WO2006119623A1

WO2006119623A1 - System and method for generating dynamic ct images of a target within a subject that experiences motion

Info

Publication number: WO2006119623A1
Application number: PCT/CA2006/000741
Authority: WO
Inventors: Marcin Wierzbicki; Terry Peters
Original assignee: Robarts Research Institute
Priority date: 2005-05-09
Filing date: 2006-05-09
Publication date: 2006-11-16

Abstract

A system and method of generating a dynamic CT image of a target within a subject that experiences motion is provided. During the method, CT radiograph data of the target is acquired during substantially its entire phase at varying exposure levels and the CT radiograph data is reconstructed to yield at least one high-quality image and a set of lesser-quality images of the target. The high-quality and lesser-quality images are registered to generate a set of image transforms. At least one deformed high-quality image is generated based on the high-quality image and the set of image transforms.

Description

SYSTEM AND METHOD FOR GENERATING DYNAMIC CT IMAGES OF A TARGET WITHIN A SUBJECT THAT EXPERIENCES MOTION

Field of the Invention

The present invention relates generally to the field of X-ray computed tomography (CT) imaging and more specifically, to a method and system for generating dynamic (four-dimensional (4D)) CT images of a target within a subject that experiences motion.

Background of the Invention

X-ray computed tomography, including conventional, helical and electron-beam forms, is commonly used to produce cross-sectional and three- dimensional (3D) images of the chest, including the heart and greater vessels. In general, cardiac tomography (also called cardiac CT scan and coronary artery scanning) is useful to evaluate aortic disease (such as aortic dissection), cardiac masses and pericardial disease. As is well known, during image acquisition the heart is in a constant state of motion, which can cause image artefacts, namely blurring and streaking in reconstructed cardiac CT images. A variety of methods have been developed in the art to account for this motion so that improved cardiac CT images can be generated. In most methods some form of cardiac gating (prospective or retrospective) using electrocardiography (ECG) or pulse oximetry is required to synchronize the acquisition and/or reconstruction of CT images with the phase of the cardiac cycle. Although these methods improve CT image acquisition and/or reconstruction, they do suffer drawbacks.

Retrospective gating techniques measure ECG signals during acquisition of the CT radiograph data over the entire cardiac cycle. A computer is then used to retrospectively select data from the acquired data set that correspond to a particular part of the cardiac cycle (e.g. around the mid-diastole phase (MD)) to reconstruct the image of the heart at that time point. Unfortunately, CT image quality varies depending on the part of the cardiac cycle the acquired data set being sampled represents. Images reconstructed from CT radiograph data acquired during the diastolic phase of the cardiac cycle are relatively free of motion artefacts and are considered to be good quality CT images. Images reconstructed from CT radiograph data acquired during the other phases of the cardiac cycle are subject to greater image artifacts from organ motion, and thus are much lower quality CT images. These retrospective gating techniques also result in a high X-ray dose or exposure to the patient, and thus are unsuitable for repeated imaging procedures, and image-guided surgical applications.

In prospective gating techniques, ECG signals acquired from the subject during CT scanning are used to trigger image acquisition at specific times during the cardiac cycle when the heart is relatively stationary such as during MD. Using this method, only radiographs collected during the "quasi- static" phase of the cardiac cycle are used in image reconstruction thereby attempting to minimize the effect of heart motion. Although this method reduces the X-ray dose administered to the patient when compared to retrospective gating techniques, it limits visualization of the heart only to those times when there is little motion (i.e. during MD). No image information is available at other times during the cardiac cycle when the heart is in motion, and there is no ability to visualize the 3D heart anatomy in dynamic motion (i.e. 4D).

Other methods in the field have proposed to "gate" the CT images on the basis of characteristics of the observed radiographic data (kymograms) as described by Kachelrieβ M., Sennst D., Maxlmoser W. and Kalender W. (Medical Physics - - July 2002 - - Volume 29, Issue 7, pp. 1489- 1503).

