基于图编码与小样本学习的精神分裂症分析方法

A Method for Analyzing Schizophrenia Based on Graph Encoding and Few-Shot Learning

在基于图神经网络的脑功能性疾病研究中,构建脑网络之后不再变化,且一般需要大量的数据进行建模训练。为了解决这两个问题,文章提出一种基于图编码与小样本学习的分析诊断模型。该模型采用皮尔逊相关系数和自注意力机制构建自适应脑网络,并将时域特征、频域特征和脑网络特征作为图卷积神经网络的输入,进行动态学习自适应脑网络和图编码特征。将图编码特征作为图原型网络的输入,进行小样本学习并实现分类。将该模型应用于精神分裂症的分类诊断,实验结果表明,精神分裂症的识别准确率达到83.4%,为脑网络研究提供一种全新的思路和方法,为小样本学习在精神分裂症研究中的应用开辟了新的方向。

In research on functional brain diseases based on graph neural networks,the constructed brain networks remain static and typically require a large amount of data for modeling and training.To address these two problems,this paper proposes an analysis and diagnostic model based on graph encoding and Few-shot learning.The model utilizes Pearson correlation coefficient and Self-Attention mechanism to construct an adaptive brain network,and takes temporal features,frequency domain features,and brain network features as inputs to a graph convolutional neural network,thereby dynamically learning the adaptive brain network and graph encoding features.The graph encoding features are used as inputs to a graph prototype network for Fewshot learning and classification.Applying this model to the classification and diagnosis of schizophrenia,experimental results demonstrate an accuracy rate of 83.4%in schizophrenia identification.This provides a novel perspective and approach for brain network research,and opens up new directions for the application of Few-shot learning in schizophrenia studies.