Computerized characterization of cardiac motion in medical diagnostic ultrasound

• US 10,321,892 B2
• Filed: 09/16/2011
• Issued: 06/18/2019
• Est. Priority Date: 09/27/2010
• Status: Active Grant

• Current Assignee: Siemens Medical Solutions USA Incorporated (Siemens AG)
• Original Assignee: Siemens Medical Solutions USA Incorporated (Siemens AG)
• Inventors: Wang, Yang, Georgescu, Bogdan, Houle, Helene C., Comaniciu, Dorin
• Primary Examiner(s): Kish, James M
• Application Number:
  US13/234,697
• Publication Number:
  US 20120078097A1
• Time in Patent Office: 2,832 Days
• Field of Search: 600437
• CPC Class Codes: A61B 8/00 Diagnosis using ultrasonic, sonic or infrasonic waves imaging of objects using sonar G01S15/00
• CPC Class Codes: A61B 8/0883 for diagnosis of the heart
• CPC Class Codes: A61B 8/483 involving the acquisition of a 3D volume of data
• CPC Class Codes: A61B 8/5223 for extracting a diagnostic or physiological parameter from medical diagnostic data for algorithms to analyse biomedical images G06T7/0012
• CPC Class Codes: G06T 2207/10136 3D ultrasound image
• CPC Class Codes: G06T 2207/30048 Heart; Cardiac
• CPC Class Codes: G06T 7/251 involving models

Abstract

Computerized characterization of cardiac wall motion is provided. Quantities for cardiac wall motion are determined from a four-dimensional (i.e., 3D+time) sequence of ultrasound data. A processor automatically processes the volume data to locate the cardiac wall through the sequence and calculate the quantity from the cardiac wall position or motion. Various machine learning is used for locating and tracking the cardiac wall, such as using a motion prior learned from training data for initially locating the cardiac wall and the motion prior, speckle tracking, boundary detection, and mass conservation cues for tracking with another machine learned classifier. Where the sequence extends over multiple cycles, the cycles are automatically divided for independent tracking of the cardiac wall. The cardiac wall from one cycle may be used to propagate to another cycle for initializing the tracking. Independent tracking in each cycle may reduce or avoid inaccuracies due to drift.

Claims

1. A method for computerized characterization of cardiac motion from ultrasound data in an automatic manner, the method comprising:

acquiring a sequence of frames of ultrasound data, the frames comprising three-dimensionally distributed voxels as data representing a three-dimensional volume of a patient at different times;

identifying, by a processor, a myocardium in a first frame;

estimating, by the processor, non-rigid deformation of a spatial location of the myocardium over time through the sequence, the estimating of the non-rigid deformation of the spatial location of the myocardium being a function of a machine-learned model of speckle patterns, or intensity and speckle patterns specific to the left ventricle myocardium using discriminative classifiers, the machine-learned model trained by the machine processing annotated training data of a plurality of samples from other patients using machine-learning to select learned features and the machine learned model comprising a motion prior learned from other examples of typical myocardium positions for different phases, the motion prior used in a Bayesian objective function with integration of a plurality of cues, including myocardium boundary detection, speckle tracking and mass conservation, and the estimating being a function of volumetric tracking in three dimensions of the myocardium from the first frame to a second frame of the sequence of the frames;

calculating, by the processor, a first myocardial mechanical quantity as a function of the non-rigid deformation, the first myocardial mechanical quantity computed in both a Cartesian coordinate system and a local heart coordinate system; and

outputting a display of the first myocardial mechanical quantity.

2. The method of claim 1 further comprising:

determining an end-diastolic or end-systolic frame as a function of a volume within the myocardium, wherein the volume is based on the motion prior.

3. The method of claim 1 wherein obtaining comprises scanning the patient and wherein identifying comprises identifying in real time with the scanning.

4. The method of claim 1 wherein identifying comprises locating with the motion prior.

5. The method of claim 4 wherein performing the identifying and estimating comprises reducing dimensionality by maintaining relative location information and avoiding use of absolute coordinates.

6. The method of claim 1 wherein estimating comprises volumetric tracking between pairs of the frames in a forward direction and volumetric tracking between the pairs of frames in a reverse direction, the forward and reverse directions being temporal directions through the sequence, and comprises combining the trackings from the forward and reverse directions for the same pairs.