Previous work has also focused on the development of enhanced cardiac visualization tools for minimally invasive surgery and therapy applications. For example, the publication entitled "Validation of dynamic heart models obtained using non-linear registration for virtual reality training, planning, and guidance of minimally invasive cardiac surgeries" authored by Marcin Wierzbicki, Maria Drangova, Gerard Guiraudon and Terry Peters (Medical Image Analysis. VoI 8 (3), 387-401 , 2004) describes the validation of a non-linear registration algorithm designed to extract the motion of the heart from 4D CT and MR images. The long-term goal of the motion extraction procedure is to create a dynamic, geometric surface model of the heart to aid a cardiac surgeon in training for, planning for, and guiding a minimally invasive cardiac surgery. Initially, a static model is generated by segmenting one of the image frames in a given 4D data set. The dynamics of the static model are then extracted from the remaining image frames using a non-linear, intensity-based registration algorithm with a choice of six (6) different similarity metrics. The registration algorithm is validated on an artificial CT image set created using an excised porcine heart, on CT images of canine subjects, and on MR images of human volunteers. With the appropriate choice of similarity metric, the registration algorithm extracts the motion of the epicardial surface in CT images, or of the myocardium, right atrium, right ventricle, aorta, left atrium, pulmonary arteries, vena cava and epicardial surface in MR images, with a root mean square error in the one millimeter range.

The publications entitled "Determining Epicardial Surface Motion Using Elastic Registration: Towards Virtual Reality Guidance of Minimally Invasive Cardiac Interventions" authored by Marcin Wierzbicki and Terry Peters (Nov 2003 "Medical Image Computing - Computer-assisted Intervention" (MICCAI-2003) Montreal Nov 15-18, 2003. Lecture Notes in Computer Science (LNCS) 2878, RE Ellis TM Peters (eds)), and "Four- dimensional modeling of the heart for image guidance of minimally invasive cardiac surgeries" authored by Marcin Wierzbicki, Maria Drangova, Gerard Guiraudon and Terry Peters (Proceedings of SPIE, Volume 5367, Medical Imaging 2004: Visualization, Image-Guided Procedures, and Display, pp. 713- 723) describe the validation of an algorithm used to extract the motion of a heart in 4D helical CT images. The extracted motion is used to animate a virtual surface model over the cardiac cycle for eventual use in training for, planning for, and guiding a minimally invasive cardiac surgery.

The publication entitled "Cardiac Endoscopy Enhanced by Dynamic Organ Modeling for Minimally-invasive Surgery Guidance" authored by Szpala S, Guiraudon G and Peters, TM ("Medical Image Computing - Computer-assisted Intervention" (MICCAI-2003) Montreal Nov 15-18, 2003; Lecture Notes in Computer Science (LNCS) 2878, RE Ellis, TM Peters (eds), -A-

499-506, Springer-Verlag Heidelberg, 2003) describes the integration of a cardiac model based on sequential CT scans with an endoscopic view of the actual dynamic model. This paper serves to demonstrate the level of artefact seen in a dynamic sequence of CT images. The conference paper entitled "A High Resolution Dynamic

Heart Model Based on Averaged MRI Data" by John Moore, Maria Drangova, Marcin Wierzbicki, John Barron, and Terry Peters (Nov 2003 "Medical Image Computing - Computer-assisted Intervention" (MICCAI-2003) Montreal Nov 15-18, 2003. Lecture Notes in Computer Science (LNCS) 2878, RE Ellis TM Peters (eds). pp 549-555) describes the use of a registration algorithm in aligning images of the same volunteer and averaging the results to create a high resolution and high-signal-to-noise ratio image. The purpose of the registration algorithm is to assemble an excellent quality image for various applications. U.S. Patent No. 6,393,091 to Slack et al. describes a method for imaging a heart of a patient utilizing a CT imaging system. The method includes the step of assigning a scanning priority to phases as a representative cardiac cycle of the patient's heart, selecting phases of the cardiac cycle for scanning in accordance with the assigned scanning priority, and obtaining image slices of the patient's heart corresponding to the selected phases of the cardiac cycle. The method can be performed by a CT imaging system including an EKG machine to record EKG data.

