Medical Image Segmentation Based On Self-Organizing Map

The medical images play an important role in research of humanorgans and their functions due to its non-invasion and more information.Along with more demand for 3D reconstruction, quantitative analysis andvisualization, the more precise segmentation of medical image is required.So, the segmentation of medical images becomes a key issue in medicalimages research. Thus, this paper aims to do more research in thesegmentation of medical images with neural networks.Since medical images always include considerable uncertaintyand unknown noise, this generally leads to further difficulties withsegmentation. Most methods are not robust to noisy images.This paperevaluted median filter in company with wavelet transformation to processmedical image that combined Gauss noise and impulse noise at the sametime.Analyzed experiment result.it showed that the method can betterkeep the edge inforamtion of the original and wipe off noise synchronously. It's result was better, compared with applied median filteror wavelet transformation only.we proposed a new approach to segment the medical imagesusing a multi-scale and adaptive spatial fuzzy self-organizing feature map.Kohonen's self-organizing feature map (SOM) is a two-layer feedforwardcompetitive learning network, and has been used as a competitivelearning clustering algorithm in medical images segmentation. However,most medical images always present overlapping gray-scale intensitiesfor different tissues, Therefore, fuzzy methods are integrated with SOMin this paper to overcome this problem. Moreover, for image data, there isstrong correlation between neighboring pixels. To produce meaningfulsegmentation, we proposed a multi-scale and adaptive spatial fuzzyself-organizing feature map (MSFSOM) for medical image segmentation,in which we consider the spatial relationships between image pixels andmulti-scale processing method to reduce the noise efect and theclassification ambiguity. The efficacy of our approach is validated byextensive experiments using both simulated and real MRI images. AnIntensified Fuzzy Kohonen Clustering Network is proposed, for the slowspeed of convergence in texture segmentation by Fuzzy KohonenClustering Network.