Key Technologies Research On Non-reference MRI Thermometry Based On Statistical Methods

Cancer threatens human health, and it has become one of the larger public health problems worldwide, and continues to be derived as the first killer to human health in the New Century. Hyperthermia is one of the many physical therapies for cancer, in which high-intensity focused ultrasound(High Intensity Focused Ultrasound, HIFU) is being called one of the best non-invasive thermal treatment technology in vitro. Accurate temperature monitoring is the key scientific and technological issue to ensure that hyperthermia to be effective and safe. Currently magnetic resonance imaging(MRI) technology has become the best choice for accurate temperature measurement during the hyperthermia for tumor. Although the MRI thermometry has developed a variety of temperature measurement algorithms, in actual use, the temperature measuring environment is affected usually by factors including magnetic field drift, analyte tissue characteristics, and the human body internal environment, resulting these temperature measurement algorithms have large errors, which is difficult to meet the precise temperature measurement requirements for MRI(currently the MRI temperature measurement error <2.5℃ as international standard). In this thesis, a new MRI temperature measurement model has been studied and established based on the statistical methods, in order to deal with problems existing in practical application of the temperature measurement algorithm based on the chemical shift of the proton resonance frequency(referred to as "proton resonance chemical shift"). The main research contents and innovations are as follows:1. This paper presents an implementation method of the proton resonance chemical shift temperature algorithm without reference temperature measurement, and establishes a MRI temperature measurement model based on statistical methods of linear regression and the BP neural network. By using the test data for method validation, the result shows that the linear regression model for MRI temperature measurement network has a good stability. The temperature measuring effect of new model(maximum absolute temperature difference of 1.82℃, the standard deviation is 0.51℃) is better than that of the traditional numerical algorithm PRF method(maximum absolute temperature difference of 3.70℃, the standard deviation is 0.79℃). The temperature measurement accuracy reached the requirement(±2℃) of National Science and Technology Support Program of China(No.: 2012BAI15B07).2.This study raises a phase compensation model based on the black box theory in order to deal with phase drift, induced by non-temperature factors, existing in the practical application of the proton resonance chemical shift temperature algorithm. The phase compensation model achieves drift correction without reference by using statistical methods. At room temperature, the phase changes of the six kinds of materials consist of vitro liver, vitro pig heart, agar gel, soft tofu, vitro beef and vitro pig thigh, have been studied. The experimental results show that at high field superconducting environment(1.5T), the above organic tissues have extremely small impact on the phase shift of temperature measurement, and the phase shift of MRI temperature measuring comes mainly from the MRI device parameters and environmental impacts.3. This study has made theoretical and experimental research on key technical issues such as exact phase information collection and calculation. After analysing the main noise sources and noise distribution model during MRI process, it raised a non-local average filtering method(Non-Local Means, NLM). The experimental results show that, NLM method effectively improves the peak signal noise ratio while protecting local information of the image. For multi-channel parallel imaging technology, it proposed using amplitude size of each channel magnitude image as the weights to get the weighted average linear phase difference. The experimental results show that the multi-channel phase synthesis can improve temperature measurement accuracy and reduce standard values of temperature measurement errors effectively.