U.S. Patent No. 6,353,653 to Edic describes a method and apparatus for use with a computed tomography (CT) system that collects CT radiograph data for every view of the CT gantry so that a data set corresponding to all views of the CT gantry is available for use in reconstructing an image of a patient's heart and coronary vasculature. For each view of the CT gantry, the CT radiograph data associated with the view is collected at different instants in time with respect to the period of the cardiac cycle and each revolution of the CT gantry. Prior to data acquisition, the patient's heart rate is measured and the period of the CT gantry is set such that data is acquired at a different time with respect to the period of the cardiac cycle for every view of the CT gantry and for each revolution of the CT gantry. Therefore, for each revolution of the CT gantry and for each view of the CT gantry, the instant in time in the period of the cardiac cycle at which any given detector element of the detector array is sampled will be different from the instant in time in the period of the cardiac cycle at which the same detector element was sampled in the previous revolution. After all of the CT radiograph data is collected, the radiographs are processed by an interpolation algorithm that interpolates radiographs to a selected instant in time with respect to the period of the cardiac cycle. A reconstruction algorithm is then used to process and back-project the interpolated radiographs to produce a 3-dimensional (3D) image of the heart and coronary vasculature. The interpolation algorithm may be performed repeatedly to interpolate radiographs to more than one instant in time, and then corresponding reconstructions may be performed to generate a A- dimentionsal (4D) of the heart and coronary vasculature.

U.S. Patent Application Publication No. US2004/0077941 to Reddy et al. describes a method and system for associating ECG waveform data with medical imaging data using ECG gating for dose reduction and image improvement by generating the ECG waveform data using an electrocardiogram device. The ECG data is first validated and then QRS complexes are detected using a detection function. An underlying cardiac rhythm based on the detected QRS complexes is analyzed and an even number N of substantially normally shaped consecutive QRS complexes are selected. An RR interval between consecutive QRS complexes is computed to yield N-1 intervals. Duration of a representative cardiac cycle by averaging at least a plurality of N-1 intervals is determined. Once a representative cardiac cycle is determined, a method to control power and improve image quality with the presence of patients having arrhythmias is performed. Although, the above-references disclose various imaging techniques, these techniques fail to acquire image data over much of the cardiac cycle at reduced radiation exposure levels to patients and to generate entire dynamic CT imaging sequences. As will be appreciated, improvements in CT imaging are desired.

It therefore an object of the present invention to provide a novel method and system for generating dynamic CT images of a target within a subject that experiences motion.

Summary of the Invention

Accordingly, in one aspect of the present invention there is provided a method of generating a dynamic CT image of a target within a subject that experiences motion comprising: acquiring CT radiograph data of said target during substantially its entire phase at varying exposure levels to yield at least one high-quality image and a set of lesser-quality images of said target; registering the high-quality and lesser-quality images to generate a set of image transforms; and generating at least one deformed high-quality based on the high-quality image and the set of image transforms.

In one embodiment, a CT scanner with prospective ECG gating is used to acquire 3D images of the heart at different time points in the cardiac cycle. In particular, projections are acquired during the mid-diastole phase of the cardiac cycle when there is less inherent heart motion, with sufficient X- ray dose and number of projections, to acquire CT radiograph data to reconstruct a high-quality static CT image of the heart. Fewer projections and/or lower X-ray dose projections are then obtained during other defined phases of the cardiac cycle, where there is greater inherent motion, resulting in production of a set of lesser-quality CT cardiac images. The total dose used to acquire these projections is then considerably lower than that used to acquire the mid-diastole projections. The lesser-quality images are not of sufficient quality to extract high level detail of the heart anatomy, however, there is sufficient detail to characterize the motion of the heart during the cardiac cycle. An image registration method is then used to successively register the high-quality CT cardiac image to each lesser-quality CT cardiac image in the set, resulting in the generation of a series of vector fields representing the non-linear movement/warping of the heart tissue from phase to phase in the cardiac cycle. Due to the known continuity of the motion of the heart during the cardiac cycle, additional vector fields can be interpolated at times between the acquired data sets, thus approximating the effects of obtaining greater temporal resolution. The complete set of vector fields (acquired and interpolated) are then used to non-linearly warp the high-quality CT cardiac image dynamically to simulate the entire movement of the heart throughout the full cardiac cycle, based on the actual heart movement. The resulting dynamic heart image benefits from the spatial resolution of the high- quality CT cardiac image, and has higher temporal resolution, with no need for increased scan time, and reduced X-ray dose to the subject. In an alternate embodiment, rather than automatically interpolating between the vector fields with a fixed time step to create a new set of vector fields, representing images equi-spaced throughout the cardiac cycle, the user can select any set of arbitrary time points during the cardiac cycle to reconstruct 3D views of the heart anatomy. Vector fields corresponding to the position and shape of the CT cardiac image at these specific (user selected) time points are then interpolated from the originally acquired vector field set, and images of the heart are obtained by non-linearly warping the high-quality CT cardiac image based on the interpolated vector field(s). According to another aspect of the present invention there is provided a system for generating a dynamic CT image of a target within a subject that experiences motion comprising: a CT scanner acquiring CT radiograph data of said target at varying exposure levels to yield at least one high-quality image and a set of lesser-quality images of said target; and processing structure registering the high-quality and lesser- quality images to generate a set of image transforms and generating at least one deformed high-quality based on the high-quality image and the set of image transforms.

