Development and optimization SPECT RNA cluster analysis for the prediction of CRT outcome

1678 Objectives Phase analysis of radionuclide angiography (RNA) has been investigated for its potential to predict the outcome of cardiac resynchronization therapy (CRT). With this approach, information about the complex shape of the time-activity curve (TAC) is lost as it uses a simple sinusoidal fit and this may reduce its predictive value. We propose a new method, cluster analysis, which directly evaluates the TAC. In this study, cluster analysis is developed and optimized as an alternative method for predicting outcome to CRT. Methods 48 patients (LVEF 120 ms) underwent a SPECT RNA scan prior to and after CRT. A semi-automated segmentation algorithm sampled the LV wall to produce 568 TACs which were individually compared to a 50-subject normal database. The number of abnormal TAC segments was used to predict response. Based on a 5-segment heart model, receiver-operating characteristic (ROC) analysis was used to calculate the area under the curve (AUC) for several abnormal cluster definitions. AUC and concordance (κ) were compared for all patients and ischemic (N=26) and non-ischemic sub-populations with response assessed at follow-up RNA. Results The best cluster analysis results were obtained from septal wall analysis. Cluster analysis produced an ROC AUC of 0.82, and an optimal operating point of 94% sensitivity and 67% specificity in the ischemic population (κ = 0.64). Optimal settings produced ROC AUC values of 0.67 and concordance values less than 0.35 for both the whole and non-ischemic populations. Conclusions SPECT RNA cluster analysis algorithms were successfully developed and optimized for the prediction of CRT outcome. Results suggest an improved concordance in the ischemic group over SPECT RNA phase analysis (κ = 0.37) and PET scar size analysis (κ = 0.42), but the sample size was not large enough to show a significant difference (p>0.05).