Random field models: a new option of textural analysis in ultrasonic images of the liver

In ultrasonic images of the liver conventional two dimensional texture parameters (CTP) were compared with parameters derived from a new stochastic model i.e. auto-regressive periodic random field models (APRFM). By fitting the model to a given textural pattern, the estimated model parameters are suitable texture features to distinguish between images of the liver with and without microfocal lesions. The APRFM approach produces classification results equivalent or even better than those obtained by use of CTP parameters. Taking advantage of the principle of analysis by synthesis the possibility of comparing visually the re-synthesized image with the original ultrasonic image is important for clinical acceptance