Association of deep learning-derived temporalis sarcopenia with mortality in acute ischemic stroke.

Abstract

Background: Sarcopenia is associated with mortality following acute ischemic stroke (AIS), but diagnosis is time-consuming. Computed tomography (CT) measures of temporal muscle volume (TMV) and density (TMD) can be automatically extracted from existing scans using deep learning (DL). This study sought to demonstrate the incremental value of DL-derived TMV and TMD from CT scans on AIS mortality and length of stay (LOS).

Methods: In this retrospective, single-center cohort study, consecutive AIS patients admitted from 2014 to 2023 were included. TMV and TMD were quantified on admission CT scans by a novel DL model and represented as continuous or categorical variables. Patients were classified as non-sarcopenic, pre-sarcopenic (low TMV or TMD), or sarcopenic (low TMV and TMD) using 5th-percentile thresholds derived from 50 healthy adults. The primary outcome was 30-day all-cause mortality. Secondary outcomes were 365-day all-cause mortality and LOS. Multivariable logistic and linear regression were used.

Results: The cohort consisted of 2285 patients with 1151 (50%) females, and a mean (SD) age of 74.7 (13.7) years. There were 877 (38%) non-sarcopenic, 838 (37%) pre-sarcopenic, and 570 (25%) sarcopenic patients. Adjusted ORs (95% CIs) for 30-day mortality were 2.70 (1.64-4.46) and 2.91 (1.72-4.91) for pre-sarcopenia and sarcopenia. Findings were consistent across secondary outcomes. The association between sarcopenia and mortality was preserved after adding the Hospital Frailty Risk Score (HFRS) to the models. Findings were consistent in sensitivity analyses using cohort-derived threshold definitions.

Conclusions: TMV and TMD extracted using a novel DL model were incrementally predictive of AIS mortality.