布 樹輝¹, 山川 誠², 椎名 毅¹

Shuhui BU¹, Makoto YAMAKAWA², Tsuyoshi SHIINA¹

¹京都大学医学研究科人間健康科学科, ²京都大学先端医工学研究ユニット

¹Human Health Science, Graduate School of Medicine, Kyoto University, ²Advanced Biomedical Engineering Research Unit, Kyoto University

キーワード :

【目的】
Accurate assessment of local myocardial strain is important for diagnosing ischemic heart diseases because decreased myocardial motion often appears in the early stage. Calculating the spatial derivation of displacement is a necessary step in strain calculation, but the numerical calculation is extremely sensitive to noise. Commonly used smoothing methods are moving average and median filters; however, these methods have a trade-off between the spatial resolution and accuracy. We propose a novel fitting method to overcome the problem. In this method, the detected displacement vectors are discrete at mesh nodes, and virtual springs are connected between adjacent nodes. By controlling virtual springs’ elasticity, miss-detected displacements are fitted without above problem. Further improvements can be achieved by applying Kalman filter for position tracking, and then calculating strain from accumulated displacement vectors.
【方法】
In this method, a 2-D array transducer acquires echo data from the myocardium through volumetric sector scanning. RF signals for each scan line are received at all elements in the probe. The phase shifts at every measuring point between two consecutive frames are calculated by combined autocorrelation method, and the displacement vectors are calculated by weighted phase gradient method from obtained phase shifts. Meanwhile, the myocardial endocardium boundaries are segmented from the B-mode image, whereas the epicardium boundary is calculated from the endocardium boundary, by adding a fixed radius. Next, the myocardium meshes are generated from the myocardial boundaries. The detected displacement vectors and myocardial meshes are then input into a displacement re-sampling module. The displacement vectors in each mesh node are calculated by bilinear interpolation, their coordinates are translated from Cartesian coordinates to cylindrical coordinates, and ADGI then fits the displacement vectors. Each nodes’ position vectors are tracked by Kalman filter, and then displacement vectors are accumulated. Finally, the strain tensors are calculated from the processed displacement vectors.
【結論】
From the resulting images, we can see that our proposed method does not have the problem of angle dependency. There are no instances of distinct accuracy decrease for the infarction region with different positions. In simulation models, although the noise level is high, strain data which calculated from accumulated displacement have better accuracy and stability. From the simulation results, we can conclude that the images generated by proposed method are more accurate, robust, and have high spatial resolution. Therefore, the proposed method is expected to provide more accurate diagnostic information.