Research Of Brain MRI Image Segmentation Algorithm Based On SOFM

As a new diagnosis means of medical imaging, magnetic resonance imaging can show brain dissection frame exactly, and can confirm or eliminate most pathological tissues, so it is widely used in disease diagnosis. With the development of the science and technology, the demand for higher resolution of imaging is more urgent. Thus, the research on the MRI segmentation which has higher precision,faster speed and better clinic application effect is very important in actual work. Nerve network has faster collateral capacity and stronger learning capacity, which plays an important role in magnetic resonance imaging. This paper mainly research on the application of SOFM in brain MRI image segmentation.Based on deeply study the research status of image segmentation algorithms at home and abroad, this paper discusses technology factors which infect image segmentation, and the image segmentation algorithms which based SOFM in clustering are compared. For the matter that it is slow and the precision is not ideal when SOFM dispose high resolution images, MWSOFM, an image segmentation algorithm based on multi-wavelet is proposed. This algorithm is based on the traditional SOFM, exact and fast segmentation to high resolution of MRI image is the main concern. It uses multi-wavelet to analyze the images, and then uses SOFM to clustering to the multi-wavelet modulus pyramid; finally the segmented result is got. Then for the issue that MRI images include noise, traditional SOFM must set parameter beforehand, and network frame is solidified in training process and can't adjust neatly, BTSOFM, an image segmentation algorithm based on the improved TGSOM is proposed. This algorithm is based on the improved TGSOM algorithm, uses balanced multi-wavelet to denoise the MRI image, then uses the improved TGSOM to clustering to the multi-wavelet modulus pyramid, and finally the segmented result is got. Simulation experiment on MRI brain images indicates that both MWSOFM and BTSOFM not only can accelerate the speed, but also can enhance the precision of clustering, and the segmentation effect is improved obviously.