Development of a Simplified Protocol for Respiratory Muscle Segmentation in Unenhanced Chest CT and Identification of New Radiomic Biomarkers of Sarcopenia in Lung Diseases: A Retrospective Study

Abstract

Background/Objectives: Respiratory muscle sarcopenia worsens outcomes in chronic lung disease, and quantitative Computed Tomography (CT) may provide objective biomarkers; this study aimed to develop a time-efficient segmentation protocol and identify radiomic biomarkers of respiratory muscle sarcopenia. Methods: This retrospective study analyzed 30 unenhanced chest CT from adult patients. The whole volume of the pectoralis major (PM), pectoralis minor (Pm), serratus anterior (SA), and fourth intercostal (4I) muscles was manually segmented. Patients were classified as sarcopenic or non-sarcopenic. Radiomics features and mean muscle density were extracted using PyRadiomics. Features associated with sarcopenia were selected using Least Absolute Shrinkage and Selection Operator (LASSO) regression and backward stepwise selection. Four sets of slices consisting of one, three, five, and seven slices were then sampled from each muscle around a fixed anatomical landmark. Deviations of each set of slices from whole-muscle metrics were evaluated using Mean Absolute Error (MAE) and Mean Absolute Percentage Error (MAPE). Results: Features selection identified 25 biomarkers of sarcopenia in PM, 24 in Pm, and 34 in SA. Variability-related features were significantly associated with sarcopenia (OR = 2.26; p = 0.012), while structural features showed an inverse association (OR = 0.18; p = 0.004). Mean muscle density and most radiomic features were well represented by single slice for every muscle. In the PM and Pm, eight and six radiomic features were better approximated segmenting more than one slice (p < 0.05). Conclusions: Radiomics enables quantitative assessment of sarcopenia. For SA, a simplified segmentation protocol consisting of a single slice enables approximating muscle density and radiomics of whole muscle volume. For PM and Pm, three or more slices allow a better representation of 8 and 6 radiomic features, respectively.

Affiliated Institutions

• University of Palermo IT
• National Research Council IT
• University of Genoa IT
• Istituto Mediterraneo per i Trapianti e Terapie ad Alta Specializzazione IT
• Institute of Electronics, Computer and Telecommunication Engineering IT
• Ospedale Policlinico San Martino IT
• Politecnico di Milano IT
• Azienda Ospedaliera Universitaria Policlinico "Paolo Giaccone" di Palermo IT

Related Publications