Freehand Three-Dimensional Ultrasound Calibration

Freehand three-dimensional (3D) ultrasound is a technique for acquiring ultrasonic data of a 3D volume with application in radiotherapy and surgery planning. As the clinician sweeps the ultrasound probe over the anatomy, the trajectory of the probe is recorded using an attached position sensor. However, the position sensor tracks the position of the sensor, rather than the ultrasound B-scan. It is therefore necessary to find the rigid body transformation from the coordinate system of the B-scan to that of the position sensor. This transformation is found by probe calibration. During probe calibration, the user usually scans an object with known dimensions (a phantom). An image of the phantom is reconstructed in 3D space. The correct rigid body calibration transformation is found when the reconstructed image matches its model. There are several designs of the phantom, each with its own strengths and weaknesses. In a clinical environment, some of the factors considered when choosing a phantom are accuracy, reliability, speed and ease of use. If probe calibration is inaccurate, the reconstructed image of the anatomy will be distorted, leading to incorrect distance and volume measurements. A difficult calibration routine may result in an unreliable calibration if the user is not suitably trained. In this thesis, we investigate how to achieve an accurate, reliable and rapid calibration. Our first step is to improve the reliability of an accurate calibration technique from the literature. The Cambridge phantom, a variant of plane-based calibration, has been shown to be one of the most accurate calibration techniques. Unfortunately, calibrations performed using plane-based techniques are often unreliable. The user is required to scan the phantom in a complex way. Inexperienced users often neglect one of the required scanning motions which results in an unreliable calibration. We show how it is possible to provide feedback on the reliability of the calibration. This allows the user to rectify an unreliable calibration. Having achieved an accurate and reliable calibration, we now search for a fast and easy calibration technique. We study a class of two-dimensional alignment phantoms—the Z-fiducial phantom. Probe calibration using such a phantom only requires a single image of the phantom. However, calibration speed using this phantom is impeded by the necessity of segmenting isolated points on the phantom reliably, which requires human intervention. We solve this problem by mounting a thin rubber membrane on top of the phantom. The membrane is segmented automatically and the phantom features can be easily located as they are at known positions relative to the membrane. This enables us to segment isolated points automatically at the full PAL frame rate of 25Hz, enabling calibration to be completed in a few seconds. In addition to improving existing calibration techniques, we present two novel phantoms—the cone phantom and the Cambridge stylus. They are both simple in design, easy to use and produce accurate calibrations. The cone phantom produces calibrations with accuracies matching the Cambridge phantom. The Cambridge stylus is small in size and can be carried around conveniently. These phantoms offer alternatives to the Cambridge phantom and the Z-phantom, ensuring calibration reliability and simplicity, while producing accurate calibrations. In this dissertation, we show how to achieve a good reliability with one of the most accurate calibration techniques—the plane-based calibrations. We also show how to achieve rapid calibration by using the Z-phantom, allowing calibration to be performed in a few seconds. We present the cone phantom and the Cambridge stylus. Both of them are easy to use and are capable of achieving a good accuracy.