segmentation of ct angiography based on a combination of segmentation methods university of west...

16
Segmentation of CT angiography based on a combination of segmentation methods University of West Bohemia in Pilsen Czech republic Ing. Ivan Pirner Ing. Miroslav Jiřík Ing. Miloš Železný, Ph.D.

Upload: marilyn-atkins

Post on 30-Dec-2015

214 views

Category:

Documents


0 download

TRANSCRIPT

Segmentation of CT angiography basedon a combination of segmentation methods

University of West Bohemia in PilsenCzech republic

Ing. Ivan PirnerIng. Miroslav Jiřík

Ing. Miloš Železný, Ph.D.

Contents:

medical imagessegmentationtechniques thresholding edge detection region growingprocessconclusion

PRIA 2010, Saint Peterburg 8th December 2 of 16

Segmentation of CT angiography Ing. Ivan Pirner

Sources of medical images:CT, MRI, USG, X-ray, PET, etc.

Example:

CT (picture source: http://www.jnch.nic.in)

Contents:

medical imagessegmentationtechniques thresholding edge detection region growingprocessconclusion

PRIA 2010, Saint Peterburg 8th December 3 of 16

Segmentation of CT angiography Ing. Ivan Pirner

Image properities:

Image F(i,j) is a 2-dimensional array of pixels. Each pixel on the position i,j is characterized by its value – the density. The density is a non-negative integer value belonging to a known finite range, usually 8 or 16bit.

Remark: Most of the medical images are grayscale. Methods used in this work may be generalized for color images.

Contents:

medical imagessegmentationtechniques thresholding edge detection region growingprocessconclusion

PRIA 2010, Saint Peterburg 8th December 4 of 16

Segmentation of CT angiography Ing. Ivan Pirner

Definition:

Segmentation = labeling the pixels of an image in such way, that the labels have a strong correlation with real objects observed in the image.

Purposes:removing unwanted regions of datacounting regionsmeasuring regions

Contents:

medical imagessegmentationtechniques thresholding edge detection region growingprocessconclusion

PRIA 2010, Saint Peterburg 8th December 5 of 16

Segmentation of CT angiography Ing. Ivan Pirner

Useful segmentation techniques:

thresholding(threshold value?)

edge-based methodsedge image thresholding (threshold value?)

region-based methodsregion growing (homogeneity rule?)

graph cut segmentationenergy minimization (model?, parameters?)

Contents:

medical imagessegmentationtechniques thresholding edge detection region growingprocessconclusion

PRIA 2010, Saint Peterburg 8th December 6 of 16

Segmentation of CT angiography Ing. Ivan Pirner

Thresholding:

Thresholded image is a binary image G(i,j), where

G(i,j) = 1 if F(i,j) > T andG(i,j) = 0 otherwise

i, j – spatial coordinatesF(i,j) – original image pixels

Conditions of use:

Object to be segmented has other pixel values range than its background.

The output segmentation needs often postprocessing, many dummy segments due to image noise.

The threshold value must be chosen properly.

Contents:

medical imagessegmentationtechniques thresholding edge detection region growingprocessconclusion

PRIA 2010, Saint Peterburg 8th December 7 of 16

Segmentation of CT angiography Ing. Ivan Pirner

Example:

left: original CT sliceright: double thresholded image

Contents:

medical imagessegmentationtechniques thresholding edge detection region growingprocessconclusion

PRIA 2010, Saint Peterburg 8th December 8 of 16

Segmentation of CT angiography Ing. Ivan Pirner

Edge detection:

The purpose is to find places in the image with significant discontinuities in the image function (big differences in values between neighborning pixels).

There are many similar operators, which approximate the first derrivative of the image function. We used Sobel’s operator.

first two of four Sobels’ operators (the basic mask is rotating)

Contents:

medical imagessegmentationtechniques thresholding edge detection region growingprocessconclusion

PRIA 2010, Saint Peterburg 8th December 9 of 16

Segmentation of CT angiography Ing. Ivan Pirner

Proceeding:

The edge image is set as an output of a 2D-correlation between the mask and the original image. Results for different directions are summed and the output image then thresholded into a binary image.

Conditions of use:

The seeked region must be bordered by a “sharp” edge.

The threshold value must be chosen properly.

Contents:

medical imagessegmentationtechniques thresholding edge detection region growingprocessconclusion

PRIA 2010, Saint Peterburg 8th December 10 of 16

Segmentation of CT angiography Ing. Ivan Pirner

Example:

left: original CT sliceright: edge image

Contents:

medical imagessegmentationtechniques thresholding edge detection region growingprocessconclusion

PRIA 2010, Saint Peterburg 8th December 11 of 16

Segmentation of CT angiography Ing. Ivan Pirner

Region growing:

In this part we used the modified confidence connected algorithm:

1) Set a seed (1 or multiple points) and make it the current region.2) Find all pixels neighboring upon the current region.3) For all of this neighboring pixels decide, whether they fulfill the

homogeneity criteria, if yes, append them to the current region.4) If no points added in step 3, END, else GOTO 2.

We chose as homogeneity criteria K(p) a double inequality:

K(p) = 1 if p>T_min && p<T_maxK(p) = 0 otherwise

p – tested pixelT_min – chosen minimum valueT_max – chosen maximum value

Sketch:

region growing (image source: http://www.cs.cf.ac.uk)

Conditions of use:

Seeked region must be homogenous.The seed set must be chosen within the region.

Contents:

medical imagessegmentationtechniques thresholding edge detection region growingprocessconclusion

PRIA 2010, Saint Peterburg 8th December 12 of 16

Segmentation of CT angiography Ing. Ivan Pirner

Proceeding:

Contents:

medical imagessegmentationtechniques thresholding edge detection region growingprocessconclusion

PRIA 2010, Saint Peterburg 8th December 13 of 16

Segmentation of CT angiography Ing. Ivan Pirner

original image

edge imagesegmented bones+vessels

bone image

region growing edge detection

morphologicaloperations

segmented vessels

subtraction

Visualization of the 3D data:3D model using volume rendering:

Contents:

medical imagessegmentationtechniques thresholding edge detection region growingprocessconclusion

PRIA 2010, Saint Peterburg 8th December 14 of 16

Segmentation of CT angiography Ing. Ivan Pirner

Conclusion:

The CT angiography vessel segmentation may be made using “simple” methods when combining them together.

Parameters of each of the used segmentation methods can be easily interpreted and either directly determined or experimentally measured.

Future work:

Graph cuts could bring more precise results, although we need to determine a proper model and estimate its parameters.

Contents:

medical imagessegmentationtechniques thresholding edge detection region growingprocessconclusion

PRIA 2010, Saint Peterburg 8th December 15 of 16

Segmentation of CT angiography Ing. Ivan Pirner

Thank you for your attention.

Contents:

medical imagessegmentationtechniques thresholding edge detection region growingprocessconclusion

PRIA 2010, Saint Peterburg 8th December 16 of 16

Segmentation of CT angiography Ing. Ivan Pirner