The Research And Application Of Medical Image Segmentation Based On Convolutional Neural Networks

In medical image diagnosis,doctors usually need to analyze and process 2D slices or 3D image sequences with the help of computer.In this way,they can make qualitative and quantitative analysis of the region of interest.The research mainly focuses on the medical image segmentation algorithm,and investigates the medical image registration and its application,including automatic fetal brain extraction,left ventricle segmentation from cardiac MRI,and parotid delivered dose estimation based on deformable registration of computed tomography?CT?and cone-beam CT?CBCT?.Fetal brain extraction is one of the most essential steps for prenatal brain magnetic resonance imaging?MRI?reconstruction and analysis.However,due to the fetal movement within the womb,it is a challenging task to extract fetal brains from sparsely acquired imaging stacks typically with motion artifacts.To address this problem,we propose an automatic brain extraction method for fetal MRI using multi-stage 2D U-Net with deep supervision?DS U-net?.Specifically,we initially employ a coarse segmentation derived from DS U-net to define a 3D bounding box for localizing the position of the brain.The DS U-net is trained with deep supervision loss to acquire more powerful discrimination capability.Then,another DS U-net focuses on the extracted region to produce finer segmentation.The final segmentation results are obtained by performing refined segmentation.We validate the proposed method on 80 stacks of training images and 43 testing stacks.The experimental results illustrate that the precision and robustness of our method outperforms other state-of-the-art methods.Validation metrics,percentage of good contours?%?,average dice metric?ADM?and average perpendicular distance?APD?were computed as 97.44?3.17?,0.94?0.02?,and 1.49?0.16?mm,respectively.Segmentation of the left ventricle?LV?from MRI images is an essential step for calculation of clinical indices.In this work,we proposed a new methodology for automatic LV segmentation on short-axis MRI.The method was carried out in three stages:?1?a single neural network was trained to detect the LV chamber and extract regions of interest?ROIs?.?2?full convolutional networks were employed to learn the task of LV segmentation from the ground true data.?3?cardiac MRI image with ROI which was located by the detector was provided for the segmentation model for prediction.The method was trained and evaluated using 45 cardiac MRI datasets from the MICCAI 2009 LV segmentation challenge.The evaluation results revealed that our method outperformed other state-of-the-art methods.Validation metrics,percentage of good contours?%?,average dice metric?ADM?and average perpendicular distance?APD?were computed as 97.65?3.04?,0.95?0.01?,and 1.35?0.16?mm,respectively.Xerostomia induced by radiotherapy is a common toxicity for head and neck carcinoma patients.In this study,the deformable image registration of planning CT and weekly CBCT was used to override the Hounsfield unit value of CBCT,and the modified CBCT was introduced to estimate the radiation dose delivered during course of treatment.Herein,the beams from each patient's treatment plan were applied to the modified CBCT to construct the weekly delivered dose.Then,weekly doses were summed together to obtain the accumulated dose.A total of 42 parotid glands?PGs?of 21 nasopharyngeal carcinoma patients were analyzed.The results demonstrated that the p-value of V20,V30,D50,and D_mean difference of the delivery dose between patients with xerostomia and patients without xerostomia was less than 0.05.However,for the planning dose,the significant dosimetric difference between the two groups only existed in D50 and D_mean.Xerostomia is closely related to V20,V30,D50,and D_mean.