According to yet another aspect of the present invention there is provided a method of generating a dynamic CT image of an organ comprising: imaging a subject's organ at differing exposure levels to yield a high-quality image of the organ when at rest and a set of lesser-quality images of the organ when in motion; registering the high-quality and lesser-quality images to generate a motion description of the organ; and generating at least one warped high-quality image based on the motion description and high-quality image.

The method and system of generating dynamic CT images provides advantages in that visualization of a high-quality dynamic model (4D) of the heart can be achieved with significantly lower X-ray exposure levels to the patient. Also, a 4D CT cardiac image can be created with minimal image artifacts normally caused by organ motion. Further, a user is able to generate high-quality images of the heart at arbitrary time points during the cardiac cycle, different from those originally acquired. In addition, an apparent increase in the temporal resolution of the 4D image data set is obtained without increasing the X-ray dose to the patient or the scan time. Extracted cardiac motion information can be used directly to deform geometric models

(points, lines, surfaces, volumes) representing the patient's heart anatomy over the cardiac cycle for added visualization, diagnosing of heart conditions, or other applications.

Brief Description of the Drawings

Embodiments will now be described more fully with reference to the accompany drawings in which:

Figure 1 is a flowchart showing the general steps performed to generate a dynamic CT image of a target within a subject that experiences motion; Figure 2a shows a system for performing the steps of Figure 1 comprising a CT scanner with a patient on the scanner table, and an ECG machine monitoring the patient's heart;

Figure 2b shows an X-ray source and filter forming part of the CT scanner;

Figure 3 shows a prospective gated acquisition scheme over a single heart beat characterized by the R-R interval employed by the system of Figure 2;

Figures 4a to 4c are flowcharts showing the steps performed during the acquisition, registration of images, image generation steps of Figure 1 ;

Figure 5 are images at various stages during the method of Figure 1 ; and

Figure 6 is a flowchart showing alternative image generation steps.

Detailed Description of the Embodiments

Turning now to Figure 1 , a flowchart showing the overall method of generating dynamic CT images of a patient's heart is shown. Initially, CT radiograph data is acquired in a manner to reduce a patient's exposure to radiation and the acquired CT radiograph data is reconstructed to yield a high- quality 3D CT cardiac image around mid-diastole, and a series of lesser- quality 3D CT cardiac images elsewhere during the cardiac cycle (step 100). Thus, the high-quality 3D CT cardiac image is reconstructed from CT radiograph data acquired while the heart is substantially stationary and the lesser-quality 3D CT cardiac images are reconstructed from CT radiograph data acquired while the heart is in motion. The high-quality 3D CT cardiac image and the lesser-quality 3D CT cardiac images are then registered to extract the motion of the heart throughout the cardiac cycle (step 200). The CT cardiac image registration yields one high-quality 3D CT cardiac image and a 4D cardiac motion description. An evenly spaced in time set of high- quality 3D CT cardiac images (i.e. a dynamic (4D) CT image) is then created using the original high-quality 3D CT cardiac image and the 4D cardiac motion description (step 300).

