Ensuring robust methods for maintaining high levels of medical data security is crucial in the Medical Internet of Things(IoT)for the protection of sensitive patient data during real-time transmission and analysis.Ele...Ensuring robust methods for maintaining high levels of medical data security is crucial in the Medical Internet of Things(IoT)for the protection of sensitive patient data during real-time transmission and analysis.Electroencephalography(EEG)signals in medical IoT systems are transmitted through cloud and edge networks,which create risks of cyber threats,unauthorized access,and data breaches.Consequently,there is an urgent need for efficient encryption methods to ensure the confidentiality of EEG signals during classification and prediction processes,as several state-of-the-art models either neglect security during classification or suffer from increased computational overhead that limits real-time applicability.In this paper,an innovative framework is proposed for secured and accurate detection of seizures from EEG signals based on Quantum Hilbert Encryption-assisted hybrid deep learning and machine learning methods.Raw EEG signals are first converted into 2D spectrogram images to facilitate visual feature analysis.These spectrogram images are then encrypted using the newly developed Quantum Hilbert Encryption Scheme to preserve the sensitive medical data while processing or transmission occurs.In addition,deep learning methods used encrypted EEG representations to extract discriminant features while maintaining the signal integrity and data confidentiality.Then,machine learning classifiers are used for seizure classification,efficiently and accurately doing so.Experimental evaluations highlight the system's strong performance:Support Vector Machine(SVM)achieved top accuracies of 87.63%with ResNet50 as features extractor and 83.51%with VGG19 as feature extractor,while RF excelled in precision with scores of 88.61%(SVM with ResNet50 as features extractor)and 87.91%(RF with Xception as features extractor).These results confirm the system's superior capability in seizure detection using encrypted EEG data.To enhance model interpretability,we employed Gradient-weighted Class Activation Mapping(Grad-CAM)and Local Interpretable Model-agnostic Explanations(LIME)to visualize and explain the decision-making process of the proposed hybrid AI model.By combining cutting-edge encryption with hybrid AI models and Explainable AI,the proposed method holds promising potential for application in Medical IoT(MIoT)environments,where secure real-time automatic EEG analysis is paramount.展开更多
目的旨在构建一套具备可解释性与置信度分析功能的前列腺癌(orostate cancer,PCa)良恶性分类模型,以提升诊断准确性并降低临床误诊风险。方法回顾性分析267例PCa患者和143例非PCa患者的双参数磁共振数据,采用VGG-16网络构建分类模型,通...目的旨在构建一套具备可解释性与置信度分析功能的前列腺癌(orostate cancer,PCa)良恶性分类模型,以提升诊断准确性并降低临床误诊风险。方法回顾性分析267例PCa患者和143例非PCa患者的双参数磁共振数据,采用VGG-16网络构建分类模型,通过梯度加权类激活映射(gradient-weighted class activation mapping,Grad-CAM)方法实现可视化解释,并使用蒙特卡洛Dropout(Monte Carlo Dropout,MC-Dropout)法进行不确定性估计,引入拒绝机制;最后通过受试者工作特性(receiver operating characteristic,ROC)曲线和曲线下面积(area under the curve,AUC)评估模型性能。结果相比于原始VGG-16网络,本次提出的置信度模型提高了正确分类比例(94.6%vs 79.3%),并减少了假阳性(5.3%vs 15.3%),同时漏诊率接近零(0.1%),AUC值提高(P<0.05);模型正确分类比例高于高年资医生(94.6%vs 90.8%),高置信度激活区域与真实病灶区域高度吻合。结论本次提出的PCa分类模型,结合可视化与拒绝机制,无需像素级标签,也可准确识别PCa病灶并输出置信度,显著提高临床决策的准确性与安全性。展开更多
有效地分析处理癫痫脑电信号并对其准确分类可以进一步完善癫痫检测问题。因此,各种深度学习方法逐渐应用到该问题中,如使用BiLSTM模型对癫痫脑电的一维时间序列数据进行处理。为进一步提高癫痫脑电分类的准确率,本文将癫痫脑电的一维...有效地分析处理癫痫脑电信号并对其准确分类可以进一步完善癫痫检测问题。因此,各种深度学习方法逐渐应用到该问题中,如使用BiLSTM模型对癫痫脑电的一维时间序列数据进行处理。为进一步提高癫痫脑电分类的准确率,本文将癫痫脑电的一维时间序列数据转换为二维图像,使用EfficientNetV2模型来实现癫痫检测的二分类。同时,引入梯度加权类激活映射(Gradient⁃weighted class activation mapping,Grad⁃CAM)对二维图像分类进行可视化分析。对德国伯恩大学脑电癫痫脑电信号数据集的预处理版本进行分类实验,EfficientNetV2模型的准确率达到了98.69%,优于BiLSTM模型。结果表明,EfficientNetV2模型可以有效通过二维脑电图像实现癫痫脑电分类,而且分类准确率更高。展开更多
基金funded by Princess Nourah bint Abdulrahman University Researchers Supporting Project number(PNURSP2026R197)Princess Nourah bint Abdulrahman University,Riyadh,Saudi Arabia.
