This study describes and validates a new method for automatic segmentation of left ventricular mass (LVM) in myocardial perfusion SPECT (MPS) images. This is important for estimating the size of a perfusion defect as percentage of the left ventricle. METHODS: A total of 101 patients with known or suspected coronary artery disease underwent both rest and stress MPS and MRI. A new automated algorithm was trained in 20 patients (40 MPS studies) and tested in 81 patients (162 MPS studies). The algorithm, which segmented the left ventricle in the MPS images, is based on Dijkstra's algorithm and finds an optimal mid-mural line through the left ventricular wall. From this line, the endocardium and epicardium are identified on the basis of an individually estimated wall thickness and signal intensity. The algorithm was validated by comparing LVM in both stress and rest MPS, with LVM of the manually segmented left ventricle from MRI as the reference standard. For comparison, LVM was quantified using the software quantitative perfusion SPECT (QPS). RESULTS: The mean difference +/- SD in LVM between MPS and MRI was lower for the new method (6% +/- 15% LVM) than for QPS (18% +/- 19% LVM) for both mean difference (P < 0.001) and SD (P = 0.015). Linear regression analysis of LVM, comparing MPS and MRI, yielded R(2) = 0.83 using the new method and R(2) = 0.80 using QPS. Interstudy variability, measured as the coefficient of variance between rest MPS and stress MPS, was 6% for both the new method and QPS. Both the new algorithm and QPS systematically overestimated LVM in hearts with thin myocardium and underestimated LVM in hearts with thick myocardium. CONCLUSION: The new segmentation algorithm quantifies LVM with a significantly lower bias and variability than does the commercially available QPS software, when compared to manually segmented LVM by MRI. This makes the new algorithm an attractive method to use for estimating the size of the perfusion defect when expressing it as percentage of the left ventricle. This study shows that inaccurate estimation of wall thickness is the main source of error in automatic segmentation.
- Radiologi och bildbehandling