Using Anatomical Markers for Left Ventricular Segmentation of Long Axis Ultrasound Images
This is an incremental improvement for medical imaging analysis, specifically for measuring left ventricular function indices.
The study tackled left ventricular segmentation in long axis ultrasound images by using anatomical markers to create an initial contour guess, resulting in good agreement with manually traced contours on 21 clips.
Left ventricular segmentation is essential for measuring left ventricular function indices. Segmentation of one or several images requires an initial guess of the contour. It is hypothesized here that creating an initial guess by first detecting anatomical markers, would lead to correct detection of the endocardium. The first step of the algorithm presented here includes automatic detection of the mitral valve. Next, the apex is detected in the same frame. The valve is then tracked throughout the cardiac cycle. Contours passing from the apex to each valve corner are then found using a dynamic programming algorithm. The resulting contour is used as an input to an active contour algorithm. The algorithm was tested on 21 long axis ultrasound clips and showed good agreement with manually traced contours. Thus, this study demonstrates that detection of anatomic markers leads to a reliable initial guess of the left ventricle border.