Texture analysis of ultrasonic liver images based on wavelet denoising and histogram equalization

Visual criteria for diagnosing diffused liver diseases through ultrasonic image is time-confusing and subjective. This paper proposes a method for ultrasonic images quantitative feature extraction. We employ wavelet denoising and histogram equalization to preprocess the ultrasonic liver images, then classification feature are extracted by the image texture analysis method, gray level difference statistic (GLDS), lastly quantitative feature parameters are extracted from GLDS. These features are fed to a neural network classification. The experiments show that the ultrasonic images performed by wavelet denoising and histogram equalization are conductive to further texture analysis and classify the fatty liver from normal liver. On the contrary for the fatty and normal ultrasonic images without wavelet denoing and histogram equalization, the feature parameters extracted from GLDS have no significant difference.