7. The method of claim 1 further comprising dividing the sequence into first and second heart cycles based on a volume within the myocardium and then performing the identifying of the myocardium in one frame for each of the first and second heart cycles, the first frame being the one frame for the first heart cycle, wherein identifying for the first heart cycle comprises identifying with the motion prior, wherein identifying for the second heart cycle comprises identifying with propagation of myocardium locations from the first frame of the first heart cycle to a second frame of the second heart cycle, and wherein estimating comprises volumetric tracking the myocardium through the first and second heart cycles independently.

8. The method of claim 1 wherein calculating comprises calculating volume, displacement, velocity, twist, torsion, strain, strain rate, principal strain, radius, curvature, contraction front map, relaxation front map, coronary map, or combinations thereof from the estimate of the non-rigid deformation of the myocardium, the calculating occurring in both the Cartesian coordinate system and the local heart coordinate system without user input for tracing or tracking.

9. The method of claim 1 further comprising detecting multi-planar reconstruction planes with an additional machine-trained classifier.

10. A non-transitory computer readable storage media having stored therein data representing instructions executable by a programmed processor for characterizing cardiac motion from ultrasound information, the storage media comprising instructions for:

acquiring a sequence of frames of ultrasound data, the frames comprising three-dimensionally distributed voxels as data representing a three-dimensional volume of a patient at different times;

identifying first and second cycles of the heart from the sequence of the frames of the ultrasound data;

propagating a first location of a myocardium from a first frame of the frames of the first cycle to a second location of the myocardium in a second frame of the frames of the second cycle, the second location being an initial location of the myocardium used in volumetric tracking of the myocardium through the frames of the second cycle, the volumetric tracking through the frames of the second cycle using the initial location and other processes so that the volumetric tracking is independent between the first cycle and the second cycle other than the initial location, the propagation comprising temporally interpolating the second location of the second cycle from the first location of the first cycle;

displaying a quantity based on the second location;

tracking as a function of a speckle tracking, boundary detection, mass conservation, and a motion prior;

combining a detection score from a machine learning model with a plurality of scores, including a speckle tracking score, a mass conservation score, and a motion prediction score; and

determining a deformation of the myocardium over time as a function of the combining.

11. The non-transitory computer readable storage media of claim 10 wherein the tracking comprise tracking with first machine-learned knowledge specific to a left ventricle myocardium with the speckle tracking, boundary detection, mass conservation, and motion prior as input features, the motion prior comprises second machine-learned knowledge.

12. The non-transitory computer readable storage media of claim 10 wherein propagating comprises performing the tracking from the first frame to the second frame without the motion prior.

13. The non-transitory computer readable storage media of claim 10 further comprising locating the myocardium through the sequence with the machine-learned model, the machine-learned model comprising a machine-learned detector, and wherein identifying comprises identifying as a function of variation in a volume of the heart over time, the volume calculated from the myocardium located through the sequence.

14. A non-transitory computer readable storage media having stored therein data representing instructions executable by a programmed processor for characterizing cardiac motion from ultrasound information in an automatic manner, the storage media comprising instructions for:

acquiring a sequence of frames of ultrasound data, the frames comprising three-dimensionally distributed voxels as data representing a three-dimensional volume of a patient at different times;

locating a heart wall from a first frame of ultrasound data of three-dimensional voxels in a sequence representing the heart wall in three dimensions, the locating of the heart wall being a function of a first machine-learned knowledge of the heart wall locations in training data, the first machine-learned knowledge gained by a machine processing an annotated version of the training data of a plurality of samples from other patients using machine-learning;

tracking the heart wall automatically through the sequence, the tracking being a function of a second machine-learned knowledge of a plurality of cues as an input feature vector, the plurality of cues including the first machine-learned knowledge with speckle tracking, boundary detection, mass conservation, and a motion prior learned from other examples of typical myocardium positions for different phases, the first and second machine-learned knowledge being different models;

calculating a cardiac parameter, statistical index of the cardiac motion, or both the cardiac parameter and the statistical index of the cardiac motion as a function of the tracking; and

displaying the cardiac parameter.

15. The non-transitory computer readable storage media of claim 14 wherein locating comprises locating a position, orientation, and scale of a boundary of a left ventricle as a function of steerable features.

16. The non-transitory computer readable storage media of claim 14 wherein tracking comprises tracking with the second machine-learned knowledge comprising a Bayesian objective function.

17. The non-transitory computer readable storage media of claim 14 wherein the first machine-learned knowledge comprises 4D generalized procrustes analysis and manifold learned knowledge with K-means clustering.

18. The non-transitory computer readable storage media of claim 14 wherein the tracking is performed separately for different heart cycles in the sequence and wherein the locating is performed for one of the different heart cycles based on the heart wall propagated from another of the different heart cycles without the first machine-learned knowledge.