基于3D注意力的混合神经网络用于脑影像认知评估OA
Hybrid Neural Network with 3D Attention for Brain Image-Based Cognitive Assessment
阿尔茨海默病(Alzheimer's Disease,AD)是一种以进行性认知功能障碍为主要特征的神经退行性疾病.临床认知分数(如简易精神状态量表、临床痴呆评定量表)是刻画疾病严重程度、判断病理阶段及预测疾病进展的重要依据.然而,传统量表评估过程耗时且易受医生主观判断及受试者个体差异影响,导致评估结果存在一定主观偏差.为提高认知评估的客观性与准确性,该研究提出一种基于结构磁共振成像(structural MRI,sMRI)的混合深度网络,实现临床认知分数的自动估计.该方法构建了一个融合 3D 卷积神经网络(3D-CNN)与 Transformer 注意力机制的回归模型,用于联合预测多种认知分数.实验基于 ADNI 数据集开展评估,结果表明所提方法能够有效捕捉与认知衰退相关的大脑结构特征,在认知分数估计任务中取得了较为稳定且可靠的预测性能,为 AD 的客观评估与辅助诊断提供了一种可行方案.
Alzheimer's disease(AD)is a neurodegenerative disease characterized by progressive cognitive dysfunction.Clinical cognitive scores(such as the Mini-Mental State Examination and the Clinical Dementia Rating)are important criteria for characterizing disease severity,determining pathological stages,and predicting disease progression.However,traditional scale-based assessment procedures are time-consuming and are easily influenced by physicians'subjective judgment and individual differences among subjects,resulting in a certain degree of subjectivity in the assessment results.To improve the objectivity and accuracy of cognitive assessment,this study proposes a hybrid deep network based on structural magnetic resonance imaging(sMRI)to achieve automatic estimation of clinical cognitive scores.This method constructs a regression model that integrates a three-dimensional convolutional neural network(3D-CNN)with a Transformer attention mechanism for the joint prediction of multiple cognitive scores.Experiments are conducted on the ADNI dataset.The results demonstrate that the proposed method can effectively capture brain structural features related to cognitive decline and achieve stable and reliable predictive performance in cognitive score estimation tasks,providing a feasible solution for the objective assessment and assisted diagnosis of AD.
黄基;朱可欣;徐宾
福州大学附属省立医院 福建省医疗大数据工程重点实验室,福建 福州 350001台湾中山大学 资讯工程系,台湾 高雄 80424福建师范大学 泉港石化研究院,福建 泉州 362800
信息技术与安全科学
认知评估脑影像注意力机制神经网络
cognitive assessmentbrain imageattention mechanismneural network
《现代信息科技》 2026 (17)
31-35,5
福建省医疗大数据工程重点实验室开放课题项目(KLKF202304)
评论