![[Go Back]](../IMAGES/next.gif){kind=small}
Marek
| Image Data | Artificial image data Angiography data sets, IVUS, ICUS |
| Goals | Develop a self-learning system that will generate appropriate cost functions from manually traced borders - followed by graph-search based border detection |
| Methods I, IP | Matching filters, edge detectors, matching etc. to get border features |
| Methods I, AI | NN and/or fuzzy to get local costs |
| Validation | In 20 images from each set not used for training, calculate rms border positioning errors and area errors. |
Hanli
| Image Data | Milled slices of a tooth |
| Goals | Develop a program for 3D tooth surface detection |
| Methods I, IP | Region growing/pixel classification; if the "inside border is the final border, border detection will have to follow the region growing |
| Methods I, AI | RGB supervised class/clustering/fuzzy clustering |
| Validation | Manually trace the desired borders in 20 randomly selected frames, determine rms border positioning error, area error |
Richard
| Image Data | Transthoracic LV ultrasound |
| Goals | Determine realistic segmentation of LV from short & long views using a feedback self-adjusting process |
| Methods I, IP | Construction of 3D LV surface model from manual tracings in individual images using optimal surface fitting, mapping the model back in the image data, detecting locations of mismatch and automated feedback-based modification of traced borders |
| Methods I, AI | Genetic algorithm approach to optimal surface fitting |
| Validation | Validate against independently traced surfaces in at least 5 LV studies |
Paul
| Image Data | IVUS, 3D |
| Goals | Develop a 3D segmentation for IVUS pullback sequences |
| Methods I, IP | Develop a multi-stage approach to segment approximately in stage 1 and improve the borders in stage 2 using different information than was already used in stage 1 |
| Methods I, AI | 3D snakes will be optimized using some intelligent way |
| Validation | 80 IVUS images, 10 slices each, comaparison with manual tracings |
Zeng
| Image Data | IVUS - tissue data |
| Goals | Plaque tissue classification |
| Methods I, IP | Wavelet-based texture description |
| Methods I, AI | NN feature selectioon/compression followed by NN-classification |
| Validation | 40 (or so) DeJong-traced IVUS images against manual classification |
Sung
| Image Data | B/W images of 2D parts, 1 part per image, 20 different part kinds |
| Goals | Part classification |
| Methods I, IP | Fourier contour descriptors |
| Methods I, AI | Linear and one other NN classifier |
| Validation | At least 10 images per part kind that were not used for training and on-line presentation of program functionality under the class camera |
Yefei
| Image Data | Rotationally symmetric 3D objects on real-life background |
| Goals | Detect the long axis and true contour of the object(s) |
| Methods I, IP | Double Hough transform |
| Methods I, AI | Smart color-based selection of true long axes and contours |
| Validation | In real images based on success/failure, minimum 50 images, and present the functionality on real data during the final presentation |
Carmen
| Image Data | IVUS and brachial ultrasound data sequences |
| Goals | Develop a 3D snake and PDM methods for surface detection |
| Methods I, IP | Snake, PDM, 3D shape description |
| Methods I, AI | Assess the feasibility/probability of intermediate 3D surface shape and intelligently modify the snake/PDM segmentation parameters to achieve a good solution |
| Validation | Against manual tracing and/or graph searching, in brachial arteries by analyzing the diameter function |
Weidong
| Image Data | Brachial ultrasound |
| Goals | Develop a near-real-time, robust quality control and correction approach for brachial border detection in short and long time sequences |
| Methods I, IP | Border division in sections, analysis along t-axis, matching with cardiac cycle - expected diameter function |
| Methods I, AI | Correlation, feature extraction and detection of unlikely phases, NN-based (or other intelligent) outlier detection |
| Validation | Approximately 50 subjects from the Muscatine study |
Xiang
| Image Data | Artificial image data Angiography data sets, IVUS, ICUS |
| Goals | Develop a self-learning system that will generate appropriate cost functions from manually traced borders - followed by graph-search based border detection |
| Methods I, IP | Edge detectors, edge patterns |
| Methods I, AI | Representing the individual profiles as features (using, e.g., location and amplitude of max. edge considering direction - then apply a radial basis function NN = 2 stages: clustering on the front end, followed by nonlinear classification |
| Validation | In 20 images from each set not used for training, calculate rms border positioning errors and area errors |
Tony
| Image Data | Follicle ultrasound |
| Goals | Develop a fully automated intelligent approach to follicle border detection |
| Methods I, IP | Low resolution image-watershed segmentation, identification of follicles, identification of the inside walls, followed by graph-search identification of the outside walls |
| Methods I, AI | Intelligent detection of common walls between follicles using feature-based classification of follicle border segments, applying "common-border-specific" cost function in these segments |
| Validation | In approximately 49 follicle images against manual tracing |
Last Modified: February 7, 1997