Automatic generation of parametric myocardial blood flow images using a basis function method and cluster analysis showed high agreement with manually derived nonlinear least-squares regression (ICC 0.984).
Observational (n=19)
No
Does automatic generation of parametric MBF images using cluster analysis and BFM agree with manually derived MBF using NLR in patients undergoing [15O]H2O PET/CT?
Parametric MBF images of diagnostic quality can be generated automatically using cluster analysis and a basis function method with RV spillover correction, showing high agreement with manual methods.
Effect estimate: ICC 0.984
PURPOSE: Parametric imaging of absolute myocardial blood flow (MBF) using (15)OH(2)O enables determination of MBF with high spatial resolution. The aim of this study was to develop a method for generating reproducible, high-quality and quantitative parametric MBF images with minimal user intervention. METHODS: Nineteen patients referred for evaluation of MBF underwent rest and adenosine stress (15)OH(2)O positron emission tomography (PET) scans. Ascending aorta and right ventricular (RV) cavity volumes of interest (VOIs) were used as input functions. Implementation of a basis function method (BFM) of the single-tissue model with an additional correction for RV spillover was used to generate parametric images. The average segmental MBF derived from parametric images was compared with MBF obtained using nonlinear least-squares regression (NLR) of VOI data. Four segmentation algorithms were evaluated for automatic extraction of input functions. Segmental MBF obtained using these input functions was compared with MBF obtained using manually defined input functions. RESULTS: The average parametric MBF showed a high agreement with NLR-derived MBF intraclass correlation coefficient (ICC) = 0.984. For each segmentation algorithm there was at least one implementation that yielded high agreement (ICC > 0.9) with manually obtained input functions, although MBF calculated using each algorithm was at least 10% higher. Cluster analysis with six clusters yielded the highest agreement (ICC = 0.977), together with good segmentation reproducibility (coefficient of variation of MBF <5%). CONCLUSION: Parametric MBF images of diagnostic quality can be generated automatically using cluster analysis and a implementation of a BFM of the single-tissue model with additional RV spillover correction.
Harms et al. (2011) conducted an observational in Evaluation for coronary artery disease (n=19). Automatic parametric imaging using basis function method and cluster analysis vs. Nonlinear least-squares regression with manually defined input functions was evaluated on Agreement between average parametric MBF and NLR-derived MBF (ICC 0.984). Automatic generation of parametric myocardial blood flow images using a basis function method and cluster analysis showed high agreement with manually derived nonlinear least-squares regression (ICC 0.984).