Abstract 14436: Effects of Uncertainty in Border Delineation on GLS BL versus LVEF

Abstract

Background: There are many clinical scenarios in which early detection of small changes in cardiac function can be critical in determining a patient’s treatment. Two of the most common imaging-based metrics in the clinic are LVEF and border-length-based GLS (GLS BL ). The aim of this project was to analyze the inherent biases and uncertainties in GLS BL relative to LVEF with respect to uncertainties in delineation of the endocardial border. Methods: I n silico 3D human LV endocardial surface models were created at end-diastole and end-systole based on MRIs of a healthy human volunteer. Twelve virtual long-axis imaging slices (in a radially-prescribed) were projected through the surface models to generate noise-free endocardial contour points. To simulate border delineation uncertainty, the set of endocardial contour points was subjected to inward and outward shifts of 1-5 mm, incrementally (Fig 1A). The expected uncertainty when using EC is around 3 mm. The % error vs. offsets were calculated for bi-plane GLS BL , bi-plane area-length LVEF AL , and bi-plane method-of-disks LVEF MOD , with 3D conformal surface-fit LVEF SURF as ground-truth. Results: The GLS BL %error vs. offset was linear (R 2 = 1.000) with a slope of 4.92 (Fig. 1B). Each of the LVEF measures were quadratic with offset (R 2 = 1.000), with slopes: LVEF AL : 6.68; LVEF MOD : 6.71; and LVEF SURF : 10.26. Conclusions: GLS BL is less sensitive to boundary delineation uncertainty than LVEF, due to the linear vs. quadratic dependence on offset. LVEF AL and LVEF MOD consistently underestimate true LVEF, making them less sensitive to boundary uncertainty. However, if border uncertainty is negligible, then LVEF is more sensitive than GLS BL to small changes in LV volume, and LVEF SURF is more sensitive than either LVEF AL and LVEF MOD .