基于残差神经网络的大脑磁共振图像配准

Registration of brain magnetic resonance images based on residual neural network

医学影像技术在现代医疗中的作用越来越大,但是不同时间或不同模态下的图像移动不可避免。医学图像配准对于病情诊断和治疗具有重要价值,但是传统配准算法迭代优化时间较长,还容易陷入局部最优。针对大脑核磁共振图像中的运动伪影,本文采用基于残差神经网络的刚性变换配准模型,预测刚性变换参数。模型采用无监督的方法,不需要变换参数作为标签,通过相似性度量作为损失函数约束模型的训练。实验结果表明,对于大脑核磁共振图像配准,模型具有非常好的配准效果,并且配准速度比传统方法有数十倍的提升,对于临床大脑核磁共振图像分析具有重要意义。

Medical imaging technology plays an increasingly important role in modern medicine,but image movement in different time or different modes is unavoidable.Medical image registration is of great value for disease diagnosis and treatment,but the traditional registration algorithm takes a long time to optimize parameters,and it is easy to fall into local optimization.A registration model based on residual neural network is used to predict rigid transformation parameters of the brain MRI images.The unsupervised method is used in the proposed model,because it didn′t require transformation parameters as labels.However,it needs similarity measurement as loss function to train the model.The experimental results show that the model has a very good registration effect for brain MRI Image,and the registration speed is dozens of times higher than that of traditional methods,which is of great significance for clinical brain MRI image analysis.