Freehand 3D Ultrasound Reconstruction Algorithms—A Review

Abstract

Three-dimensional (3D) ultrasound (US) is increasingly being introduced in the clinic, both for diagnostics and image guidance. Although dedicated 3D US probes exist, 3D US can also be acquired with the still frequently used two-dimensional (2D) US probes. Obtaining 3D volumes with 2D US probes is a two-step process. First, a positioning sensor must be attached to the probe; second, a reconstruction of a 3D volume can be performed into a regular voxel grid. Various algorithms have been used for performing 3D reconstruction based on 2D images. Up till now, a complete overview of the algorithms, the way they work and their benefits and drawbacks due to various applications has been missing. The lack of an overview is made clear by confusions about algorithm and group names in the existing literature. This article is a review aimed at explaining and categorizing the various algorithms into groups, according to algorithm implementation. The algorithms are compared based on published data and our own laboratory results. Positive and practical uses of the various algorithms for different applications are discussed, with a focus on image guidance. (E-mail: [email protected])

Introduction

Minimally invasive surgery or image-guided surgery is an important field in therapy, becoming more and more widespread. To minimize the intervention, high demands must be made to the imaging modalities used due to image quality and accuracy. During surgery, intraoperative imaging is needed in addition to preoperative images, since changes occur during surgery. Even though the more commonly used imaging technologies such as magnetic resonance imaging (MRI) and computed tomography (CT) are possible to use during surgery, there are still some significant practical limitations, due to costs, equipment adaptation in the magnetic field, user friendliness, image quality and radiation doses. Three-dimensional (3D) ultrasound (US) is already being introduced alone or together with preoperational images for guidance of surgical applications.

Two-dimensional (2D) US is being extensively used for a variety of clinical applications and 3D US is now also more frequently demonstrated in the clinic. The main advantage of 3D US is that arbitrary 2D images through the volume may be visualized and not only images in the same plane as the US acquisition is performed, which is the only option with 2D US. 3D allows views not possible with 2D. In addition, 3D US also allows a 3D volume rendered view and 3D segmentations of objects. Two different main approaches for 3D US creation exist: using a dedicated 3D US probe or using a regular 2D US probe for acquiring the images and combining these 2D slices to a 3D volume. A 3D US probe may be a 2D array acquiring 3D volumes directly or a mechanical 3D probe consisting of a regular one-dimensional (1D) array acquiring multiple 2D images with a motor that sweeps the 1D array over the scanned area in a certain manner: linear, tilt or rotational (Fenster et al. 2001). The 3D reconstructions with motorized probes are very similar to a freehand 3D reconstruction, although the 2D image positions have a more regular pattern than that for freehand 3D US. A 2D array US probe may also be used to acquire real-time 3D volumes (also called 4D).

3D volume reconstruction from free-hand acquired 2D images usually needs position data of the 2D slices. The most common method for obtaining positioning data is to attach a position sensor to the probe: electromagnetic, optical, mechanical arm or acoustic. [See Cinquin et al. (1995) for a description of these positioning systems.] However, some systems use alternative methods such as the I-beam probe (Hossack et al. 2000), where the image positions are tracked with respect to each other with a special probe configuration and the US data. Other methods do not use any external positioning measurements at al.: the predefined operator probe movement (Downey and Fenster 1995), speckle decorrelation (Tuthill et al. 1998), frame distance estimation (Lee et al. 2001) or linear regression (Prager et al. 2003). In the literature so far, the sensorless methods have not been shown to give the same accuracy as tracking systems.

In addition to the reconstruction algorithm itself, several factors affect the 3D volume reconstruction accuracy. High quality 3D reconstructions depend on both the quality of the input 2D images and the accuracy of the position data. Tracking system inaccuracy, the ultrasound probe calibration process, sound of speed variation and tissue movement are all important error sources that are handled elsewhere (Lindseth et al 2002, Treece et al 2002, Mercier et al 2005).

Earlier work has been done to explain 3D US in general (Nelson and Pretorius 1998, Fenster et al 2001), 3D US in neurosurgery specifically (Unsgaard et al. 2006) and some of the different reconstruction algorithms and a grouping of these (Rohling et al. 1999). The article by Rohling et al. (1999) is often referred to by others for demonstrating examples of different algorithms and is sometimes also given as reference to specific algorithms as well. Some confusion exists in the literature about algorithm and group names along with some unclear algorithm origins. We believe that these problems arise from lack of a clear overview of freehand 3D US reconstruction algorithms.

The present article will, therefore, provide a thorough description and grouping of the various freehand 3D reconstruction algorithms with focus on the recently published. Although choice of positioning system and probe calibration also affects reconstruction accuracy, this will not be the focus of this article. For a comprehensive review, see the article by Mercier et al. (2005). Benefits and drawbacks of the various 3D reconstruction algorithms will be discussed with emphasize on time usage, quality, practical implementation and usefulness in image-guided surgery applications.