FIELD OF THE INVENTION

The following generally relates to cardiac imaging and more particularly to machine learning based cardiac rest phase determination and is described with particular application to computed tomography (CT).

BACKGROUND OF THE INVENTION

Computed tomography (CT) has provided useful information about the interior characteristics of an object or subject under examination. For example, cardiac CT, which includes Coronary CT Angiography (CTA), can be used to assess the coronary arteries. For a CTA examination, a subject is first intravenously administered a radiocontrast agent such as an iodine, barium, gadolinium, etc. based contrast agent. The heart is then scanned while the radiocontrast agent is present in the coronary arteries. The radiocontrast agent absorbs X-rays, which results in brighter (visually enhanced) pixels/voxels in the coronary arteries in the displayed reconstructed volumetric image data, relative to volumetric image reconstructed with CT data acquired without the radiocontrast agent present in the coronary arteries.

The heart, however, is a moving object, beating (i.e. contracting and relaxing), e.g., 60-100 beats per minute, on average, for an adult at rest, and the motion manifests in the displayed reconstructed volumetric image data as artifact (i.e. blur, such as blurred arteries, vessels and/or other tissue). An approach to reduce this artifact and improve image quality of the coronary arteries is to reconstruct the volumetric image data during a sub-time period / phase of the cardiac cycle during which motion is smallest as the motion of the heart is not the same during each phase of the cardiac cycle. This phase has been referred to as a“quiet” phase.

Unfortunately, the location of the“quiet” phase varies from subject to subject, heart rate to heart rate, and/or segment to segment of a coronary artery, which makes it difficult if not impossible to predict the“quiet” phase.

As a consequence, for cardiac CT, volumetric image data has been reconstructed at multiple different phases, and the multiple sets of reconstructed volumetric image data, each corresponding to a different phase, has been evaluated, e.g., either by visual inspection or based on a quality measure, to identify the volumetric image data of the multiple sets with the least motion artifact for the coronary arteries. Unfortunately, initially reconstructing multiple sets of volumetric image data, each corresponding to a different phase, for each subject increases overall processing time and consumption of processing resources, such as processing cycles and memory, as well as examination time and/or clinician time.

SUMMARY OF THE INVENTION

Aspects described herein address the above-referenced problems and others.

The following describes an approach to determine, using machine learning, a cardiac CT reconstruction phase with a least amount of motion artifact (a“quiet” phase), relative to other phases of the cardiac cycle, from cardiac CT volumetric image data reconstructed only once and at a pre-determined initial reconstruction phase.

In one aspect, a system includes a processor and a memory storage device. The memory storage device includes at least a reconstruction phase determiner. The processor is configured to execute the reconstruction phase determiner to determine, with a trained statistical model, a cardiac cycle reconstruction phase with a least amount of motion artifact based on volumetric image data reconstructed with an initial predetermined reconstruction phase. A reconstructor is configured to reconstruct the volumetric image data with the initial

predetermined reconstruction phase and reconstruct volumetric image data with the cardiac cycle reconstruction phase with the least amount of motion artifact. A display is configured to visually present at least the volumetric image data reconstructed with the cardiac cycle reconstruction phase with the least amount of motion artifact.

In another aspect, a method includes reconstructing, with a reconstructor, volumetric image data with an initial predetermined reconstruction phase, determining, with a processing and a trained statistical model, a cardiac cycle reconstruction phase with a least amount of motion artifact based on volumetric image data reconstructed with an initial predetermined reconstruction phase, reconstructing, with the reconstructor, volumetric image data with the cardiac cycle reconstruction phase with a least amount of motion artifact, and displaying the volumetric image data reconstructed with the cardiac cycle reconstruction phase with the least amount of motion artifact.

In another aspect, a computer-readable storage medium stores instructions that when executed by a processor of a computer cause the processor to reconstruct volumetric image data with an initial predetermined reconstruction phase, determine, with a trained statistical model, a cardiac cycle reconstruction phase with a least amount of motion artifact based on volumetric image data reconstructed with an initial predetermined reconstruction phase, reconstruct volumetric image data with the cardiac cycle reconstruction phase with a least amount of motion artifact, display the volumetric image data reconstructed with the cardiac cycle reconstruction phase with the least amount of motion artifact. Those skilled in the art will recognize still other aspects of the present application upon reading and understanding the attached description.

