![[Univ of Cambridge]](uniban-s.gif){kind=small}
![[Dept of Engineering]](engban-s.gif){kind=small}
The objectives of this project are:
To develop a system for the acquisition of real-time, freehand 3D radio-frequency (RF) ultrasound data. This is a prerequisite for the remainder of the project.
To develop and evaluate a system for metric 3D ultrasound reconstruction without the need for an intrusive position sensor. This work will improve the prospects of freehand 3D ultrasound moving out of the research laboratory and into everyday clinical use.
To investigate possibilities into improving the traditional B-scan image with additional information from the 3D RF ultrasound data, looking in particular at deconvolution and superresolution, detection and correction of shadow and enhancement artifacts and elastography imaging.
We have successfully developed the real-time, RF 3D ultrasound acquisition system [24]. This already constitutes a unique and powerful research facility which we have incorporated into our Stradx and Stradwin systems for other groups to use. Analogue RF ultrasound signals from a minimally modified Dynamic Imaging Diasus scanner are digitised after receive focusing and time-gain compensation, but before log-compression and envelope detection, using a Gage Compuscope CS14100 14-bit digitiser. Whole frames are stored in on-board Gage memory before transferring to PC memory at 75 MB/s. The system operates in real-time, with acquisition rates exceeding 30 frames per second. Sampling is at 66.67 MHz, synchronous with the ultrasound machine's internal clock: this synchronization minimises phase jitter between vectors. The RF data can be matched with positions from an external position sensor and converted into a number of forms for display, including amplitude, phase or strain images. The example below shows 3D amplitude and phase images of a 0.15 mm diameter pinhead.
The goal here is to develop a freehand 3D ultrasound acquisition system which does not require an intrusive position sensor. Our approach starts with the established theory of speckle decorrelation.
The above figure illustrates this theory. The in-plane motion
between scans A and B (translation in the x and
y directions, roll around the plane normal) is readily
determined using conventional 2D image registration techniques. This
leaves three degrees of freedom: translation in the elevational
direction, tilt and yaw. Consider corresponding patches in scans A
and B (the shaded ellipses). Because of the imperfect
elevational focusing, the contents of the patches depend on
scatterers within overlapping resolution cells (the hollow
ellipsoids) and are therefore correlated. The correlation coefficient
depends on the degree of overlap and hence the elevational
separation. It follows that, given a suitable decorrelation curve, a
measured correlation
can be used to look up the corresponding separation
.
Repeating this process for three (or more) non-collinear patches
determines the elevational separation, tilt and yaw of A
relative to B.
A weakness of this theory is that it holds only for uncompressed RF scans of fully developed speckle. The RF limitation is addressed through our use of an RF acquisition system. However, we have demonstrated that there is precious little fully developed speckle in typical clinical data [6, 18, 36]. We have therefore extended the theory to allow elevational distance estimation in the presence of coherent scattering [7, 38]. The figure below shows metric reconstructions of a phantom and animal tissue using the new theory. Distance errors are less than 4% in all cases.
A parallel line of research aims to improve the ease with which sensorless 3D ultrasound data can be acquired. Currently, all speckle decorrelation systems assume monotonic motion of the scan plane. We have developed novel reconstruction algorithms to relax this constraint [1, 2, 15, 20, 23, 31, 33]. The new algorithms can cope with non-monotonic scanning and intersecting frames. The figure below shows a non-monotonic scanning pattern (a) and reconstructions using a position sensor (b, dotted) and speckle decorrelation (b, solid). (c) and (e) are reslices through the position sensor-based reconstruction, (d) and (f) are reslices through the sensorless reconstructions. While the sensorless reconstructions exhibit inevitable drift (see below), the fine scale detail is better, with no reconstruction jitter.
Speckle decorrelation algorithms perform well over small scales, but since they locate each B-scan relative to its predecessor, and not relative to a fixed datum, they are susceptible to error accumulation (drift) over large scales. In contrast, position sensors suffer small scale jitter, but many are immune to large scale drift. To remedy both these effects, we have developed a hybrid system [25] which corrects the tracking drift using sparse position sensor readings without compromising the fine-scale detail. We have developed novel yet straightforward calibration techniques for the sensor which are far easier to use than existing alternatives [3, 9, 11, 12, 19, 27, 34, 41]. With careful calibration of the sensor, our hybrid system is capable of sub-millimetre large-scale geometrric accuracy, and image-based registration (using adaptive speckle decorrelation) improves the fine-scale detail.
