Classification of liver diseases from ultrasound images using a hybrid kohonen SOM and LPND speckle reduction method

Abstract

Ultrasound imaging has several advantages over other medical imaging modalities (e.g., X-rays, computed tomography, and magnetic resonance). It is safe, relatively low cost, and allows real-time imaging. Tissue characterization with ultrasound has become important topic since computer facilities have been available for the analysis of ultrasound signals. In this paper, an unsupervised neural network learning technique-Hybrid Kohonen Self Organising Map (SOM) is proposed for the recognition of liver diseases from ultrasound images. Three kinds of liver diseases cyst, hepatoma and hemangioma are identified as the most critical liver diseases. The diagnosis scheme includes three steps: Speckle reduction, feature extraction and classification. A speckle is removed from ultrasound images by Laplacian pyramid nonlinear diffusion method and also preserves prominent edge information for classification. For nonlinear diffusion in each pyramid layer, a gradient threshold is automatically determined by a variation of Median Absolute Deviation (MAD) estimator. Then the features are extracted from mean gray level, entropy, local variance, co-occurrence matrix, first order statistics, gradient features and fractal dimensions are collected from the normal and abnormal ultrasound images. The unsupervised neural network learning technique called Hybrid Kohonen Self organizing map is proposed to classify normal and abnormal liver diseases from ultrasound images.