PhD Oral Exam - Morteza Mirzaei, Electrical and Computer Engineering
Accurate and Precise Time-Delay Estimation for Ultrasound Elastography
This event is free
School of Graduate Studies
When studying for a doctoral degree (PhD), candidates submit a thesis that provides a critical review of the current state of knowledge of the thesis subject as well as the student’s own contributions to the subject. The distinguishing criterion of doctoral graduate research is a significant and original contribution to knowledge.
Once accepted, the candidate presents the thesis orally. This oral exam is open to the public.
Elastography is a technique for detecting pathological tissue alterations by extracting mechanical properties of the tissue. It can be performed using different imaging modalities, including magnetic resonance imaging and ultrasound. Unlike biopsy that is invasive and considers a small portion of tissue, elastography is a non-invasive technique that interrogates a larger part of the tissue and reduces the probability of missing abnormalities.
Ultrasound Elastography (USE) is an approach for detecting mechanical properties of tissue by using ultrasound imaging. Ultrasound as an imaging tool has emerged in the latter half of the 20th century and has become one of the most popular imaging modalities. The main advantages of ultrasound imaging lie in its noninvasive nature, low cost, convenience, and wide availability. USE may help in early diagnosis which substantially increases the success probability of treatment. In recent years, USE has been explored for several clinical applications including ablation guidance and monitoring, differentiating benign thyroid nodules from malignant ones and breast lesion characterization. Surgical treatment of liver cancer, assessment of non-alcoholic fatty liver disease, assessment of fibrosis in chronic liver diseases, detecting prostate cancer, differentiating abnormal lymph nodes in benign conditions and brain tumor surgery are other relevant clinical applications of USE.
An important challenging step for USE is Time Delay Estimation (TDE) between pre- and post-deformed tissue. TDE is an ill-posed problem since the 2D displacement of one sample cannot be uniquely calculated based on its intensity. Moreover, presence of noise due to speckles, out-of-plane movement, blood flow and other biological motions affect the accuracy of TDE. The other limiting factors for TDE are low resolution of ultrasound data, low sampling rate and lack of carrier signal in the lateral direction.
In Chapter 2, a 3D window-based technique is proposed. Window-based tracking methods are one of the most popular tracking techniques that assume displacements of spatially neighboring samples are the same and look for a similar window in the other image. In traditional window-based methods, 2D spatial windows were considered and information of spatially neighbors was utilized. In this chapter, we use 3D windows to benefit from additional information made available by including samples in the third dimension. In this chapter, two approaches named Spatial Temporal Normalized Cross Correlation (STNCC) and Channel data Normalized Cross Correlation (CNCC) are proposed. In STNCC, the third dimension is temporally neighboring samples and in CNCC time-delayed pre-beam-formed data that is collected by neighboring channels are utilized as the third dimension.
In Chapter 3, tOtal Variation rEgularization and WINDow-based (OVERWIND) time-delay estimation method as a regularized optimization-based method is proposed. In OVERWIND, a cost function containing two parts is introduced for optimal displacement estimation. In the first part of the proposed cost function, the difference of two images as pre- and post-compressed RF data are penalized. By assuming that neighboring samples have the same displacement, small windows are considered around each sample and all samples in these windows are forced to have the same displacement in the pre- and post-compression RF data. Therefore, this method increases robustness against noise by utilizing information from neighboring samples. The second term of cost function deals with regularization which penalizes displacement of neighbor samples. Herein, we use L1 norm regularization which assigns a smaller penalty to discontinuities compared to the more commonly used L2 norm regularization. This modification adds nonlinear terms to the cost function, and therefore, we use iterative methods to optimize the cost function.
In Chapter 4, Channel data for GLobal Ultrasound Elastography (CGLUE) is proposed for USE. The novelty of CGLUE is using time-gain and time delay corrected pre-beamformed channel data instead of RF data. The channel data contains more information than RF data which intuitively informs that their comparison results in better displacement estimation than RF data. We also prove that utilizing channel data in data term decreases bias and variance of error.
In Chapter 5, we beamform the row data utilizing Virtual Source Synthetic Aperture (VSSA). Despite the capability of proposed USE methods in estimating both axial and lateral displacements, the latter is of lower quality compared to the former for three main reasons: low sampling rate, lack of carrier signal and low resolution in the lateral direction. In this chapter, we propose to use VSSA imaging that implements Synthetic Aperture (SA)-based beamforming on focused transmitted signals. On the one hand, this enables us to benefit from advantages of SA such as high resolution and the capability to increase the sampling frequency to increase the resolution and number of A-lines. On the other hand, we can take advantageous of line-by-line imaging in high penetration depth.
In Chapter 6, we propose a novel method for 2D displacement estimation method for more precise and accurate lateral displacement estimation. In MUSCULAR, we add physics-based constraints as a regularization part to the cost function, as well as absolute intensity differences and displacement continuity of neighboring samples. We show that the additional physics-based constraints substantially improve the accuracy of the 2D displacement estimation, and outline a computationally efficient way to exploit physics-based constraints.
The last chapter concludes the thesis and proposes some ideas for future work. Despite the proposed USE methods to increase accuracy of tracking tissue motion, still the estimation is lateral direction is not as good as the axial one. Low resolution and lack of carrier signal in lateral direction is two of the important factors. However, increasing resolution in lateral direction does not necessarily increase the estimation accuracy. We outline avenues for future work to address these issues.