Research On Magnetic Resonance Image Processing Method Of Prostate Cancer

Magnetic resonance imaging is the most effective method for the non-invasive diagnosis of prostate cancer,and medical imaging processing has a crucial role in the qualitative and grading evaluation of prostate tumors and in the precise planning and postoperative assessment of surgical implementation protocols.With continued advances medical imaging and iterations in imaging equipment and imaging contrast agents,the development of medical image processing technology and multimodal imaging fusion has provided a variety of magnetic resonance images for prostate cancer diagnosis.With the continuous research of informatization and intelligent machine learning algorithms and the development of quantitative features of tumor images,artificial intelligence in medical fields such as intelligent image processing has developed vigorously.These new technological advances described above bring new opportunities and challenges to the application of imaging in precision oncology medicine.In response to the above hot issues,this dissertation combines advanced imaging technology,image processing,artificial intelligence,and deep convolutional neural networks,and conducts a more in-depth study of medical image analysis based on multimodal magnetic resonance imaging techniques and deep learning algorithms from the following four aspects:(1)A magnetic resonance image fusion algorithm based on the U-Net network and Gaussian pyramid is proposed,which solves the problem that multimodal magnetic resonance images of prostate cancer cannot be effectively fused,and overcomes the difficulty of artificially designing robust pixel activity level measurement and weight distribution strategies.By analyzing the relationship between the image feature information and the training target,the fusion weight map automatically generated by U-Net is applied to the Gaussian pyramid weighted fusion strategy based on the weight map,which avoids the generation of artifacts and energy loss.In this dissertation,an effective image fusion algorithm is designed to combine the effective information of multimodal magnetic resonance images of prostate cancer to form a fusion image that is clearer than that displayed separately.The experimental results show that the proposed method is superior to similar fusion algorithms in the objective evaluation indexes of axial T2 W sequence and ADC sequence magnetic resonance image fusion of 10 patients with prostate cancer,as well as the visual perception of prostate cancer region definition and edge details.(2)A multi-modal prostate cancer image segmentation algorithm based on deformation convolution is proposed,which solves the problem of inaccurate segmentation of prostate region with large geometric changes and irregular shape by using fixed receptive field with ordinary convolution kernel,by making the convolution operator learn the offset of the target segmentation region to change the spatial sampling position,the fixed receptive field of the traditional convolution operator is changed into an adaptive receptive field that can feel the change of features,to improve the segmentation accuracy of the target region.It is also designed to simulate the training process of physicians’ integration from simple to complex read mapping methods into convolutional neural network models,solves the problem of using convolutional neural networks for accurate segmentation of prostate cancer from multimodal small sample data sets with overfitting.Experiments show that the segmentation accuracy of prostate tumor region trained on the gray image formed by multimodal MRI image fusion can reach about 0.92,which improves the segmentation effect of prostate cancer multimodal MRI image,especially the accuracy of irregular small tumor segmentation.The overall accuracy and reliability of prostate cancer segmentation are better than the existing segmentation algorithms.(3)The processing algorithm of fitting Z-spectrum model for prostate cancer area in chemical exchange saturation metastasis imaging is proposed and the prostate tumor tissue is quantitatively analyzed,which solves the problem that the current pulse sequence function model for detecting prostate cancer area,including the two variables of pulse amplitude and offset frequency,is not conducive to quantitative analysis.In this dissertation,we plotted a model fitted by the Lorentzian function of z-spectral curves in diseased and normal regions from 18 patients with prostate cancer under four different pre-saturation pulse amplitude conditions,followed by an asymmetric analysis of z-spectral curves to measure protein activity in regions of interest and thereby detect tumors.Experimental results show that this marker is independent of the pre-saturation pulse amplitude,and the use of markers can distinguish between normal peripheral zone and prostate cancer tissue when the presaturation pulse amplitude changes.(4)An ergonomic driver is designed for magnetic resonance elastic imaging,which solves the problem that it is not convenient to carry out the research of tissue magnetic resonance elastic imaging technology due to the lack of elastic imaging driver design patent in China.The performance of soft tissue elastic modulus measurement is verified by using the designed driver.The driver designed in this dissertation has a simple structure and can be placed in the magnetic resonance scanning coil.In order to verify the effectiveness of the driving equipment,an axial simulation slice is designed,and the slice is analyzed by finite element mechanics.The results show that the shear wave generated by the driver can form effective propagation in the soft tissue region.Effective spread is formed in the area.Based on the effective excitation generated by the driver equipment,the elastic modulus diagram of tissue is obtained under different driving frequencies.The analysis of the elastic modulus image shows that the elastic modulus of different tissues is different at the same frequency,and the measured value of elastic modulus has a positive correlation with the frequency.The faster the frequency,the higher the measured value.Ten volunteers underwent two MRI elastography scans,and the experiment proved that the generated elastic modulus map has good reproducibility.The design scheme of this driver can be easily transplanted into magnetic resonance elastography of prostate cancer for further verification.To sum up,this dissertation takes magnetic resonance image processing as the main line and studies the two systems of deep learning for clinical application mature multimodal magnetic resonance image sequence analysis and magnetic resonance functional imaging technologies for prostate cancer detection.This dissertation explores the effective fusion algorithm of multimodal magnetic resonance images,uses the magnetic resonance image segmentation algorithm of prostate cancer patients,uses the Z-spectrum model to measure the difference between prostate cancer tissues and normal tissues,and explores the feasibility of using selfdesigned pneumatic driving equipment to measure tissue elastic modulus.This research helps to improve the effect of MRI assisted diagnosis,and can provide doctors with more diversified image data as an important basis for diagnosing diseases in the field of precise treatment of tumors.