Method and System for Dynamic Pulmonary Trunk Modeling and Intervention Planning

• US 20100239147A1
• Filed: 03/16/2010
• Published: 09/23/2010
• Est. Priority Date: 03/18/2009
• Status: Active Grant

• Current Assignee: Siemens Healthcare GMBH (Siemens AG)
• Original Assignee: Siemens AG, Siemens Corp. (Siemens AG)
• Inventors: Georgescu, Bogdan, Comaniciu, Dorin, Ionasec, Razvan Ioan, Huber, Martin, Vitanovski, Dime
• Granted Patent: US 8,538,109 B2
• Time in Patent Office: Days
• US Class Current: 382/131
• CPC Class Codes: G06F 18/2323 based on graph theory, e.g. minimum spanning trees [MST] or graph cuts
• CPC Class Codes: G06T 2207/10076 4D tomography; Time-sequential 3D tomography
• CPC Class Codes: G06T 2207/20081 Training; Learning
• CPC Class Codes: G06T 2207/20108 Interactive selection of 2D slice in a 3D data set
• CPC Class Codes: G06T 2207/30048 Heart; Cardiac
• CPC Class Codes: G06T 2207/30101 Blood vessel; Artery; Vein; Vascular
• CPC Class Codes: G06T 7/0012 Biomedical image inspection
• CPC Class Codes: G06V 10/457 by analysing connectivity, e.g. edge linking, connected component analysis or slices
• CPC Class Codes: G06V 10/7635 based on graphs, e.g. graph cuts or spectral clustering
• CPC Class Codes: G06V 2201/031 of internal organs

Abstract

A method and system for modeling the pulmonary trunk in 4D image data, such as 4D CT data, and model-based percutaneous pulmonary valve implantation (PPVI) intervention is disclosed. A patient-specific dynamic pulmonary trunk data is generated from 4D image data of a patient. The patient is automatically classified as suitable for PPVI intervention or not suitable for PPVI intervention based on the generated patient-specific dynamic pulmonary trunk model.

Claims

1. A method for determining whether a patient is suitable for percutaneous pulmonary valve implantation (PPVI) intervention, comprising:

receiving 4D cardiac image data of a patient comprising a plurality of frames, each frame comprising volumetric image data;

generating a patient-specific dynamic pulmonary trunk model based on the 4D cardiac image data of the patient; and

automatically classifying the patient as suitable or unsuitable for PPVI intervention based on the generated patient-specific dynamic pulmonary trunk model.

2. The method of claim 1, wherein said step of automatically classifying the patient as one of suitable for PPVI intervention based on the generated patient-specific dynamic pulmonary trunk model comprises:

extracting shape features from the patient specific dynamic pulmonary trunk model; and

classifying the patient as suitable or unsuitable for PPVI intervention based on the extracted shape features using a trained discriminative distance function.

3. The method of claim 2, wherein the trained discriminative distance function clusters the patient-specific dynamic pulmonary trunk model with similar pulmonary trunk models in training data based on the extracted shape features, and said step of classifying the patient as suitable or unsuitable for PPVI intervention based on the extracted shape features using a trained discriminative distance function comprises:

if the patient-specific dynamic pulmonary trunk model is clustered with pulmonary trunk models that are suitable for PPVI intervention, classifying the patient as suitable for PPVI intervention; and

if the patient-specific dynamic pulmonary trunk model is clustered with pulmonary trunk models that are unsuitable for PPVI intervention, classifying the patient as unsuitable for PPVI intervention.

4. The method of claim 2, wherein the trained discriminative distance function is trained using one of a Random Forest in product space and an intrinsic Random Forest similarity of a Random Forest trained in an original space.

5. The method of claim 1, wherein said step of generating a patient-specific dynamic pulmonary trunk model based on the 4D cardiac image data of the patient comprises:

detecting bounding boxes in the frames of the 4D cardiac image data;

detecting anatomic landmarks in the frames of the 4D cardiac image data based on the detected bounding boxes;

detecting ribs of a pulmonary artery in the frames of the 4D cardiac image data based on the detected anatomic landmarks; and

generating the patient-specific dynamic pulmonary trunk model by fitting a physiological pulmonary trunk model to the frames of the 4D cardiac image data based on the detected ribs and anatomic landmarks.