BRIEF DESCRIPTION OF THE DRAWINGS

The invention may take form in various components and arrangements of components, and in various steps and arrangements of steps. The drawings are only for purposes of illustrating the embodiments and are not to be construed as limiting the invention.

FIG. 1 diagrammatically illustrates an imaging system with a reconstruction phase determiner, in accordance with an embodiment s) herein.

FIG. 2 depicts an example electrocardiogram.

FIG. 3 diagrammatically illustrates an example of the reconstruction phase determiner, in accordance with an embodiment s) herein.

FIG. 4 depicts an example of a display of volumetric image data, in accordance with an embodiment(s) herein.

FIG. 5 depicts another example of a display of the volumetric image data, in accordance with an embodiment s) herein.

FIG. 6 illustrates an example method, in accordance with an embodiment s) herein.

DETAILED DESCRIPTION OF EMBODIMENTS FIG. 1 diagrammatically illustrates an imaging system such as a computed tomography (CT) scanner 102. The scanner 102 includes a stationary gantry 104 and a rotating gantry 106, which is rotatably supported by the stationary gantry 104 and rotates around an examination region 108 about a longitudinal or z-axis (“Z”).

A subject support 110, such as a couch, supports a subject or object in the examination region 108. The subject support 110 is movable in coordination with performing an imaging procedure so as to guide the subject or object with respect to the examination region 108 for loading, scanning, and/or unloading the subject or object.

A radiation source 112, such as an X-ray tube, is supported by and rotates with the rotating gantry 106 around the examination region 108. The radiation source 112 emits X- ray radiation that is collimated e.g., by a source collimator (not visible) to form a generally fan, wedge, cone or other shaped X-ray radiation beam that traverses the examination region 108.

A radiation sensitive detector array 114 subtends an angular arc opposite the radiation source 112 across the examination region 108. The detector array 114 includes one or more rows of detectors that are arranged with respect to each other along the z-axis direction and detects radiation traversing the examination region 108. The radiation sensitive detector array 114 produces projection data (line integrals).

A reconstructor 116 reconstructs the projection data to generate volumetric image data. As described in greater detail below, for cardiac CT, the reconstructor 116 initially reconstructs the projection data only once and at a pre-determined reconstruction phase of the cardiac cycle and then subsequently at a reconstruction phase(s) of the cardiac cycle predicated to have a least amount of motion artifact (a“quiet” phase), which is predicted based on machine learning and the volumetric image data reconstructed with the initial pre-determined

reconstruction phase.

An operator console 118 includes input/output (I/O) interfaces 120 for I/O devices, including a human readable output device 122 such as a display monitor, a filmer, etc., and an input device 124 such as a keyboard, mouse, etc., as well as peripheral devices, such as an electrocardiograph (ECG) 126 and/or other device that provides signals indicative of the electrical activity of the heart.

The operator console 118 further includes a processor 128 (e.g., a central processing unit (CPU), a microprocessor, etc.) and computer readable storage medium

(“memory”) 130 (which excludes transitory medium) such as physical memory like a memory storage device, etc. The computer readable storage medium 130 includes computer readable instructions 132, including instructions for scanner control 134, a reconstruction phase determiner 136, and a rendering engine 138.

The scanner control 134 controls, e.g., data acquisition. In one instance, for cardiac CT, the scanner control 134 analyzes the ECG signal from the ECG 126 and initiates data acquisition only during a pre-determined set of one or more reconstruction phases determined from the ECG signal. This is referred to as a gated acquisition. An example is discussed in connection with FIG. 2, which shows an example ECG waveform 202. In this example, the scanner control 134 detects the beginning of an R-R interval 204 by sensing an R wave 206 and triggers data acquisition only for a pre-determined sampling window around a pre determined percent of the R-R interval.

