Detection of landmarks and key-frames in cardiac perfusion MRI using a joint spatial-temporal context model

1. A method, comprising:

receiving an image sequence, wherein the image sequence includes a plurality of two-dimensional (2D) image frames of an organ arranged in a time sequence;

constructing a three-dimensional (3D) volume by stacking a plurality of the 2D image frames in time order;

detecting a best bounding box for a target of interest in the 3D volume, wherein the best bounding box is specified by a plurality of parameters including spatial and temporal information contained in the 3D volume; and

determining the target of interest from the best bounding box,

wherein detecting a best bounding box for a target of interest comprises:

(a) setting values for a nine-parameter set, wherein the nine-parameter set defines a box-like sub-volume in the 3D volume and the nine parameter set includes three positions, x,y,z, three orientations γ,φ,ψ and three scales sx,sy,sz, wherein the orientation parameters γ,φ,ψ reduce to γ only,

wherein <x an ,y an > and <x in ,y in > denote true positions of right ventricle (RV) insertion anterior and RV insertion inferior, respectively, time frame indices for a key-frame and a peak-frame of RV are expressed as t RV and t pRV , respectively

wherein a parameter set for RV insertion is:

positions {(x an +x in )/2,(y an +y in )/2,t RV }; orientations {γ: orthogonal to a line segment connecting the RV insertion anterior and the RV insertion inferior, and pointing toward a left ventricle (LV), φ=0, ψ=0}; and scales {s ins ,s ins ,(t pRV −t RV )*α}, where s ins =√{square root over ((x an −x in ) 2 +(y an −y in ) 2 )}{square root over ((x an −x in ) 2 +(y an −y in ) 2 )}*β and α and β are factors used to adjust range.

2. The method of claim 1 , wherein the image sequence includes a perfusion magnetic resonance imaging (MRI) sequence.

3. The method of claim 2 , wherein the organ includes a heart.

4. The method of claim 1 , wherein the target of interest includes an LV center point, RV insertion points, an LV key-frame or an RV key-frame.

5. The method of claim 1 , wherein detecting a best bounding box for a target of interest comprises:

(b) extracting the sub-volume from the 3D volume;

(c) applying the sub-volume to a pre-learned model for the target of interest; and

(d) assigning a classification value to the sub-volume using the pre-learned model.

6. The method of claim 5 , further comprising repeating a-d for a predetermined number of times, each time with at least one different value for the nine-parameter set, wherein the sub-volume with a highest classification value is the best bounding box.

7. The method of claim 5 , wherein the pre-learned model uses a machine learning algorithm to automatically extract features to distinguish the target of interest from background.

8. The method of claim 1 , wherein the target of interest is determined from the best bounding box by using a one-to-one mapping between the target of interest and the best bounding box.

9. A computer program product, comprising:

a non-transitory computer readable storage medium having computer readable program code embodied therewith, the computer readable program code comprising:

computer readable program code configured to perform the steps of:

receiving an image sequence, wherein the image sequence includes a plurality of two-dimensional (2D) image frames of an organ arranged in a time sequence;

constructing a three-dimensional (3D) volume by stacking a plurality of the 2D image frames in time order;

detecting a best bounding box for a target of interest in the 3D volume, wherein the best bounding box is specified by a plurality of parameters including spatial and temporal information contained in the 3D volume; and

determining the target of interest from the best bounding box,

wherein detecting a best bounding box for a target of interest comprises:

(a) setting values for a nine-parameter set, wherein the nine-parameter set defines a box-like sub-volume in the 3D volume and the nine parameter set includes three positions, x,y,z, three orientations γ,φ,ψ and three scales sx,sy,sz, wherein the orientation parameters γ,φ,ψ reduce to γ only,

wherein <x an ,y an > and <x in ,y in > denote true positions of right ventricle (RV) insertion anterior and RV insertion inferior, respectively, time frame indices for a key-frame and a peak-frame of RV are expressed as t RV and t pRV , respectively

wherein a parameter set for RV insertion is:

positions {(x an +x in )/2,(y an +y in )/2,t RV }; orientations {γ: orthogonal to a line segment connecting the RV insertion anterior and the RV insertion inferior, and pointing toward a left ventricle (LV), φ=0, ψ=0}; and scales {s ins ,s ins ,(t pRV −t RV )*α}, where s ins =√{square root over ((x an −x in ) 2 +(y an −y in ) 2 )}{square root over ((x an −x in ) 2 +(y an −y in ) 2 )}*β and α and β are factors used to adjust range.

10. The computer program product of claim 9 , wherein the image sequence includes a perfusion magnetic resonance imaging (MRI) sequence.

11. The computer program product of claim 10 , wherein the organ includes a heart.

12. The computer program product of claim 9 , wherein the target of interest includes an LV center point, RV insertion points, an LV key-frame or an RV key-frame.

13. The computer program product of claim 9 , wherein detecting a best bounding box for a target of interest comprises:

(b) extracting the sub-volume from the 3D volume;

(c) applying the sub-volume to a pre-learned model for the target of interest; and

(d) assigning a classification value to the sub-volume using the pre-learned model.

14. The computer program product of claim 13 , wherein the computer readable program code is further configured to perform the step of repeating a-d for a predetermined number of times, each time with at least one different value for the nine-parameter set, wherein the sub-volume with a highest classification value is the best bounding box.

15. The computer program product of claim 9 , wherein the target of interest is determined from the best bounding box by using a one-to-one mapping between the target of interest and the best bounding box.

16. A method, comprising:

receiving a cardiac perfusion magnetic resonance imaging (MRI) sequence, wherein the cardiac perfusion MRI sequence includes a plurality of two-dimensional (2D) image frames of a heart arranged in a time sequence;

constructing a three-dimensional (3D) volume by stacking a plurality of the 2D image frames in time order;

detecting a bounding box enclosing a landmark in the 3D volume, wherein the bounding box is specified by a plurality of parameters including spatial and temporal information contained in the 3D volume; and

determining the landmark and its associated key-frame from the bounding box,

wherein <x an ,y an > and <x in ,y in > denote true positions of right ventricle (RV) insertion anterior and RV insertion inferior, respectively, time frame indices for a key-frame and a peak-frame of RV are expressed as t RV and t pRV , respectively

wherein a parameter set for RV insertion is:

positions {(x an +x in )/2,(y an +y in )/2,t RV }; orientations {γ: orthogonal to a line segment connecting the RV insertion anterior and the RV insertion inferior, and pointing toward a left ventricle (LV), φ=0, ψ=0}; and scales {s ins ,s ins ,(t pRV −t RV )*α}, where s ins =√{square root over ((x an −x in ) 2 +(y an −y in ) 2 )}{square root over ((x an −x in ) 2 +(y an −y in ) 2 )}*β and α and β are factors used to adjust range.

17. The method of claim 16 , wherein the landmark and its associated key-frame are determined simultaneously.

18. The method of claim 16 , wherein the landmark includes a center part of the LV and the key-frame is a 2D image frame in which an upslope along a signal intensity curve for the LV reaches a maximum in the time sequence.

19. The method of claim 16 , wherein the landmark includes RV insertion points and the key-frame is a 2D image frame in which an upslope along a signal intensity curve for the RV reaches a maximum in the time sequence.

20. The method of claim 19 , wherein the RV insertion points include the RV anterior insertion point and the RV inferior insertion point.