Method and system for bone segmentation and landmark detection for joint replacement surgery

Patent metadata

Field	Value
Publication number	US-9646229-B2
Application number	US-201314041029-A
Country	US
Kind code	B2
Filing date	Sep 30, 2013
Priority date	Sep 28, 2012
Publication date	May 9, 2017
Grant date	May 9, 2017

Abstract

A method and system for automatic bone segmentation and landmark detection for joint replacement surgery is disclosed. A 3D medical image of at least a target joint region of a patient is received. A plurality bone structures are automatically segmented in the target joint region of the 3D medical image and a plurality of landmarks associated with a joint replacement surgery are automatically detected in the target joint region of the 3D medical image. The boundaries of segmented bone structures can then be interactively refined based on user inputs.

First claim

The invention claimed is:

1. A method for bone segmentation and landmark detection for joint replacement surgery, comprising:
receiving a 3D medical image of at least a target joint region of a patient;
automatically segmenting a plurality bone structures in the target joint region of the 3D medical image by estimating, using trained discriminative classifiers, a shape for each of the plurality of bone structures in a respective learned shape space trained based on a database of training data; and
automatically detecting a plurality of landmarks associated with a joint replacement surgery in the target joint region of the 3D medical image using respective trained landmark detectors.

2. The method of claim 1 , wherein the target joint region is a knee region.

3. The method of claim 2 , wherein automatically segmenting a plurality bone structures in the target joint region of the 3D medical image by estimating, using trained discriminative classifiers, a shape for each of the plurality of bone structures in a respective learned shape space trained based on a database of training data comprises;
automatically segmenting a femur, tibia, fibula, and patella in the 3D medical image.

4. The method of claim 3 , wherein automatically detecting a plurality of landmarks associated with a joint replacement surgery in the target joint region of the 3D medical image using respective trained landmark detectors comprises:
automatically detecting a femur medial most distal, femur lateral most distal, femur lateral posterior condyle point, femur anterior cortex point, femur medial posterior condyle point, femoral head, and ankle center in the 3D medical image.

5. The method of claim 1 , wherein automatically segmenting a plurality bone structures in the target joint region of the 3D medical image by estimating, using trained discriminative classifiers, a shape for each of the plurality of bone structures in a respective learned shape space trained based on a database of training data comprises:
independently segmenting each of the plurality of bone structures in the 3D medical image.

6. The method of claim 1 , wherein automatically segmenting a plurality bone structures in the target joint region of the 3D medical image by estimating, using trained discriminative classifiers, a shape for each of the plurality of bone structures in a respective learned shape space trained based on a database of training data comprises, for each of the plurality of bone structures:
generating a mesh representing a boundary of the bone structure by estimating, in the 3D medical image, the shape in the learned shape space for the bone structure; and
refining the mesh using a trained boundary detector.

7. The method of claim 6 , wherein generating a mesh representing a boundary of the bone structure by estimating, in the 3D medical image, a shape in a learned shape space for the bone structure comprises:
estimating pose parameters and shape space parameters to align the mesh for the bone structure to the 3D medical image.

8. The method of claim 6 , wherein refining the mesh using a trained boundary detector comprises:
adjusting each of a plurality of vertices of the mesh in a normal direction using the trained boundary detector.

9. The method of claim 1 , wherein automatically segmenting a plurality bone structures in the target joint region of the 3D medical image by estimating, using trained discriminative classifiers, a shape for each of the plurality of bone structures in a respective learned shape space trained based on a database of training data comprises:
jointly segmenting the plurality of bone structures using prior spatial constraints to prevent overlaps between the plurality of bone structures.

10. The method of claim 1 , wherein automatically detecting a plurality of landmarks associated with a joint replacement surgery in the target joint region of the 3D medical image using respective trained landmark detectors comprises:
automatically detecting the plurality of landmarks using the respective trained landmark detectors, wherein a search space for at least one of the respective trained landmark detectors is constrained based on at least one other landmark detection result.

11. The method of claim 10 , wherein a search space for at least one of the respective trained landmark detectors is constrained based on the segmented bone structures.

12. The method of claim 1 , wherein automatically detecting a plurality of landmarks associated with a joint replacement surgery in the target joint region of the 3D medical image using respective trained landmark detectors comprises:
(a) for each undetected one of the plurality of landmarks, determining a search space for a corresponding trained landmark detector based on at least one detected one of the plurality of landmarks;
(b) selecting the trained landmark detector having the smallest search space;
(c) detecting, using the selected trained landmark detector, a corresponding one of the plurality of landmarks within the search space determined for the selected trained landmark detector; and
(d) repeating steps (a)-(c) until no undetected ones of the plurality of landmarks remain.

13. The method of claim 1 , further comprising:
refining the segmented bone structures based on user inputs.

14. The method of claim 13 , wherein refining the segmented bone structures based on user inputs comprises:
receiving user inputs corresponding to seed points in the segmented bone structures; and
refining the segmented bone structures by minimizing an energy functional based on the automatically segmented bone structures, the seed points, and image intensities of the 3D medical image.

15. The method of claim 13 , wherein refining the segmented bone structures based on user inputs comprises:
receiving user inputs corresponding to seed points in a selected slice of the segmented bone structures; and
locally refining the segmented bone structures in the selected slice based on the received user inputs.

16. The method of claim 1 , further comprising:
automatically segmenting metal structures in the target joint region of the 3D medical image.

17. An apparatus for bone segmentation and landmark detection for joint replacement surgery, comprising:
means for receiving a 3D medical image of at least a target joint region of a patient;
means for automatically segmenting a plurality bone structures in the target joint region of the 3D medical image by estimating, using trained discriminative classifiers, a shape for each of the plurality of bone structures in a respective learned shape space trained based on a database of training data; and
means for automatically detecting a plurality of landmarks associated with a joint replacement surgery in the target joint region of the 3D medical image using respective trained landmark detectors.

18. The apparatus of claim 17 , wherein the target joint region is a knee region.

19. The apparatus of claim 18 , wherein the means for automatically segmenting a plurality bone structures in the target joint region of the 3D medical image by estimating, using trained discriminative classifiers, a shape for each of the plurality of bone structures in a respective learned shape space trained based on a database of training data comprises;
means for automatically segmenting a femur, tibia, fibula, and patella in the 3D medical image.

20. The apparatus of claim 19 , wherein the means for automatically detecting a plurality of landmarks associated with a joint re

Assignees

• Sofka Michal
• Liu Meizhu
• Wu Dijia
• Zhou Shaohua Kevin
• Siemens Medical Solutions Usa Inc

Inventors

• Sofka Michal
• Liu Meizhu
• Wu Dijia
• Zhou Shaohua Kevin

Classifications

• G06T7/10 Primary

  Segmentation; Edge detection (motion-based segmentation G06T7/215) · CPC title

• G06T2207/10088

  Magnetic resonance imaging [MRI] · CPC title

• G06T7/0014

  using an image reference approach · CPC title

• G06K9/66 Primary

  Physics · mapped topic

• G06T2207/20081

  Training; Learning · CPC title