MRI KNEES SEGMENTATION

Cheng, R., Jackson, J. N., McCreedy, E. S., Gandler, W., Eijkenboom, J. J. F. A., van Middelkoop, M., McAuliffe, M.J., & Sheehan, F. T. (2016, March). Patellar segmentation from 3D magnetic resonance images using guided recursive ray-tracing for edge pattern detection. In SPIE Medical Imaging (pp. 97880C-97880C). International Society for Optics and Photonics.

Through MIPAV plug-ins, BIRSS has collaborated with NIH IRP scientists on dozens of projects, helping to advance their work through agile development and a close working relationship. One such collaboration was the research BIRSS team members performed in concert with Dr. Frances Gavelli of National Institutes of Neurological Disorders and Stroke. BIRSS developers worked very closely with Dr. Gavelli and her team to design and implement as a MIPAV plug-in an automatic segmentation methodology for the patellar bone, based on 3D gradient recalled echo and gradient recalled echo with fat suppression magnetic resonance images. Constricted search space outlines are incorporated into recursive ray-tracing to segment the outer cortical bone. A statistical analysis based on the dependence of information in adjacent slices is used to limit the search in each image to between an outer and inner search region. A section based recursive ray-tracing mechanism is used to skip inner noise regions and detect the edge boundary. The proposed method achieves higher segmentation accuracy (0.23mm) than the other methods with the average dice similarity coefficient of 96.0% (SD 1.3%) agreement between the auto-segmentation and ground truth surfaces.

The average dice similarity coefficient was excellent, demonstrating 96.0% (SD 1.3%) agreement between the surfaces.

The bone and cartilage models were segmented in MIPAV and then using our registration process they can be animated to demonstrate cartilage contact patterns.

We are expanding this project to look at the shape of the scapula and to expand into motion.

1. The accuracy of proposed algorithm is 50% better than what has been reported for previous algorithms [1, 5]. Dice similarity coefficients well above the 80% reported for previous knee segmentation algorithm [7].
2. The decaying and growing ratio used to create the inner and outer search contours captures the shape variation into narrow band regions, save large amount of recursion based computation.
3. Recursive ray-tracing based edge pattern detection algorithm on both GRE and GRE-FS images enhances the algorithm’s ability to accurately detect the outer cortical boundary, and overcome the noise and intensity related issues on MRI GRE image alone.
4. The proposed edge patterns detection is a novel approach to guide MRI knees segmentation.