Study On Temperature Measurement Based On Multi Feature Of B-ultrasound Image In HIFU Treatment

High-intensity focused ultrasound(HIFU)is a non-invasive treatment that produce high temperature to ablate diseased tissue.During the treatment,excessive temperature is easy to damage normal tissue,low temperature will result in incomplete ablation of the diseased tissue,so temperature monitoring becomes the key to treatment.Because of the characteristics of real-time,easy integration and low cost of ultrasonic imaging,temperature measurement based on ultrasonic image has become a research hotspot in recent years.In this paper,the B-ultrasound image of fresh isolated pork tissue before and after HIFU irradiation was obtained and processed.The characteristic parameters which can well characterize the temperature were extracted to realize the non-invasive temperature measurement during HIFU treatment.The main task are as follows:1.Preprocessing of B-ultrasound image.It mainly includes experimental system designing,image grayscale processing and image cropping.2.Registration of B-ultrasound image.The edge feature points are manually selected,the direction is assigned to each feature point by the SURF algorithm,and the description vector of each feature point is calculated.Through similarity measurement,the matching point pairs ofB-ultrasound images before and after HIFU irradiation were obtained for registration,which is prepare for image difference processing.3.Feature parameter extraction.The gray mean value of the B-ultrasound image,the characteristic parameters of gray-level co-occurrence matrix and the gray gradient co-occurrence matrix,which reflect the image texture information,were extracted from the B-ultrasound image.Combined with the bidimensional empirical mode decomposition(BEMD),the image was processed to select the characteristic parameters that can well represent the temperature.The ultrasonic difference image was decomposed into a series of IMF components and a residual from high frequency to low frequency by BEMD method.It was found that the gray average,gray entropy and mixing entropy of the remaining image were well correlated with the temperature.Non-invasive temperature measurement method based on B-ultrasonic image in single feature parameters and multiple feature parameters.Based on the approximate linear relationship between characteristic parameter and temperature,a linear regression model of temperature and individual characteristic parameter was constructed to measure temperature.By using the data mining techniques,three machine learning models including random forest,support vector machine and BP neural network were employed.Based on the regression functions of the three models,the temperature estimation model was constructed with the three characteristic parameters that arestrongly correlated with temperature.Through the research and analysis,it is found that the gray mean value,the gray entropy and the mixed entropy of the gray gradient co-occurrence matrix of the remaining image obtained by the BEMD,have a strong correlation with the temperature.Based on these three parameters,the random forest model can accurately measure the tissue temperature during HIFU treatment,which is of great significance for the non-invasive temperature measurement research.