55:247 Image Analysis and Understanding

Project Listing - Spring 2001


IP = Image processing part

AI = Artificial intelligence part


 

 

Steve Mitchell

Image Data IVUS sequences the Fusion Project Pig IVUS and IVUS sequences provide by Dijkstra at the LUMC
Goals Develop a methodology for the segmentation of lumen and lamina borders of IVUS frames
Methods I, IP Dynamic Programming and Mahalanobis edge weighting for probable borders.
Methods II, AI Active Shape Model extended in to include adjacent frames for prior knowledge about plausible IVUS border shapes
Validation In the IVUS sequences provided by the LUMC, the data was hand segmented by an expert. Training will be applied to one half of the series and validation against manually traced borders will be applied to the other half.

Jianchun He

Image Data CT images of human inner ear
Goals To automatically (semi-automatically) detect the major objects of interest, including cochlea, IAC(inner auditory canal), Ossicles, semi-circular canals and utricle etc.
Methods I, IP Volume growing approach possiblyusing 3D edges for better delineatiuon of borders
Methods II, AI 3D location of individual objects, their shapes, and similar used as a priori knowledge
Validation 6 volumes with manually traced objects of interest. Assess segmentation accuracy in comparison to manualy defined independent standard. If some learning is used, the leave-one-out validation approach will be used.

Kyoung-Mi Lee

Image Data Breast Cancer Cell (cytological) images
Goals Develop a methodology for cell detection, segmentation and classification based on shape features
Methods I, IP Memory-efficient template matching (iterative GHT) for detection and for segmentation
Methods II, AI Neural network for pattern classification - benign/malignant, especially suitable in on-line modeling of non-stationary process
Validation 800 training cells and 300 testing cells AND using leave-100-out approach.

John Dill

Image Data 3D segmentations of the uterus
Goals Develop a method for registering the uterus
Methods I, IP Generation of a point based model from the surface of the segmentation
Methods II, AI Principal Component Analysis for generation of surface model Validation : Use a known deformation with a known correspondence of the segmentations and see how well the registration matches the transformation
Validation  

 

Tim Irwin

Image Data 4D ultrasound data sets of the left ventricle, 6 normals and 7 or so stenotic aortic valve patients taken preop, postop, and 6 months postop; 6 or so balloon data sets used in volume validation.
Goals To develop an automated method to segment the epicardium of the left ventricle
Methods I, IP Suboptimal 3D graph searching
Methods II I, AI Use the position of the endocardium as priori information
Validation Compare the epicardial models of those found by the above method to those modeled by an expert. Calculate the border positioning errors, and possibly volume errors, to give a quantitative error measurement.

Seonho Shin

Image Data CT images of heart(3D heart image)
Goals Define Coronary Artery of heart
Methods I, IP ??
Methods II, AI ??
Validation ??

Xiaoxu Han

Image Data Cardiac MR images and image sequences
Goals Compare PCA and ICA performance
Methods I, IP

Incorporate ICA in the existing AAM/AAMM framework.

Methods II, AI Implement ICA approach to replace the currently used PCA approach in AAM and AAMM modeling
Validation Compare 2D and 2D+T AAM - results between PCA/ICA/manual using standard comparison indices used in the past.

Junghyun Byun

Image Data CT, MRI, and SPECT images
Goals Improve a manual and an automated coregistrations
Methods I, IP Rigid body transformation rotation method for manual coregistration.
Methods II, AI Mutual information method for automated coregistration
Validation Using multimodality brain images, compare the coregistration routine by the above methods to commercially used one.

Ryan Long

Image Data CT volumes of lungs
Goals Develop a reliable 3D tree representation
Methods I, IP Identify primary airway trees from volumetric CT images together with some approximate bifurcation points
Methods II, AI develop tree-structure-based representation allowing node identification, tree traversal, basic voxel-based measurements
Validation Comparison with manual tracings using quantitative indices

Igor Okulist

Image Data 4D ultrasound data sets of the left ventricle, 6 normals and 7 or so stenotic aortic valve patients taken preop, postop, and 6 months postop;
Goals To develop automated method to detect heart twisting motion. Find out the "normal" twisting pattern and compare patient data against normal.
Methods I, IP

Suboptimal 3D graph searching, pattern matching.

Methods II, AI Use the gray level values, position of the endocardium and valve location as priori information.
Validation Compare the heart twisting pattern to the one found in MR studies.

Mark Olszewski

Image Data

3D IVUS data (2D sequence)

Goals Develop a rudimentary 3D approach to IVUS segmentation
Methods I, IP 3D - considering temporal sequence context
Methods II, AI training from examples
Validation against manually defined independent standard

David Rodriguez

Image Data

Sequences of MR Images of arteries that (may) have plaque

Goals Segmentation of vesssel lumen, plaque and wall
Methods I, IP Edge based segmentation, generalized hough transform
Methods II, AI Graph search dynamic programming with automated cost functions
Validation Provided images, manually traced. Some of the images will be used for training and some for validation

 


Last Modified: 2001

[Go Back]