Automated computed tomography measures predict functional decline in cystic fibrosis

<bold>Objective:</bold> Clinical trials in cystic fibrosis (CF) are limited by the measurement variability of forced expiratory volume in the first second (FEV1). Computer analysis tools that objectively and accurately quantify airways, fat, and muscle on computed tomography (CT) imaging may identify patients likely to show FEV1 decline. <bold>Methods:</bold> Quantitation of representative skeletal muscle and subcutaneous fat volumes at the 9th thoracic vertebrae level and patent airway trees were automatically segmented using a deep learning model (Figure). A computational pipeline, AirQuant, measured intersegmental airway tapering and intrasegmental tortuosity in patients with FEV1>40% (mild/moderate disease). FEV1% decline at one year was predicted using linear mixed-effects models. Thresholds of muscle, fat, airway measures predicting FEV1 decline >5% identified in a derivation cohort (n=84) by maximizing the Youden Index were tested in a validation cohort (n=74). Prediction accuracy was expressed as area under the curve (AUC). <bold>Results:</bold> 25/84 (derivation cohort) and 24/74 (validation cohort) patients had an FEV1 decline >5%. The CT measures with the highest predictive AUC in the validation cohort were volume of fat (0.56), volume of muscle (0.63), airway intertapering (0.57) and airway tortuosity (0.60). <bold>Conclusion:</bold> Automated CT measurements that predict decline in CF could represent a cohort enrichment tool for clinical trials. <fig><object-id>erj;64/suppl_68/OA5470/F1</object-id><object-id>F1</object-id><object-id>F1</object-id><graphic></graphic></fig>