Egyptian Informatics Journal最新文献_第4页

Advancing personalized neurology with explainable AI in Alzheimer’s classification using NeuroX-DualFusion framework 使用NeuroX-DualFusion框架，在阿尔茨海默氏症分类中使用可解释的人工智能推进个性化神经学

IF 4.3 3区计算机科学 Q1 COMPUTER SCIENCE, ARTIFICIAL INTELLIGENCE

Egyptian Informatics Journal

Pub Date : 2026-03-01 Epub Date: 2026-03-04 DOI: 10.1016/j.eij.2026.100926

A.P. Ponselvakumar, S. Anandamurugan

Alzheimer’s disease classification from MRI slices is a cornerstone of personalized neurology, enabling patient-specific diagnosis and treatment planning. Traditional machine learning approaches often fail in this domain due to class imbalance, limited feature representation, and poor interpretability, which restrict their clinical adoption and typically leading to biased predictions and unstable subject-level outcomes. This research introduces NeuroX-DualFusion, a hybrid framework that integrates a local attention stream and a global convolutional stream to capture both fine-grained and contextual features. The pipeline begins with standardized preprocessing and data augmentation to enhance anatomical clarity and mitigate class imbalance. Segmentation via attention-based U-Net isolates critical brain regions, while proposed NeuroX-DualFusion, dual-stream feature extraction enables robust representation learning. Additionally, Grad-CAM visualizations provide transparent, class-specific interpretability, highlighting discriminative regions aligned with clinical markers. Quantitative evaluation across Accuracy (97.5%), Precision (96.5%), Recall (97.5%), F1-score (96.5%), and Specificity (98.5%), demonstrates that NeuroX-DualFusion outperforms individual models, achieving subject-level accuracy. These findings underscore the potential of NeuroX-DualFusion to advance personalized neurology by delivering reliable, interpretable, and patient-centered dementia stage classification using MRI data, bridging the gap between computational innovation and clinical practice.

从MRI切片中对阿尔茨海默病进行分类是个性化神经学的基石，可实现患者特异性诊断和治疗计划。由于类别不平衡、有限的特征表示和较差的可解释性，传统的机器学习方法往往在这一领域失败，这限制了它们的临床应用，并通常导致有偏见的预测和不稳定的学科水平结果。本研究介绍了NeuroX-DualFusion，这是一个混合框架，集成了局部注意力流和全局卷积流，以捕获细粒度和上下文特征。管道从标准化预处理和数据增强开始，以提高解剖清晰度并减轻类别不平衡。基于注意力的U-Net分割分离了关键的大脑区域，而提出的NeuroX-DualFusion双流特征提取实现了鲁棒的表征学习。此外，Grad-CAM可视化提供了透明的、特定类别的可解释性，突出了与临床标记一致的区分区域。准确性（97.5%）、精确度（96.5%）、召回率（97.5%）、f1评分（96.5%）和特异性（98.5%）的定量评估表明，NeuroX-DualFusion优于单个模型，达到了受试者水平的准确性。这些发现强调了NeuroX-DualFusion通过使用MRI数据提供可靠、可解释和以患者为中心的痴呆分期分类来推进个性化神经学的潜力，弥合了计算创新和临床实践之间的差距。

{"title":"Advancing personalized neurology with explainable AI in Alzheimer’s classification using NeuroX-DualFusion framework","authors":"A.P. Ponselvakumar, S. Anandamurugan","doi":"10.1016/j.eij.2026.100926","DOIUrl":"10.1016/j.eij.2026.100926","url":null,"abstract":"<div><div>Alzheimer’s disease classification from MRI slices is a cornerstone of personalized neurology, enabling patient-specific diagnosis and treatment planning. Traditional machine learning approaches often fail in this domain due to class imbalance, limited feature representation, and poor interpretability, which restrict their clinical adoption and typically leading to biased predictions and unstable subject-level outcomes. This research introduces NeuroX-DualFusion, a hybrid framework that integrates a local attention stream and a global convolutional stream to capture both fine-grained and contextual features. The pipeline begins with standardized preprocessing and data augmentation to enhance anatomical clarity and mitigate class imbalance. Segmentation via attention-based U-Net isolates critical brain regions, while proposed NeuroX-DualFusion, dual-stream feature extraction enables robust representation learning. Additionally, Grad-CAM visualizations provide transparent, class-specific interpretability, highlighting discriminative regions aligned with clinical markers. Quantitative evaluation across Accuracy (97.5%), Precision (96.5%), Recall (97.5%), F1-score (96.5%), and Specificity (98.5%), demonstrates that NeuroX-DualFusion outperforms individual models, achieving subject-level accuracy. These findings underscore the potential of NeuroX-DualFusion to advance personalized neurology by delivering reliable, interpretable, and patient-centered dementia stage classification using MRI data, bridging the gap between computational innovation and clinical practice.</div></div>","PeriodicalId":56010,"journal":{"name":"Egyptian Informatics Journal","volume":"33 ","pages":"Article 100926"},"PeriodicalIF":4.3,"publicationDate":"2026-03-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"147396703","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":3,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