In general, two cardiac phases associated with lower cardiac motion (“quiet” phases) occur around forty percent (40%) 208 and seventy-five percent (75%) 210 of the R-R interval 204, depending on heart rate. In this example, at 40%, ventricular systole 212 has ended and the ventricles cease contracting and ventricular diastole 214 begins and the ventricles relax and begin filling ( after the T wave), and at 75%, the R-R interval 204 is still in ventricular diastole but just before atrial contraction (P wave). The set of angular views acquired for reconstruction for any phase (e.g., 45%, 75%, etc.) is at least 180 degrees plus a source fan.

As previously discussed, the actual“quiet” phase varies from subject to subject, heart rate to heart rate, and/or segment to segment of a coronary artery and/or the heart. As such, the scanner control 134 controls data acquisition for either the 40% reconstruction phase or the 75% reconstruction phase and triggers data acquisition during an acquisition sampling window (e.g., 10, 12, 15%, 18%, etc.) around the 40% or 75% reconstruction phase. For example, in one instance, data acquisition begins at 30% and ends at 50% (for the 40% phase) or begins at 60% and ends at 90% (for the 75% phase) of each R-R interval. In another embodiment, the scanner control 134 controls data acquisition to acquire data for the entire cardiac cycle.

Returning to FIG. 1, the reconstruction phase determiner 136 is configured to select the initial reconstruction phase and determine the reconstruction phase(s) of interest based on the volumetric image data reconstructed with the initial reconstruction phase. As described in greater detail below, in one instance, the reconstruction phase determiner 136 employs machine learning to determine the reconstruction phase(s) of interest from volumetric image data reconstructed at the initial reconstruction phase. With this approach, only a single initial reconstruction is performed at the initial reconstruction phase, and the reconstructor 116 then reconstructs the projection data at the reconstruction phase(s) of interest. In one instance, this reduces overall processing time and consumption of processing resources, such as processing cycles and memory, as well as examination time and/or clinician time, relative to a configuration in which the reconstruction phase determiner 136 is not utilized or omitted.

The rendering engine 138 displays the volumetric image data. As described in greater detail below, in one instance the rendering engine 138 constructs a user-interactive graphical user interface with soft controls and image display regions, displays the user- interactive graphical user interface via a display monitor of the output device(s) 120, and displays the volumetric image data in the image display regions. The soft controls allow a user to control the volumetric image data displayed.

In a variation, the reconstruction phase determiner 136 and/or the rendering engine 138 are executed by a processor in a different computing system, such as a dedicated workstation located remotely from the scanner 102,“cloud” based resources, etc. The different computing system can receive volumetric image data from the scanner 102, another scanner, a data repository (e.g., a radiology information system (RIS), a picture archiving and

communication system (PACS), a hospital information system (HIS), etc.), etc. The different computing system additionally or alternatively can receive projection data from the scanner 102, another scanner, a data repository, etc. In this instance, the different computing system may include a reconstructor configured similar to the reconstructor 116 in that it can process the projection data and generate the spectral and/or non-spectral volumetric image data.

FIG. 3 diagrammatically illustrates an example of the reconstruction phase determiner 136. The illustrated reconstruction phase determiner 136 includes a reconstruction phase point selector / adjuster 302, an initial reconstruction phase algorithm 304 and a reconstruction phase delta determiner 306. The reconstruction phase point selector / adjuster 302 selects the initial reconstruction phase for the initial reconstruction and subsequently adjusts the initial reconstruction phase, if needed, to reconstruct volumetric image data at a“quiet” reconstruction phase of the subject under examination.

For the initial reconstruction, the reconstruction phase point selector / adjuster 302 selects a reconstruction phase based on an initial reconstruction phase algorithm 304. In one instance, the initial reconstruction phase algorithm 304 selects a pre-determined default reconstruction phase. For example, in one instance the initial reconstruction phase algorithm 304 includes a rule that indicates that the initial reconstruction phase is 40% of the R-R interval for heart rates that exceed a given threshold (e.g., 70 bpm) and 75% otherwise. In other examples, other percentages, thresholds, etc. are contemplated.

For the subsequent reconstruction, the reconstruction phase point selector / adjuster 302 adds/subtracts a reconstruction phase delta to the initial reconstruction phase. The reconstruction phase delta is determined by the reconstruction phase delta determiner 306 and indicates, e.g., in terms of % of the R-R interval, a distance between the initial reconstruction phase and the reconstruction phase(s) with the least amount of motion / the“quiet” phase(s).