Turning now to Figures 2 to 4a, the steps performed during CT radiograph data acquisition will be further described. In order to acquire the CT radiograph data, a CT scanner 50 as shown in Figure 2 is utilized. As can be seen, CT scanner 50 comprises a housing 52 having an opening 54 into which a patient 56 is positioned. A scanner table 60 supports the patient 56 and is moveable axially relative to the housing 52 allowing the patient to be inserted and removed from the housing 52. The patient 56 is connected to an ECG machine 62 via a plurality of electrodes 64 placed on the patient's chest. The CT scanner 50 is also connected to the ECG machine 62. An X-ray source 65, filter 66 and X-ray detector are accommodated by the housing 52. The filter 66 is moveable between a retracted position and an extended position. In the retracted position, the filter 66 does not impede X-rays emitted by the X-ray source 65 allowing for full (i.e. 100%) X-ray exposure. In the extended position, the filter 66 blocks a portion of the X-rays emitted by the X-ray source 65 thereby to reduce X-ray exposure. A computer 70 is coupled to the CT scanner 50 so that acquired CT radiograph data can be processed. During CT radiograph data acquisition, the patient 56 is placed on the scanner table 60 and is connected to the ECG machine 62 via the electrodes 64. The ECG machine 62 is then initiated so that ECG recording commences (see step 102 in Figure 4a). The patient 56 is then instructed to hold their breath and the scanner table 60 is moved axially to position the patient within the opening 54 of the housing 52. The CT scanner 50 is then operated so that helical CT projections are acquired (step 104).

During helical CT projection acquisition, the number of photons passed through the patient is modulated using prospective ECG gating (i.e. the X-ray exposure level to the patient 56 is varied). In particular, for a given heart beat, the R to R time is estimated from the ECG recording and divided into two parts, namely the times immediately surrounding the mid-diastole phase of the cardiac cycle, and the remaining times. For each helical CT projection being acquired, a check is made to determine if the heart is in the mid-diastole phase (step 106). During the mid-diastole phase, the filter 66 is mechanically moved to the retracted position to allow the full photon dose to be delivered to the patient 56 and a complete set of helical CT projections are acquired (step 108). During the remaining times, the filter 66 is moved to the extended position to block a portion of the radiation delivered to the patient 56 and a complete set of helical CT projections are acquired (step 110). In this manner, the patient is exposed to reduced radiation levels during helical CT projection acquisition outside of the mid-diastole phase. Steps 104 to 110 are repeated until the desired number of helical CT projections have been acquired. The overall result is that the number of photons (dose) used to create the mid-diastole high-quality CT cardiac image is 100% of those normally used, while only a small fraction "f" of photons is used to create the lesser-quality CT cardiac images at other time points in the cardiac cycle. The acquired helical CT projections are reconstructed using standard retrospective reconstruction software from the CT scanner manufacturer (step 112), producing one high-quality CT cardiac image around mid-diastole from the helical CT projections acquired at full X- ray exposure levels and a series of lesser-quality 3D CT cardiac images from the helical CT projections acquired at reduced X-ray exposure levels elsewhere (step 114) thereby to yield a 4D image set. Although the lesser- quality 3D CT cardiac images are not of sufficient quality to extract high-level detail of the heart anatomy, the lesser-quality 3D CT cardiac images include enough detail to charcterize the motion of the heart outside of the mid-diastole phase.

At step 110, if desired, rather than using the filter 66 to block emitted X-ray radiation, in order to reduce exposure levels, the number of helical CT projections acquired outside of the mid-diastole phase can be reduced, thereby to reduce overall X-ray exposure levels. Of course, the filter 66 can be used to block emitted X-ray radiation and the number of helical CT projections acquired outside of the mid-diastole phase can be reduced to further reduce exposure levels. Rather than using a mechanical filter 66 to block emitted X-ray radiation, an electronic filter that modulates the current to the X-ray source 65 can be used.

Figure 3 illustrates the above acquisition scheme over a single heart beat characterized by the R-R interval. As will be appreciated, the acquisition scheme is similar to a typical retrospectively gated acquisition scheme. The photon levels employed to acquire the helical CT projections are shown below the ECG signal. In this example, full dose (100%) photon levels are used during helical CT projection acquisition in the mid-diastole phase and reduced photon levels are used during helical CT projection acquisition outside of the mid-diastole phase. The R-R interval is divided into reconstruction bins as shown below the photon levels for the purpose of 4D image reconstruction. In this example, the R-R interval is divided into nine (9) bins.

