Semi-automated border detection for right ventricular volume estimation from MR images

Two different methods for semi-automated right ventricular (RV) endocardial border detection from MR images, based on different implementation of level set technique, have been developed and validated. Dynamic, ECG-gated, steady-state free precession short axis images were obtained in 26 consecutive patients. An expert cardiologist provided the “gold standard” for RV dimensions, by manually tracing the endocardial contours. Semi-automated detection was applied to obtain RV end-diastolic and end-systolic volumes, as well as stroke volume and ejection fraction. Comparison with “gold standard” was performed by linear regression and Bland-Altman analyses. Results showed high correlations and small biases and narrow limits of agreement with the “gold standard” values. Both methods provided reliable measurements of RV dimensions; however, better accuracy is related to higher manual interaction.