The reconstructor 116 then reconstructs volumetric image data at the reconstruction phase(s) with the least amount of motion / the“quiet” phase(s).

For this, the reconstruction phase delta determiner 306 receives, as input, at least the volumetric image data reconstructed with the initial reconstruction phase. In the illustrated example, the reconstruction phase delta determiner 306 also receives, as input, data indicating segmentation tissue of interest. For example, in one instance, the segmentation tissue of interest is one or more coronary arteries and/or a coronary tree. In another, the segmentation tissue of interest is alternatively or additionally a heart chamber(s), valve(s), and/or tissue. In another example, the segmentation tissue of interest is instead a pre-determined default.

The reconstruction phase delta determiner 306 includes a tissue of interest segmentor 308. The tissue of interest segmentor 308 employs a tissue of interest segmentation algorithm from a tissue of interest segmentation algorithm bank 310. An examples of suitable algorithm includes, but is not limited to, a model-based segmentation such as the segmentation described in US 8,160,332 B2, filed September 28, 2007, and entitled“Model-Based Coronary Centerline Localization,” which is incorporated by reference herein in its entirety. The tissue of interest segmentor 308 segments the tissue(s) of interest from the initial reconstruction and outputs the segmentation, including a centerline, which represents a central axis of the vessel.

The reconstruction phase delta determiner 306 further includes a segmentation divider 312. The segmentation divider 312 employs a segmentation dividing algorithm from a segmentation dividing algorithm bank 314. In one instance, e.g., for the coronary arteries, the segmentation dividing algorithm divides the coronary artery into multiple segments. For example, in one instance the segmentation dividing algorithm divides the segmentation into pre determined equal or different length segments (e.g., each three, seven, ten, twelve or other millimeter (3, 7, 10, 12, etc. mm). In another example, the segmentation dividing algorithm 314 divides the segmentation based on an American College of Cardiology (ACC), American Heart Association (AHA), Society of Cardiovascular Computed Tomography (SCCT) and/or other classification. Examples are discussed in Austen et al.,“A Reporting System on Patients Evaluated for Coronary Artery Disease,” AHA GRADING COMMITTEE 7-40, and Leipsic et al,“SCCT guidelines for the interpretation and reporting of coronary CT angiography: A report of the Society of Cardiovascular Computed Tomography Guidelines Committee,” Journal of Cardiovascular Computed Tomography 8(2014) 342- 358.

The reconstruction phase delta determiner 306 further includes a feature extractor 316. The feature extractor 316 receives, as input, the segments and extracts features from each of the segments based on a feature(s) of interest 318. In one instance, the feature(s) of interest 318 includes artifact such as motion artifact. In another instance, e.g., where the reconstructed volumetric image data includes spectral (energy dependent) volumetric image data, the feature(s) of interest 318 additionally or alternatively includes spectral information such as contrast concentration.

The reconstruction phase delta determiner 306 further includes a phase delta determiner 320. The phase delta determiner 320 receives, as input, the extracted features along with the segments and employs machine learning to determine the phase delta. In this example, the phase delta determiner 320 employs a learned statistical model 322. The learned statistical model 322 maps the extracted feature(s) for a segment to a phase delta from the initial reconstruction phase to a“quiet” phase. For representation learning and/or deep learning, the model 322 itself extracts feature(s) from the volumetric input patches. The phase delta determiner 320 outputs the phase delta for each segment. As discussed herein, the reconstruction phase point selector / adjuster 302 adds/subtracts a reconstruction phase delta to the initial reconstruction phase for each segment. The reconstructor 116 then reconstructs the projection data at each of the updated reconstruction phases. As such, the set of volumetric image data will include image data for each segment reconstructed at the“quiet” phase. The set of volumetric image data can be variously displayed, e.g., as described next.

FIGS. 4 and 5 show examples of the display of the volumetric image data, which includes a curved multiplanar reconstruction (CMPR) of a coronary artery. In both examples, the rendering engine 138 displays a slice of the volumetric image data as a two-dimensional (2- D) image 400 corresponding to a particular vessel 401 and including a center line 403. In FIG.

