As of March 2022, there have been nearly 450 million SARS-CoV-2 infections and over 6 million COVID-19 related deaths. Whether an infection escalates into the need for mechanical ventilation or results in death is highly dependent on a number of parameters and conditions.
“Several recent studies identified clinical parameters associated with a higher vulnerability to severe clinical outcome,” said René Hosch, MSc, a research assistant at Essen University Hospital in Germany, who co-authored a Scientific Reports article on the topic. “These studies use general characteristics, including Body Mass Index, to better predict the clinical course severity of individual SARS-CoV-2 patients.”
According to Hosch, who spoke during a Monday session, one reason Body Mass Index (BMI) is the characteristic of choice for predicting disease severity is that the information is readily accessible.
“Although convenient, BMI is a very shallow approximation of body composition, as its interpretability can be negatively affected by physical anomalies or unusual body proportions,” he said.
A Shift Towards Body Tissue Composition
To address this shortcoming, researchers, including Hosch, are now using more detailed features that describe a patient’s body composition.
“Body tissue composition is an example of a biomarker with high diagnostic and prognostic value, not only in cardiovascular, oncological and orthopedic disease, but also in rehabilitation medicine and drug dosage,” Hosch explained.
The problem is that obtaining those biomarkers requires one to conduct a Body Composition Analysis (BCA), which is a manual and thus time-consuming process. As a result, many BCA methods have been semi-automated or use only reference regions, such as the L3 lumbar vertebra, for assessment.
But, as Hosch explained, a 2D reference image at the L3 level is only a rough estimation of the tissue composition, which may differ throughout the volume.
“Furthermore, the L3 region is typically not captured on a regular chest CT, rendering this method unsuitable for assessing patients with SARS-CoV-2 pneumonia in clinical routine,” he added.
Automation to the Rescue
Frustrated by these limitations, Hosch and the team at Essen University Hospital turned to deep learning. Specifically, they used a fully automatic 3D semantic segmentation convolutional neural network (CNN) that the research group previously developed for an earlier project.
“This allowed us to precisely quantify relevant body tissues like bone, muscle and multiple adipose tissues and combine them as potential predictive biomarkers to determine the clinical outcome of SARS-CoV-2 patients,” Hosch said.
The segmentation network was used for the extraction of multiple BCA features for 918 newly admitted patients, each of whom received a chest CT scan shortly after admission. Researchers used this data to investigate the statistical significance of the relationship between various biomarkers and two endpoints, namely severity and mortality. What they found was that CT-based body composition analysis biomarkers can be used to assess the risk for mortality and disease severity in patients admitted with COVID-19.
“These findings demonstrate that CT-based BCA biomarkers contain considerable information that is highly relevant to COVID-19 patients,” Hosch explained. “For example, patients with higher cardiac markers and lower sarcopenia markers are at higher risk of severe illness or death.”
Improving Risk Stratification
Hosch says the findings also show that, when it comes to assessing the risk of mortality and disease severity for COVID-19 patients, detailed BCA features are more relevant than, and thus should replace, the more commonly used Body Mass Index.
“BCA markers based on raw BCA parameters extracted automatically from CT scans have the potential to improve risk stratification in patients with acute SARS-Cov-2 infection,” Hosch concluded. “With the support of our robust and fully automatable methodology, these parameters should be considered when developing new risk scores.”
Access the presentation, “Biomarkers Extracted by Fully Automated Body Composition Analysis From Chest CT Correlate with SARS-CoV-2 Outcome Severity,” (M7-SSCH09-3) on demand at Meeting.RSNA.org.