For our hybrid system to be clinically functional, we required the sensor to be unobtrusive to the clinician, thus ruling out optical systems (which require clear lines of sight) and magnetic systems (which must be kept away from metal). We have experimented with the ShapeTape six-degree-of-freedom sensor and the Xsens (MEMS gyro) three-degree-of-freedom sensor and developed novel calibration techniques for each of these. Our experiments demonstrated that drift correction is feasible with either sensor, although the performance ShapeTape sensor was found to be adequate only when its motion was highly constrained and thus it could only be used with linear scanning protocols.
Acquisition of raw RF data with the requisite phase information has made possible the application of deconvolution to improve the spatial resolution of the B-scan. Initial work in this direction looked at estimating the point-spread function of the imaging system from RF measurements of the reflections off a line target [37]. Subsequently, we developed a novel deconvolution algorithm that incorporates wavelet-based denoising to produce speckle-bearing B-scans with substantially higher resolution and better contrast and speckle-free images for gross feature examination. As far as we are aware, this algorithm is the first of its kind to solve the deconvolution problem and the despeckling problem simultaneously. We have been successful at applying this algorithm on 2D images [5, 21, 30] and have secured follow-on funding to extend the work to 3D.
We have also investigated attenuation correction using RF measurements (centre frequency analysis and spectral division) in standard B-scans [8]. Some in vitro results are presented below. (a) shows a conventional B-scan of an olive in gelatine, including a severe attenuation artefact. (b) shows the backscatter image after attenuation correction by spectral division. The corresponding attenuation estimate is shown in (c).
The 3D RF system has been invaluable in our research into state-of-the-art elastography (strain) imaging. By acquiring two B-scans of a probe with slightly different pressures applied to the probe each time, a strain image or elastogram can be computed in which hard inclusions, in particular, are especially well highlighted. The elastogram is a useful diagnostic complement to the conventional B-scan image and, using the 3D RF system, we have been able to extend the capabilities of 2D elastography [13, 14, 17, 26, 29, 32] and pioneer 3D elastography (including freehand 3D strain imaging) [4, 22, 35]. An in vitro example of a 3D elastogram is shown below.
Lastly, we have used our 3D RF system for probe pressure correction in freehand 3D scanning [10] and for fluid detection in small vessels [25].
[1] R.J. Housden, A.H. Gee, G.M. Treece and R.W. Prager. Sensorless reconstruction of unconstrained freehand 3D ultrasound data. To appear in Ultrasound in Medicine and Biology
[2] R.J. Housden, A.H. Gee, G.M. Treece and R.W. Prager. Subsample interpolation strategies for sensorless freehand 3D ultrasound. Ultrasound in Medicine and Biology, 32(12):1897-1904, December 2006.
[3] P-W. Hsu, R.W. Prager, A.H. Gee and G.M. Treece. Rapid, easy and reliable calibration for freehand 3D ultrasound. Ultrasound in Medicine and Biology, 32(6):823-835, June 2006.
[4] J.E. Lindop, G.M. Treece, A.H. Gee and R.W. Prager. 3D elastography using freehand ultrasound. Ultrasound in Medicine and Biology, 32(4):529-545, April 2006.
[5] J.K.H. Ng, R.W. Prager, N.G. Kingsbury, G.M. Treece and A.H. Gee. Modelling ultrasound imaging as a linear, shift-variant system. IEEE Transactions on Ultrasonics, Ferroelectrics, and Frequency Control, 53(3):549-563, March 2006.
[6] P. Hassenpflug, R.W. Prager, G.M. Treece and A.H. Gee. Speckle classification for sensorless freehand 3D ultrasound. Ultrasound in Medicine and Biology, 31(11):1499-1508, November 2005.
[7] A.H. Gee, R.J. Housden, P. Hassenpflug, G.M. Treece and R.W. Prager. Sensorless freehand 3D ultrasound in real tissue: speckle decorrelation without fully developed speckle. Medical Image Analysis, 10(2):137-149, April 2006.
[8] G. Treece, R. Prager and A. Gee. Ultrasound attenuation measurement in the presence of scatterer variation for reduction of shadowing and enhancement. IEEE Transactions on Ultrasonics, Ferroelectrics, and Frequency Control, 52(12):2346-2360, December 2005.
[9] A.H. Gee, N.E. Houghton, G.M. Treece and R.W. Prager. A mechanical instrument for 3D ultrasound probe calibration. Ultrasound in Medicine and Biology, 31(4):505-518, April 2005.
[10] G.M. Treece, R.W. Prager and A.H. Gee. RF and amplitude-based probe pressure correction for 3D ultrasound. Ultrasound in Medicine and Biology, 31(4):493-503, April 2005.
