Research Of Key Technologies In Four-Dimensional Medical Image Reconstruction And Segmentation

Medical imaging and medical imaging analysis have greatly advanced the progress of clinical diagnosis.With the popularity of digital medical imaging and the advancement of computer computing capabilities,4D(4 Dimensional)medical imaging has begun to play a role in this field.It is essential for quantifying organ displacement,observing organ movements in the body,and assessing the mechanical function of the organ.The analysis of organ function before and after treatment of dynamic imaging,and the dynamic tracking of lesions during radiotherapy are the basis for the implementation of 4D radiation therapy.Accurate segmentation of organs and lesions in 4D medical imaging can minimize damage to normal tissue,maximize radiation dose to the target,and protect tissue adjacent to the target.The reconstruction and analysis of 4D medical images is generally divided into five processes: 1.acquisition of dynamic medical image data,2.reconstruction of 4D medical images,3.segmentation of ROI(Region of interesting)in 4D medical images,4.make measurements to quantify,5.Analysis based on quantitative measurements.This paper refers to the whole process of imaging and analysis from 4D medical imaging,research on gating information acquisition for 4D medical image construction and segmentation technology for medical image analysis,and tests the proposed algorithm in clinical applications.The main research contents of this paper are as follows:1.The key to existing chest MRI(Magnetic Resonance Imaging)dynamic images based on retrospective acquisition slices is the accurate collection of respiratory gating information.This paper proposes a gating technique based on optical flow method to track biomarkers in the body to automatically mark the end of exhalation(EE)and the end of inspiration(EI).This gating technology allows the patient to breathe freely during medical image acquisition.It requires minimal human interaction and can perform large amounts of data annotation in a short time.The method of automatic labeling has a small error compared to the labeling results of manual labeling.In addition,this paper proposes an abnormal respiratory signal detection algorithm.During chest MRI dynamic image reconstruction,abnormal respiratory cycles can disrupt the integrity of dynamic imaging and affect the doctor’s assessment of the patient’s condition.On this basis,the paper also designed a more advanced complete CNN(Convolutional Neural Networks)model and LSTM(Long Short-Term Memory)model to label the respiratory nodes and detect abnormal respiratory cycle signals in free breathing acquisition slices.This framework overcomes the defect that the respiratory node labeling algorithm based on the optical flow method is sensitive to the labeling position and improves the accuracy of the labeling.The abnormal respiratory signal detection algorithm under this framework does not require manual design features to detect abnormal respiratory cycle signals.This paper tested the availability of the framework on clinical data to meet clinical requirements.2.The existing segmentation model cannot utilize multiple modal medical images or the specificity of different enhanced medical images of the same modality.This paper designs a multi-channel MC-FCNs(Mutil-Channel Fully Convolutional Networks)network based on FCN(Fully Convolutional Networks).Different phases of multi-phases enhanced CT(Computed Tomography)have different imaging effects of lesions.The MC-FCNs network can use this feature to perform feature learning for different enhanced phases by using more channels,training independent model parameters for each enhanced phase,and merging lesion features of different phases.In this paper,the algorithm was tested in the task of segmenting liver lesions in the clinical data.Compared with single-channel FCN,MC-FCNs achieved better segmentation accuracy of liver lesions.3.This paper designs a general 4D medical image segmentation framework to study the effects of different segmentation directions on the accuracy of 4D medical imaging based on two 4D MRI data sets.Although the segmentation algorithm in 2D medical imaging has been widely used in medical images,there is no research work to discuss the influence of different segmentation directions(time and space)on the segmentation accuracy of 4D medical images.The 4D medical image algorithm designed in this paper consists of two independent T networks for time direction segmentation and S network for spatial direction segmentation.Users can freely select any time point or arbitrary position point of 4D medical image as the starting point of segmentation.The direction of time or space can be changed to complete the entire segmentation of the 4D medical image.Under this general framework,the basic structure U-Net(U-Network)for segmentation can be replaced by other segmentation structures.Based on the test results obtained from two 4D MRI datasets,this paper can provide a reference for the segmentation accuracy that can be obtained in different segmentation directions when using the framework to segment 4D medical images.This study included the reconstruction and segmentation of 4D medical image analysis.In the reconstruction stage,this paper solved the key problems in the reconstruction of thoracic dynamic MRI,realized the automatic labeling of respiratory time points,detected abnormal respiratory signals,and improved the reconstruction efficiency.In the stage of medical image analysis,the segmentation algorithm based on 2D(2 Dimensional)medical image and 4D medical image was implemented respectively.