کنفرانس مهندسی برق ایران

صفحه اصلی / سی امین کنفرانس بین المللی مهندسی برق

Displacement Estimation for Ultrasound Elastography based on a Robust Uniform Stretching Method

نویسندگان :

Zahra Hosseini¹ Ali Khadem² Mohammadreza Hassannejad Bibalan³

1- دانشگاه صنعتی خواجه نصیرالدین طوسی 2- دانشگاه خواجه نصیر‌الدین طوسی 3- دانشگاه بین‌المللی امام خمینی (ره)

کلمات کلیدی :

Elastogram،normalized cross-correlation (NCC)،robust uniform stretching (RUS)،strain،ultrasound elastography.

چکیده :

This paper proposes a novel approach to estimate the displacement fields, which is a common and challenging step in ultrasound elastography. Maximization of normalized cross-correlation (NCC) is one of the literature’s widely used displacement estimation methods. This algorithm is simple to implement and computationally efficient; however, its main drawback is signal decorrelation which is the main source of distortions in elastogram images. In this work, a novel approach to increase the robustness of NCC to signal decorrelation and false peak error is proposed. This technique is according to the stretching-based methods that improve the correlation between pre- and post-compression data. In other words, in this method, the optimum stretching factor producing the highest NCC will be obtained from a two-step algorithm. The first step provides a rough estimation of the stretching factor, and in the second step, the exact value of the stretching parameter will be obtained. Then, the final displacement field is computed by providing a relationship between the stretching factor and the time delay. Finally, the strain image is estimated using the Kalman least square method, ensuring axial and lateral continuity. We call the proposed approach Robust Uniform Stretching (RUS), and its performance is investigated using the simulated and In-vivo data in terms of contrast-to-noise ratio (CNR), signal-to-noise ratio (SNR), and root mean square error (RMSE) criteria. The proposed method had superior performance compared with some conventional window-based algorithms by reaching the highest CNR values of 13.91 and 3.99 and SNR values of 14.50 and 3.61 for the simulation and In-vivo experiments, respectively.