[11] P-W. Hsu, R.W. Prager, A.H. Gee, and G.M. Treece. Real-time freehand 3D ultrasound calibration. In Proceedings of SPIE Vol. 6513, Medical Imaging 2007, San Diego, CA, March 2007.
[12] P-W. Hsu, R.W. Prager, N.E. Houghton, A.H. Gee, and G.M. Treece. Accurate fiducial location for freehand 3D ultrasound. In Proceedings of SPIE Vol. 6513, Medical Imaging 2007, San Diego, CA, March 2007.
[13] J.E. Lindop, G.M. Treece, A.H. Gee, and R.W. Prager. Location estimation increases the accuracy of ultrasonic strain estimates. In Proceedings of the Fifth International Conference on the Ultrasonic Measurement and Imaging of Tissue Elasticity, page 104, Snowbird, Utah, October 2006.
[14] G.M. Treece, J.E. Lindop, A.H. Gee, and R.W. Prager. Efficient elimination of dropouts in displacement tracking. In Proceedings of the Fifth International Conference on the Ultrasonic Measurement and Imaging of Tissue Elasticity, page 68, Snowbird, Utah, October 2006.
[15] R.J. Housden, A.H. Gee, G.M. Treece and R.W. Prager. Sensorless reconstruction of freehand 3D ultrasound data. In Medical Image Computing and Computer-Assisted Intervention - MICCAI'06, volume 2, pages 356-363, Copenhagen, Denmark, October 2006 (LNCS 4191, Springer).
[16] J.K.H. Ng, R.W. Prager, N.G. Kingsbury, G.M. Treece and A.H. Gee. An iterative, wavelet-based deconvolution algorithm for the restoration of ultrasound images in an EM framework. In Proceedings of SPIE Vol. 6147, Medical Imaging 2006, pages 77-88, San Diego, California, February 2006.
[17] J.E. Lindop, G.M. Treece, A.H. Gee and R.W. Prager. Frame filtering for improved freehand 3D US elastography. In Proceedings of the Fourth International Conference on the Ultrasonic Measurement and Imaging of Tissue Elasticity, page 74, Austin, USA, October 2005.
[18] P. Hassenpflug, R.W. Prager, G.M. Treece and A.H. Gee. Speckle classification for sensorless freehand 3D ultrasound. In Proceedings of Medical Image Understanding and Analysis, pages 119-122, Bristol, UK, July 2005.
[19] P-W. Hsu, R.W. Prager, A.H. Gee and G.M. Treece. Rapid, easy and reliable calibration for freehand 3D ultrasound. In Proceedings of Medical Image Understanding and Analysis, pages 91-94, Bristol, UK, July 2005.
[20] R.J. Housden, A.H. Gee, R.W. Prager and G.M. Treece. Sensorless freehand 3D ultrasound for non-monotonic and intersecting frames. In Proceedings of Medical Image Understanding and Analysis, pages 127-130, Bristol, UK, July 2005.
[21] A. Marianski, R.W. Prager, A.H. Gee and G.M. Treece. Ultrasound point-spread function estimation using higher order statistics. In Proceedings of Medical Image Understanding and Analysis, pages 223-226, Bristol, UK, July 2005.
[22] J.E. Lindop, G.M. Treece, A.H. Gee and R.W. Prager. 3D elastography using freehand ultrasound. In Medical Image Computing and Computer-Assisted Intervention - MICCAI'04, volume 2, pages 1103-1104, Saint-Malo, France, September 2004 (LNCS 3217, Springer).
[23] P. Hassenpflug, R. Prager, G. Treece and A. Gee. Distance measurement for sensorless 3D US. In Medical Image Computing and Computer-Assisted Intervention - MICCAI'04, volume 2, pages 1087-1088, Saint-Malo, France, September 2004 (LNCS 3217, Springer).
[24] G. Treece, R. Prager and A. Gee. Freely available software for 3D RF ultrasound. In Medical Image Computing and Computer-Assisted Intervention - MICCAI'04, volume 2, pages 1099-1100, Saint-Malo, France, September 2004 (LNCS 3217, Springer).
[25] R. Chung, R.W. Prager, A.H. Gee and G.M. Treece. High frequency image-based flow detection. Journal of Physics, Conference Series I, Advanced Metrology for Ultrasound in Medicine, pages 193-198, National Physical Laboratory, UK, April 2004.
[26] R.J. Housden, A.H. Gee, G.M. Treece and R.W. Prager. Hybrid systems for reconstruction of freehand 3D ultrasound data. Technical report CUED/F-INFENG/TR 574, Cambridge University Department of Engineering, March 2007.