As mentioned previously, registration between the high-quality 3D CT cardiac image and the lesser-quality 3D CT cardiac images is required to extract the motion of the heart throughout the cardiac cycle, thereby to yield one high-quality 3D CT cardaic image together with a 4D cardiac motion description. In this embodiment, the registration is performed in a free form deformation (FFD) frame work where a 3D grid of points (nodes) is overlaid on the high-quality 3D CT cardiac image to be registered (i.e. the source image). Each node is then assigned a displacement vector estimating the deformation required in the surrounding region to match this part of the source image to a target lesser-quality 3D CT cardiac image.

The process of finding the vector displacement at any given node in the FFD grid is an optimization problem that balances an image similarity metric and the transform probability terms. The optimization itself is performed using the downhill simplex method. For each node in the FFD grid, the displacement vector is determined by minimizing the following function:

C(x, y, z) j

where:

C is the cost associated with the 3D displacement vector (x,y,z); MSD is the mean squared difference between source image intensities l_s translated by (x,y,z) and target image intensities l_τ calculated over the sub-volume V centered on the current FFD node (measures degree of image alignment); and BE is the value of the regulahzation term, weighted by the constant α (measures the probability of x,y,z).

The overall optimization process is divided into three main stages, each with an increasingly dense grid of vectors. At each stage, the vector grid is refined and applied to the source image iteratively, allowing the source image to slowly progress towards the optimal registration. After a given iteration, the improvement in global image alignment is measured using the MSD and the average vector displacement magnitude is calculated. If the magnitude of the average vector displacement is less than a specified constant, the optimization process proceeds to the next stage where a higher density grid is used. If image alignment is reduced after a given registration, the optimization process reverses the iteration and proceeds to the next stage with the previous result. Between stages, the FFD grid is subdivided using linear interpolation.

The registration steps required for motion estimation are shown in Figure 4b. Each image in the 4D image set is represented by a number n from 0 to N, where n = M is reserved for the source image and N is the total number of images in the 4D image set (step 202). The registration between the source image and each target lesser-quality image n is performed as a combination of serial registrations. In particular, an iteration variable i is set to zero (0) and the transform T₀ for iteration i=0 is set to null (step 204). The high-quality source image is then deformed by the transform T₁ and the result is registered to the (M¹ (i+1 )) target lesser-quality image to produce a transform T (step 206). Transform T₁ is then combined with transform T to yield transform T_i+1 (step 208). A check is then made to determine if M¹ (i+1 )=n (step 210). If not, the iteration value i is incremented (step 212) and the process reverts back to step 206. If so, the transform Ti₊₁ is stored and a check is made to determine if n=M (step 214). If not, the value n is incremented, skipping n=M (step 216) and the process reverts to step 204. If so, the process is deemed to be complete yielding a set of N-1 transforms (step 218).

For example, if n = 2, M = 5, and N = 10, the high-quality source image is first registered to the n = 4 target lesser-quality image. The resulting n = 4 transform is used to generate a deformed source image which is then registered to the n = 3 target lesser-quality image. The two previous transforms (n = 4 and n = 3) are then combined and used to deform the original high-quality source image, which is then registered to the n = 2 target lesser-quality image. Finally, the three previous transforms (n = 4, 3, 2) are combined to give the direct transformation between the high-quality source image and the n = 2 target lesser-quality image. This process is repeated for each target lesser-quality image, resulting in as many 3D vector fields as target lesser-quality images (N - 1 ). As will be appreciated, the above registration method successively registers the high-quality CT cardiac image to each lesser- quality CT cardiac image in the image set resulting in the generation of a series of vector fields representing the non-linear movement/warping of the heart tissue from phase to phase in the cardiac cycle. Due to the known continuity of the motion of the heart during the cardiac cycle, additional vector fields can be interpolated at times between the acquired image data set.

