Volume Measurement in Sequential Freehand 3-D Ultrasound

Abstract

SummaryThree-dimensional (3D) acquisition and visualisation techniques are increasingly being incorporated into commercial ultrasound scanners. The diagnostic benefit of such techniques is not yet convincing, compared to the added inconvenience of using them. Although it is straightforward to use special probes to acquire 3D data, these are more restrictive than conventional 2D probes. The only 3D technique which retains the scanning flexibility and wide area of application of 2D ultrasound, is freehand 3D ultrasound, where a 2D probe is moved manually and its location sensed remotely. However, it is hard to generate a regular volume of data from the resulting non-parallel ultrasound images.Probably the largest body of evidence justifying the use of 3D ultrasound relates to the accurate measurement of volume. However, volume measurement with 3D ultrasound requires segmentation (i.e. outlining of the area of interest), which is a time consuming, manual task. Both this problem, and that of creating a regular volume of data, are eased by the use of sequential freehand 3D ultrasound, a recent technique amplified in this thesis, where all the processing is performed on the original ultrasound images. Thus a regular array of data is not required, and segmentation is performed on images whose features and artifacts are recognisable to the clinician.In this thesis, novel volume measurement and surface visualisation algorithms are developed for sequential freehand 3D ultrasound. A particular emphasis is placed on limiting the number of segmented cross-sections required for a given measurement accuracy. Inherent accuracy is demonstrated by simulation to be within ±2%, from 10 or fewer cross-sections. In vivo precision, including registration and segmentation errors, is shown to be within ±7%, even on complicated objects such as the liver or a foetus, from similar numbers of cross-sections. The volume can be updated in real-time as each image is segmented, and surfaces can be interpolated and visualised within a few seconds. Such visualisation can reveal errors in the segmentation which are otherwise hard to see.In addition, a framework for multiple-sweep data is developed, which allows volume measurement and surface visualisation of anatomy that can only be scanned by several sweeps of the ultrasound probe. Accurate volume measurements can therefore be made of all areas observable by conventional ultrasound. These measurements can be accomplished within approximately 5 minutes of examining the patient.The surface visualisation algorithms can also produce high quality renderings of data from a wide range of sources in medical imaging and other fields. The interpolation of cross-sections is an improvement on shape-based interpolation, and the triangular mesh created from the data is of a much higher quality than that using marching cubes. An extension of the surface interpolation algorithm can also be used to gradually change, or 'morph' one 3D surface to another, a technique commonly used in the film industry.

Geometric Methods for Vessel Visualization and Quantification — A Survey

“…This problem has been addressed by many authors, trying to align centroids before interpolation, and scale the cross-sections to match bounding rectangles, etc, but still not generate correct surfaces in complex cases. Treece [99] recognized that the problem is in the definition of connectivity, because the correspondence evaluating entire contours is too coarse. He proposed the disc-guided interpolation, where the interpolation is guided by using correspondence of regions of the cross-sections.…”

“…Much effort is required for detection and handling of special cases to allow a correct triangulation. An alternative to the direct reconstruction approach is to use the crosssections to estimate a 3D function that represents the measure of distance from any point to the surface [99]. Among these method belongs the shapebased interpolation proposed by Raya and Udupa [85], which uses the cityblock distance to the surface.…”

CAVASS: a computer assisted visualization and analysis software system - visualization aspects

“…Such operations are labeled Type 1. Examples of such operations are: image gray level slice interpolation methods (linear, spline-based methods) [57], shape-based (binary as well as gray-level) interpolation [57][58][59][60][61], image-based registration (via mutual information/correlation) [62,63], diffusive filtering [64][65][66], inhomogeneity correction [67], all non-user-steered slice-byslice segmentation methods (such as clustering techniques), non-connected isosurface detection, and structure manipulation [46,54]. There are other CAVA operations, which work (chunk-by-chunk) in the above sense but some further operation is needed to combine the outputs produced by the chunks to yield the final output.…”

Introducing CAVASS: a Computer-Assisted Visualization and Analysis Software System

“…Such operations are labeled Type 1. Examples of such operations are: image gray level slice interpolation methods (linear, spline-based methods) [30], shape-based (binary as well as gray-level) interpolation [30][31][32][33][34], image-based registration (via mutual information/correlation) [35,36], diffusive filtering [37][38][39], inhomogeneity correction [40], all non-user-steered slice-byslice segmentation methods (such as clustering techniques), non-connected isosurface detection, and structure manipulation [17,24]. There are other CAVA operations, which work (chunk-by-chunk) in the above sense but some further operation is needed to combine the outputs produced by the chunks to yield the final output.…”