引用次数: 0

Epileptic seizure detection using information Gain-Based hybrid Features: Deep Neural network and comparative Machine learning approaches 基于信息增益的混合特征的癫痫发作检测：深度神经网络和比较机器学习方法

IF 4.3 3区计算机科学 Q1 COMPUTER SCIENCE, ARTIFICIAL INTELLIGENCE

Egyptian Informatics Journal

Pub Date : 2026-03-01 Epub Date: 2026-01-24 DOI: 10.1016/j.eij.2026.100889

Nuri Ikizler, Gunes Ekim

Automatic detection of epileptic seizures is crucial in clinical diagnosis to enable early intervention and ensure patient safety. However, systematic comparisons across multi-class combinations and quantitative evaluation of discriminative features remain limited in the literature. This study aims to identify the most effective features for seizure detection and to develop a high-accuracy classification model. Statistical, spectral, and wavelet-based features from time, frequency, and time–frequency domains were selected using the Information Gain method, and four models were integrated into a hybrid framework. The approach was evaluated on 26 class combinations using Random Forest, Support Vector Machines, k-Nearest Neighbors, Gradient Boosting, and a Deep Neural Network. The proposed method achieved an average accuracy of 99%, with the Deep Neural Network reaching 99.69% in combinations including class E, demonstrating strong generalizability in multi-class scenarios. The main novelty of this work lies in combining Information Gain-based hybrid feature selection with a systematic multi-class analysis, a gap not fully addressed in previous studies. This approach enhances accuracy, interpretability, and generalizability, thereby contributing to improved clinical decision-making in epilepsy diagnosis.

癫痫发作的自动检测在临床诊断中至关重要，可以实现早期干预并确保患者安全。然而，在文献中，跨多类组合的系统比较和判别特征的定量评估仍然有限。本研究旨在找出最有效的特征，以检测癫痫发作，并开发一个高精度的分类模型。利用信息增益方法从时间、频率和时频域选择基于统计、频谱和小波的特征，并将四个模型集成到一个混合框架中。该方法使用随机森林、支持向量机、k近邻、梯度增强和深度神经网络对26个类组合进行了评估。该方法的平均准确率达到99%，其中Deep Neural Network在包括E类在内的组合中达到99.69%，在多类场景中表现出较强的泛化能力。这项工作的主要新颖之处在于将基于信息增益的混合特征选择与系统的多类分析相结合，这是以往研究中没有完全解决的空白。这种方法提高了准确性、可解释性和通用性，从而有助于改善癫痫诊断的临床决策。

{"title":"Epileptic seizure detection using information Gain-Based hybrid Features: Deep Neural network and comparative Machine learning approaches","authors":"Nuri Ikizler, Gunes Ekim","doi":"10.1016/j.eij.2026.100889","DOIUrl":"10.1016/j.eij.2026.100889","url":null,"abstract":"<div><div>Automatic detection of epileptic seizures is crucial in clinical diagnosis to enable early intervention and ensure patient safety. However, systematic comparisons across multi-class combinations and quantitative evaluation of discriminative features remain limited in the literature. This study aims to identify the most effective features for seizure detection and to develop a high-accuracy classification model. Statistical, spectral, and wavelet-based features from time, frequency, and time–frequency domains were selected using the Information Gain method, and four models were integrated into a hybrid framework. The approach was evaluated on 26 class combinations using Random Forest, Support Vector Machines, k-Nearest Neighbors, Gradient Boosting, and a Deep Neural Network. The proposed method achieved an average accuracy of 99%, with the Deep Neural Network reaching 99.69% in combinations including class E, demonstrating strong generalizability in multi-class scenarios. The main novelty of this work lies in combining Information Gain-based hybrid feature selection with a systematic multi-class analysis, a gap not fully addressed in previous studies. This approach enhances accuracy, interpretability, and generalizability, thereby contributing to improved clinical decision-making in epilepsy diagnosis.</div></div>","PeriodicalId":56010,"journal":{"name":"Egyptian Informatics Journal","volume":"33 ","pages":"Article 100889"},"PeriodicalIF":4.3,"publicationDate":"2026-03-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"146037844","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":3,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

引用次数: 0

Predicting preterm birth with privacy-preserving AI models: Federated learning and explainable AI 用保护隐私的人工智能模型预测早产：联邦学习和可解释的人工智能

