Search Contact information
University of Cambridge Home Department of Engineering
University of Cambridge > Engineering Department

Hybrid 3D Ultrasonic Imaging

EPSRC Grant EP/F016476/1


Most of the ultrasound machines in hospitals today work in two dimensions. They send high frequency sound pulses into the body and display the echos that come back as a two-dimensional (2D) picture. They produce an image that shows the sound reflectors in one slice through the body. However, there are some applications where doctors would like to be able to gather ultrasound data as a three-dimensional (3D) block rather than a two-dimensional slice. For example, when there is a need for volume measurement or the analysis of complex geometry. Two different types of 3D ultrasound have been developed to meet this requirement. One type involves a special probe that can record a fixed block of data, either by having an internal sweeping mechanism or by using electronic beam steering. The other type of 3D ultrasound uses a conventional 2D ultrasound machine and is called freehand 3D ultrasound. In this approach, an optical or magnetic position sensor is attached to the probe and the precise 3D trajectory of the probe (including the angles) is recorded as the doctor performs the scan. The position information from the sensor enables the 2D ultrasound slices from the ultrasound machine to be interpreted as 3D data using a computer.
The aim of this project is to produce a hybrid of these two types of 3D ultrasound. The new system will incorporate benefits from both the existing strategies and also offer some additional unique advantages. It will record dense regular data like the integrated 3D probe described above and will also acquire large data-sets as is possible with the freehand approach. There will be no need for an inconvenient external position sensor attached to the probe as much of the information required to calculate the probe trajectory can be inferred by matching the 3D blocks of recorded data. A miniature inertial orientation sensor will be used to guide the matching algorithms, increasing their speed and reliability. Such a sensor could eventually be incorporated in the probe housing, hence it will not be inconvenient in a clinical context. The project will focus on: tracking the trajectory of the probe based on the the acquired data and the output of the inertial position sensor; calibration of the hybrid system; correction of artifacts in the data caused by variations of the pressure from the probe during the scan; and development and evaluation of software tools to enable the system to be used effectively in a hospital environment.


We have addressed two more open-ended pieces of research. In the Engineering Department, we have explored the potential for exploiting the richness of the dense overlapping data that the new machine will record. The fact that the probe can be easily used to scan some points from more than one direction will be particularly useful. In the hospital, we have mounted a complete version of the system on a trolley and are currently using it to explore the range of applications in which this type of scanner could offer particular benefits to the doctors. We are also in the process of measuring the size and precision of scans than can be reliably acquired using the system in a busy ultrasound clinic. In the future, as it becomes more and more common to produce intrinsically three-dimensional ultrasound probes, we believe that a hybrid system, such as we propose here, will become the natural form of ultrasound scanner. It will offer 3D tools that in some cases will replace CT (X-ray computed tomography) with greater safety and replace MRI (magnetic resonance imaging) at lower cost. Thus, as well as our main goal of developing new imaging techniques based on current technology, this project has addressed the design issues that target the future of ultrasonic imaging in general.

Hybrid 3D Ultrasonic Imaging System

Ultrasonix Sonix RP hardware that provides a complete solution to hybrid 3D ultrasonic imaging.

Modified keypad controls for Ultrasonix Sonix RP, fully integrated with the developed software. The data acquisition process is as simple as holding the 3D probe with one hand and clicking the acquire volume button (bottom right) with the other hand.

3D probes with jigs and attached accelerometers.

Software developed for Ultrasonix Sonix RP. The figure shows volumetric data of an arm and a reslice across the ROI. This software uses image-based registration algorithms in combination with measurements provided by the attached inertial sensor to stitch the volumes together.



