融合残差块注意力机制和生成对抗网络的海马体分割

研究一种基于改进的生成对抗网络的深度学习方法对海马体进行分割。提出不同的卷积配置,以捕获由分割网络获得的信息。提出以Pixel2Pixel为基本架构的生成对抗网络模型,生成模型结合残差网络以及注意力机制的编解码结构以捕获更多细节信息。判别网络采用卷积神经网络对生成模型的分割结果和专家分割结果进行判别。经过生成模型和对抗模型不断地传递其损失,使生成模型达到分割海马体的最优状态。使用来自ADNI数据集130名健康受试者的T1加权MRI扫描和相关海马标签作为训练和测试数据,以相似度系数作为评价指标,准确率达到89.46%。试验结果表明,该网络模型可以实现高效地自动分割海马体,对于阿尔茨海默症等疾病的正确诊断具有重要的现实意义。

Hippocampal segmentation combining residual attention mechanism and generative adversarial networks

This research discussed a deep learning method based on an improved generative adversarial network to segment the hippocampus. Different convolutional configurations were proposed to capture the information obtained by the segmentation network. The generative adversarial network based on Pixel2Pixel was proposed. The generator was a codec structure combining residual network and attention mechanism to capture more detailed information. The discriminator used a convolutional neural network to discriminate the segmentation results of the generated model and the expert segmentation results. Through generator and discriminator continuously transmitted losses, the generator reached the optimal state of segmenting the hippocampus. Using the T1-weighted MRI scans and related hippocampus labels of 130 healthy subjects from the ADNI data set as training and test data, and the similarity coefficient as the evaluation index, the accuracy rate reached 89.46%. Results showed that the network model could achieve efficient automatic segmentation of hippocampus, which had important practical significance for the correct diagnosis of diseases such as Alzheimer's disease.