基于DBCNet的TOF-MRA中脑动脉树区域自动分割方法

An Automatic Segmentation Method of Cerebral Arterial Tree in TOF-MRA Based on DBCNet

从脑部医学影像中划分动脉树区域是诊断和评估许多脑血管疾病的早期步骤.现有的区域分割方法多依赖人工辅助,本文中提出了一种基于双分支连通网络(dual branch connected network,DBCNet)的脑动脉树自动分区方法,可以将时间飞跃磁共振血管造影(time of flight-magnetic resonance angiography,TOF-MRA)中的动脉树分割为6个主要区域.DBCNet中引入了分支特征解耦模块和Swin Transformer机制的全局与局部特征融合模块,训练采用先定位后分割的两步训练策略.本研究使用了111例TOF-MRA数据,其中81例作为训练集,20例作为验证集,10例作为测试集,模型在测试集上的平均Dice系数为74.72%,95%豪斯多夫距离(HD95)为3.89 mm.和其他先进分割网络相比较,该网络能更准确地分割出各个主要区域,并具有一定的鲁棒性.

Arterial tree region segmentation from medical images of the brain is an early step in the diagnosis and evaluation of many cerebrovascular diseases.Most of the existing region segmentation methods rely on manual assistance.In this paper,we propose an automatic brain arterial tree partitioning method based on a dual branch connected network(DBCNet),which can partition the arterial tree in time of flight-magnetic resonance angiography(TOF-MRA)into six main regions.The branch feature decoupling module and the global and local feature fusion module based on Swin Transformer mechanism were used for DBCNet.The two-step training strategy of localization followed by segmentation was used for training.In this study,111 cases of TOF-MRA data were used,of which 81 cases as the training set,20 cases as the validation set,and 10 cases as the test set.The average Dice coefficient of the model on the test set was 74.72%and 95%Haus dorff distance(HD95)was 3.89 mm.Compared with other advanced segmentation networks,the network reported in this paper can segment each major region more accurately with robustness.