Presentation is loading. Please wait.

Presentation is loading. Please wait.

TU/e, BMIA & PMS, X-Ray Predevelopment, Erik Franken, 22-09-2004 1 Context-Enhanced Detection of Electrophysiology Catheters in Noisy Fluoroscopy Images.

Similar presentations


Presentation on theme: "TU/e, BMIA & PMS, X-Ray Predevelopment, Erik Franken, 22-09-2004 1 Context-Enhanced Detection of Electrophysiology Catheters in Noisy Fluoroscopy Images."— Presentation transcript:

1 TU/e, BMIA & PMS, X-Ray Predevelopment, Erik Franken, 22-09-2004 1 Context-Enhanced Detection of Electrophysiology Catheters in Noisy Fluoroscopy Images Erik Franken Final presentation Master’s project Technische Universiteit Eindhoven 22 September 2004

2 TU/e, BMIA & PMS, X-Ray Predevelopment, Erik Franken, 22-09-2004 2 Outline 1.Introduction 2.Local feature detection 3.Context enhancement 4.EP catheter extraction 5.Evaluation 6.Conclusions

3 TU/e, BMIA & PMS, X-Ray Predevelopment, Erik Franken, 22-09-2004 3 1. Introduction Application Approach

4 TU/e, BMIA & PMS, X-Ray Predevelopment, Erik Franken, 22-09-2004 4 1.1. Application: Cardiac Electrophysiology Treatment of heart rhythm disorders 1.Insertion of EP catheters 2.Recording of intracardiac electrograms 3.Ablation of problematic spot, or blocking undesired conduction path

5 TU/e, BMIA & PMS, X-Ray Predevelopment, Erik Franken, 22-09-2004 5 1.2. X-ray guidance

6 TU/e, BMIA & PMS, X-Ray Predevelopment, Erik Franken, 22-09-2004 6 1.3. Project goal: finding the EP catheters Restrict to spatial context Focus on noise robustness No initial seed position 

7 TU/e, BMIA & PMS, X-Ray Predevelopment, Erik Franken, 22-09-2004 7 1.4. Algorithm steps ABC A.Detect local image features (ridges, blobs) B.Enhance local feature information C.The decision step: group image features to object of interest

8 TU/e, BMIA & PMS, X-Ray Predevelopment, Erik Franken, 22-09-2004 8 2. Local feature detection Background equalization Ridge detection Blob detection

9 TU/e, BMIA & PMS, X-Ray Predevelopment, Erik Franken, 22-09-2004 9 2.1. Background equalization Using morphological closing operation Original imageBackground imageBackground normalized image 

10 TU/e, BMIA & PMS, X-Ray Predevelopment, Erik Franken, 22-09-2004 10 2.2. Ridge detection Catheter is locally ridge-shaped. Profile function: Class of filters

11 TU/e, BMIA & PMS, X-Ray Predevelopment, Erik Franken, 22-09-2004 11 2.3. Ridge detection Orientations  Ridgeness Example We use steerable filters 

12 TU/e, BMIA & PMS, X-Ray Predevelopment, Erik Franken, 22-09-2004 12 2.4. Blob detection Based on second eigenvalue of the Hessian matrix 

13 TU/e, BMIA & PMS, X-Ray Predevelopment, Erik Franken, 22-09-2004 13 2.5. Local features are too unreliable Source imageLocal ridgeness …in case of noisy images 

14 TU/e, BMIA & PMS, X-Ray Predevelopment, Erik Franken, 22-09-2004 14 2.6. The importance of spatial context Local filter Context filter

15 TU/e, BMIA & PMS, X-Ray Predevelopment, Erik Franken, 22-09-2004 15 3. Context enhancement Introduction to tensor voting Steerable tensor voting Repeated tensor voting

16 TU/e, BMIA & PMS, X-Ray Predevelopment, Erik Franken, 22-09-2004 16 3.1. Tensor voting components Tensor voting Input: local feature data encoded in tensor field Model: voting field Operation: tensor communication Output: context enhanced tensor field …versus Political elections Input: people with the right to vote Model: electoral system Operation: collection of votes from the polling stations Output: the parliament (with the elected people)

17 TU/e, BMIA & PMS, X-Ray Predevelopment, Erik Franken, 22-09-2004 17 3.2. Encoding in tensor field 1 - 2 = orientation certainty 2 = orientation uncertainty For each pixel position, we have a tensor in which the local features are encoded. Graphical representation:

18 TU/e, BMIA & PMS, X-Ray Predevelopment, Erik Franken, 22-09-2004 18 3.3. Voting field Is a model for the continuation of line structures Most likely Least likely V(x,y)