4, soft controls 402 include a drop-down menu 404 with a list of coronary arteries 406, ... , 408, where“clicking” (e.g., with a computer mouse) on a particular artery causes the processor 128 to a slice with the selected artery. In FIG. 5, soft controls 502 includes a 3-D rendering 504 that can be rotated, panned, magnified, scrolled through using a cut plane, etc., where the user uses these tools to visually locate a vessel and then, likewise,“clicks” on the tissue of interest.

In one instance, the rendering engine 138 is configured to employ a multiplanar reconstruction (MPR) algorithm(s) to extract the slice the includes the coronary artery. In general, the MPR algorithm(s) creates a 2-D slice from the volumetric image data in a plane orthogonal to the acquisition plane, a plane non-orthogonal (oblique) to the acquisition plane, and/or a curved plane (CMPR). The slice has a given thickness, which could be the same or different from the acquisition thickness. Curved MPR is well-suited for vascular imaging since it straightens curved vessels, which allows the entire length of a curved vessels to visualized in the same or a few images, even though the entire vessel is not, anatomically, in a same plane.

In the example of FIGS. 4 and 5, the entire slice (i.e. each segment) is reconstructed with only one reconstruction phase. In one instance, this corresponds to the adjusted reconstruction phase for a particular segment (e.g., a first, a last, a middle, etc.) of the displayed vessel. In another instance, this corresponds to the initial reconstruction phase where the adjusted reconstruction phase for at least one of the segments is the same phase as the initial reconstruction phase. In another instance, this corresponds to a pre-determined default, a user specified, a computer determined, and/or other reconstruction phase.

In the example of FIGS. 4 and 5, the entire displayed slice is reconstructed at an adjusted reconstruction phase of 75%. However, the adjusted reconstruction phase for at least one of the segments is 78%, which means the displayed reconstruction for this segment does not necessarily include the least amount of motion artifact for that segment. In this example, the rendering engine 138 visually highlights this segment by outlining a region 410 containing the segment (e.g., an ellipse with a border and/or shading, as shown) and provides a textual indication that the adjusted reconstruction phase for the segment (as shown) and/or the delta to the adjusted reconstruction phase for the segment.

In FIGS. 4 and 5, the textual indication is a soft tool 412. As such, the user “clicks” on or“hovers” over the soft tool 412 to change the displayed slice to a slice that is reconstructed at an adjusted reconstruction phase of 78%. In this instance, the textual indication is instead used to indicate that the adjusted reconstruction phase of the other parts of the vessel is 75%. The user likewise can“click” on or“hover” over the soft tool 412 to revert back to the display of the 75% reconstruction. The illustrated example includes two different adjusted reconstruction phase, but it is to be understood that each segment in the displayed slice could have a different adjusted reconstruction phase.

Returning to FIG. 3, to train the statistical model 322, projection data is obtained for a population of subjects suspected of coronary artery disease. In one instance, this population does not include any subjects with acute myocardial infarction or healthy hearts. Each set of projection data for each subject in the population is reconstructed at one or more reconstruction phases. For example, in one instance, a set of projection data for subject is reconstructed at 40% or 75%. In another example, a set of projection data for subject is reconstructed at adjacent reconstruction phases about the 40% phase (e.g., 25% to 55%, including 40%) and/or about 75% phase (e.g., 60% to 90%, including 75%), in predetermined increments (e.g., 1%, 2%, 5%, etc.).

The tissue of interest is segmented from the reconstructions and divided into segments, e.g., as described herein. In one instance, one or more cardiac experts evaluates (e.g., using a five-point Likert scale) each segment of each of the reconstruction phases for each of the subjects to determine which reconstruction phase for each segment for each subject contains a least amount of motion artifact. The one or more cardiac experts label / annotate each reconstruction to indicate how far way, in terms of % of the R-R interval, each reconstruction is from this“quiet” phase. For example, if the expert(s) determines the 63% phase reconstruction has the least motion artifact and the initial reconstruction is for the 75% phase reconstruction, the reconstruction at 63% is annotated as being -12% (i.e. distance and direction) away from the initial reconstruction phase.