During dynamic (4D) CT cardiac image generation at step 300, an evenly, spaced in time, set of high-quality 3D CT cardiac images (i.e. a high-quality 4D CT cardiac image) is created. The steps performed to generate the set of high-quality 3D CT cardiac images are shown in Figure 4c. With the N - 1 transforms obtained at step 218, a variable n is set to zero (0) (step 302) and the transform T_n is applied to the high-quality source image (step 304). The resultant deformed high-quality image is saved (step 306) and a check is made to determine if n=N (step 308). If not, the variable n is incremented, skipping n=M, (step 310) and the process reverts back to step 304. If so, the process is deemed to be complete yielding the high-quality 4D CT cardiac image comprising N - 1 deformed high-quality CT cardiac images and the original high-quality source image (step 312).

As will be appreciated, during the dynamic CT cardiac image generation process described above, the complete set of vector fields, acquired and interpolated during registration, are used to non-linearly warp the original high-quality CT cardiac image dynamically to simulate the entire movement of the heart throughout the full cardiac cycle, based on actual heart movement. The resulting dynamic high-quality 4D CT heart image benefits from the spatial resolution of the original high-quality CT cardiac image and has higher temporal resolution with no need for increased scan time and reduced X-ray dose to the patient.

Figure 5 shows images at various stages during the method. As can be seen, the CT cardiac images in the first row are high-quality and acquired at full or 100% photon exposure levels. The CT cardiac images in the second row are lesser-quality simulating an acquisition at 1 % photon exposure levels. The CT cardiac images in the last row are deformed high- quality images reconstructed using the corresponding high-quality and lesser- quality images.

Although a particular method of dynamic (4D) CT image generation is described above, alternatives are available. For example, turning now to Figure 6, another embodiment of the steps performed to generate high-quality 3D images are shown. In this embodiment, initially, the variable n is set to zero (0) and a user selects one or more particular time points t in the cardiac cycle to be imaged (step 402). The selected points in time do not need to coincide with acquired CT cardiac images. For each selected time point t, all of the transforms in the set are fitted using linear or cardinal spline interpolants over time (step 404) and a new transform T_t at the user selected point of time t is interpolated (step 406). The transform T_t is then applied to the high-quality source image (step 408) and the resultant deformed high-quality CT cardiac image at time t is saved (step 410). As will be appreciated, the above process allows the vector fields corresponding to the position and shape of the CT cardiac image at the specified time points to be interpolated from the originally acquired vector field set and images of the heart are obtained by non-linearly warping the high-quality CT cardiac image based on the interpolated vector fields.

The system and method for generating dynamic (4D) CT images provide a number of benefits. These benefits include for example, the ability to generate high-quality dynamic CT images with minimal increase in dose to patient when compared to standard methods, the ability to increase temporal resolution without increase in dose/scan time, thus allowing visualization of high-quality 3D representations of the heart anatomy at arbitrary time-points other than those originally acquired, the ability to create high-quality 3D images of the heart anatomy at arbitrary time points in the cardiac cycle, different from those times originally acquired, and the capability to use the extracted motion information directly to deform geometric models (points, lines, surfaces, volumes) representing the patient's heart anatomy over the cardiac cycle for added visualization, diagnosing heart conditions, or other application.

While the above methods to generate dynamic 4D CT images have been described in the context of CT cardiac imaging, it will be apparent to those of skill in the art that the methods have equal application to the imaging of other organs and tissues. Thus, for example the approach could be easily modified for generation of dynamic 4D CT images of the liver, or other organs affected by respiratory or cardiac motion.

Although embodiments have been described with reference to the accompanying drawings, those of skill in the art will appreciate that variations and modifications may be made without departing from the spirit and scope thereof as defined by the appended claims.

Claims

What is claimed is:

1. A method of generating a dynamic CT image of a target within a subject that experiences motion comprising: acquiring CT radiograph data of said target during substantially its entire phase at varying exposure levels and reconstructing the CT radiograph data to yield at least one high-quality image and a set of lesser- quality images of said target; registering the high-quality and lesser-quality images to generate a set of image transforms; and generating at least one deformed high-quality image based on the high-quality image and the set of image transforms.

