Research On Segmentation And Measurement Algorithm Of Aortic Root Structure Based On Deep Learning

Aortic valve disease is the most common type of valvular heart disease in the elderly population.Accurate medical imaging evaluation is very important for its clinical diagnosis and treatment.However,there is often a lot of noise and residual image in medical images,which brings great obstacles to the preoperative image segmentation and recognition.With the development of computer technology and artificial intelligence,deep learning algorithm has been widely used in the field of medical images,such as lesion detection,organ segmentation,benign and malignant classification,intelligent diagnosis and treatment,which brings new inspiration to solve the difficult problems in preoperative image evaluation.However,due to the particularity of medical image data,the deep learning model still faces many challenges,such as local optimization,insufficient learning and huge network parameters.To solve these problems,this paper studies the application of deep learning algorithm in CT image segmentation of aortic root.We propose an image segmentation algorithm based on residual and dual attention mechanism,and based on this segmentation algorithm,we propose an automatic measurement algorithm scheme for aortic root parameters.The main work and achievements of this paper are as follows:1.This paper constructs a CT image data set of aortic root,and completes the manual annotation of the image under the guidance of professional imaging doctors.In addition,aiming at the problems of different CT image quality and unclear organ boundary,this paper carries out preprocessing operations such as image clipping,histogram optimization and window width adjustment,so as to provide a good data basis for the research of segmentation algorithm.2.This paper proposes an image segmentation model based on residual attention and dual attention mechanism(RAU net).The algorithm adds a dual attention module integrating channel and spatial dimension to the baseline model u-net,which makes the model focus on the regions that are more valuable for segmentation,so as to improve the segmentation effect of the model on small targets such as aortic valve and calcification.Secondly,by adding a large number of residual structures to the network for multi-scale feature fusion,we can reduce the burden of network training and improve the accuracy of model segmentation.In addition,this paper proposes a weighted loss function focaldiceloss based on dice loss and focal loss.While considering the regional coincidence degree,the loss of easily divided pixels is attenuated,so that the model focuses on the difficult boundary pixels.The experimental results show that the final average Dice coefficient of RAU net in aortic root structure segmentation task proposed in this paper reaches 75.74%,which is 1.91%higher than that of u-net,and is better than that of other advanced networks on this data set.3.In this paper,an automatic measurement algorithm of aortic root structural parameters is designed and implemented.The algorithm reconstructs the three-dimensional model based on the segmentation results of RAU net,and realizes the functions of centerline extraction,plane positioning of aortic valve ring,multi plane parameter measurement and visualization.The experimental results show that the average relative error between the measurement results of the algorithm and the manual measurement results is 3.77%.The average measurement time of a case is shortened from 15 minutes to less than 3 minutes,which greatly saves the evaluation time on the premise of ensuring the measurement accuracy.Therefore,this algorithm provides a fully automated scheme for the preoperative CT image evaluation of transcatheter aortic valve replacement(TAVR).