Research Of Medical Image Segmentation Based On PCNN

With the development of the medical treatment technology, people's requirement to the medical image processing is heightening too. Therefore, rapid and exact image processing algorithm becomes the hotspot and difficulty in medical image processing recently. However, because of the complexity, incertitude and different individuals existing widely otherness in the imaging process of MRI, it has important academic meaning and practical value to segment the interesting regions automatically and accurately.As a new neural network, pulse-coupled neural network has more advantages in image processing than other segmented means, it can offset incoherence in spatial and exiguous changes in scope, so it can reserve region information of images rather perfectly. This paper bases the research in-depth on the characteristics of basic PCNN model, in order to implement the best segmented result, combines the characteristics of medical images, aims at the problems that traditional PCNN model needs many parameters that need manual enactment and threshold attenuation fashion is instability etc, finally an improved PCNN algorithm is proposed which is used on the segmentation of medical images. The proposed algorithm predigests the acceptance parts of PCNN, improves the selection fashion of the linking input L, and changes the threshold attenuation fashion of PCNN, and assures the ignition nerve cells maintain ignition state; therefore, it decreases the number of initialization parameter, improves the description ability of pixel special information, and expedites the constringency speed of the model. Emulational experiment indicates that the improved model has the characteristics of faster speed and better detail disposal. Aims at the traditional PCNN model can't determine the best iterative time automatically, based on the improved model, make the implementation of the accurate and fast segmentation as the goal, this paper combines the fuzziness of the medical images and the special information of the pixels, finally gives the maximum fuzzy entropy and the minimum 2D cross entropy as the best determinant rules which can assure the best circular iterative time atomically. Compared with the PCNN models based the Shannon entropy and the minimum cross entropy, emulational experiment indicates that the two given rules both have better segmentation results for medical images. Thereinto, the maximum fuzzy entropy has higher robustness and has better segmented effect between background and the goal region; the minimum 2D cross entropy has lower time complexity and has better effect in detail disposal of images.