Making Every CT Count: Automated Body Composition Analysis for Clinical Use in COPD and Beyond

<bold>Background:</bold> Muscle loss in COPD worsens prognosis and increases morbidity and hospitalizations. While diagnostic CT scans contain valuable unused body composition (BC) data, current assessment methods are time-consuming and impractical for routine use, highlighting the need for automated solutions. <bold>Aim:</bold> To evaluate the agreement between a fully automated and an expert-driven semi-automated segmentation method for BC analysis in thoracic and abdominal CT scans. <bold>Methods:</bold> CT images from trauma patients without chronic disease were retrospectively analyzed. One axial slice per vertebral level (T2–L5) was segmented using a semi-automated method with manual corrections (viking Slice) and a fully automated method (DAFS Express). HU thresholds, based on Alberta protocol, muscle(MUS), visceral(VAT), intermuscular(IMAT) and subcutaneous fat (SAT). VAT from T2–T10 was excluded due to the lack of manual segmentation. Dice scores assessed agreement between methods. <bold>Results:</bold> 21patients (11 males; median age 29 years) were included, with 332 images analyzed. DAFS processed images: average of 18 seconds/ image (total 1 hour 37 minutes) compared to 1–9 minutes/image for manual segmentation. <table-wrap><object-id>erj;66/suppl_69/PA977/TB1</object-id><object-id>T1</object-id><object-id>TB1</object-id><caption> Table 1 Mean Dice score (95% confidence interval) </caption><table><colgroup><col></col><col></col><col></col><col></col><col></col></colgroup><tbody><tr><td></td><td>Number of images</td><td>R1 vs R2</td><td>DAFS vs R1</td><td>DAFS vs R2</td></tr><tr><td>MUS</td><td>335</td><td>.958 (CI .957; .960)</td><td>.943 (CI 940; .946)</td><td>.943 (CI .940; .946)</td></tr><tr><td>VAT</td><td>144</td><td>.972 (CI.961; .982)</td><td>.960 (CI .947; .974)</td><td>.965 (CI .956; .974)</td></tr><tr><td>SAT</td><td>335</td><td>.992 (CI .990; .993)</td><td>.926 (CI .922; .931)</td><td>.922 (CI .917; .928)</td></tr></tbody></table></table-wrap> <bold>Conclusion:</bold> The automated method showed excellent agreement with the semi-automated method, supporting its use as a rapid reproducible tool for BC analysis to aid early muscle loss detection in COPD and other diseases