Estimation of marbling score in live cattle based on dynamic ultrasound image using a neural network

In the present paper, we propose a new method to estimate the beef marbling standard (BMS) number by dynamic ultrasound imaging of live beef cattle. The proposed method consists of four processes: time series texture analysis, extraction of dynamic features, principal component analysis, and estimation of BMS number by a neural network. Highly accurate estimation is expected by frequency analysis of time series texture features extracted from dynamic ultrasound imaging and averaging the texture features. In addition, a neural network model can flexibly represent a non-linear mapping relationship between image features and BMS number. The correlation coefficient between the actual BMS number and the BMS number estimated using dynamic image features by the leave-one-out method is r = 0.75(P <; 0.01), and the mean estimation error is 1.09. These results suggest that dynamic image features extracted from dynamic ultrasound images have the potential to accurately estimate the BMS number.