For training, in one instance, each segment is sub-divided into centerline points that represent the coronary artery centerline. For each centerline point of each segment, a patch of voxels of the volumetric image data (e.g., perpendicular or otherwise) to the centerline is identified. For a segment, the patches are provided as input to a regression model and the annotations are provided as target values to the regression mode. The statistical model 322 processes the patches and the annotations for each segment to learn to predict the distance (i.e. the phase delta) between the reconstruction with the least motion artifact and the initial reconstruction phase based on features of the volumetric image data. The learned / trained statistical model 322 is stored in the memory 130 (FIG. 1).

In another example, a combination of expert labeling and automatic labeling (e.g. using a set of motion artifact measures) is employed. Examples of suitable automatic labeling include, but are not limited to, Rohkohl et al.,“Improving best-phase image quality in cardiac CT by motion correction with MAM optimization,” Medical Physics 40.3 (2013), and Lossau et al,“Motion artifact recognition and quantification in coronary CT angiography using convolutional neural networks,” Medical Image Analysis 52 (2019): 68-79. During training, the expert labelling (or the automatic labeling) can be weighted higher than the automatic labeling (expert labeling), or the expert and automatic labeling can be equally weighted.

In one instance, the statistical model 322 includes a convolutional neural network (CNN). Examples of algorithms are discussed in Gouk, et al.,“Fast Sliding Window

Classification with Convolutional Neural Networks,” IVNVZ‘14 Proceedings of the 29th International Conference on Image and Vision Computing New Zealand, Pages 114-118, November 19-21, 2014,“Fully convolutional networks for semantic segmentation,” Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition, 2015, and Ronneberger, et al,“U-Net: Convolution Networks for Biomedical Image Segmentation,” Medical Image Computing and Computer-Assisted Intervention (MICCAI), Springer, LNCS, Vol. 9351 : 234— 241, 2015. Other statistical models are contemplated herein.

FIG. 5 illustrates an example method in accordance with an embodiment(s) herein.

It is to be appreciated that the ordering of the acts in the method is not limiting. As such, other orderings are contemplated herein. In addition, one or more acts may be omitted, and/or one or more additional acts may be included.

At 502, the learned statistical model 322 is trained as described herein and/or otherwise.

At 504, a cardiac scan is performed, as described herein and/or otherwise.

At 506, projection data from the scan is reconstructed at a pre-determined initial reconstruction phase, as described herein and/or otherwise. At 508, the learned statistical model 322 predicts a reconstruction phase with a least amount of motion artifact based on the volumetric image data reconstructed at the pre determined initial reconstruction phase, as described herein and/or otherwise.

At 510, the projection data is reconstructed with the predicted reconstruction phase, as described herein and/or otherwise.

At 512, the volumetric image data is displayed, as described herein and/or otherwise.

The above may be implemented by way of computer readable instructions, encoded or embedded on computer readable storage medium, which, when executed by a computer processor(s), cause the processor(s) to carry out the described acts. Additionally, or alternatively, at least one of the computer readable instructions is carried out by a signal, carrier wave or other transitory medium, which is not computer readable storage medium.

While the invention has been illustrated and described in detail in the drawings and foregoing description, such illustration and description are to be considered illustrative or exemplary and not restrictive; the invention is not limited to the disclosed embodiments. Other variations to the disclosed embodiments can be understood and effected by those skilled in the art in practicing the claimed invention, from a study of the drawings, the disclosure, and the appended claims.

The word“comprising” does not exclude other elements or steps, and the indefinite article“a” or“an” does not exclude a plurality. A single processor or other unit may fulfill the functions of several items recited in the claims. The mere fact that certain measures are recited in mutually different dependent claims does not indicate that a combination of these measured cannot be used to advantage.

A computer program may be stored/distributed on a suitable medium, such as an optical storage medium or a solid-state medium supplied together with or as part of other hardware, but may also be distributed in other forms, such as via the Internet or other wired or wireless telecommunication systems. Any reference signs in the claims should not be construed as limiting the scope.