Method and system for catheter detection and tracking in a fluoroscopic image sequence

1. A method for detecting and tracking an ablation catheter tip in a fluoroscopic image sequence including a plurality of frames, comprising:

detecting catheter tip candidates in each frame of the fluoroscopic image sequence using marginal space learning, wherein the marginal space learning detects the catheter tip candidates in a series of marginal space learning levels of increasing dimensionality; and

tracking the catheter tip candidates over all of the frames of the fluoroscopic image sequence to determine an ablation catheter tip location in each frame of the fluoroscopic image sequence.

2. The method of claim 1 , wherein said step of detecting tip candidates in each frame of the fluoroscopic image sequence comprises, for each frame:

detecting positions of catheter tip candidates in the frame in a first marginal space learning level; and

detecting catheter tip candidates including position and orientation in the frame in a second marginal space learning level.

3. The method of claim 2 , wherein said step of detecting positions of catheter tip candidates is performed by a first classifier trained using a Probabilistic Boosting Tree (PBT), and said step of detecting catheter tip candidates including position and orientation is performed by a second classifier trained using a PBT.

4. The method of claim 2 , wherein said step of detecting tip candidates in each frame of the fluoroscopic image sequence further comprises, for each frame:

reducing a number of catheter tip candidates using non-maximal suppression.

5. The method of claim 4 , wherein said step of reducing a number of catheter tip candidates using non-maximal suppression comprises:

selecting a catheter tip candidate with a highest probability;

discarding catheter tip candidates in a neighborhood of the selected catheter tip candidate; and

repeating said selecting and discarding steps until each catheter tip candidate has been selected or discarded.

6. The method of claim 1 , wherein said step of detecting tip candidates in each frame of the fluoroscopic image sequence comprises, for each frame:

detecting a number of position candidates at quarter resolution using a first classifier;

scaling said position candidates at quarter resolution to generate a plurality of full resolution potential position candidates;

detecting a number of position candidates at full resolution from said full resolution potential position candidates using a second classifier;

augmenting each of said position candidates at full resolution using a plurality of discrete angles to generate a plurality of potential position and orientation candidates;

detecting a number of catheter tip candidates from said potential position and orientation candidates using a third classifier; and

reducing the number of catheter tip candidates using non-maximal suppression.

7. The method of claim 6 , wherein said first, second, and third classifiers are Probabilistic Boosting Tree (PBT) classifiers.

8. The method of claim 1 , wherein said step of tracking the catheter tip candidates over all of the frames of the fluoroscopic image sequence to determine an ablation catheter tip location in each frame of the fluoroscopic image sequence comprises:

determining a trajectory between catheter tip candidates over all of the frames in the fluoroscopic image sequence having a minimal cost; and

selecting a catheter tip candidate in each frame in the trajectory having the minimal cost as the ablation catheter tip location for that frame.

9. The method of claim 8 , wherein said step of determining a trajectory between catheter tip candidates over all of the frames in the fluoroscopic image sequence having a minimal cost comprises at each frame:

calculating a best trajectory ending at each catheter tip candidate in that frame;

determining a catheter tip candidate that has the best trajectory with a lowest cost; and

if the lowest cost is less than a lowest cost determined at a previous frame, selecting the catheter tip candidate as the ablation catheter tip location for the frame.

10. An apparatus for detecting and tracking an ablation catheter tip in a fluoroscopic image sequence including a plurality of frames, comprising:

means for detecting catheter tip candidates in each frame of the fluoroscopic image sequence using marginal space learning, wherein the marginal space learning detects the catheter tip candidates in a series of marginal space learning levels of increasing dimensionality; and

means for tracking the catheter tip candidates over all of the frames of the fluoroscopic image sequence to determine an ablation catheter tip location in each frame of the fluoroscopic image sequence.

11. The apparatus of claim 10 , wherein said means for detecting tip candidates in each frame of the fluoroscopic image sequence comprises:

means for detecting positions of catheter tip candidates in each frame in a first marginal space learning level; and

means for detecting catheter tip candidates including position and orientation in each frame in a second marginal space learning level.

12. The apparatus of claim 11 , further comprising:

means for training a first classifier using a Probabilistic Boosting Tree (PBT) corresponding to said first marginal space learning level; and

means for training a second classifier using a PBT corresponding to said second marginal space learning level.

13. The apparatus of claim 11 , wherein said means for detecting tip candidates in each frame of the fluoroscopic image sequence further comprises:

means for reducing a number of catheter tip candidates detected in each frame using non-maximal suppression.

14. The apparatus of claim 10 , wherein said means for detecting tip candidates in each frame of the fluoroscopic image sequence comprises, for each frame:

means for detecting a number of position candidates at quarter resolution using a first classifier;

means for scaling said position candidates at quarter resolution to generate a plurality of full resolution potential position candidates;

means for detecting a number of position candidates at full resolution from said full resolution potential position candidates using a second classifier;

means for augmenting each of said position candidates at full resolution using a plurality of discrete angles to generate a plurality of potential position and orientation candidates;

means for detecting a number of catheter tip candidates from said potential position and orientation candidates using a third classifier; and

means for reducing the number of catheter tip candidates using non-maximal suppression.

