AI-Supported Semiquantitative CT Characterization of COVID-19 Pneumonia Across SARS-CoV-2 Wild-Type, Alpha, and Delta Variants and Its Association with Clinical Parameters
Year: 2026
Type: article
Abstract: Objective: An AI-based software prototype for the detection of coronavirus disease 2019 (COVID-19) pneumonia was evaluated regarding its potential to quantify pulmonary opacities across Severe Acute Respiratory Syndrome Coronavirus 2 (SARS-CoV-2) variant groups, incorporating clinical and radiological parameters to facilitate faster diagnosis and therapy. Methods: Chest computed tomography (CT) examinations of 255 patients with confirmed COVID-19 pneumonia were analyzed using an AI-supported semiquantitative CT software prototype between November 2020 and December 2021. The results were correlated with clinical and laboratory parameters. The cohort consisted of wild-type (n = 93), Alpha (n = 66), Beta/Gamma (n = 3), and Delta (n = 93). Beta/Gamma cases were excluded from subgroup comparisons due to the small number of patients. Results: The mean AI-derived COVID-19 probability was 0.84 ± 0.33. After exclusion of five technically non-evaluable measurements, global mean opacity attenuation differed significantly among the wild-type (−475.12 ± 116.47 Hounsfield Units (HU)), Alpha (−464.68 ± 114.80 HU), and Delta groups (−423.92 ± 112.56 HU; Kruskal–Wallis H(2) = 12.062, p = 0.002). The Volume of Opacity showed no significant differences between wild-type (984.64 ± 908.92 mL), Alpha (1161.17 ± 933.55 mL), and Delta (988.19 ± 866.20 mL) (p = 0.339). The extent of pulmonary involvement (Volume of Opacity) correlated with the duration of hospital stay (r = 0.311, p < 0.001**), duration of intensive care treatment (r = 0.468, p < 0.001**), and the laboratory parameters CRP, leukocyte count, relative lymphocyte count, and AST (p < 0.001**). It showed no significant correlation with the duration of invasive ventilation (p > 0.05). A total of 37 (14.5%) patients died and 92 (36.1%) required intensive care unit (ICU) admission. Manual corrections of lung lobe and infiltrate boundaries were performed in 221/255 datasets and were required in all assessed datasets (221/221, 100%). Conclusions: The CT Pneumonia Analysis Prototype enabled semiquantitative assessment of pulmonary opacities across different SARS-CoV-2 variants but required radiological oversight and manual correction of the automated segmentations. Given the retrospective observational design, the findings should be interpreted in light of potential confounding and heterogeneity in CT acquisition protocols. The observed associations suggest that combining semiquantitative CT parameters with clinical and laboratory data may be of potential relevance for future prognostic models, which requires further evaluation in multivariable analyses.
Source: Diagnostics
Language: English
Topic: COVID-19 diagnosis using AI
Subfield: Radiology, Nuclear Medicine and Imaging
Field: Medicine
Domain: Health Sciences
Open Access status: gold
Estimated APC paid: $3,137
OpenAlex ID: W7225627264