Improving three‐dimensional mechanical imaging of breast lesions with principal component analysis

Abstract

Purpose: Elastography has emerged as a new tool for detecting and diagnosing many types of diseases including breast cancer. To date, most clinical applications of elastography have utilized two-dimensional strain images. The goal of this paper is to present a new quasi-static elastography technique that yields shear modulus images in three dimensions. Methods: An automated breast volume scanner was used to acquire ultrasound images of the breast as it was gently compressed. Cross-correlation between successive images was used to determine the displacement within the tissue. The resulting displacement field was filtered of all but compressive motion through principal component analysis. This displacement field was used to infer spatial distribution of shear modulus by solving a 3D elastic inverse problem. Results: Three dimensional shear modulus images of benign breast lesions for two subjects were generated using the techniques described above. It was found that the lesions were visualized more clearly in images generated using the displacement data de-noised through the use of principal components. Conclusions: We have presented experimental and algorithmic techniques that lead to three- dimensional imaging of shear modulus using quasi-static elastography. This work demonstrates feasibility of this approach, and lays the foundation for images of other, more informative, mechanical parameters.

Non-invasive imaging of Young’s modulus and Poisson’s ratio in cancers in vivo

“…6). However, experimental estimation of the elevational strain component requires data acquisition along the elevational direction [13], [58], [59], which is difficult to accomplish in free-hand elastography, in general, and particularly challenging in poroelastography, where temporal data is required. Currently, most clinical studies involving ultrasound imaging use linear arrays, which allow only planar data acquisition similarly to what we have reported in this manuscript.…”

“…Among these two techniques, the first one is a recent one [15] and the other one is a classical technique [36]. Other reconstruction methods have been proposed in the literature [58], [59], which are also three-dimensional and may have some advantages with respect to the methods in Refs. [15] [36] chosen for the statistical comparison carried out in this paper.…”

Physics-guided machine learning for 3-D quantitative quasi-static elasticity imaging

“…Richards et al automated the data acquisition process by attaching a US transducer to a positioning system and collecting volumetric RF data after compression with platens (Richards et al 2009), greatly reducing jitter errors between successive RF frames while utilizing full 3D echo information for displacement estimation. Similarly, other investigators adapted an automated breast volume scanning (ABVS) system to collect pre-and post-compression RF data (Wang et al 2017, Hendriks et al 2016, Hendriks et al 2018, Tyagi et al 2017. Even though these automated methods reduce decorrelation caused by uncertainty of the US probe position, they require long acquisition times and are thus susceptible to motion artifacts in displacements estimated via speckle-tracking.…”

3-D Single Breath-Hold Shear Strain Estimation for Improved Breast Lesion Detection and Classification in Automated Volumetric Ultrasound Scanners

“…First steps in the implementation of elastography in the ABVS were shown in recent studies [15], [26]. Tyagi et al [26] proposed a method to obtain shear moduli in breast tissue by conventional ABVS scanning. Although initial results are promising, this method has some practical limitations.…”