Application of neural network in ultrasound tissue characterization using backscattered signal parameters

Ultrasonic techniques have shown good potential for estimating tissue composition for noninvasive imaging, diagnosis, and meat evaluation. Research carried out to define useful ultrasonic spectral parameters for tissue characterization, and to test multiparameter pattern recognition methods for tissue classification. The feasibility of two pattern recognition methods, linear discriminant analysis and artificial neural networks was studied for their applications to beef quality grading. The differentiating criteria were the tissue composition (% fat) and the tissue inhomogeneity (fat marbling). The discriminant analysis and neural network both showed good potential for evaluating marbling grades and % fat for beef rib-eye quality grading using ultrasonic backscattered spectral parameters