Heart Visualization from MRI Marek Zimnyi Julius Parulek Faculty - - PowerPoint PPT Presentation

Heart Visualization from MRI

Marek Zimányi Julius Parulek

Faculty of Mathematics, Physics and Informatics Comenius University, Bratislava

and International Laser Center Bratislava

# 2

Marek Zimányi, DAI CU

Goal of this work

n Create Heart surface model

from MRI data set

n Input MRI data set

# 3

Marek Zimányi, DAI CU

Three main problems:

n MRI Image Enhancement n Heart segmentation n Surface modeling from contours

# 4

Marek Zimányi, DAI CU

Load DICOM Data

n Data

n MRI – Dicom FILES (not parallel too)

n Loading using DCMTK n Computing time period for every slice and

group slices with the same period value

DataLoad Enhancement Segmentation Modeling

# 5

Marek Zimányi, DAI CU

Image Enhancement

n Enhance contrast and histogram equalization n Bias correction (Estimation of inhomogeneities)

n Work in progress

DataLoad Enhancement Segmentation Modeling

# 6

Marek Zimányi, DAI CU

Image Enh. - Bias correction

n Bias correction (Estimation of inhomogeneities)

DataLoad Enhancement Segmentation Modeling

# 7

Marek Zimányi, DAI CU

Image Enh. - Bias correction

Bias field Measured Restored

# 8

Marek Zimányi, DAI CU

POSSIBLE USE SENARIOS

n

Image Preprocessing

n

• riginal image -> IntensityCorrector -> preprocessed image

n

Bias Field Estimation

n

• riginal image (or preprocessed image) -> BiasFieldEstimator ->

coefficients of the bias field estimate

n

Bias Correction

n

• riginal image + the coefficients of the bias field estimate (from

BiasFieldEstimator) -> BiasCorrector -> bias field corrected image

n

Bias Image Generation

n

the coefficients of the bias field estimate + result image dimension and size -> BiasImageGenerator -> bias image

DataLoad Enhancement Segmentation Modeling

# 9

Marek Zimányi, DAI CU

Heart Segmentation

n Small changes – median filter, sharpen etc … n Then segmentation

DataLoad Enhancement Segmentation Modeling

# 10

Marek Zimányi, DAI CU

Heart Segmentation

n Canny/Deriche n than Snake

DataLoad Enhancement Segmentation Modeling

# 11

Marek Zimányi, DAI CU

Heart Segmentation

DataLoad Enhancement Segmentation Modeling

# 12

Marek Zimányi, DAI CU

Heart Segmentation - next

n Automatic segmentation n Create heart contour when ventricle(s)

contour is known

DataLoad Enhancement Segmentation Modeling

# 13

Marek Zimányi, DAI CU

Heart Segmentation

n Our added value:

n Add value for extracted pixel of contour, “how

sure we are that it is a contour point”

n Segmetation of heart when ventricles is known

DataLoad Enhancement Segmentation Modeling

# 14

Marek Zimányi, DAI CU

Heart Modeling

n Input: contours n Ouput: Surface model

DataLoad Enhancement Segmentation Modeling

# 15

Marek Zimányi, DAI CU

Heart Modeling by Implicit Surfaces

n Set of points { c1, c2, … ck } - contour n Set of constraints { h1, h2, … hk } n f(ci)= hi, n Minimization of energy:

DataLoad Enhancement Segmentation Modeling

# 16

Marek Zimányi, DAI CU

Heart Modeling by Implicit Surfaces

DataLoad Enhancement Segmentation Modeling

n Equestion E can be solved using radial basis

functions

n ci is localization of points, di are weights

and P(x) if polynomial of deg 1

# 17

Marek Zimányi, DAI CU

Heart Modeling by Implicit Surfaces

DataLoad Enhancement Segmentation Modeling

n f(ci)= hi, than n Solving by symmetric LU

decomposition

# 18

Marek Zimányi, DAI CU

Heart Modeling by Implicit Surfaces

n Problems:

n Correct setting of constrains n Contours don’t have to intersect

• points with constrain value 0

can be in the object

DataLoad Enhancement Segmentation Modeling

# 19

Marek Zimányi, DAI CU

Heart Modeling by Implicit Surfaces

n Solution:

n Add new contours of L/R ventricle as an

interior of heart

DataLoad Enhancement Segmentation Modeling

# 20

Marek Zimányi, DAI CU

Heart Modeling

n Our added value:

n Create mechanism for creating implicit surface

when points with constrain value 0 can be in the object.

DataLoad Enhancement Segmentation Modeling

# 21

Marek Zimányi, DAI CU

Next work

n Finnish correct setting of constrains fo implicit

surface generation

n (Semi)Automatic segmentation of heart n Add motion info to segmentation

# 22

Marek Zimányi, DAI CU

Literature

n

Jorgen Ahlberg, Active Contours in Three Dimension, research report, 1996

n

Zhukov et al, Dynamic Deformable Models for 3D MRI Heart Segmentation

n

Sorgel W., Vaerman V., Automatic heart localization from a 4D MRI dataset

n

Majcenic Z., Loncaric S., Algorithm for spatio-temporal hear segmentation

n

Uschler M., Image-Based verification of parametric models in heart- ventricle volumetry, Graz 2001,

n

Cipolla R., Giblin P., Visual Motion of Curves and Surfaces, book

n

Rucker D., Segmentation and Tracking in Cardiovascular MR Images using Geometrically Deformable Models and Templates, PhD work 1997

n

M-P Jolly, N.Duta, G F-Lea, Segmentation of Left Ventricle in Cardiac MR Images, ICCV 01

n

Greg Turk, J F O’Brien, Shape Transformation Using Variatonal Implicit Functions, Siggraph’99