6. The method of claim 1, wherein the 4D cardiac image data is a 4D computed tomography (CT) sequence.

7. The method of claim 5, wherein the 4D cardiac image data is a 4D computed tomography (CT) sequence, and:

said step of detecting bounding boxes comprises detecting bounding boxes in frames of the 4D CT sequence representing end-diastolic (ED) and end-systolic (ES) phases of a cardiac cycle and extending the detected bounding boxes to remaining frames of the 4D CT sequence;

said step of detecting anatomic landmarks comprises detecting anatomic landmarks in the frames of the 4D CT sequence representing the ED and ES phases and extending the detected anatomic landmarks to the remaining frames of the 4D CT sequence;

said step of detecting ribs comprises detecting ribs of the pulmonary artery in the frames of the 4D CT sequence representing the ED and ES phases and extending the detected ribs to the remaining frames of the 4D CT sequence; and

said step of generating a dynamic pulmonary trunk model comprises fitting the physiological pulmonary trunk model a frames of the 4D CT sequence representing one of the ED and ES phases and extending the physiological pulmonary trunk model to the remaining frames of the 4D CT sequence using one-step-forward prediction for smooth motion estimation.

8. The method of claim 5, wherein said step of detecting bounding boxes comprises:

detecting bounding boxes corresponding to an RVOT region, a main bifurcation region, a left bifurcation region, and a right bifurcation region in the frames of the 4D cardiac data.

9. The method of claim 8, wherein said step of detecting bounding boxes corresponding to an RVOT region, a main bifurcation region, a left bifurcation region, and a right bifurcation region in the frames of the 4D cardiac data:

detecting the bounding boxes corresponding to an RVOT region, a main bifurcation region, a left bifurcation region, and a right bifurcation region using trained detectors for each bounding box, wherein the trained detectors are trained using marginal space learning (MSL), such that each of the trained detectors comprises marginal space classifiers trained using a probabilistic boosting tree (PBT) with at least one of 3D Haar features and steerable features.

10. The method of claim 8, wherein said step of detecting anatomic landmarks comprises:

detecting trigone and RVOT locations in the frames of the 4D cardiac data based on the detected bounding box corresponding to the RVOT region;

detecting a main-bifurcation location in the frames of the 4D cardiac data based on the detected bounding box corresponding to the main-bifurcation region;

detecting a left-bifurcation location in the frames of the 4D cardiac data based on the detected bounding box corresponding to the left-bifurcation region; and

detecting a right-bifurcation location in the frames of the 4D cardiac data based on the detected bounding box corresponding to the right-bifurcation region.

11. The method of claim 10, wherein the trigone, RVOT, main-bifurcation, left-bifurcation, and right-bifurcation locations are detected using trained detectors for each anatomic landmark, wherein the trained detectors are trained using a probabilistic boosting tree (PBT).

12. The method of claim 10, wherein said step of detecting ribs of a pulmonary artery comprises:

detecting a main-rib based on the detected RVOT and main-bifurcation locations, a left-rib based on the detected main-bifurcation and left-bifurcation locations, and a right-rib based on the detected main-bifurcation and right bifurcation locations.

13. The method of claim 12, wherein said step of detecting a main-rib based on the detected RVOT and main-bifurcation locations, a left-rib based on the detected main-bifurcation and left-bifurcation locations, and a right-rib based on the detected main-bifurcation and right bifurcation locations comprises:

detecting the main-rib, left-rib, and right rib with trained circle detectors for each rib.

14. The method of claim 5, wherein said step of generating a dynamic pulmonary trunk model comprises:

generating an initial estimate of the pulmonary trunk model in the frames of the 4D cardiac image data by fitting the physiological pulmonary trunk model based on the detected ribs and anatomic features; and

detecting a boundary of the pulmonary trunk in the frames of the 4D image data to refine a boundary of the initial estimate of the pulmonary trunk model.

15. The method of claim 14, wherein said step of detecting a boundary of the pulmonary trunk in the frames of the 4D image data comprises:

detecting the boundary of the pulmonary trunk using a trained boundary detector, wherein the trained boundary detector is trained using PBT with steerable features.

16. An apparatus for determining whether a patient is suitable for percutaneous pulmonary valve implantation (PPVI) intervention, comprising:

means for receiving 4D cardiac image data of a patient comprising a plurality of frames, each frame comprising volumetric image data;

means for generating a patient-specific dynamic pulmonary trunk model based on the 4D cardiac image data of the patient; and

means for automatically classifying the patient as suitable or unsuitable for PPVI intervention based on the generated patient-specific dynamic pulmonary trunk model.

