tetano
Editor, Senior Moderator
J Thorac Dis
. 2025 Oct 31;17(10):7487-7497.
doi: 10.21037/jtd-2025-353. Epub 2025 Sep 25. Development and validation of a pneumonia severity prediction model using AI analysis of abdominal and paravertebral intramuscular fat on chest CT in COVID-19 patients
Bin Lin[SUP] #[/SUP][SUP] 1 [/SUP], Xiao Luo[SUP] #[/SUP][SUP] 1 [/SUP], Liangji Lu[SUP] 1 [/SUP], Kaipeng Jin[SUP] 2 [/SUP], Zhiyuan Gao[SUP] 2 [/SUP], Baoxin Qian[SUP] 3 [/SUP], Shuyue Wang[SUP] 1 [/SUP], Peiyu Huang[SUP] 1 [/SUP], Minming Zhang[SUP] 1 [/SUP]
Affiliations
Background: Pneumonia severity assessment is critical for clinical intervention, yet traditional methods often lack precision. This study addresses the clinical need for noninvasive, accurate tools by evaluating artificial intelligence (AI)-driven analysis of chest computed tomography (CT) scans to predict pneumonia severity and disease progression in coronavirus disease 2019 (COVID-19) patients, leveraging body composition metrics.
Methods: This retrospective study included 314 COVID-19 patients who underwent chest CT scans. Inclusion criteria were confirmed COVID-19 diagnosis and availability of CT imaging. Pneumonia severity, determined as the gold standard by expert radiologists assessing lesion percentage on CT, was categorized into Grade 1 (1-24%) and Grades 2-4 (≥25%). AI software measured subcutaneous adipose tissue (SAT), visceral adipose tissue (VAT), paravertebral muscle (PM) area, mean CT values of PM, and the proportion of paravertebral intramuscular fat (PIMF). The dataset was split into training (70%) and validation (30%) samples to develop and test the prediction model. Significant parameters (P<0.05) were analyzed using Spearman correlation, and binary logistic regression identified independent predictors.
Results: The cohort included 314 patients {median age, 77 years [interquartile range (IQR): 69-85.2 years], 199 men}. Significant differences in PIMF proportion were observed across severity groups (P<0.05). PM area and mean CT values negatively correlated with PIMF proportion (r=-0.452, -0.395, all P<0.05). Binary logistic regression identified PIMF proportion (coef. =20.512) and PM mean CT values (coef. =-0.059) as independent predictors. The AUC for PIMF proportion was 0.700 [95% confidence interval (CI): 0.623-0.777], with sensitivity of 65% and specificity of 68% in the validation set.
Conclusions: These findings support the use of AI-based CT analysis in clinical settings to enhance early risk stratification and guide targeted interventions for COVID-19 patients.
Keywords: Obesity; arcopenia; computed tomography (CT); pneumonia.
. 2025 Oct 31;17(10):7487-7497.
doi: 10.21037/jtd-2025-353. Epub 2025 Sep 25. Development and validation of a pneumonia severity prediction model using AI analysis of abdominal and paravertebral intramuscular fat on chest CT in COVID-19 patients
Bin Lin[SUP] #[/SUP][SUP] 1 [/SUP], Xiao Luo[SUP] #[/SUP][SUP] 1 [/SUP], Liangji Lu[SUP] 1 [/SUP], Kaipeng Jin[SUP] 2 [/SUP], Zhiyuan Gao[SUP] 2 [/SUP], Baoxin Qian[SUP] 3 [/SUP], Shuyue Wang[SUP] 1 [/SUP], Peiyu Huang[SUP] 1 [/SUP], Minming Zhang[SUP] 1 [/SUP]
Affiliations
- PMID: 41229871
- PMCID: PMC12603512
- DOI: 10.21037/jtd-2025-353
Background: Pneumonia severity assessment is critical for clinical intervention, yet traditional methods often lack precision. This study addresses the clinical need for noninvasive, accurate tools by evaluating artificial intelligence (AI)-driven analysis of chest computed tomography (CT) scans to predict pneumonia severity and disease progression in coronavirus disease 2019 (COVID-19) patients, leveraging body composition metrics.
Methods: This retrospective study included 314 COVID-19 patients who underwent chest CT scans. Inclusion criteria were confirmed COVID-19 diagnosis and availability of CT imaging. Pneumonia severity, determined as the gold standard by expert radiologists assessing lesion percentage on CT, was categorized into Grade 1 (1-24%) and Grades 2-4 (≥25%). AI software measured subcutaneous adipose tissue (SAT), visceral adipose tissue (VAT), paravertebral muscle (PM) area, mean CT values of PM, and the proportion of paravertebral intramuscular fat (PIMF). The dataset was split into training (70%) and validation (30%) samples to develop and test the prediction model. Significant parameters (P<0.05) were analyzed using Spearman correlation, and binary logistic regression identified independent predictors.
Results: The cohort included 314 patients {median age, 77 years [interquartile range (IQR): 69-85.2 years], 199 men}. Significant differences in PIMF proportion were observed across severity groups (P<0.05). PM area and mean CT values negatively correlated with PIMF proportion (r=-0.452, -0.395, all P<0.05). Binary logistic regression identified PIMF proportion (coef. =20.512) and PM mean CT values (coef. =-0.059) as independent predictors. The AUC for PIMF proportion was 0.700 [95% confidence interval (CI): 0.623-0.777], with sensitivity of 65% and specificity of 68% in the validation set.
Conclusions: These findings support the use of AI-based CT analysis in clinical settings to enhance early risk stratification and guide targeted interventions for COVID-19 patients.
Keywords: Obesity; arcopenia; computed tomography (CT); pneumonia.