![[Go Back]](../IMAGES/next.gif){kind=small}
IP = Image processing part
AI = Artificial intelligence part
Deniz Bilgen
| Image Data | CT images of lungs/ segmented bronchial trees. |
| Goals | Develop a methodology for centerline detection in intrathoracic airway trees. |
| Methods I, IP | 3D skeletonization, 3D connectivity. |
| Methods I, AI | Traversing a 3D graph, identifying bifurcations. |
| Validation | 3D trees - bronchial, at least 5 volumes of in vivo images plus phantoms, assess sensitivity/specificity of branch point identification, as well as positioning accuracy with respect to manual analysis. |
Hans Johnson
| Image Data | CT images of human head.. |
| Goals | To automatically (semi-automatically) detect the 5 major suture lines of the skull {Saggital, L-R coronal, L-R lambdoid}. |
| Methods I, IP | Unwrap the spherical projection of the skull surface into a 2D image, and use Linear Elastic Transformations to match a template image with the target. |
| Methods I, AI | Implement 2D graph search on 2D projections of each of the sutures to find X-Y coordinates of 3D line. Follow this by projecting this back into the 3D volume. |
| Validation | Quantitatively compare automatically detected suture lines with manually determined suture lines in at least 10 subjects. |
Yin Peng
| Image Data | CT images of pelvis |
| Goals | Develop a methodology for segmentation of soft tissue objects. |
| Methods I, IP | PDM in 2D to represent shape of the tissue objects and its relationship with respect to the body outline. |
| Methods I, AI | Use a) graph representation of your objects in 2D, then b) graph representation in 3D. Train in at least 5 volumes. |
| Validation | In at least 5 volumetric data sets, calculate rms border positioning errors and volume errors. |
Jaehoon Seol
| Image Data | |
| Goals | |
| Methods I, IP | |
| Methods I, AI | |
| Validation |
Rob Stefancik
| Image Data | Contrast Enhanced MRA Images of lower abdomen and upper legs. |
| Goals | Develop a methodology for segmentation and labelling of vascular tree. |
| Methods I, IP | Mathematical morphology operators for tree graph generation. (Bounded, iterative region growing) |
| Methods I, AI | Use vascular atlas as a priori knowledge for vessel identification and segmentation. |
| Validation | In at least 10 MRA data volumes, perform segmentation algorithm and qualitatively assess anatomical accuaracy of the resulting vascular trees. Determine the method's sensitivity/specificity in large-enough venous and arterial segments. |
Siying Yang
| Image Data | CT images of brain |
| Goals | Cross-modality registration using mutual information |
| Methods I, IP | Develop/modify/implement 3D warping methodology. |
| Methods I, AI | Use existing mutual information objective criteria, and develop at least two other objective functions based on mutual information. |
| Validation | In at least 5 volumetric data set pairs, determine the warping accuracy in user-defined landmarks (at least 20/volume), compare the performance for the 3 or more designed objective functions. Experiment with initial mutual position of the two (MR&CT) volumes. |
Juerg Tschirren
| Image Data | Brachial ultrasound image sequences + Doppler image data. |
| Goals | Determine the envelope function of the blood flow velocity diagram and use it for determination of "current" blood flow velocity. |
| Methods I, IP |
Edge detection for determining the envelope function. Determination of the horizontal baseline. Possibly graph searching or snake segmentation for the envelope. |
| Methods I, AI | Shape recognition of the envelope function (for validation purpose). Optical character recognition of the scale information. |
| Validation | At least 10brachial ultrasound image sequences. Verification with gated and with ungated sequences as well. Validation against manually obtained data. |
Shaolong Wang
| Image Data | IVUS image sequences |
| Goals | Develop a methodology for longitudinal and cross-sectional IVUS segmentation. |
| Methods I, IP | Graph searching & possibly simultaneous graph searching for border pairs. |
| Methods I, AI | Usage of combined information from longitudinal and cross-sectional images to achieve reliable segmentation overall. |
| Validation | In 10 IVUS sequences, validate against manually traced contours in selected longitudinal and cross-sectional images, at least 50 frames and 20 longitudinal images, randomly selected. |
Li Zhang
| Image Data | EBCT thorax images |
| Goals | Develop a methodology for segmentation of lung lobes. |
| Methods I, IP | Suboptimal 3D graph searching or 3D balloons for lobar surfaces |
| Methods I, AI | Use some 3D model of the lobes and lungs as a priori information |
| Validation | In 5 volumetric lung data sets, calculate rms border positioning errors and volume errors. |
Last Modified: February 7, 1997