[27] J.E. Lindop, G.M. Treece, A.H. Gee and R.W. Prager. Dynamic resolution selection in ultrasonic strain imaging. Technical report CUED/F-INFENG/TR 566, Cambridge University Department of Engineering, September 2006.
[28] P-W. Hsu, R.W. Prager, A.H. Gee and G.M. Treece. Real-time freehand 3D ultrasound calibration. Technical report CUED/F-INFENG/TR 565, Cambridge University Department of Engineering, September 2006.
[29] G.M. Treece, R.W. Prager and A.H. Gee. Ultrasound compounding with automatic attenuation compensation using paired angle scans. Technical report CUED/F-INFENG/TR 558, Cambridge University Department of Engineering, June 2006.
[30] J.E. Lindop, G.M. Treece, A.H. Gee and R.W. Prager. Phase-based ultrasonic deformation estimation. Technical report CUED/F-INFENG/TR 555, Cambridge University Department of Engineering, May 2006.
[31] J.K.H. Ng, R.W. Prager, N.G. Kingsbury, G.M. Treece and A.H. Gee. Wavelet restoration of medical pulse-echo ultrasound images in an EM framework. Technical report CUED/F-INFENG/TR 554, Cambridge University Department of Engineering, May 2006.
[32] R.J. Housden, A.H. Gee, G.M. Treece and R.W. Prager. Sensorless reconstruction of unconstrained freehand 3D ultrasound data. Technical report CUED/F-INFENG/TR 553, Cambridge University Department of Engineering, May 2006.
[33] J.E. Lindop, G.M. Treece, A.H. Gee and R.W. Prager. Estimation of displacement location for enhanced strain imaging. Technical report CUED/F-INFENG/TR 550, Cambridge University Department of Engineering, March 2006.
[34] R.J. Housden, A.H. Gee, G.M. Treece and R.W. Prager. Sub-sample interpolation strategies for sensorless freehand 3D ultrasound. Technical report CUED/F-INFENG/TR 545, Cambridge University Department of Engineering, January 2006.
[35] P-W. Hsu, R.W. Prager, A.H. Gee and G.M. Treece. Rapid, easy and reliable calibration for freehand 3D ultrasound. Technical report CUED/F-INFENG/TR 534, Cambridge University Department of Engineering, August 2005
[36] J.E. Lindop, G.M. Treece, A.H. Gee and R.W. Prager. 3D elastrography using freehand ultrasound. Technical report CUED/F-INFENG/TR 531, Cambridge University Department of Engineering, July 2005.
[37] P. Hassenpflug, R.W. Prager, G.M. Treece and A.H. Gee. Speckle classification for sensorless freehand 3D ultrasound. Technical report CUED/F-INFENG/TR 513, Cambridge University Department of Engineering, March 2005.
[38] J.K.H. Ng, R.W. Prager, N.G. Kingsbury, G.M. Treece and A.H. Gee. Tomographic estimation of the point-spread function of an ultrasound imaging system for deconvolution. Technical report CUED/F-INFENG/TR 529, Cambridge University Department of Engineering, January 2005.
[39] A.H. Gee, R.J. Housden, P. Hassenpflug, G.M. Treece and R.W. Prager. Sensorless freehand 3D ultrasound in real tissue: speckle decorrelation without fully developed speckle. Technical report CUED/F-INFENG/TR 510, Cambridge University Department of Engineering, January 2005.
[40] J.K.H. Ng, R.W. Prager, N.G. Kingsbury, G.M. Treece and A.H. Gee. Modelling ultrasound imaging as a linear, shift-variant system. Technical report CUED/F-INFENG/TR 509, Cambridge University Department of Engineering, January 2005.
[41] G.M. Treece, R.W. Prager and A.H. Gee. Ultrasound attenuation measurement in the presence of scatterer variation for reduction of shadowing and enhancement. Technical report CUED/F-INFENG/TR 502, Cambridge University Department of Engineering, November 2004.
[42] A.H. Gee, N.E. Houghton, G.M. Treece and R.W. Prager. 3D ultrasound probe calibration without a position sensor. Technical report CUED/F-INFENG/TR 488, Cambridge University Department of Engineering, September 2004.
Dr. Andrew Gee - Reader
Dr. Richard Prager - Reader
Dr. Graham Treece - Royal Academy of Engineering Research Fellow
Mr James Housden - PhD student
Mr Benny Hsu - PhD student
Mr Joel Lindop - PhD student
Mr James Ng - PhD student and Research Associate
Mr. Peter Hassenpflug - Research Associate
Dr. Laurence Berman - Consultant Radiologist
Mr. Allan Findlay - Managing Director, Dynamic Imaging
[ Cambridge University | CUED | MI Lab | Medical Imaging Group ]