19 TU/e, BMIA & PMS, X-Ray Predevelopment, Erik Franken, 22-09-2004 19 3.4. Tensor communication Voting field is used to let tensors vote for each other.  Amplification of smooth and elongated structures  Filling of gaps in structures V V V (x’,y’) (x,y)(x,y)(x,y)(x,y) (x,y)(x,y) (x’,y’) = sender and (x,y) = recipient

20 TU/e, BMIA & PMS, X-Ray Predevelopment, Erik Franken, 22-09-2004 20 3.5. Rotation of the voting field Tensor field rotation: By choosing an appropriate voting field, tensor voting can be written in a steerable form where

21 TU/e, BMIA & PMS, X-Ray Predevelopment, Erik Franken, 22-09-2004 21 3.6. Steerable tensor voting scheme Using steerability, tensor voting boils down to (e.g.) Consists of complex-valued convolutions More efficient with ,

22 TU/e, BMIA & PMS, X-Ray Predevelopment, Erik Franken, 22-09-2004 22 3.7. Example - input Source imageLocal ridgeness  

23 TU/e, BMIA & PMS, X-Ray Predevelopment, Erik Franken, 22-09-2004 23 3.8. Example - result  Context enhanced ridgeness * * * * * + + + + U 2 (x,y)= |U 2 | 

24 TU/e, BMIA & PMS, X-Ray Predevelopment, Erik Franken, 22-09-2004 24 3.9. Repeated tensor voting Tensor voting  thinning  tensor voting Result after first stepResult after second step 

25 TU/e, BMIA & PMS, X-Ray Predevelopment, Erik Franken, 22-09-2004 25 4. Catheter extraction Overview Step by step explanation on an example

26 TU/e, BMIA & PMS, X-Ray Predevelopment, Erik Franken, 22-09-2004 26 4.1. Overview

27 TU/e, BMIA & PMS, X-Ray Predevelopment, Erik Franken, 22-09-2004 27 4.2. Example image Source imageBackground equalized image

28 TU/e, BMIA & PMS, X-Ray Predevelopment, Erik Franken, 22-09-2004 28 Context enhanced ridgenessBlobness 4.3. Result of tensor voting (used as input)

29 TU/e, BMIA & PMS, X-Ray Predevelopment, Erik Franken, 22-09-2004 29 Local ridge maximaExtracted most salient paths 4.4. Extraction of paths

30 TU/e, BMIA & PMS, X-Ray Predevelopment, Erik Franken, 22-09-2004 30 Electrode candidatesExtracted catheter tips 4.5. Extraction of catheter tips

31 TU/e, BMIA & PMS, X-Ray Predevelopment, Erik Franken, 22-09-2004 31 4.6. Extension of catheter tips Selection of the best extension candidate for each tip. Result:

32 TU/e, BMIA & PMS, X-Ray Predevelopment, Erik Franken, 22-09-2004 32 5. Evaluation Evaluation questions Evaluation results

33 TU/e, BMIA & PMS, X-Ray Predevelopment, Erik Franken, 22-09-2004 33 5.1. Quantitative evaluation – questions Is there an added value of the tensor voting step? ? What is the robustness to noise? How feasible is extraction of tip, tip + additional segment, and entire EP catheter in clinical images?

34 TU/e, BMIA & PMS, X-Ray Predevelopment, Erik Franken, 22-09-2004 34 5.2. Quantitative evaluation – clinical images Low noiseHigh noise n catheters =103

35 TU/e, BMIA & PMS, X-Ray Predevelopment, Erik Franken, 22-09-2004 35 6. Conclusions and recommendations Conclusions Recommendations

36 TU/e, BMIA & PMS, X-Ray Predevelopment, Erik Franken, 22-09-2004 36 Application: Tensor voting makes EP catheter extraction more robust Detection of tip quite successful, detection of entire catheters still error-prone Algorithms still far too slow 6.1. Conclusions Context Enhancement methodology: Derived an efficient scheme for tensor voting Context enhancement methods will be useful for a lot of other (medical) image analysis problems

37 TU/e, BMIA & PMS, X-Ray Predevelopment, Erik Franken, 22-09-2004 37 6.2. Recommendations Application: The use of temporal information Parameter optimization with larger test set More efficient implementation Context Enhancement methodology: Include curvature Improve voting field Improve communication scheme Vote with other |m|-components

38 TU/e, BMIA & PMS, X-Ray Predevelopment, Erik Franken, 22-09-2004 38 Acknowledgements Prof. Paul van den Bosch, prof. Bart ter Haar Romeny Markus van Almsick, Peter Rongen Other colleagues at TU/e Other colleagues at PMS Family Friends

39 TU/e, BMIA & PMS, X-Ray Predevelopment, Erik Franken, 22-09-2004 39 Questions


Download ppt "TU/e, BMIA & PMS, X-Ray Predevelopment, Erik Franken, 22-09-2004 1 Context-Enhanced Detection of Electrophysiology Catheters in Noisy Fluoroscopy Images."

Similar presentations


Ads by Google