Genetic Algorithms in Search, Optimization and Machine Learning
Genetic Algorithms in Search, Optimization and Machine Learning
3D-interaction techniques for planning of oncologic soft tissue operations
GRIN'01 No description on Graphics interface 2001
Segmentation of medical images using a genetic algorithm
Proceedings of the 8th annual conference on Genetic and evolutionary computation
| Hi-index | 0.00 |
The aim of this study was to find a semi-automatic method of bone segmentation on the basis of computed tomography (CT) scan series in order to recreate corresponding 3D objects. So, it was crucial for the segmentation to be smooth between adjacent scans. The concept of graphics pipeline computing was used, i.e. simple graphics filters such as threshold or gradient were processed in a manner that the output of one filter became the input of the second one resulting in so called pipeline. The input of the entire stream was the CT scan and the output corresponded to the binary mask showing where a given tissue is located in the input image. In this approach the main task consists in finding the suitable sequence, types and parameters of graphics filters building the pipeline. Because of the high number of desired parameters (in our case 96), it was decided to use a slightly modified genetic algorithm. To determine fitness value, the mask obtained from the parameters found through genetic algorithms (GA) was compared with those manually prepared. The numerical value corresponding to such a comparison has been defined by Dice's coefficient. Preparation of reference masks for a few scans among the several hundreds of them was the only action done manually by a human expert. Using this method, very good results both for trabecular and cortical bones were obtained. It has to be emphasized that as no real border exists between these two bone types, the manually prepared reference masks were quite conventional and therefore charged with errors. As GA is a non-deterministic method, the present work also contains a statistical analysis of the relations existing between various GA parameters and fitness function. Finally the best sets of the GA parameters are proposed.