Use of 18F-FDG PET/CT texture analysis to diagnose cardiac sarcoidosis

Purpose(18)F-fluorodeoxyglocose positron emission tomography (FDG PET) plays a significant role in the diagnosis of cardiac sarcoidosis (CS). Texture analysis is a group of computational methods for evaluating the inhomogeneity among adjacent pixels or voxels. We investigated whether texture analysis applied to myocardial FDG uptake has diagnostic value in patients with CS.MethodsThirty-seven CS patients (CS group), and 52 patients who underwent FDG PET/CT to detect malignant tumors with any FDG cardiac uptake (non-CS group) were studied. A total of 36 texture features from the histogram, gray-level co-occurrence matrix (GLCM), gray-level run length matrix (GLRLM), gray-level zone size matrix (GLZSM) and neighborhood gray-level difference matrix (NGLDM), were computed using polar map images. First, the inter-operator and inter-scan reproducibility of the texture features of the CS group were evaluated. Then, texture features of the patients with CS were compared to those without CS lesions.ResultsTwenty-eight of the 36 texture features showed high inter-operator reproducibility with intraclass correlation coefficients (ICCs) over 0.80. In addition, 17 of the 36 showed high inter-scan reproducibility with ICCs over 0.80. The SUVmax showed no difference between the CS and non-CS group [7.362.77 vs. 8.78 +/- 4.65, p=0.45, area under the curve (AUC)=0.60]. By contrast, 16 of the 36 texture features could distinguish CS from non-CS grsoup with AUC>0.80. Multivariate logistic regression analysis after hierarchical clustering concluded that long-run emphasis (LRE; P=0.0004) and short-run low gray-level emphasis (SRLGE; P=0.016) were significant independent factors that could distinguish between the CS and non-CS groups. Specifically, LRE was significantly higher in CS than in non-CS (30.1 +/- 25.4 vs. 11.4 +/- 4.6, P<0.0001), with high diagnostic ability (AUC=0.91), and had high inter-operator reproducibility (ICC=0.98).Conclusions p id=Par4 The texture analysis had high inter-operator and high inter-scan reproducibility. Some of texture features showed higher diagnostic value than SUVmax for CS diagnosis. Therefore, texture analysis may have a role in semi-automated systems for diagnosing CS.