15. The apparatus of claim 14 , wherein said first, second, and third classifiers are Probabilistic Boosting Tree (PBT) classifiers.

16. The apparatus of claim 10 , wherein said means for tracking the catheter tip candidates over all of the frames of the fluoroscopic image sequence to determine an ablation catheter tip location in each frame of the fluoroscopic image sequence comprises:

means for determining a trajectory between catheter tip candidates over all of the frames in the fluoroscopic image sequence having a minimal cost; and

means for selecting a catheter tip candidate in each frame in the trajectory having the minimal cost as the ablation catheter tip location for that frame.

17. The apparatus of claim 16 , wherein said means for determining a trajectory between catheter tip candidates over all of the frames in the fluoroscopic image sequence having a minimal cost comprises:

means for calculating a best trajectory ending at each catheter tip candidate in each frame;

means for determining a catheter tip candidate in each frame that has the best trajectory with a lowest cost; and

means for selecting the catheter tip candidate as the ablation catheter tip location for a frame if the lowest cost is less than a lowest cost determined at a previous frame.

18. A non-transitory computer readable medium encoded with computer program instructions for detecting and tracking an ablation catheter tip in a fluoroscopic image sequence including a plurality of frames, the computer program instructions defining steps comprising:

detecting catheter tip candidates in each frame of the fluoroscopic image sequence using marginal space learning, wherein the marginal space learning detects the catheter tip candidates in a series of marginal space learning levels of increasing dimensionality; and

tracking the catheter tip candidates over all of the frames of the fluoroscopic image sequence to determine an ablation catheter tip location in each frame of the fluoroscopic image sequence.

19. The non-transitory computer readable medium of claim 18 , wherein the computer program instructions defining the step of detecting tip candidates in each frame of the fluoroscopic image sequence comprise computer program instructions defining the following steps for each frame:

detecting positions of catheter tip candidates in the frame in a first marginal space learning level; and

detecting catheter tip candidates including position and orientation in the frame in a second marginal space learning level.

20. The non-transitory computer readable medium of claim 19 , wherein the computer program instructions defining the step of detecting tip candidates in each frame of the fluoroscopic image sequence further comprise computer program instructions defining the following step for each frame:

reducing a number of catheter tip candidates using non-maximal suppression.

21. The non-transitory computer readable medium of claim 20 , wherein the computer program instructions defining the step of reducing a number of catheter tip candidates using non-maximal suppression comprise computer program instructions defining the following steps:

selecting a catheter tip candidate with a highest probability;

discarding catheter tip candidates in a neighborhood of the selected catheter tip candidate; and

repeating said selecting and discarding steps until each catheter tip candidate has been selected or discarded.

22. The non-transitory computer readable medium of claim 18 , wherein the computer program instructions defining the step of detecting tip candidates in each frame of the fluoroscopic image sequence comprise computer program instructions defining the following steps for each frame:

detecting a number of position candidates at quarter resolution using a first classifier;

scaling said position candidates at quarter resolution to generate a plurality of full resolution potential position candidates;

detecting a number of position candidates at full resolution from said full resolution potential position candidates using a second classifier;

augmenting each of said position candidates at full resolution using a plurality of discrete angles to generate a plurality of potential position and orientation candidates;

detecting a number of catheter tip candidates from said potential position and orientation candidates using a third classifier; and

reducing the number of catheter tip candidates using non-maximal suppression.

23. The non-transitory computer readable medium of claim 22 , wherein said first, second, and third classifiers are Probabilistic Boosting Tree (PBT) classifiers.

24. The non-transitory computer readable medium of claim 18 , wherein the computer program instructions defining the step of tracking the catheter tip candidates over all of the frames of the fluoroscopic image sequence to determine an ablation catheter tip location in each frame of the fluoroscopic image sequence comprise computer program instructions defining the following steps:

determining a trajectory between catheter tip candidates over all of the frames in the fluoroscopic image sequence having a minimal cost; and

selecting a catheter tip candidate in each frame in the trajectory having the minimal cost as the ablation catheter tip location for that frame.

25. The non-transitory computer readable medium of claim 24 , wherein the computer program instructions defining the step of determining a trajectory between catheter tip candidates over all of the frames in the fluoroscopic image sequence having a minimal cost comprise computer program instructions defining the following steps for each frame:

calculating a best trajectory ending at each catheter tip candidate in that frame;

determining a catheter tip candidate that has the best trajectory with a lowest cost; and

if the lowest cost is less than a lowest cost determined at a previous frame, selecting the catheter tip candidate as the ablation catheter tip location for the frame.