Medical ultrasound images have traditionally been presented as two-dimensional images of essentially raw ultrasound data, as in familiar fetal ultrasound images. Three-dimensional volumetric images could conceivably be constructed by appropriately combining multiple, two-dimensional ultrasound images slices. The research being conducted under this project aims at examining data acquisition schemes and data processing algorithms for appropriately registering, assembling, and displaying the volumetric ultrasound data. The focus of this project is the development of a 'free-hand' 3D ultrasound scanning system that can work with existing clinical scanners. This work allows us to demonstrate our capabilities to medical researchers and clinicians, for addressing a clinically relevant research issue.

A true volumetric real-time ultrasound scanning technique employs a novel 2D matrix array of piezoelectric crystals that generates a pyramidal burst of ultrasound to scan the region-of-interest and to create a dynamic 3D representation. Real-time 3D echocardiography requires sophisticated parallel-processing system for very rapid beam formation and data processing. Such specialized ultrasound systems and transducers are still under development, and are not widely available. In contrast, the 3D ultrasound approach using existing ultrasound scanners can be easily adapted in the clinical settings for studies in near future.

The 'free-hand' scanning approach for 3D ultrasound uses magnetic-field based motion sensor and has been used by many researchers and a few recent commercial products. CNDE has already assembled such a system using initiative funds from the Roy J. Carver Charitable Trust. Use of this system allows CNDE scientists to develop and implement newer algorithms, whereas commercial systems restrict access to their proprietary software. The current research issues in 3D ultrasound include calibration and registration, processing non-parallel slices, compounding and interpolation, visualization, image and volume segmentation, and automatic measurements.

In laboratory settings, multiple parallel 2D image slices can be collected using a stepper-motor controlled movement of the transducer. However, for the user (physician or ultrasonographer) in a clinical setting, the most intuitive and acceptable technique would be to setup a transducer (position and orientation) locating mechanism so that a ‘free-hand’ movement of the transducer can be used for 3D data collection. As illustrated in figure 1, a position sensor is attached on the ultrasound transducer, and the operator scans the internal organs in a manner similar to a standard 2D ultrasound exam. The 2D images are collected, each image originating from a unique location and orientation within the body, and the position sensor simultaneously feeds this information to the computer. We use an electromagnetic positioner that contains a spatially varying magnetic field transmitter placed rigidly in the vicinity of the transducer, and a receiver containing three orthogonal coils to measure field strength, which is mounted rigidly on the transducer.

Figure 1.  Concept of free-hand scanning for volumetric data acquisition using conventional ultrasound scanner and the probe.  A small receiver, mounted on the ultrasound probe, detects the electromagnetic field generated by the transmitter to detect six degree-of-freedom motion.  Each image slice is tagged with the position information for later use in 3D reconstruction.

CNDE scientists have developed individual software components to allow image capturing with position information, to preprocess data and to visualize using volume and surface rendering. A 3D ultrasound phantom has been used for assessment of 3D data acquisition schemes and the accuracy of algorithms for registration, reconstruction, and volumetric measurement. Additional calibration experiments are done using the stepper-motor controlled motion of the transducer to collect ideal parallel slices. Examples of 2D image and 3D reconstructed volume of phantom data are shown in figure 2.

Figure 2.  Data from a 3D ultrasound tissue-mimicking phantom containing two egg-shaped contrast objects. Left - 2D ultrasound image showing cross-sections of the two egg-shaped contrast objects. Right - 3D visualization of the two egg-shaped objects reconstructed from multiple 2D image slices and object segmentation.

The processing and visualization of volumetric ultrasound data have been challenging research tasks due to significant noise and speckle, lower dynamic range than MRI and CT images, blurred boundaries around anatomical features, and variable resolution through the volume. The techniques to overcome these challenges include enhancing the quality of the original data using filtering, interpolation, and spatial compounding. A registration algorithm has been developed using homogenous matrix transformation, from image pixel position to receiver coordinate, then to transmitter coordinate, and finally to reconstruction volume space. A calibration procedure for position sensor and image registration is developed. For calibration, an object such as a small stainless-steel sphere in a fixed position is scanned from multiple angles, positions, and directions to derive a set of coefficients to be used in the coordinate transformations. For visualization, volumetric and surface rendering techniques, for both parallel and non-parallel data, are implemented using OpenGL and VTK graphics standards in C++ environment. Under guidance of Dr. Amin, an Electrical Engineering graduate student, Bo Wang, has made significant contributions to this project.

After initial experiments in the CNDE lab, potential application of this prototype system have been sought. With help of University of Iowa collaborators, Milan Sonka and Ronald Lauer, a significant application of the 3D ultrasound approach for carotid and brachial artery scanning for atherosclerosis risk and disease evaluation has been identified. Milan Sonka, PhD, professor in Electrical and Computer Engineering, is an experienced medical imaging researcher, and Ronald Lauer, MD, professor in Pediatric Cardiology and Epidemiology, leads long-term NIH-funded research study on atherosclerosis risk in young adults. Our demonstration of early developments in 'free-hand' 3D ultrasound system to these colleagues has generated very enthusiastic response for its application to atherosclerosis risk evaluation. Trial scans of carotid and brachial arteries have been performed using a clinical ultrasound scanner connected with our 3D ultrasound system. Software tools have been developed for acquisition, processing, and visualization of the volume data. We have been developing our system further towards the carotid and brachial scanning for atherosclerosis risk evaluation.

