Texture Analysis on Myocardial Perfusion SPECT to Diagnose Myocarditis

Objectives: Myocardial perfusion SPECT(MPS) has been demonstrated to be a useful noninvasive technique to detect inflammation and necrosis in viral myocarditis(VMC)¹. Texture analysis comprises a wide range of heterogenetic spatial distributions of pixel grey-levels characterizing the underlying tissue texture, and has been recently validated in cardiac studies^{2, 3}.We aim to evaluate whether the texture analysis on MPS can offer incremental diagnostic information over myocardial perfusion for identifying patients with VMC.

Methods: Twenty-one patients with clinically confirmed VMC (male, 12; age, 27 ± 11 years) and twenty-one normal controls (male, 10; age, 47 ± 20 years) were referred for rest ^99mTc-sestamibi MPS. Planar images were reconstructed and reoriented into LV short-axis images with Emory Cardiac Toolbox. For each patient, the short-axis images were submitted to an automatic sampling algorithm which searched in 3D for the maximal count circumferential profiles to represent the regional perfusion level. Texture analysis was applied to the perfusion samples by using a freely available software package; texture features, such as 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 previously reported³. Total perfusion defect (TPD) was obtained from QPS software. Parameters were compared between the two groups, and the diagnostic performance was evaluated by the area under the curve (AUC) derived from the receiver operating characteristic (ROC) analysis.

Results: 27 of the 38 texture features could distinguish VMC from control group. Multivariable logistic regression analysis after hierarchical clustering showed that Dissimilarity of gray-level co-occurrence matrix (DissimilarityGLCM) was the only independent predictors of VMC (P<0.001). Compared with the control group, DissimilarityGLCM and TPD were both significantly higher in VMC group (P<0.001, P<0.05; respectively), whereas DissimilarityGLCM was the better diagnostic parameter (AUC: 0.87, sensitivity: 81%, specificity: 81%).

Conclusions: Texture analysis applied on MPS has promise to increase the diagnosis of VMC.