2. The method of claim 1 wherein said CT radiograph data is reconstructed to yield a single high-quality image and a set of lesser-quality images.

3. The method of claim 2 wherein said single high-quality image is reconstructed from CT radiograph data acquired when said target is substantially at rest and wherein said set of lesser-quality images is reconstructed from CT radiograph data acquired when said target is in motion.

4. The method of claim 3 wherein said high-quality image is acquired during cardiac mid-diastole.

5. The method of claim 4 wherein during cardiac mid-diastole, CT radiograph data is acquired at a substantially full photon exposure levels.

6. The method of claim 5 wherein outside of cardiac mid-diastole, CT radiograph data is acquired in a manner to minimize photon exposure levels.

7. The method of claim 6 wherein outside of cardiac mid-diastole

CT radiograph data is acquired by at least one of reducing the number of acquired projections and reducing X-ray beam power.

8. The method of claim 7 wherein outside of cardiac mid-diastole,

CT radiograph data is acquired by reducing the number of acquired projections and by reducing X-ray beam power.

9. The method of claim 7 wherein X-ray beam power is reduced by placing a filter between a source of X-ray radiation and the subject.

10. The method of claim 7 wherein X-ray beam power is reduced by modulating current flow to a source of X-ray radiation.

1 1. The method of claim 3 wherein during registration, said high- quality image is registered to each lesser-quality image resulting in a series of vector fields representing movement/warping of said target over its phase.

12. The method of claim 11 wherein during registration, N-1 transforms are generated where N is the total number of images.

13. The method of claim 12 wherein during generating, N-1 deformed high-quality images are generated, said N-1 deformed high-quality images and said acquired high-quality image yielding a dynamic (4D) image of said target.

14. The method of claim 13 wherein said N-1 deformed high-quality images are generally evenly spaced in time.

15. The method of claim 12 wherein during generating, a deformed high-quality image at each user specified point in time is generated.

16. A system for generating a dynamic CT image of a target within a subject that experiences motion comprising: a CT scanner acquiring CT radiograph data of said target at varying exposure levels and reconstructing the CT radiograph data to yield at least one high-quality image and a set of lesser-quality images of said target; and processing structure registering the high-quality and lesser- quality images to generate a set of image transforms and generating at least one deformed high-quality image based on the high-quality image and the set of image transforms.

17. The system of claim 16 wherein said CT scanner exposes the patient to substantially full exposure levels when said target is substantially at rest to yield said at least one high-quality image and reduces patient exposure when said target is in motion to yield said set of lesser-quality images.

18. The system of claim 17 wherein said CT scanner reduces patient exposure by at least one of reducing the number of acquired projections and reducing X-ray beam power.

19. The system of claim 18 wherein said CT scanner reduces patient exposure by reducing the number of acquired projections and by reducing X-ray beam power.

20. The system of claim 17 further comprising an ECG machine to monitor the patient's heart, said CT scanner exposing the patient to substantially full exposure levels during cardiac mid-diastole.

21. The system of claim 17 wherein said processing structure generates N-1 transforms during registration, where N is the total number of images.

22. The system of claim 21 wherein said processing structure generates N-1 deformed high-quality images, said N-1 deformed high-quality images and said acquired high-quality image yielding a dynamic (4D) image of said target.

23. The system of claim 21 wherein said processing structure generates a deformed high-quality image at each user specified point in time.

24. The system of claim 20 wherein CT radiograph data is acquired using a prospective ECG gating scheme.

25. The system of claim 17 wherein said processing structure registers the high-quality image to each lesser-quality image resulting in a series of vector fields representing movement/warping of said target over its phase.

26. A method of generating a dynamic CT image of an organ comprising: imaging a subject's organ at differing exposure levels to yield a high-quality image of the organ when at rest and a set of lesser-quality images of the organ when in motion; registering the high-quality and lesser-quality images to generate a motion description of the organ; and generating at least one warped high-quality image based on the motion description and high-quality image.

27. The method of claim 26 wherein during said generating, a set of warped high-quality images is generated.

28. The method of claim 27 wherein the warped high-quality images are evenly spaced in time.

29. The method of claim 27 wherein the warped high-quality images are at user specified points in time.