The following information gives a sneak-preview of those features in Stradwin that are developed for our hybrid three-dimensional clinical ultrasound system i.e. the "Registration" tab. The following videos are generated in real time and demonstrate the registration capabilities.  Double-click to see these videos in fullscreen.  
Registration results for 3D arm data set. The data is recorded with a mechanically-swept 3D probe and comprises a sequence of 2D frames. The currently selected B-scan is shown in red outline in the top-right window with its contents displayed in the top-left window. Here, 10 overlapping volumes make up the in vivo scan of the right arm. The probe was moved in the lateral direction while acquiring the volumes. After loading the data, the clinician can just click on the "Register Volumes" button without needing to configure anything. The details of the registration algorithms are given in [1,2]. The volumes move on the screen during the registration process. This was done with the intention of helping the clinician see whether the algorithms are converging to the same positions or not. After the registration process is complete, the clinician can click on the "Locate for minimum voxel" button which fits a reslice plane around the registered volumes (shown with green outline in the top-right window). The lower half of the visualisation area displays the content of this reslice. There are two ways to visualise this information: Dividing plane view, in which the data on the either side of the plane (shown with a blue line) comes from different volumes; and a blended volumes view, in which the opacity of the overlapping volumes can be altered to analyse the registration performance. The registration is always done in pairs and using the blended volumes, each pair can also be analysed separately.
Demonstration of robustness and speed with the benchmark datasets. In this video, we run one of the fastest algorithms on 16 benchmark datasets in turn (comprising both in vivo and phantom scans as used in [1,2]). It can be seen that the registration is successful on every attempt with an average registration time between 6 and 8 seconds. Other algorithms are also available for cases where the fast algorithm is unsuccessful although these take longer to run.
Tools for manual registration. Should every algorithm fail, one can rely on his instincts to register the volumes. For this purpose, the manual alignment tool is provided on the "Registration" tab. By double-clicking on any of the ortho reslices, the tool gets aligned automatically to the orientation of that slice. One can then use different blending options in conjunction with the manual registration. Registration results with a curvilinear probe. Here, we have used a convex probe to scan the abdomen. The imaging depth was 12.7cm. The probe was moved in the lateral direction and three volumes were acquired.

Multi-directional scattering

Once the registration is complete, there are potentially useful things that can be done to exploit the richness of overlapping data with different directions of insonification. An ultrasound image is created from back-scattered echoes originating both from diffuse and directional scattering. We can separate these two components for the purpose of tissue characterization and have developed novel algorithms [4] allowing for both 2D and 3D variations in insonfication direction.

2D simulated data with specular scattering on an arc. We have generated 31 ultrasound images using Field II simulation software with angular variation from -15 to +15 degrees. The algorithm has separated the diffuse and specular components and correctly identified the scattering direction along the arc.

3D phantom data with specular scattering on a plane. The scanning subject is a speckle phantom consisting of an agar cylinder with uniform distribution of aluminium oxide powder providing scattering. Specular scattering was achieved by inserting a thin sheet of aluminium foil. The results are generated using a 3D algorithm on a reslice through the centre.


  1. UZ Ijaz, RW Prager, AH Gee, and GM Treece. Rapid hybrid ultrasound volume registration.  Technical Report CUED/F-INFENG/TR 644, Cambridge University Department of Engineering, March 2010.
  2. UZ Ijaz, RW Prager, AH Gee, and GM Treece. Optimisation strategies for ultrasound volume registration. Measurement Science and Technology, 21(8): 085803 (17pp), July 2010.
  3. RW Prager, UZ Ijaz, AH Gee, and GM Treece. Three-dimensional ultrasound imaging. Proceedings of the Institution of Mechanical Engineers Part H-Journal of Engineering in Medicine, 224(2):193-223, 2010.
  4. UZ Ijaz, RJ Housden, GM Treece, RW Prager, and AH Gee. Multi-directional scattering models for 3D ultrasound. Technical Report CUED/F-INFENG/TR 661, Cambridge University Department of Engineering, January 2011.


Cambridge UniversityCUEDMedical Imaging Group  |

Department of Engineering at the University of Cambridge
Information provided by Dr Umer Zeeshan Ijaz (uzi20)
Last updated: January 2011