IF 4.3 3区计算机科学 Q1 COMPUTER SCIENCE, ARTIFICIAL INTELLIGENCE

Egyptian Informatics Journal

Pub Date : 2026-03-01 Epub Date: 2026-02-09 DOI: 10.1016/j.eij.2026.100901

Md Tanjum An Tashrif , Shahariar Hossain Mahir , Dipanjali Kundu , Anichur Rahman , Fahmid Al Farid , Sarina Mansor , Abu Saleh Musa Miah

Preterm birth remains a significant public health challenge, closely associated with infant mortality and long-term morbidity. The complexity of its causes complicates accurate prediction. In this study, we present an AI-driven model designed to predict preterm birth, integrating federated learning (FL), deep learning (DL), and explainable artificial intelligence (XAI) to prioritize both data privacy and interpretability. We utilized a primary dataset of 58 electrohysterogram (EHG) recordings from pregnant women, each collected over 1000-second intervals, and applied the Synthetic Minority Oversampling Technique (SMOTE) to address class imbalance. To rigorously assess generalizability, we performed external validation on the independent TPEHGDB dataset comprising 300 EHG recordings from a different institution and time period Our approach evaluated a range of models, from established machine learning algorithms like XGBoost, LightGBM, and CatBoost, to advanced frameworks such as a Transformer-based architecture and quantum convolutional neural networks (QCNN). By leveraging FL, we enabled secure, collaborative training across institutions while maintaining patient data confidentiality. Additionally, XAI techniques, particularly SHAP, were employed to elucidate the key risk factors influencing predictions, thereby enhancing clinical transparency. XGBoost and Transformer models achieved 96.17% and 94.94% accuracy on internal validation, respectively, and demonstrated robust generalization with 88.67% and 89.33% accuracy on external validation, maintaining clinically critical recall rates of 78.95% and 81.58% for preterm detection. Critically, federated learning introduced minimal performance degradation (more than 2%) compared to centralized training, validating privacy-preserving collaborative learning. Although QCNN showed promise as an innovative approach, its performance lagged slightly behind classical models on external data. This underscores the potential of our approach as a scalable, privacy-preserving, and interpretable tool for early detection of preterm birth, with demonstrated generalizability across independent clinical populations.

早产仍然是一项重大的公共卫生挑战，与婴儿死亡率和长期发病率密切相关。其成因的复杂性使准确预测变得复杂。在这项研究中，我们提出了一个人工智能驱动的模型，旨在预测早产，整合联邦学习（FL）、深度学习（DL）和可解释人工智能（XAI），以优先考虑数据隐私和可解释性。我们利用了来自孕妇的58个子宫电图（EHG）记录的主要数据集，每个记录的收集间隔超过1000秒，并应用合成少数过采样技术（SMOTE）来解决类别不平衡问题。为了严格评估泛化性，我们对独立的TPEHGDB数据集进行了外部验证，该数据集包含来自不同机构和时间段的300个EHG记录。我们的方法评估了一系列模型，从已建立的机器学习算法（如XGBoost、LightGBM和CatBoost）到高级框架（如基于transformer的架构和量子卷积神经网络（QCNN））。通过利用FL，我们实现了跨机构的安全协作培训，同时保持了患者数据的机密性。此外，XAI技术，特别是SHAP，被用于阐明影响预测的关键风险因素，从而提高临床透明度。XGBoost和Transformer模型在内部验证的准确率分别为96.17%和94.94%，在外部验证的准确率分别为88.67%和89.33%，在早产检测方面保持了78.95%和81.58%的临床关键召回率。关键的是，与集中式训练相比，联邦学习引入了最小的性能下降（超过2%），验证了保护隐私的协作学习。尽管QCNN显示出作为一种创新方法的希望，但它在外部数据上的表现略落后于经典模型。这强调了我们的方法作为一种可扩展、隐私保护和可解释的早产早期检测工具的潜力，并证明了在独立临床人群中的普遍性。

