基于多头注意力机制的多模态帕金森病安全检测系统

A multimodal safety detection system for Parkinson"s disease based on Multi-headattention mechanism

摘要:

在实际的帕金森病远程诊断过程中,应用单模态数据检测帕金森病存在误诊率较高的问题,且远程诊断的安全性问题突出；为提高帕金森病远程诊断准确率与安全性,设计一种具有隐私保护功能的帕金森病多模态安全远程辅助检测系统；使用帕金森病语音和步态双模态数据,在传统卷积神经网络后融合多头注意力机制与多层感知机,有效提高模型的特征提取、融合与识别能力；为了保证数据传输过程的安全性,使用基于余弦混沌的差分隐私加噪方式扰动随机拆分的数据编号,提高帕金森病数据传输安全性；通过两模态消融实验和对比实验结果表明,提出的基于多头注意力机制的帕金森病多模态远程检测模型实际测试准确率达到0.913,且模型的各项评估指标和收敛速度等均高于传统模型,具备良好的帕金森病智能辅助检测效果,能够满足帕金森病早期智能安全筛查与诊断需求。

Abstract:

In the actual remote diagnosis process of Parkinson"s disease, the application of single-mode data to detect Parkinson"s disease has a high misdiagnosis rate, and the safety of remote diagnosis is prominent.? To improve the accuracy and security of remote diagnosis of Parkinson"s disease, a multimodal secure remote auxiliary detection system with a privacy protection function was designed.? The dual-modal data of Parkinson"s disease speech and gait were used to integrate the multi-head attention mechanism with multi-layer perceptron after the traditional convolutional neural network, which effectively improved the feature extraction, fusion, and recognition ability of the model.? To ensure the security of the data transmission process, the differential privacy and noise method based on cosine chaos is used to disturb the randomly separated data numbers to improve the data transmission security of Parkinson"s disease.? The two-mode ablation experiment and comparison experiment results show that the actual test accuracy of the proposed multi-mode remote detection model of Parkinson"s disease based on the multi-head attention mechanism reaches 0.913.? The evaluation indicators and convergence speed of the model are higher than that of the traditional model, which has a good effect on intelligently assisted detection of Parkinson"s disease.? It can meet the needs of early innovative safety screening and diagnosis of Parkinson"s disease.