基于独立成分分析与LSTM网络的患者病情预测模型研究

Research on patient condition prediction model based on independent component analysis and LSTM network

摘要

为提高脑疾病患者病情预测的精度,以阿尔茨海默脑疾病为研究对象,提出一种基于独立成分分析法与LSTM的预测模型。模型分为3个阶段,首先采用独立成分分析法,将患者功能磁共振成像数据分解成若干状态网络,获取个体独立成分。然后利用滑动窗口构建动态功能脑网络。最后将动态功能脑网络输入长短期记忆(long short-term memory,LSTM)网络,通过LSTM网络分类预测患者病情。在仿真平台上对基于独立成分分析法与LSTM的预测模型进行仿真,结果表明,基于独立成分分析法与LSTM的预测方法,可有效分类预测健康、患病、早期轻度认知障碍3类阿尔茨海默脑疾病,且具有较高的预测精度,平均分类准确率、敏感性、特异性和F₁值分别达到87.37%、87.48%、89.15%、87.65%,相较于支持向量机(support vector machine,SVM)、图神经网络(graph neural network,GNN)构建的预测模型,具有明显优势。由此得出,基于独立成分分析法与LSTM的预测方法,提高了脑疾病患者病情的预测精度,为脑疾病患者病情的精确预测奠定了基础。

Abstract

To improve the accuracy of predicting the condition of patients with brain diseases, a prediction model based on independent component analysis and LSTM is proposed with Alzheimer′s disease as the research object. The model is divided into three stages. Firstly, independent component analysis is used to decompose the patient′s functional magnetic resonance imaging data into several state networks and obtain individual independent components. Then sliding windows is used to construct dynamic functional brain networks. Finally, the dynamic functional brain network is input into a long short term memory (LSTM) network, which is used to classify and predict the patient′s condition. On the simulation platform, the prediction model based on independent component analysis and LSTM is simulated. The results show that the prediction method based on independent component analysis and LSTM can effectively classify and predict three types of Alzheimer′s brain diseases: healthy, diseased, and early mild cognitive impairment, with high prediction accuracy. The average classification accuracy, sensitivity, specificity, and F₁ score reaches 87.37%, 87.48%, 89.15%, and 87.65%, respectively. Compared with the prediction models constructed by SVM, GNN, etc., it has obvious advantages. From this, it can be concluded that the prediction method based on independent component analysis and LSTM improves the accuracy of predicting the condition of patients with brain diseases, laying the foundation for accurate prediction of the condition of patients with brain diseases.

关键词

深度学习 / LSTM网络 / 独立成分分析 / 病情预测 / 阿尔茨海默症 / 脑疾病

Key words

deep learning / LSTM network / independent component analysis / disease prediction / Alzheimer′s disease / brain disease

[1] 孙磊, 张义宁, 薛艳芳, 等. 自适应功能连接网络学习及其在脑疾病识别中的应用[J]. 中国科学技术大学学报, 2020, 50(8): 1102-1109, 1155.
[2] 张格, 林岚, 吴水才. 基于脑网络的图卷积神经网络在脑疾病中的研究进展[J]. 医疗卫生装备, 2021, 42(8): 85-92.
[3] 李涛, 邱震钰, 李瑶, 等. 基于时变特性的多层脑网络拓扑属性分析及脑疾病分类[J]. 科学技术与工程, 2023, 23(19): 8114-8123.
[4] 李懿, 吴凯, 李文豪, 等. 基于脑电的精神疾病自动分类研究进展[J]. 生物医学工程与临床, 2023, 27(2): 238-246.
[5] 罗屹, 陈丽梅, 孙继飞, 等. 基于独立成分分析探索经皮耳迷走神经刺激对难治性抑郁症视觉网络的即刻调节效应[J]. 中国临床新医学, 2023, 16(4): 317-321.
[6] 单帅帅, 马清峰, 谢雯鑫. 基于小波包能量谱和ICA的模拟电路故障特征提取方法[J]. 电子测量技术, 2021, 44(18): 19-23.
[7] 宋昊, 徐颂, 刘国明, 等. 基于独立成分分析的非侵入式脑-机接口眼电伪迹自动去除算法[J]. 生物医学工程学杂志, 2022, 39(6): 1074-1081.
[8] 林邹卿, 汪腾龙, 王国强. 基于低频振幅和独立成分分析的缓解期抑郁症静息态磁共振研究[J]. 临床精神医学杂志, 2022, 32(5): 357-361.
[9] 陈逸嘉, 陶力, 李凯, 等. 基于IMIFS-VMD和ROA-LSTM的日前电价预测方法[J]. 自动化技术与应用, 2025, 44(9): 23-28.
[10] 吉兴全, 曾若梅, 张玉敏, 等. 基于注意力机制的CNN-LSTM短期电价预测[J]. 电力系统保护与控制, 2022, 50(17): 125-132.
[11] 翟政中, 黄永涛, 刘华云. 基于LSTM的电气化铁路牵引供电故障识别研究[J]. 自动化技术与应用, 2025, 44(6): 95-98.
[12] 谭海旺, 杨启亮, 邢建春, 等. 基于XGBoost-LSTM组合模型的光伏发电功率预测[J]. 太阳能学报, 2022, 43(8):75-81.
[13] 姜林鸿, 刘晓丹, 孙萍萍, 等. 运动想象训练对卒中后患者上肢运动功能恢复的影响及静息态fNIRS脑网络特征研究[J]. 中国康复医学杂志, 2023, 38(11):1505-1513.
[14] 李祎升, 王纯, 张宁. 焦虑障碍不同亚型间脑网络功能异常的相似性和异质性研究进展[J]. 广西医科大学学报, 2023, 40(10): 1603-1607.
[15] 冯思同, 宁艳哲, 郑思思, 等. 基于f MRI技术研究针刺神门穴对急性睡眠剥夺后工作记忆相关脑网络的影响[J]. 中华中医药杂志, 2023, 38(9): 4512-4516.