{"title":"Predicting preterm birth with privacy-preserving AI models: Federated learning and explainable AI","authors":"Md Tanjum An Tashrif , Shahariar Hossain Mahir , Dipanjali Kundu , Anichur Rahman , Fahmid Al Farid , Sarina Mansor , Abu Saleh Musa Miah","doi":"10.1016/j.eij.2026.100901","DOIUrl":"10.1016/j.eij.2026.100901","url":null,"abstract":"<div><div>Preterm birth remains a significant public health challenge, closely associated with infant mortality and long-term morbidity. The complexity of its causes complicates accurate prediction. In this study, we present an AI-driven model designed to predict preterm birth, integrating federated learning (FL), deep learning (DL), and explainable artificial intelligence (XAI) to prioritize both data privacy and interpretability. We utilized a primary dataset of 58 electrohysterogram (EHG) recordings from pregnant women, each collected over 1000-second intervals, and applied the Synthetic Minority Oversampling Technique (SMOTE) to address class imbalance. To rigorously assess generalizability, we performed external validation on the independent TPEHGDB dataset comprising 300 EHG recordings from a different institution and time period Our approach evaluated a range of models, from established machine learning algorithms like XGBoost, LightGBM, and CatBoost, to advanced frameworks such as a Transformer-based architecture and quantum convolutional neural networks (QCNN). By leveraging FL, we enabled secure, collaborative training across institutions while maintaining patient data confidentiality. Additionally, XAI techniques, particularly SHAP, were employed to elucidate the key risk factors influencing predictions, thereby enhancing clinical transparency. XGBoost and Transformer models achieved 96.17% and 94.94% accuracy on internal validation, respectively, and demonstrated robust generalization with 88.67% and 89.33% accuracy on external validation, maintaining clinically critical recall rates of 78.95% and 81.58% for preterm detection. Critically, federated learning introduced minimal performance degradation (more than 2%) compared to centralized training, validating privacy-preserving collaborative learning. Although QCNN showed promise as an innovative approach, its performance lagged slightly behind classical models on external data. This underscores the potential of our approach as a scalable, privacy-preserving, and interpretable tool for early detection of preterm birth, with demonstrated generalizability across independent clinical populations.</div></div>","PeriodicalId":56010,"journal":{"name":"Egyptian Informatics Journal","volume":"33 ","pages":"Article 100901"},"PeriodicalIF":4.3,"publicationDate":"2026-03-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"146187895","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":3,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

引用次数: 0

Movie Recommendation system with sentiment analysis using deep learning algorithms 使用深度学习算法进行情感分析的电影推荐系统

IF 4.3 3区计算机科学 Q1 COMPUTER SCIENCE, ARTIFICIAL INTELLIGENCE

Egyptian Informatics Journal

Pub Date : 2026-03-01 Epub Date: 2026-02-06 DOI: 10.1016/j.eij.2026.100905

Agboola A.O., Ladoja K.T., Onifade O.F.W.

In the era of digital media saturation, recommendation systems have become essential tools for delivering personalized content to users. While traditional approaches rely on user–item interactions and content similarity, they often overlook the emotional nuances expressed in user reviews. This study presents a sentiment-aware hybrid recommendation system that integrates deep learning-based sentiment classification with user demographics and item features to enhance movie recommendation accuracy. The proposed model employs Bidirectional Encoder Representations from Transformers (BERT) to classify user reviews into five nuanced sentiment polarities viz positive, slightly positive, neutral, slightly negative, and negative. These sentiment scores are embedded into a Deep Factorization Machine (DeepFM) architecture, which captures complex relationships among users, items, and emotional cues. A multi-filtering strategy incorporating user age, gender, occupation, location, and movie genre is utilized to mitigate cold-start problems and refine recommendations. Experimental evaluation using the MovieLens dataset, complemented with IMDb user reviews, demonstrates improvements in ROC-AUC (84.47%), Balanced Accuracy (76.36%), and PR-AUC (82.13%) compared to traditional systems. The findings highlight the effectiveness of integrating fine-grained sentiment analysis into the recommendation process, offering deeper insights into user intent and improving the personalization of suggestions. The proposed framework presents a scalable and efficient solution for building emotionally intelligent recommendation systems, fostering deeper user engagement, informed decision-making, and more meaningful media experiences.

在数字媒体饱和的时代，推荐系统已经成为向用户提供个性化内容的重要工具。虽然传统的方法依赖于用户与物品的交互和内容的相似性，但它们往往忽略了用户评论中表达的情感上的细微差别。本研究提出了一种情感感知混合推荐系统，该系统将基于深度学习的情感分类与用户人口统计和项目特征相结合，以提高电影推荐的准确性。该模型采用来自变形金刚的双向编码器表示（BERT），将用户评论分为五种细微的情绪极性，即积极、略积极、中性、略消极和消极。这些情绪得分被嵌入到深度分解机器（DeepFM）架构中，该架构捕捉用户、项目和情感线索之间的复杂关系。采用结合用户年龄、性别、职业、位置和电影类型的多重过滤策略来缓解冷启动问题并改进推荐。使用MovieLens数据集和IMDb用户评论进行的实验评估表明，与传统系统相比，ROC-AUC（84.47%）、平衡精度（76.36%）和PR-AUC（82.13%）有所提高。研究结果强调了将细粒度情感分析整合到推荐过程中的有效性，提供了对用户意图的更深入的见解，并提高了建议的个性化。所提出的框架为构建情感智能推荐系统、促进更深层次的用户参与、知情决策和更有意义的媒体体验提供了可扩展和高效的解决方案。

