Localization of anatomical structures using learning-based regression and efficient searching or deformation strategy
View Patent ↗Disclosed herein is a framework for localizing anatomical structures. In accordance with one aspect, the framework receives a learned regressor and image data of a subject. The learned regressor may be invoked to predict a first spatial metric from a seed voxel to a target anatomical structure in the image data. The learned regressor may further be invoked to predict second spatial metrics from candidate voxels to the target anatomical structure. The candidate voxels may be located around a search region defined by the first spatial metric. The candidate voxel associated with the smallest second spatial metric may then be output as a localized voxel.
1. A non-transitory computer-readable medium embodying a program of instructions executable by machine to perform steps for localizing an anatomical structure, the steps comprising:
(i) receiving a learned regressor and image data of a subject;
(ii) invoking the learned regressor to predict a first distance from a seed voxel to a target anatomical structure in the image data based on appearance features of the seed voxel;
(iii) selecting candidate voxels in the image data located at the first distance from the seed voxel;
(iv) invoking the learned regressor to predict second distances from the candidate voxels to the target anatomical structure based on appearance features of the candidate voxels;
(iv) in response to a smallest second distance being less than a predetermined threshold, outputting the candidate voxel associated with a smallest second distance as a localized voxel; and
(v) in response to the smallest second distance being more than the predetermined threshold, setting the seed voxel to the candidate voxel associated with the smallest second distance and repeating at least steps (ii), (iii) and (iv).
2. The computer-readable medium of claim 1 embodying further instructions executable by the machine to train the regressor to predict the first and second distances.
3. The computer-readable medium of claim 2 embodying further instructions executable by the machine to train the regressor by using random forest regression.
4. The computer-readable medium of claim 2 embodying further instructions executable by the machine to train the regressor by using Adaboost regression.
5. A computer-implemented method of localizing an anatomical structure, comprising:
(i) receiving a learned regressor and image data of a subject;
(ii) invoking the learned regressor to predict a first spatial metric from a seed voxel to a target anatomical structure in the image data;
(iii) selecting candidate voxels in the image data located around a search region defined by the first spatial metric;
(iv) invoking the learned regressor to predict second spatial metrics from the candidate voxels to the target anatomical structure; and
(v) in response to a smallest second spatial metric being less than a predetermined threshold, outputting the candidate voxel associated with the smallest second spatial metric as a localized voxel.
6. The method of claim 5 further comprising setting the seed voxel to the candidate voxel associated with the smallest second spatial metric and repeating at least steps (ii), (iii) and (iv) in response to the smallest second spatial metric being more than the predetermined threshold.
7. The method of claim 5 further comprising training the regressor to predict the spatial metric.
8. The method of claim 7 wherein the spatial metric comprises a scalar distance.
9. The method of claim 7 wherein the spatial metric comprises a vector displacement.
10. The method of claim 7 wherein the spatial metric comprises a tensor displacement.
11. The method of claim 7 wherein training the regressor to predict the spatial metric comprises performing random forest regression.
12. The method of claim 7 wherein training the regressor to predict the spatial metric comprises performing Adaboost regression.
13. The method of claim 7 further comprising selecting appearance features to predict the spatial matric.
14. The method of claim 13 wherein the appearance features comprise Haar wavelets, Gabor features, linear binary patterns, or a combination thereof.
15. The method of claim 5 wherein selecting the candidate voxels comprises selecting the candidate voxels at a boundary of a sphere centered at the seed voxel with a radius defined by the first spatial metric.
16. An anatomical structure localization system, comprising:
a non-transitory memory device for storing computer-readable program code; and
a processor in communication with the memory device, the processor being operative with the computer-readable program code to perform steps comprising
(i) receiving a learned regressor and image data of a subject,
(ii) invoking the learned regressor to predict signed distances from vertices on a surface to a target boundary surface in the image data,
(iii) in response to the signed distances being less than a predetermined threshold, outputting the surface as a segmented boundary, and
(iv) in response to the signed distances being more than the predetermined threshold, deforming the surface based the signed distances and repeating at least step (ii).
17. The system of claim 16 wherein the processor is further operative with the computer-readable program code to train the regressor to predict the signed distances.
18. The system of claim 17 wherein the processor is further operative with the computer-readable program code to train the regressor by using random forest regression.
19. The system of claim 17 wherein the processor is further operative with the computer-readable program code to train the regressor by using Adaboost regression.
20. The system of claim 16 wherein the image data comprises image data acquired by magnetic resonance (MR) imaging, computed tomography (CT), helical CT, x-ray, positron emission tomography (PET), PET-CT, fluoroscopy, ultrasound, single-photon emission computed tomography (SPECT), SPECT-CT, MR-PET, or a combination thereof.