Résumé :
|
Communication n° 460. Introduction : We present here the preliminary results of an automatic segmentation method of soft tissues in the upper or lower segment. Quantitation of muscle volume and of muscle/fat/connective tissue ratios is essential for the characterization of muscular disease natural progression and response to treatment. Manual analysis is long and cumbersome, and is not suitable to the evaluation of routine clinical protocols. Objective : In this context we are developing a method to automatically measure these parameters on axial NMR images. Methods : To perform the segmentation and extract the useful parameters, we pursue a sequential approach. First the bones, the subcutaneous and the muscular tissues with the inter-muscular fat and fascia are extracted. Clustering and morphological methods are used to achieve this and the result is then refined using a deformable model. In the second step, limb segment is segmented into muscle groups. In some instances, the lack of visible fascia led us to introduce additional information about the shape to be detected. A model of the muscle structure is registered to the image and is used as initialization of a level set deformable model. The last step concerns the extraction of specific muscles, using the previous result. Results : The extraction of bones, subcutaneous and the muscular tissues, as well as the estimate of the ratio Fat/Muscles are robust and provide anatomically consistent results. This provides the total muscular volume and the fat/muscle ratio. We are currently working on the next steps. First results obtained on the extraction of separate muscle groups look promising and confirm the interest of our approach. Conclusion : At the moment, all the experiments have been performed on healthy subjects. The next step will be to apply and to adapt the methodology for the quantitative evaluation of diseased muscles.
|