{"title":"Movie Recommendation system with sentiment analysis using deep learning algorithms","authors":"Agboola A.O., Ladoja K.T., Onifade O.F.W.","doi":"10.1016/j.eij.2026.100905","DOIUrl":"10.1016/j.eij.2026.100905","url":null,"abstract":"<div><div>In the era of digital media saturation, recommendation systems have become essential tools for delivering personalized content to users. While traditional approaches rely on user–item interactions and content similarity, they often overlook the emotional nuances expressed in user reviews. This study presents a sentiment-aware hybrid recommendation system that integrates deep learning-based sentiment classification with user demographics and item features to enhance movie recommendation accuracy. The proposed model employs Bidirectional Encoder Representations from Transformers (BERT) to classify user reviews into five nuanced sentiment polarities viz positive, slightly positive, neutral, slightly negative, and negative. These sentiment scores are embedded into a Deep Factorization Machine (DeepFM) architecture, which captures complex relationships among users, items, and emotional cues. A multi-filtering strategy incorporating user age, gender, occupation, location, and movie genre is utilized to mitigate cold-start problems and refine recommendations. Experimental evaluation using the MovieLens dataset, complemented with IMDb user reviews, demonstrates improvements in ROC-AUC (84.47%), Balanced Accuracy (76.36%), and PR-AUC (82.13%) compared to traditional systems. The findings highlight the effectiveness of integrating fine-grained sentiment analysis into the recommendation process, offering deeper insights into user intent and improving the personalization of suggestions. The proposed framework presents a scalable and efficient solution for building emotionally intelligent recommendation systems, fostering deeper user engagement, informed decision-making, and more meaningful media experiences.</div></div>","PeriodicalId":56010,"journal":{"name":"Egyptian Informatics Journal","volume":"33 ","pages":"Article 100905"},"PeriodicalIF":4.3,"publicationDate":"2026-03-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"146188597","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":3,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

引用次数: 0

A study on front vehicle collision warning method based on lightweight YOLOv8 and DeepSort 基于轻量级YOLOv8和DeepSort的前车碰撞预警方法研究

IF 4.3 3区计算机科学 Q1 COMPUTER SCIENCE, ARTIFICIAL INTELLIGENCE

Egyptian Informatics Journal

Pub Date : 2026-03-01 Epub Date: 2025-12-19 DOI: 10.1016/j.eij.2025.100861

Wenyu Zhang , Yajing Li , Jiaxuan Hu , Ning Wang

With the continuous increase in vehicle ownership, the frequency of traffic accidents has risen significantly, and higher demands have consequently been placed on active vehicle safety technologies. To address the challenges of insufficient real-time performance and high model complexity in traditional object detection methods under complex traffic conditions, an improved front-vehicle collision warning system has been proposed by integrating YOLOv8 and DeepSort. In this approach, the original YOLOv8 backbone network is replaced by the lightweight MobileNet V4, and the Convolutional Block Attention Module (CBAM) is incorporated to enhance feature extraction capabilities. A comprehensive algorithmic framework has been constructed, integrating multi-object recognition, front-vehicle distance estimation, ego-vehicle speed calculation, and hierarchical warning level output. Experimental results on the KITTI dataset have demonstrated a detection accuracy of 95.5 % and a total detection time of 2.6 ms per frame. Additionally, a 2.6 % improvement in mAP50–95 has been observed, accompanied by only a 0.1 % decrease in the recall rate. These findings suggest that the proposed method provides effective technical support for front-vehicle collision warning in intelligent transportation environments.

随着机动车保有量的不断增加，交通事故的发生频率显著上升，对车辆主动安全技术提出了更高的要求。针对传统目标检测方法在复杂交通条件下实时性不足、模型复杂度高的问题，将YOLOv8与DeepSort相结合，提出了一种改进的前车碰撞预警系统。在这种方法中，原始的YOLOv8骨干网络被轻量级的MobileNet V4取代，并加入卷积块注意模块（CBAM）来增强特征提取能力。构建了集多目标识别、前车距离估计、自车速度计算、预警等级输出于一体的综合算法框架。在KITTI数据集上的实验结果表明，检测准确率为95.5%，总检测时间为2.6 ms /帧。此外，观察到mAP50-95有2.6%的改善，同时召回率仅下降0.1%。研究结果表明，该方法为智能交通环境下的前车碰撞预警提供了有效的技术支持。

