The impact of machine-learning-derived lean psoas muscle area on prognosis of type B aortic dissection patients undergoing endovascular treatment.

Abstract

The aim of this work was to investigate the impact of machine-learning-derived baseline lean psoas muscle area (LPMA) for patients undergoing thoracic endovascular aortic repair. A retrospective study was undertaken of acute and subacute complicated type B aortic dissection patients who underwent endovascular treatment from 2010 to 2017. LPMA (a marker of frailty) was calculated by multiplying psoas muscle area and density measured at L3 level from the computed tomography. The optimal cut-off value of LPMA was determined by the Cox hazard model with restricted cubic spline. A total of 428 patients who met the inclusion criteria were included in this study. Patients were classified into low LPMA group (n = 218) and high LPMA group (n = 210) using the cut-off value of 395 cm2 Hounsfield unit. An automatic muscle segmentation algorithm was developed based on U-Net architecture. There was high correlation between machine-learning method and manual measurement for psoas muscle area (r = 0.91, P < 0.001) and density (r = 0.90, P < 0.001). Multivariable regression analyses revealed that baseline low LPMA (<395 cm2 Hounsfield unit) was an independent positive predictor for 30-day (odds ratio 5.62, 95% confidence interval 1.20-26.23, P = 0.028) and follow-up (hazard ratio 5.62, 95% confidence interval 2.68-11.79, P < 0.001) mortality. Propensity score matching and subgroup analysis based on age (<65 vs ≥65 years) confirmed the independent association between baseline LPMA and follow-up mortality. Baseline LPMA could profoundly affect the prognosis of patients undergoing thoracic endovascular aortic repair. It was feasible to integrate the automatic muscle measurements into clinical routine.