Research On Automatic Segmentation Method Of Knee Joint Image For Sar Simulation

The knee joint is the most complex joint in the human body and one of the most vulnerable parts of the human body.Magnetic Resonance Imaging(MRI)is the preferred imaging technique for knee diagnosis.High field MRI has a high signal-to-noise ratio and resolution,but the distribution of electromagnetic fields in the human body is more obvious as the field strength increases,and the specific RF power deposition in the knee joint tissue(Specific Absorption Rate,SAR)has become a key indicator to consider.This requires the creation of an individual knee joint 3D model on the scanned image for electromagnetic simulation to estimate local SAR.In order to reconstruct the three-dimensional model of the knee joint,it is necessary to segment the tissue regions of the image in the set.Due to various factors(such as signal-to-noise ratio,human-computer interaction workload),the traditional knee joint segmentation method is limited,so it is difficult to construct a three-dimensional knee joint model.Based on the superior performance of convolutional neural network,U-Net was trained on the existing knee magnetic resonance atlas to perform image segmentation and analysis.Based on the characteristics of the magnetic resonance image of the knee joint and the experimental analysis results,a U-Net based design was designed.Parallel network.The anatomical structure of the knee joint is simplified into a model of bone,muscle and fat for labeling and segmentation.Several shallow networks are used in the parallel structure to divide one or two types of tissues,and mask processing is used in the training and prediction process.Multiple network raw outputs are fused.Separating the bones and the background with a separate sub-network effectively solves the effects of regional texture similarity and organizational pixel imbalance on training and prediction.Reducing the difficulty of network training improves the segmentation accuracy.After the output of the network,according to the anatomical features of the knee joint and the gray distribution of the image,the output result is subjected to sliding processing to correct the leakage segmentation phenomenon caused by the network fusion and the hole caused by the insufficient information of the smooth region.Finally,the three-dimensional model of the individual knee joint is established by using the segmentation result,placed in the bird cage coil model for electromagnetic simulation,and the local SAR value distribution is calculated and calculated.The standard model established by manual segmentation is compared to verify the simplified model and the research segmentation.The effectiveness of the method.In the experiment,the method is roughly the same as the local SAR distribution finally calculated by manual segmentation.