| Medical ultrasound imaging is widely used in clinical detection of liver tissue due to its advantages of low cost,safety,reliability,and real-time non-invasiveness.Liver fibrosis caused by liver injury and it is a global liver disease which seriously affects people’s health and life safety.Therefore,early detection and intervention of liver fibrosis has important clinical significance.The density of scatterers is directly related to the progression of liver fibrosis.Most of the existing scatterer density quantitative ultrasound liver fiber detection algorithms are based on statistical distribution and parameter analysis,which have the problems of inaccurate detection and poor robustness.Therefore,two detection algorithms were proposed: a statistical analysis and parametric imaging detection algorithm based on quantitative ultrasound and a liver fiber detection algorithm based on deep convolutional neural networks.Experimental results of ultrasonic simulation data with different scatterers densities show that the accuracy of the algorithm is more than 5% higher than that of the traditional algorithm,especially the detection algorithm based on deep convolutional neural network,which still performs well in the case of low scatterers densities.Through the real data experiment of rabbit liver fiber,the two algorithms proposed in this paper can accurately detect liver fiber tissue,and have a good clinical application prospect.Ultrasound nonlinear coefficient is a sensitive acoustic parameter to reflect the structure and pathology of biological tissue,especially in liver diseases,compared with the normal tissue,its parameter value has obvious difference.At present,the measurement of nonlinear coefficients requires sophisticated and complex experimental equipment and is difficult to solve.Therefore,based on the simulation data which obtained by CREANUIS,two machine learning regression prediction algorithms based on fundamental wave and the second harmonic envelope signals were proposed: Deep Convolutional Neural Network(CNN),Convolutional Neural Network and Support Vector Regression hybrid algorithm(CNN-SVR),as well as a hybrid parameter optimization Support Vector Regression algorithm(HGP-SVR)which based on Nakagami dual-parameter.Experiments on simulation data show that the CNN-SVR and HGP-SVR algorithms are superior than the CNN regression model,and the CNN-SVR hybrid regression algorithm performs better.Experiments on real liver fiber data show that the prediction results are within the expected range.Compared with the existing algorithms,the methods in this paper are easy to implement,has higher prediction accuracy and are convenient for clinical practice applications. |
