Abdominal imaging associates body composition with COVID-19 severity

Area of Science:

• Medical imaging
• Body composition analysis
• Epidemiology

Background:

• Understanding COVID-19 severity drivers and long-term effects remains crucial.
• Previous medical imaging studies on COVID-19 often used small datasets and focused on severe cases.
• The UK Biobank provided pre-pandemic and follow-up imaging data for recovered SARS-CoV-2 individuals and controls.

Purpose of the Study:

• To investigate longitudinal changes in body composition after SARS-CoV-2 infection.
• To identify pre-pandemic image-derived phenotypes associated with COVID-19 severity.
• To develop a predictive model for COVID-19 severity using imaging and machine learning.

Main Methods:

• Longitudinal analysis of body composition changes in recovered SARS-CoV-2 individuals (n=967) and controls (n=913) using UK Biobank data.
• Association analysis of pre-pandemic image-derived phenotypes (visceral adipose tissue, liver fat, muscle volume) with COVID-19 severity.
• Machine learning classifier training using demographic, anthropometric, and imaging traits to predict disease severity.

Main Results:

• Longitudinal analysis indicated a decrease in lung volume associated with SARS-CoV-2 positivity in a population with mostly mild cases.
• Increased pre-pandemic visceral adipose tissue, liver fat, and reduced muscle volume were linked to higher COVID-19 disease severity.
• Visceral fat, liver fat, and muscle volume demonstrated prognostic value for COVID-19 severity beyond standard measurements.

Conclusions:

• Pre-pandemic body composition, particularly visceral fat, liver fat, and muscle volume, are significant predictors of COVID-19 severity.
• A machine learning model combining imaging phenotypes and demographic data shows potential for identifying at-risk populations.
• Abdominal MRI-derived phenotypes and ensemble learning offer potential clinical utility for predicting severe COVID-19 outcomes.