Identification of patients at increased risk for cardiovascular artery disease and atherosclerosis, before they occur, is crucial. There are no current screening methods for arteriosclerosis, except for screening for the most important risk factors such as high cholesterol levels, high blood pressure, and positive family history. Development and validation of ultrasound imaging as a measure of atherosclerosis has provided a method for visualizing the atherosclerotic process at the level of the vessel wall and is suitable for application in population-based studies. Use of carotid ultrasound has demonstrated strong relationships of carotid wall thickness to clinical CVD and to traditional CVD risk factors such as age, sex, lipids, hypertension, smoking, diabetes, obesity and dietary fat. The B-mode ultrasound makes it possible to non-invasively characterize arterial wall function in the carotid and brachial arteries by measuring wall thickness (intima-media thickness or IMT), wall stiffness, and endothelial function expressed as percent increase in vessel diameter after a stimulus (flow mediated dilatation or FMD) of the brachial artery.

The NIH-supported Muscatine Study, under the direction of Ronald Lauer, M.D., of University of Iowa, is the longest running study of childhood risk factor assessment and follow-up in later adult life. The present study measures the early atherosclerotic process by utilizing the electron beam computed tomograph to measure the presence of coronary calcification, ultrasound techniques to measure carotid artery intima-media thickness (IMT, a vessel wall thickness measurement) and brachial artery flow-mediated vasodilatation (FMD). All of these measures utilize 2-dimensional images. For this study, he scans a sample of young population (recruited in the study) at regular intervals for carotid and brachial arteries. The ultrasound scans are saved on super-VHS tape and sent to ultrasound reading center for interpretation. An important measurement of these scans is mean arterial wall intima-media thickness (IMT) at specified locations. The IMT is considered as a predictor of atherosclerosis and stroke since the atherosclerosis begins as some deposition of cholesterol and fat under the intimal layer of vessel wall that in turn increases the IMT value. Although this ultrasound procedure is standardized and done by trained technicians, some challenges exist such as repeatability and accuracy of the measurements at reference points. For example, different angles of the transducer provide different views of the vessel and interpretation technician may find it difficult to interpret references and measure repeatable IMT. An accurately reconstructed 3D ultrasound data could overcome these challenges since the interpretation is not limited by the views selected by the sonographer. Properly collected 3D data allows views and measurements from any angle after the data are acquired. Another challenge Dr. Lauer has indicated is a need to know the maximum diameter of the brachial artery for accurate and repeatable measurement. This issue can also be addressed by a 3D ultrasound system. In the long term, we seek to test hypothesis that the artery wall measurements (carotid wall thickness and change in brachial diameter during flow-mediated vasodilation test) using 3D ultrasound provide better repeatability than those derived from conventional 2D ultrasound scans.

In early demonstration, experimental carotid and brachial artery scans of a volunteer were performed at Dr. Lauer's Pediatric Cardiology clinic using a clinical ultrasound scanner with 7.5 MHz array probe and our early (uncalibrated) 3D acquisition setup. Additional experimental scans were performed at CNDE using a 10 MHz scanner and help of Dr. Booth of ISU Veterinary Clinic. The 3D ultrasound concept for carotid artery is illustrated in figure 3 using our preliminary data from these early experiments.  An accurately reconstructed 3D ultrasound data set could improve repeatability of IMT measurements and could help address a challenging task of identifying maximum diameter of brachial artery for accurate and repeatable measurements during FMD tests. Application specific tasks include ECG-gated acquisition to avoid artifacts of carotid artery pulsation during cardiac cycle and segmentation of the vessel wall layers.

Figure 3.  Illustration of 3D ultrasound concept for carotid artery.  Carotid artery anatomy (left) showing common, internal, and external carotid segments, 7.5 MHz ultrasound 2D scan in cross-section (bottom left) and longitudinal-section (bottom right); illustration of transducer position and orientation during 3D acquisition setup (bottom left); and 3D reconstruction after segmenting the vessel (bottom right).

In the future, the project will continue improvements in the free-hand data acquisition scheme, data processing algorithms for appropriately registering and displaying the volumetric ultrasound vessel scans.  In addition, techniques will be developed for measuring vessel wall IMT and diameter characteristics. The system and algorithms will be developed and validated using calibration test phantoms and in vitro scanning of atherosclerotic aorta segments in laboratory settings. The system will be evaluated for improvements in repeatability and accuracy for carotid and brachial scanning in volunteers in clinical settings, following established clinical scanning protocols.

For further information, contact Viren Amin.