Glomeruli Segmentation Based on Neural Network with Fault Tolerance Analysis

Image segmentation, which is the first essential and fundamental issue in the image analysis and pattern recognition, is a classical difficult problem in image processing. In the computer-aided diagnosis system of the renal biopsy images in microscope, the correct segmentation of glomerulus is an important step for automatic analysis. Complex characteristics of renal biopsy images lead to the difficulty in boundary features description. A kind of feature operator based on the definition of the cavum boundary is proposed in this paper. According to this operator, a nonlinear thresholding surface can be constructed by neural network, and the appropriate surface can be selected to enhance the cavum boundary by the fault tolerance analysis. After denoising, the segmentation results can be obtained. Experimental results indicate that this method can enhance the boundary and suppress noises at the same time; it can obtain good segmented results and has a fine adaptability to various sample images.