摘要Ensuring robust methods for maintaining high levels of medical data security is crucial in the Medical Internet of Things(IoT)for the protection of sensitive patient data during real-time transmission and analysis.Electroencephalography(EEG)signals in medical IoT systems are transmitted through cloud and edge networks,which create risks of cyber threats,unauthorized access,and data breaches.Consequently,there is an urgent need for efficient encryption methods to ensure the confidentiality of EEG signals during classification and prediction processes,as several state-of-the-art models either neglect security during classification or suffer from increased computational overhead that limits real-time applicability.In this paper,an innovative framework is proposed for secured and accurate detection of seizures from EEG signals based on Quantum Hilbert Encryption-assisted hybrid deep learning and machine learning methods.Raw EEG signals are first converted into 2D spectrogram images to facilitate visual feature analysis.These spectrogram images are then encrypted using the newly developed Quantum Hilbert Encryption Scheme to preserve the sensitive medical data while processing or transmission occurs.In addition,deep learning methods used encrypted EEG representations to extract discriminant features while maintaining the signal integrity and data confidentiality.Then,machine learning classifiers are used for seizure classification,efficiently and accurately doing so.Experimental evaluations highlight the system's strong performance:Support Vector Machine(SVM)achieved top accuracies of 87.63%with ResNet50 as features extractor and 83.51%with VGG19 as feature extractor,while RF excelled in precision with scores of 88.61%(SVM with ResNet50 as features extractor)and 87.91%(RF with Xception as features extractor).These results confirm the system's superior capability in seizure detection using encrypted EEG data.To enhance model interpretability,we employed Gradient-weighted Class Activation Mapping(Grad-CAM)and Local Interpretable Model-agnostic Explanations(LIME)to visualize and explain the decision-making process of the proposed hybrid AI model.By combining cutting-edge encryption with hybrid AI models and Explainable AI,the proposed method holds promising potential for application in Medical IoT(MIoT)environments,where secure real-time automatic EEG analysis is paramount.
摘要目的旨在构建一套具备可解释性与置信度分析功能的前列腺癌(orostate cancer,PCa)良恶性分类模型,以提升诊断准确性并降低临床误诊风险。方法回顾性分析267例PCa患者和143例非PCa患者的双参数磁共振数据,采用VGG-16网络构建分类模型,通过梯度加权类激活映射(gradient-weighted class activation mapping,Grad-CAM)方法实现可视化解释,并使用蒙特卡洛Dropout(Monte Carlo Dropout,MC-Dropout)法进行不确定性估计,引入拒绝机制;最后通过受试者工作特性(receiver operating characteristic,ROC)曲线和曲线下面积(area under the curve,AUC)评估模型性能。结果相比于原始VGG-16网络,本次提出的置信度模型提高了正确分类比例(94.6%vs 79.3%),并减少了假阳性(5.3%vs 15.3%),同时漏诊率接近零(0.1%),AUC值提高(P<0.05);模型正确分类比例高于高年资医生(94.6%vs 90.8%),高置信度激活区域与真实病灶区域高度吻合。结论本次提出的PCa分类模型,结合可视化与拒绝机制,无需像素级标签,也可准确识别PCa病灶并输出置信度,显著提高临床决策的准确性与安全性。
摘要有效地分析处理癫痫脑电信号并对其准确分类可以进一步完善癫痫检测问题。因此,各种深度学习方法逐渐应用到该问题中,如使用BiLSTM模型对癫痫脑电的一维时间序列数据进行处理。为进一步提高癫痫脑电分类的准确率,本文将癫痫脑电的一维时间序列数据转换为二维图像,使用EfficientNetV2模型来实现癫痫检测的二分类。同时,引入梯度加权类激活映射(Gradient⁃weighted class activation mapping,Grad⁃CAM)对二维图像分类进行可视化分析。对德国伯恩大学脑电癫痫脑电信号数据集的预处理版本进行分类实验,EfficientNetV2模型的准确率达到了98.69%,优于BiLSTM模型。结果表明,EfficientNetV2模型可以有效通过二维脑电图像实现癫痫脑电分类,而且分类准确率更高。
摘要深度学习近年来在故障诊断领域受到广泛应用,但基于深度学习的故障诊断模型缺乏工程上的物理解释性,难以保证其故障诊断结果的稳定性。以轴承为例,建立了以小波时频图像为故障诊断依据的卷积神经网络模型(convolutional neural network,CNN),提出了一种基于梯度加权类激活热力图(gradient-weighted class activation map,Grad-CAM)的网络模型鲁棒性分析方法,并利用美国凯斯西储大学(Case Western Reserve University,CWRU)轴承数据集进行验证。首先,将故障直径轴承数据以不同方式混合并训练大、小多个模型。其次,利用Grad-CAM方法,建立时频区域与故障模式之间的联系。最后,利用其他工况下的轴承故障数据,以及含噪数据进行测试,并根据结果结合模型最注重的时频区域进行分析。结果表明,基于深度学习的轴承故障诊断模型在参数较少时更加注重低频区域,并能使其具有更好的鲁棒性。