Dynamic, ECG-gated, steady-state free precession short-axis images were obtained (GE Healthcare, 1.5T) in 8-12 slices in 15 patients with previous myocardial infarction. An expert cardiologist provided the reference values for: 1) left ventricular (LV) volumes and mass, by manually tracing endo and epicardial contours; 2) regional wall motion (WM) interpretation, by grading (normal, abnormal) three slices selected at apical, mid and basal level. Custom software based on image noise distribution and on image gradient was applied, from which end-diastolic (ED) and end-systolic (ES) volumes and mass were computed, as well as regional fractional area change (RFAC), from which automated classification of regional WM abnormality was defined. Comparison with reference values was performed by: 1) linear regression and Bland-Altman analyses for LV volumes and mass; 2) levels of agreement between the cardiologist WM grades and the automated classification. Optimal correlations (r2.97) and no bias were found for ED and ES volumes, while LV mass resulted in a good correlation (ED: r2=.81; ES: r2=.74) with a minimal overestimation (ED: 15.2g; ES: 8.7g) and narrow 95% limits of agreement (ED: ±30g; ES: ±33g). The automated interpretation resulted in high sensitivity, specificity, and accuracy (78%, 85%, 82%, respectively) of WM abnormalities. Combined automated endo and epicardial border detection from MRI images provides reliable measurements of LV dimensions and regional WM classification.
Development and Validation of Automated Endocardial and Epicardial Contour Detection for MRI Volumetric and Wall Motion Analysis
Parodi O;
2010
Abstract
Dynamic, ECG-gated, steady-state free precession short-axis images were obtained (GE Healthcare, 1.5T) in 8-12 slices in 15 patients with previous myocardial infarction. An expert cardiologist provided the reference values for: 1) left ventricular (LV) volumes and mass, by manually tracing endo and epicardial contours; 2) regional wall motion (WM) interpretation, by grading (normal, abnormal) three slices selected at apical, mid and basal level. Custom software based on image noise distribution and on image gradient was applied, from which end-diastolic (ED) and end-systolic (ES) volumes and mass were computed, as well as regional fractional area change (RFAC), from which automated classification of regional WM abnormality was defined. Comparison with reference values was performed by: 1) linear regression and Bland-Altman analyses for LV volumes and mass; 2) levels of agreement between the cardiologist WM grades and the automated classification. Optimal correlations (r2.97) and no bias were found for ED and ES volumes, while LV mass resulted in a good correlation (ED: r2=.81; ES: r2=.74) with a minimal overestimation (ED: 15.2g; ES: 8.7g) and narrow 95% limits of agreement (ED: ±30g; ES: ±33g). The automated interpretation resulted in high sensitivity, specificity, and accuracy (78%, 85%, 82%, respectively) of WM abnormalities. Combined automated endo and epicardial border detection from MRI images provides reliable measurements of LV dimensions and regional WM classification.I documenti in IRIS sono protetti da copyright e tutti i diritti sono riservati, salvo diversa indicazione.
