GE Project

Views:
 
Category: Entertainment
     
 

Presentation Description

No description available.

Comments

Presentation Transcript

PowerPoint Presentation:

Amin Younis Semi-Automatic 3D Segmentation of The Left Ventricle In MSCT Presented By Supervised By GE-Healthcare Mr. Gil Kovalski Project

PowerPoint Presentation:

Clinical Approch Multislice CT analysis is useful to comprehend the left ventricular motion and volumes. Using 3D segmentation, important parameters can be calculated accuretly, such as Ejection-Fraction which correlates with the risk of morbidity and mortality in different types and stages of heart disease. Furthermore, segmentation is a powerful tool for detecting an abnormality in the regional wall motion which suggests ischemic heart disease. CT image

PowerPoint Presentation:

Clinical Approch

PowerPoint Presentation:

LV Segmentation What is it ? Implemented by the extraction of the boundaries of the left ventricle (epicardium and endocardium).

PowerPoint Presentation:

Reality vs Model A comparison is needed between the observed real LV borders and the segmented results or 3D model. No more than 84% correlation was achieved when using fully automatic segmentation softwares. Market softwares were considered efficient due to the very short running time of around ~ 30 seconds.

PowerPoint Presentation:

Reality vs Model

PowerPoint Presentation:

Reality vs Model

PowerPoint Presentation:

Segmentation Methods Thresholding Snakes Live Wire

PowerPoint Presentation:

Project Goals 1. Extracting the Epicardium and the Endocardium by segmentation. 2. Reducing running time. 3. Modeling the segmented parts in a 3D space.

PowerPoint Presentation:

General Process Overview

PowerPoint Presentation:

Stage1. Control Points Selection The segmentation process requires good initial interactive manual selection of control points on the midcardium 1- better accuracy 2- robustness 1- time consuming 2- subjection to inter and intra observer variability Advantages: Disadvantages: example

PowerPoint Presentation:

Stage2. LV Segmentation During the segmentation process two major methods are used : 1. Ellipsoid Fitting Algorithm 2. Active Contours Based Method

PowerPoint Presentation:

Ellipsoid Fitting Algorithm Assumption : The left ventricle shape can be desribed as an ellipsoid. x , y , z - point coordinates in space k1...k10 - ellipsoid coefficients

PowerPoint Presentation:

Active Contours Based Method Gradient Force Geometrical Force attracts the point to the image boundaries epi and endo cardium keeps the points together in an ellipsoid shape calculated by the image pixels gradient calculated by the distance from the fitted ellipsoid controlled point movement

PowerPoint Presentation:

Stage3. Interactive Manual correction Sometimes the segmetation results need further correction. So user can correct the unusual points manualy and re-run the aquisition. A slice which needs correction due to papillary muscles

PowerPoint Presentation:

First Conclusions 1. Good Segmentation of the basal and the central regions of the LV in the MSCT data was accomplished. 2. The ellipsoid fitting for the Endocardium suffers from fitting inaccuracies in the apical region of the LV. 3. Though running time was reduced, it still needs further reduction since its about 25 min.

PowerPoint Presentation:

Fixing The Apex Improving the segmentation of the apex region was done by applying image pre-processing to specific range of slices using higher contrast filter in order to exclude out the papillary muscles.

PowerPoint Presentation:

Zoom-In Final Epi Extraction

PowerPoint Presentation:

Epi and Endo in Long Axis View

PowerPoint Presentation:

Long Axis Short Axis 3D Validation

PowerPoint Presentation:

Conclusions 1. Good Segmentation of the basal and the central regions of the LV in the MSCT data was accomplished with 3D validation. 2. The ellipsoid fitting inaccuracies in the apical region of the LV were fixed by image pre-processing with higher contrast filter applied to specific range of slices before the main run. 3. Running time was reduced from 25 min to 18-21 min by switching some "for" loops into "recursive" ones and therefore reducing the MATLAB code complexity. 4. Applying the "live wire" method on MSCT data may lead to better results in a considered running time. (worth trying)

PowerPoint Presentation:

Thank you

PowerPoint Presentation:

Back Error produced by wrong initial points

PowerPoint Presentation:

CT Imaging Back

PowerPoint Presentation:

Back Good segmentation of the basal and central regions

PowerPoint Presentation:

Back Poor segmentation of the apex region

PowerPoint Presentation:

Back Final Epi Extraction

authorStream Live Help