{"title":"A study on front vehicle collision warning method based on lightweight YOLOv8 and DeepSort","authors":"Wenyu Zhang , Yajing Li , Jiaxuan Hu , Ning Wang","doi":"10.1016/j.eij.2025.100861","DOIUrl":"10.1016/j.eij.2025.100861","url":null,"abstract":"<div><div>With the continuous increase in vehicle ownership, the frequency of traffic accidents has risen significantly, and higher demands have consequently been placed on active vehicle safety technologies. To address the challenges of insufficient real-time performance and high model complexity in traditional object detection methods under complex traffic conditions, an improved front-vehicle collision warning system has been proposed by integrating YOLOv8 and DeepSort. In this approach, the original YOLOv8 backbone network is replaced by the lightweight MobileNet V4, and the Convolutional Block Attention Module (CBAM) is incorporated to enhance feature extraction capabilities. A comprehensive algorithmic framework has been constructed, integrating multi-object recognition, front-vehicle distance estimation, ego-vehicle speed calculation, and hierarchical warning level output. Experimental results on the KITTI dataset have demonstrated a detection accuracy of 95.5 % and a total detection time of 2.6 ms per frame. Additionally, a 2.6 % improvement in mAP50–95 has been observed, accompanied by only a 0.1 % decrease in the recall rate. These findings suggest that the proposed method provides effective technical support for front-vehicle collision warning in intelligent transportation environments.</div></div>","PeriodicalId":56010,"journal":{"name":"Egyptian Informatics Journal","volume":"33 ","pages":"Article 100861"},"PeriodicalIF":4.3,"publicationDate":"2026-03-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145791745","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":3,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

引用次数: 0

Graph-based temporal anomaly detection with self-supervised contrastive learning and dynamic adaptive thresholding for acoustic howling suppression 基于自监督对比学习和动态自适应阈值的基于图的时间异常检测用于啸叫抑制

IF 4.3 3区计算机科学 Q1 COMPUTER SCIENCE, ARTIFICIAL INTELLIGENCE

Egyptian Informatics Journal

Pub Date : 2026-03-01 Epub Date: 2026-01-29 DOI: 10.1016/j.eij.2026.100892

Xiaoqian Fan , Francisco Hernando-Gallego , Diego Martín , Mohammad Khishe

Acoustic howling due to feedback loops in audio systems is a major challenge in such fields as hearing aids or public address systems. Traditional approaches such as notch filters and adaptive feedback cancellation often have limitations such as lack of adaptability in dynamic environments, and a need for a large amount of labelled data. To overcome these shortcomings, a new deep learning approach, Dynamic Adaptive Thresholding and Self-Supervised Contrastive Learning for Graph-based Temporal Anomaly Recognition (GTAD-CL), is proposed in this paper. By representing audio signals as graphs, GTAD-CL uses graph neural networks to represent complex spatial–temporal patterns to detect howling with high precision as an anomaly. Self-supervised contrastive learning removes the requirement of having labeled datasets which improves the scalability and generalization of the AI models. A dynamic adaptive thresholding mechanism guarantees robust performance under different acoustic conditions, e.g. low signal to noise ratio environments. Integrated with neural filtering in real time, GTAD-CL makes howling suppression easy. Experimental results on a 100-hour custom dataset and six public benchmarks indicate that GTAD-CL has a precision of 0.92 (compared to 0.88, the best baseline, HybridAHS, showing a gain of 4.5%), recall of 0.90 (compared to 0.85, a gain of 5%) and F1-score of 0.91 (compared to 0.865, a gain of 4.5%). In suppression quality GTAD-CL achieves a PESQ score of 3.02 (compared to 2.68 for HybridAHS, i.e. ∼12.7% better), and a STOI of 0.90 (compared to 0.86, i.e. ∼4.7% better). Moreover, GTAD-Cl runs with a real-time factor of 0.36× which is better than HybridAHS’s 0.42× (approx. 14% faster). These results give validation to GTAD-CL as a powerful, scalable, and low-latency solution and high-fidelity solution that is superior to state-of-the-art results for varying acoustic scenarios.

音频系统中由反馈回路引起的声啸是助听器或公共广播系统等领域的主要挑战。传统的方法，如陷波滤波器和自适应反馈抵消往往有局限性，如缺乏对动态环境的适应性，需要大量的标记数据。为了克服这些缺点，本文提出了一种新的深度学习方法——动态自适应阈值和自监督对比学习的基于图的时间异常识别（GTAD-CL）。通过将音频信号表示为图形，GTAD-CL使用图形神经网络来表示复杂的时空模式，以高精度检测嚎叫作为一种异常。自监督对比学习消除了对标记数据集的要求，从而提高了人工智能模型的可扩展性和泛化性。动态自适应阈值机制保证了不同声学条件下的鲁棒性能，例如低信噪比环境。集成了实时神经滤波，GTAD-CL使嚎叫抑制容易。在100小时自定义数据集和6个公共基准上的实验结果表明，GTAD-CL的精度为0.92（与最佳基线HybridAHS的0.88相比，提高了4.5%），召回率为0.90（与0.85相比，提高了5%），f1得分为0.91（与0.865相比，提高了4.5%）。在抑制质量方面，GTAD-CL的PESQ得分为3.02（相比之下，HybridAHS为2.68，即提高了约12.7%），STOI为0.90（相比之下，0.86，即提高了约4.7%）。此外，GTAD-Cl运行时的实时因子为0.36×，优于HybridAHS的0.42×（约为0.42×）。快14%)。这些结果验证了GTAD-CL是一种功能强大、可扩展、低延迟和高保真的解决方案，在不同的声学场景中优于最先进的结果。

