Quantification of pulmonary opacities using artificial intelligence in chest CT scans during SARS-CoV-2 pandemic: validation and prognostic assessment.

Autor: Montoro, Fernando Sánchez, Gordo, María Luz Parra, Tascón, Áurea Díez, de Gracia, Milagros Martí, Velez, Silvia Ossaba, Fernández, Susana Fernández, Vallano, Rebeca Gil, Acosta Velásquez, Kevin Stephen
Předmět:
Zdroj: Egyptian Journal of Radiology & Nuclear Medicine; 9/14/2023, Vol. 54 Issue 1, p1-12, 12p
Abstrakt: Purpose: To assess whether the analysis of pulmonary opacities on chest CT scans by AI-RAD Companion, an artificial intelligence (AI) software, has any prognostic value. Background: In December 2019, a new coronavirus named SARS-CoV-2 emerged in Wuhan, China, causing a global pandemic known as COVID-19. The disease initially presents with flu-like symptoms but can progress to severe respiratory distress, organ failure, and high mortality rates. The overwhelming influx of patients strained Emergency Rooms worldwide. To assist in diagnosing and categorizing pneumonia, AI algorithms using deep learning and convolutional neural networks were introduced. However, there is limited research on how applicable these algorithms are in the Emergency Room setting, and their practicality remains uncertain due to most studies focusing on COVID-19-positive patients only. Methods: Our study has an observational, analytical, and longitudinal design. The sample consisted of patients who visited our emergency room from August 5, 2021, to September 9, 2021, were suspected of having COVID-19 pneumonia, and underwent a chest CT scan. They were categorized into COVID-19 negative and positive groups based on PCR confirmation. Lung opacities were evaluated separately by a team of radiologists and a commercial AI software called AI-Rad Companion (by Siemens Healthineers). After 5 months we gathered clinical data, such as hospital admission, intensive care unit (ICU) admission, death, and hospital stay. Results: The final sample included 304 patients (144 females, 160 males) with a mean age of 68 ± 19 std. Among them, 129 tested negative for COVID-19 and 175 tested positive. We used AI-generated opacity quantification, compared to radiologists' reports, to create receiver operating characteristic curves. The area under the curve ranged from 0.8 to 0.9 with a 95% confidence interval. We then adjusted opacity tests to a sensitivity cut-off of 95%. We found a significant association between these opacity tests and hospital admission and ICU admission (Chi-Squared, P < 0.05), as well as between the percentage of lung opacities and length of hospital stay (Spearman's rho 0.53–0.54, P < 0.05) in both groups. Conclusions: During the SARS-CoV-2 pandemic, AI-based opacity tests demonstrated an association with certain prognostic markers in patients with suspected COVID-19 pneumonia, regardless of whether a PCR-confirmed coronavirus infection was ultimately detected. [ABSTRACT FROM AUTHOR]
Databáze: Supplemental Index
Externí odkaz: