STENT MARKER DETECTION USING A LEARNING BASED CLASSIFIER IN MEDICAL IMAGING

• US 20110144480A1
• Filed: 12/06/2010
• Published: 06/16/2011
• Est. Priority Date: 12/10/2009
• Status: Active Grant

• Current Assignee: Siemens Healthineers AG (Siemens AG)
• Original Assignee: Siemens AG, Siemens Corp. (Siemens AG)
• Inventors: Comaniciu, Dorin, Chen, Terrence, Lu, Xiaoguang, Durlak, Peter, Pohl, Thomas
• Granted Patent: US 9,119,573 B2
• Time in Patent Office: Days
• US Class Current: 600/424
• CPC Class Codes: A61B 34/10 Computer-aided planning, simulation or modelling of surgical operations
• CPC Class Codes: A61B 6/12 Arrangements for detecting or locating foreign bodies
• CPC Class Codes: A61B 6/503 for diagnosis of the heart
• CPC Class Codes: A61B 6/504 for diagnosis of blood vessels, e.g. by angiography
• CPC Class Codes: G06T 2207/20064 Wavelet transform [DWT]
• CPC Class Codes: G06T 2207/20081 Training; Learning
• CPC Class Codes: G06T 7/0012 Biomedical image inspection
• CPC Class Codes: G06V 2201/03 Recognition of patterns in medical or anatomical images
• CPC Class Codes: G16H 50/50 for simulation or modelling of medical disorders

Abstract

Stent marker detection is automatically performed. Stent markers in fluoroscopic images or other markers in other types of imaging are detected using a machine-learnt classifier. Hierarchal classification may be used, such as detecting individual markers with one classifier and then detecting groups of markers (e.g., a pair) with a joint classifier. The detection may be performed in a single image and without user indication of a location.

Claims

1. A method for marker detection in fluoroscopic imaging of a stent, the method comprising:

individually detecting a plurality of possible markers for a stent in one frame of fluoroscopic data without information from other frames of fluoroscopic data, the individual detecting being with a first machine-trained classifier;

jointly detecting a pair of markers from the plurality of possible markers for the stent output from the individually detecting, the joint detecting being with a second machine-trained classifier; and

outputting an image representing information about the pair of markers.

2. The method of claim 1 wherein individually detecting comprises detecting with the first machine-trained classifier comprising a probabilistic boosting tree classifier.

3. The method of claim 1 wherein individually detecting comprises detecting with first input features comprising Haar features and wherein jointly detecting comprises detecting with second input features comprising gradient features.

4. The method of claim 1 wherein individually detecting further comprises removing some of the possible markers with a third machine-trained classifier, the third machine-trained classifier operating on an output of the first machine-trained classifier and the second machine-trained classifier operating on an output of the third machine-trained classifier.

5. The method of claim 1 wherein jointly detecting comprises fitting a rectangular box over the pair of markers and detecting as a function of an input feature set comprising a gradient magnitude, a gradient angle, or both relative to the rectangular box.

6. The method of claim 1 wherein jointly detecting comprises detecting with the second machine-trained classifier comprising a probabilistic boosting tree classifier.

7. The method of claim 1 wherein individually detecting comprises determining a first score for each of the possible markers, and wherein jointly detecting comprises determining a second score for each of paired combinations of the possible markers;

further comprising calculating a third score for each of the paired combinations, the third score being a function of the first and second scores associated with the possible markers in the paired combination, the pair of markers being the paired combination with a highest score.

8. The method of claim 1 further comprising:

steerably filtering the frame of fluoroscopic data such that guide wire regions are distinguished from background; and

wherein individually detecting comprises sampling locations along the guide wire regions.

9. The method of claim 1 further comprising:

repeating individual and joint detecting for a plurality of frames of fluoroscopic data;

aligning the frames of fluoroscopic data as a function of the respective pairs of markers; and

combining the aligned frames into a stent frame of data; and

wherein outputting the image comprises outputting the image as a function of the stent frame of data.

10. The method of claim 1 wherein outputting the image comprises outputting an image comprising a fluoroscopic image with the pair of markers highlighted.

11. The method of claim 1 further comprising:

selecting a marker reference frame of data from a plurality of frames of fluoroscopic data, including the frame of fluoroscopic data, the selecting being as a function of a distance between markers in the pair of markers for each of the plurality of the frames and as a function of probabilities of the pair of markers being correct.

12. In a non-transitory computer readable storage medium having stored therein data representing instructions executable by a programmed processor for marker detection in medical imaging, the storage medium comprising instructions for:

applying a machine learning-based marker detection to a fluoroscopy image; and

outputting a first location of a marker in the fluoroscopy image, the first location determined as a function of the machine learning-based marker detection.

13. The non-transitory computer readable storage medium of claim 12 wherein applying comprises detecting possible markers individually from context in the fluoroscopy image in a local region to the possible markers.

14. The non-transitory computer readable storage medium of claim 13 wherein individually detecting comprises detecting with a first machine-trained classifier and further comprises removing some of the possible markers with a second machine-trained classifier.

15. The non-transitory computer readable storage medium of claim 12 wherein applying comprises jointly detecting the marker as a function of another marker.

16. The non-transitory computer readable storage medium of claim 15 wherein jointly detecting comprises fitting a rectangular box over a pair of possible markers and detecting as a function of an input feature set comprising a gradient magnitude, a gradient angle, or both relative to the rectangular box.

17. The non-transitory computer readable storage medium of claim 12 wherein outputting the first location comprises outputting the fluoroscopic image with the marker highlighted.

18. A system for marker detection in medical imaging, the system comprising:

a memory operable to store data representing a vessel with at least two markers for a stent;

a processor configured to detect the at least two markers from the data by first detecting possible markers separately and then second detecting the at least two markers jointly from combinations of the possible markers; and

a display operable to display an image of the at least two markers.

19. The system of claim 18 wherein the processor is configured to detect with a first machine-trained classifier for the first detecting and is configured to detect with a second machine-trained classifier for the second detecting.

20. A system of claim 18 wherein the processor is configured to detect with the at least two markers comprising a pair of markers on a balloon for the stent, the processor configured to detect with the data comprising a single frame of data and without data from other frames.