{"title":"Graph-based temporal anomaly detection with self-supervised contrastive learning and dynamic adaptive thresholding for acoustic howling suppression","authors":"Xiaoqian Fan , Francisco Hernando-Gallego , Diego Martín , Mohammad Khishe","doi":"10.1016/j.eij.2026.100892","DOIUrl":"10.1016/j.eij.2026.100892","url":null,"abstract":"<div><div>Acoustic howling due to feedback loops in audio systems is a major challenge in such fields as hearing aids or public address systems. Traditional approaches such as notch filters and adaptive feedback cancellation often have limitations such as lack of adaptability in dynamic environments, and a need for a large amount of labelled data. To overcome these shortcomings, a new deep learning approach, Dynamic Adaptive Thresholding and Self-Supervised Contrastive Learning for Graph-based Temporal Anomaly Recognition (GTAD-CL), is proposed in this paper. By representing audio signals as graphs, GTAD-CL uses graph neural networks to represent complex spatial–temporal patterns to detect howling with high precision as an anomaly. Self-supervised contrastive learning removes the requirement of having labeled datasets which improves the scalability and generalization of the AI models. A dynamic adaptive thresholding mechanism guarantees robust performance under different acoustic conditions, e.g. low signal to noise ratio environments. Integrated with neural filtering in real time, GTAD-CL makes howling suppression easy. Experimental results on a 100-hour custom dataset and six public benchmarks indicate that GTAD-CL has a precision of 0.92 (compared to 0.88, the best baseline, HybridAHS, showing a gain of 4.5%), recall of 0.90 (compared to 0.85, a gain of 5%) and F1-score of 0.91 (compared to 0.865, a gain of 4.5%). In suppression quality GTAD-CL achieves a PESQ score of 3.02 (compared to 2.68 for HybridAHS, i.e. ∼12.7% better), and a STOI of 0.90 (compared to 0.86, i.e. ∼4.7% better). Moreover, GTAD-Cl runs with a real-time factor of 0.36× which is better than HybridAHS’s 0.42× (approx. 14% faster). These results give validation to GTAD-CL as a powerful, scalable, and low-latency solution and high-fidelity solution that is superior to state-of-the-art results for varying acoustic scenarios.</div></div>","PeriodicalId":56010,"journal":{"name":"Egyptian Informatics Journal","volume":"33 ","pages":"Article 100892"},"PeriodicalIF":4.3,"publicationDate":"2026-03-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"146078372","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":3,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

引用次数: 0

Reliability-oriented offloading of dependent tasks based on topology reconstruction in Industrial Internet Edge Computing 工业互联网边缘计算中基于拓扑重构的依赖任务面向可靠性卸载

IF 4.3 3区计算机科学 Q1 COMPUTER SCIENCE, ARTIFICIAL INTELLIGENCE

Egyptian Informatics Journal

Pub Date : 2026-03-01 Epub Date: 2026-02-06 DOI: 10.1016/j.eij.2026.100909

Qiong Gu , Huilong Wu , Jialei Liu , Bin Ning , Chunyang Hu , Qiaozhi Hua , Meng Zeng , Kexin Zhang , Yanyan Zhu , Zhiyuan Yuan , JiCheng Wu

To address the reliability challenges arising from dynamic network topology and complex task dependencies in Industrial Internet Edge Computing (IIEC) environment, we propose a topology reconstruction-based reliability-optimized computing offloading method. First, we construct a system model encompassing edge-cloud network platform, industrial cloud platform, Internet of Things (IoT) devices and IoT applications, and establish a mathematical framework integrating transmission delay and reliability models, with the IoT application completion time as the core reliability metric. Second, we creatively combine the Ford–Fulkerson approximation algorithm with a Deep Q-network to optimize microservice topology reconstruction and dynamic computing offloading, thereby reducing communication costs and enhancing service reliability. Experimental results demonstrate that the proposed method significantly outperforms existing approaches in IoT application completion time and reliability levels, providing a novel technical pathway for achieving efficient and reliable operations in IIEC.

