Rapid calibration for 3-D freehand ultrasound

Abstract

3-D freehand ultrasound is a new imaging technique that is rapidly finding clinical applications. A position-sensing device is attached to a conventional ultrasound probe so that, as B-scans are acquired, they can be labelled with their relative positions and orientations. This allows a 3-D data set to be constructed from the B-scans. A key requirement of all freehand imaging systems is calibration; that is, determining the position and orientation of the B-scan with respect to the position sensor. This is typically a lengthy and tedious process that may need repeating every time a sensor is mounted on a probe. This paper describes a new calibration technique that takes only a few minutes to perform and produces results that compare favourably (in terms of both accuracy and precision) with previously published alternatives.

Introduction

3-D ultrasound is a new imaging modality that has already been recognised as a valuable tool for a variety of clinical applications. Conventional 2-D diagnostic imaging uses a hand-held probe that transmits ultrasound pulses into the body and receives the echoes. The magnitude and timing of the echoes are used to create a 2-D grey-level image (B-scan) of a cross-section of the body in the scan plane. 3-D ultrasound extends this concept so that volumes of intensity data are created from pulse-echo information.

High-quality, rapid 3-D imaging remains a long-term research goal. One promising approach centres around the development of a new type of phased array probe, which sends and receives echoes from a 2-D array of elements (instead of the usual 1-D array). Unfortunately, several technical challenges must be overcome before such probes receive clinical acceptance (Smith et al. 1995). Alternative approaches, which make use of conventional 2-D ultrasound technology, include the freehand and swept-volume techniques Rankin et al 1993, Steiner et al 1994. Instead of taking a 3-D snapshot, these techniques construct a 3-D data set from a number of 2-D B-scans acquired in rapid succession.

The swept-volume approach uses a special mechanism inside the probe to sweep the plane of the B-scan through a volume of interest. In contrast, the freehand approach relies on the physician to guide a standard probe over the volume, while a position sensor, attached to the probe, measures the B-scans’ relative positions. Each method has its advantages. Swept-volume systems are easy to use and produce standardised volumes of densely sampled data without irregular gaps. However, they require the considerable expense of a dedicated machine, and are also limited to a maximum volume dictated by hardware constraints in the probe.

Freehand systems can be used to obtain arbitrary volumes of data because the motion of the probe is unconstrained. They are also cheaper, requiring only existing conventional ultrasound systems and relatively inexpensive additional components. However, the physician needs to learn how to move the probe to acquire regular, densely sampled data sets. The cost and flexibility of freehand imaging ensures it remains a popular choice and, with recent improvements in the speed of data acquisition combined with careful scanning practice, the quality of freehand data sets is improving. For these reasons, research into freehand systems is very active, and several commercial systems have recently become available.

A 3-D freehand examination can be broken into three stages: scanning, reconstruction and visualisation (see Fig. 1). Before scanning, some sort of position sensor is attached to the probe. This is typically the receiver of an electromagnetic position sensor, as illustrated in Fig. 1. Measurements from the position sensor are used to determine the positions and orientations of the B-scans with respect to the fixed transmitter. In the next stage, the set of acquired B-scans and their relative positions are used to fill a regular voxel array. Finally, this voxel array is visualised using, for example, any-plane slicing, volume rendering or surface rendering (after segmentation).

A key requirement of all freehand imaging systems is calibration. This involves determining the position and orientation of the B-scans with respect to the sensor mounted on the probe. The results of the calibration take the form of six constant offsets, three for position and three for orientation. These offsets must be combined with the sensor measurements to calculate the positions of the B-scans during reconstruction. Accurate calibration is essential for a consistent reconstruction that preserves true anatomical shape.

Calibration is required, regardless of the type of position sensor used in the examination. Previous work has described the use of acoustic spark gaps (King et al. 1991), mechanical arms (Ohbuchi et al. 1992), AC magnetic sensors Barry et al 1997, Hughes et al 1996, Nelson and Elvins 1993, DC magnetic sensors Detmer et al 1994, Leotta et al 1997, optical sensors State et al 1994, Trobaugh et al 1994, and multiple sensors (Leotta et al. 1995). Only a few, however, have seriously considered the question of accurate calibration.

A rough estimate of the calibration parameters can be obtained by external measurements of the probe and position sensor. However, the origin of the sensor (for example, the centre of the wire coils in an electromagnetic receiver) and the corner of the B-scan are not well defined with respect to the external cases of the sensor and probe. For these reasons, calibration is preferably performed by imaging a phantom, an artificial object with some known physical properties or dimensions. Measurements from an examination of the phantom, combined with its known physical properties, can be used to determine the six offsets.

Existing calibration techniques are not particularly easy to use. A typical calibration process, which often needs repeating every time a sensor is mounted on a probe, might take several hours for a skilled technician to perform. In this paper, we present a novel calibration technique that can be performed in a few minutes and produces results that compare favourably with previously published alternatives. The paper is organised as follows. First, we describe the reconstruction process and formulate the calibration problem. We then review existing calibration procedures and introduce the new technique that is the focus of this paper. Finally, we compare the accuracy, repeatability and usability of the techniques, and draw some conclusions from our results.