Multiresolution Texture Analysis - Tissue Characterization Based on Ultrasonic Images

Abstract: In this paper we investigate multiresolution texture analysis approaches based on the mammalian visual system. The aim is the characterization of tissue by multiresolution decomposition of medical ultrasonic images. The application is the diagnosis of the immune reactions in transplanted kidneys. The investigated approaches are the Gabor and Hermite Transforms and the Dome and Cortex filters. The best discrimination results are achieved by the Hermite Transform (1st order) and an extension of the Dome filter approach, resulting in a skew filter. Both methods take into account new neurophysical knowledge of the visual system.