terms— COVID-19, Machine learning, Classification, Segmentation, Data acquisition . I. INTRODUCTION Since December 2019, the global population has encountered a profound health crisis in the shape of the COVID-19 pandemic, triggered by the SARS-CoV-2 virus. Despite the administration of over 13 billion vaccine doses as of December 13, 2023, new variants of the virus and mutations continue to contribute to the rising number of infections worldwide. As of May 3, 2023, the confirmed cumulative number of Covid-19 cases has reached 772,386,069, with 6,987,222 reported deaths in over 200 regions and territories [1] . During the recent 28-day period from April 3 to April 30, 2023, there were a total of approximately 2.8 million novel cases and in excess of 17,000 fatalities reported globally [2] . These findings suggest that while the majority of people have been vaccinated, the emergence of new variants can still lead to an increase in infections and fatalities. The COVID-19 disease exhibits various symptoms, including but not limited to cough, fever, shortness of breath, anosmia, and ageusia [3] . The polymerase chain reaction (PCR) technique is commonly perceived as the benchmark nucleic acid test for identifying specific viruses, and the real-time reverse transcriptase-PCR (RT-PCR) testis well known for its high sensitivity, specificity, and rapid detection [4]. However, the RT-PCR test has constraints due to kit performance and sample collection, with reported inadequate sensitivity ranging from 30% to 60%[5], [6] . In addition to RT-PCR, medical imaging techniques provide vital information for the diagnosis. It should be noted that the symptoms have not necessarily correlate with the severity of the infection [3] . When compared to RT-PCR, chest computed tomography (CT) exhibits greater sensitivity in detection [5]. CT scans are favored over X-rays owing to their ability to provide a three-dimensional depiction of the lungs and their wider use in the identification of pulmonary infections [7], [8], [9] . During the early stage, COVID-19 infections in CT scans are categorized as ground-glass opacity (GGO), while in the subsequent stage, they are characterized by pulmonary consolidation [5], [10] . GGO is a hazy

FAQs

Why are X-ray and CT imaging important for COVID-19 diagnosis?

They provide vital visual information for detecting and monitoring infection. CT is often favored for higher sensitivity and three-dimensional depiction of the lungs.

What limitations affect RT-PCR testing compared with imaging?

RT-PCR performance can be constrained by kit behavior and sample collection, with reported inadequate sensitivity in some settings.

How does machine learning improve COVID-19 imaging workflows?

Machine learning can automate lesion detection and support classification and segmentation, reducing the labor-intensive annotation process and assisting clinicians in more consistent decisions.

Aug 6, 2026

Upload by Finn