Abstract:A new method based on ultrasonic radiofrequency signal has been proposed for normal liver and fatty liver steatosis degree

especially diagnose mild fatty liver. In this paper

we analyze the RF signal in two domains: the time domain and the frequency domain. Firstly

in the time domain

the power of RF signal was in accord with normal distribution

and the mathematical expectation (ME) of liver RF signal was extracted as the feature parameter. In the frequency domain

wavelet transforming was used to analyze RF signal

and low-frequency wavelet coefficients mean (LWCM) and wavelet transform modulus maximum mean (WMMM) were extracted as feature parameter. Finally

back-propagation (BP) artificial neutral network was employed to classify these RF signals. The accuracy rates with BP neural network are 90.0% for normal liver RF signal

86.7% for mild fatty liver RF signal

83.3% for moderate fatty liver RF signal

and 90.0% for severe fatty liver RF signal. The result shows that the ME

the LWCM and the WMMM can successfully describe the features of liver RF signal. This study shows that the ultrasonic RF signal is effective in diagnosing liver fatty degree

and so far

there is no reports about grading fatty liver based on ultrasonic RF signal. This paper provides a new direction for computer-aided diagnosis of fatty liver.