17. The apparatus of claim 16, wherein said means for automatically classifying the patient as one of suitable for PPVI intervention based on the generated patient-specific dynamic pulmonary trunk model comprises:

means for extracting shape features from the patient specific dynamic pulmonary trunk model; and

means for classifying the patient as suitable or unsuitable for PPVI intervention based on the extracted shape features using a trained discriminative distance function.

18. The apparatus of claim 17, wherein the trained discriminative distance function clusters the patient-specific dynamic pulmonary trunk model with similar pulmonary trunk models in training data based on the extracted shape features, and said means for classifying the patient as suitable or unsuitable for PPVI intervention based on the extracted shape features using a trained discriminative distance function comprises:

means for classifying the patient as suitable for PPVI intervention if the patient-specific dynamic pulmonary trunk model is clustered with pulmonary trunk models that are suitable for PPVI intervention; and

means for classifying the patient as unsuitable for PPVI intervention if the patient-specific dynamic pulmonary trunk model is clustered with pulmonary trunk models that are unsuitable for PPVI intervention.

19. The apparatus of claim 17, wherein the trained discriminative distance function is trained using one of a Random Forest in product space and an intrinsic Random Forest similarity of a Random Forest trained in an original space.

20. The apparatus of claim 16, wherein said means for generating a patient-specific dynamic pulmonary trunk model based on the 4D cardiac image data of the patient comprises:

means for detecting bounding boxes in the frames of the 4D cardiac image data;

means for detecting anatomic landmarks in the frames of the 4D cardiac image data based on the detected bounding boxes;

means for detecting ribs of a pulmonary artery in the frames of the 4D cardiac image data based on the detected anatomic landmarks; and

means for generating the patient-specific dynamic pulmonary trunk model by fitting a physiological pulmonary trunk model to the frames of the 4D cardiac image data based on the detected ribs and anatomic landmarks.

21. A non-transitory computer readable medium encoded with computer executable instructions for determining whether a patient is suitable for percutaneous pulmonary valve implantation (PPVI) intervention, the computer executable instructions defining steps comprising:

receiving 4D cardiac image data of a patient comprising a plurality of frames, each frame comprising volumetric image data;

generating a patient-specific dynamic pulmonary trunk model based on the 4D cardiac image data of the patient; and

automatically classifying the patient as suitable or unsuitable for PPVI intervention based on the generated patient-specific dynamic pulmonary trunk model.

22. The computer readable medium of claim 21, wherein the computer executable instructions defining the step of automatically classifying the patient as one of suitable for PPVI intervention based on the generated patient-specific dynamic pulmonary trunk model comprise computer executable instructions defining the steps of:

extracting shape features from the patient specific dynamic pulmonary trunk model; and

classifying the patient as suitable or unsuitable for PPVI intervention based on the extracted shape features using a trained discriminative distance function.

23. The computer readable medium of claim 22, wherein the trained discriminative distance function clusters the patient-specific dynamic pulmonary trunk model with similar pulmonary trunk models in training data based on the extracted shape features, and the computer executable instructions defining the step of classifying the patient as suitable or unsuitable for PPVI intervention based on the extracted shape features using a trained discriminative distance function comprise computer executable instructions defining the steps of:

if the patient-specific dynamic pulmonary trunk model is clustered with pulmonary trunk models that are suitable for PPVI intervention, classifying the patient as suitable for PPVI intervention; and

if the patient-specific dynamic pulmonary trunk model is clustered with pulmonary trunk models that are unsuitable for PPVI intervention, classifying the patient as unsuitable for PPVI intervention.

24. The computer readable medium of claim 22, wherein the trained discriminative distance function is trained using one of a Random Forest in product space and an intrinsic Random Forest similarity of a Random Forest trained in an original space.

25. The computer readable medium of claim 21, wherein the computer executable instructions defining the step of generating a patient-specific dynamic pulmonary trunk model based on the 4D cardiac image data of the patient comprise computer executable instructions defining the steps of:

detecting bounding boxes in the frames of the 4D cardiac image data;

detecting anatomic landmarks in the frames of the 4D cardiac image data based on the detected bounding boxes;

detecting ribs of a pulmonary artery in the frames of the 4D cardiac image data based on the detected anatomic landmarks; and

generating the patient-specific dynamic pulmonary trunk model by fitting a physiological pulmonary trunk model to the frames of the 4D cardiac image data based on the detected ribs and anatomic landmarks.