Bayesian constrained spectral method for segmentation of noisy medical images. Theory and applications

Abstract

The spectral method of medical images segmentation that is constrained by Bayesian inference on initial edge map detection is introduced and characterized. It is followed by discussion of the accuracy of the method, that depends on the noise that affects the data. Gaussian noise model is constructed and a method for noisy data multiscale wavelet decomposition and denoising is applied. The proposed segmentation method is tested for denoised cardiac ultrasonic data and its performance is compared for different noise clipping values. Further applications for multiple multimodal cases are presented showing the universality of the proposed method that is fixable and adaptable to the number of clinical applications. The brief discussion of the future development of the method is provided. © 2008 Springer-Verlag Berlin Heidelberg.