Global Navigation
Functional Tumor Imaging using Dynamic Contrast Enhanced Methodologies
- Prinicipal Investigator: Michael Knopp, M.D., Ph.D.
Segmentation and Registration of DCE-MR (Dynamic Contrast Enhanced - MR) images
- Co-PIs: Yootai Kim, Ph.D., Raghu Machiraju, Ph.D., Michael Knopp, M.D., Ph.D., Joel Saltz, M.D., Ph.D.
DCE-MRI techniques produce large datasets to capture the structure of internal organs non-invasively or locate extraneous growth and tumors. There is a need to segment and classify the results of the data-fitting process that the raw signal intensity data is subjected to. Segmentation has its challenges given the inability of thresholding techniques that can locate an entire region. On the other hand, the coherent geometry enables model-based approaches. These can perform better than region-growing methods. Although coherent, the regions-of-interest have at times large variation of the intensity or the DCE-MR analytic parameter.
We employed level set techniques to extract boundaries of tumors. The level set method requires the use of an edge-strength map that can identify strong boundaries that serve as the initial boundary. Much interaction is needed between the image analysis algorithms and the radiology protocols to help obtain the correct selection of pertinent edges. Fig 4 shows an overview of our procedure.
Fig 4. Overview of the level-set procedure.
A 3D level set method implementation was also realized. The results are being validated by Dr. Knopp and his associates. Since we have not completed validation of our results, no publications exist in the medical imaging literature. However, related work has appeared in computer science literature. Additional work has been conducted in co-registration of images in successive time frames to correct patient movement. A well-known algorithm from Prof. Knopp's group has been re-implemented and corrected. The algorithm uses multi-resolution Laplacian pyramids to detect rigid motion.