Texture analysis of delayed contrast-enhanced computed tomography to diagnose cardiac sarcoidosis

To investigate the diagnostic value of texture analysis to differentiate cardiac sarcoidosis (CS) from other non-ischemic cardiomyopathies (non-CS).

Twenty CS patients and 15 non-CS patients who had undergone myocardial CT delayed enhancement (CTDE) were included. A total of 36 texture features were calculated according to the CT attenuation of CTDE. We investigated the diagnostic value to differentiate CS from non-CS. We also assessed the intra- and inter-rater reproducibility for each feature and inter-observer agreement for visual assessment.

Seven extracted features had significantly higher run length non-uniformity (RLNU) values (5.4 × 10² ± 6.2 × 10² vs. 11.2 × 10² ± 4.9 × 10², p = 0.037) and significantly lower low gray-level zone emphasis (LGZE) values (7.1 × 10^-3 ± 8.6 × 10^-3 vs. 18.1 × 10^-3 ± 16.9 × 10^-3, p = 0.017) in CS than in non-CS. Intra- and inter-rater reproducibility of RLNU and LGZE were excellent (ICCs > 0.8), while inter-observer agreement of visual assessment was poor (kappa = 0.19). The accuracies of texture analysis were 69% with RLNU and 71% with LGZE, which were better than that of visual assessment.

Texture analysis of CTDE could differentiate CS from non-CS with high reproducibility.