Host: The Institute of Image Electronics Engineers of Japan
Name : Reports of the 258th Technical Conference of the Institute of Image Electronics Engineers of Japan
Number : 258
Location : [in Japanese]
Date : October 28, 2011 -
This paper presents a method for optimizing and sharing parameters of Level Set Methods (LSMs). We used Genetic Algorithm (GA) of evolutional learning for optimizing parameters of LSM as a deformable shape model. However, our former method required the information of Ground Truth (GT) for each target image. Therefore, the method was impossible to apply clinical images without GT. In this paper, we focus on a trade-off problem of sharing and separating parameters used for clinical images without GT. We applied our method to head Magnetic Resonance (MR) images of 52 patients from 30s through 70s classified into two groups under the instruction by a radiologist. For the result, the mean extraction accuracy is improved to 84.7% because we can share parameters of LSM in each group. However, iterations of contours are sensitive to the information of anatomical structures and luminance properties in each target image. We consider that the adjustment of iterations is required after the optimization.