摘要
针对阿尔茨海默病研究中纵向神经影像数据不足的问题,提出一种基于年龄约束与注意力机制的增强型条件生成对抗网络框架。该框架能够生成符合生理演化规律的高质量纵向三维磁共振成像数据,以基线结构磁共振影像为输入,预测24个月后的随访影像。网络架构采用3D U-Net生成器和条件深度卷积判别器,并引入卷积块注意力模块以强化关键脑区的特征提取。同时,通过增设年龄损失辅助预测任务,将年龄增长的生理学特征作为显式约束来指导纵向图像生成。实验基于ADNI-1数据集的112例AD患者数据,结果显示生成影像的结构相似性和峰值信噪比分别达到0.955和34.81 dB,且年龄预测精度与图像生成质量呈显著正相关,证实了该方法在AD疾病进展预测中的有效性。本研究为缓解纵向医学影像数据匮乏问题提供了切实可行的解决方案。
To address the issue of insufficient longitudinal neuroimaging data in Alzheimer’s disease(AD)research,this paper proposes an enhanced conditional generative adversarial network framework based on age constraints and attention mechanisms.This framework can generate high-quality longitudinal three-dimensional magnetic resonance imaging data that conforms to physiological evolution laws,taking baseline structural magnetic resonance images as input to predict follow-up images after 24 months.The network architecture employs a 3D U-Net generator and a conditional deep convolutional discriminator,and introduces a convolutional block attention module to strengthen feature extraction of key brain regions.Meanwhile,by adding an age loss auxiliary prediction task,the physiological characteristics of age growth are used as explicit constraints to guide longitudinal image generation.Experiments were conducted based on data from 112 AD patients in the ADNI-1 dataset.The results show that the structural similarity and peak signal-to-noise ratio of the generated images reach 0.955 and 34.81 dB,respectively,and there is a significant positive correlation between age prediction accuracy and image generation quality,confirming the effectiveness of this method in predicting AD disease progression.This study provides a practical solution to alleviate the shortage of longitudinal medical imaging data.
作者
卓冕钧
刘逸凡
许萌
付利华
张柏雯
ZHUO Mianjun;LIU Yifan;XU Meng;FU Lihua;ZHANG Baiwen(School of Computers,Beijing University of Technology,Beijing 100124,China;School of Computing Engineering and Science,Shanghai University,Shanghai 200444,China;Institute of Information and Artificial Intelligence Technology,Beijing Academy of Science and Technology,Beijing 100089,China)
出处
《微纳电子与智能制造》
2024年第4期28-34,共7页
Micro/nano Electronics and Intelligent Manufacturing
基金
北京市科学技术研究院财政项目(23CE-BG5-12,24CB12-01)
关键词
阿尔茨海默病
磁共振成像
生成对抗网络
注意力机制
年龄约束
Alzheimer's disease
magnetic resonance imaging
generative adversarial network
attention mechanism
age constraint