针对工业互联网边缘计算（IIEC）环境下动态网络拓扑和复杂任务依赖带来的可靠性挑战，提出了一种基于拓扑重构的可靠性优化计算卸载方法。首先，构建了包含边缘云网络平台、工业云平台、物联网设备和物联网应用的系统模型，并以物联网应用完成时间为核心可靠性度量，建立了传输时延和可靠性模型集成的数学框架。其次，我们创造性地将Ford-Fulkerson近似算法与Deep Q-network相结合，优化微服务拓扑重构和动态计算卸载，从而降低通信成本，提高服务可靠性。实验结果表明，该方法在物联网应用完成时间和可靠性水平上明显优于现有方法，为实现IIEC高效可靠运行提供了新的技术途径。

{"title":"Reliability-oriented offloading of dependent tasks based on topology reconstruction in Industrial Internet Edge Computing","authors":"Qiong Gu , Huilong Wu , Jialei Liu , Bin Ning , Chunyang Hu , Qiaozhi Hua , Meng Zeng , Kexin Zhang , Yanyan Zhu , Zhiyuan Yuan , JiCheng Wu","doi":"10.1016/j.eij.2026.100909","DOIUrl":"10.1016/j.eij.2026.100909","url":null,"abstract":"<div><div>To address the reliability challenges arising from dynamic network topology and complex task dependencies in Industrial Internet Edge Computing (IIEC) environment, we propose a topology reconstruction-based reliability-optimized computing offloading method. First, we construct a system model encompassing edge-cloud network platform, industrial cloud platform, Internet of Things (IoT) devices and IoT applications, and establish a mathematical framework integrating transmission delay and reliability models, with the IoT application completion time as the core reliability metric. Second, we creatively combine the Ford–Fulkerson approximation algorithm with a Deep Q-network to optimize microservice topology reconstruction and dynamic computing offloading, thereby reducing communication costs and enhancing service reliability. Experimental results demonstrate that the proposed method significantly outperforms existing approaches in IoT application completion time and reliability levels, providing a novel technical pathway for achieving efficient and reliable operations in IIEC.</div></div>","PeriodicalId":56010,"journal":{"name":"Egyptian Informatics Journal","volume":"33 ","pages":"Article 100909"},"PeriodicalIF":4.3,"publicationDate":"2026-03-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"146187892","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":3,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

引用次数: 0

SC-PPO: Spectral clustering-guided Proximal Policy Optimization for distributed workflow scheduling in cloud–fog computing 云雾计算中基于谱聚类的分布式工作流调度策略优化

IF 4.3 3区计算机科学 Q1 COMPUTER SCIENCE, ARTIFICIAL INTELLIGENCE

Egyptian Informatics Journal

Pub Date : 2026-03-01 Epub Date: 2026-02-20 DOI: 10.1016/j.eij.2026.100920

Sambeet Patro , Sangram Keshari Swain , S. Sudheer Mangalampalli

In fog–cloud computing, efficient task scheduling is crucial to meet the performance requirements of modern applications such as smart healthcare, intelligent transportation, industrial automation. It needs to process large-scale, latency-sensitive,dependency-rich tasks, which can be modeled as workflow-directed acyclic graphs (DAGs). Existing task scheduling mechanisms face difficulties in managing conflicting objectives such as makespan, energy, fault tolerance. To overcome these difficulties, we introduce a novel hybrid task scheduling framework called SC-PPO, which integrates spectral clustering with Proximal Policy Optimization (PPO), a deep reinforcement learning algorithm. The spectral clustering technique is first used to cluster structurally similar tasks, thereby simplifying task scheduling problem and obtaining a higher-level abstraction for decision-making. A PPO agent is then employed to schedule the clustered tasks based on task characteristics, virtual machine (VM) status, reliability values, resource availability. The PPO agent is trained using a multi-objective reward function that balances makespan, energy, task reliability, and trust-aware VM selection. Simulation experiments were conducted in a diverse fog–cloud simulation environment on the Google Cloud Jobs (GoCJ) dataset. The proposed SC-PPO approach was compared with three representative baselines: the Reliability-Improved Whale Optimization Algorithm (RIWOA), Deep Q-Network (DQN), and Advantage Actor–Critic (A2C) algorithm. The results obtained indicate that the proposed SC-PPO approach outperforms the baselines with more than a 20% improvement in makespan, lower energy consumption, higher reliability scores, and improved scalability for handling large-scale workloads.

在雾云计算中，高效的任务调度对于满足智能医疗、智能交通、工业自动化等现代应用的性能要求至关重要。它需要处理大规模的、延迟敏感的、依赖关系丰富的任务，这些任务可以建模为工作流导向的无环图（dag）。现有的任务调度机制在管理诸如完工时间、能量、容错等冲突目标方面存在困难。为了克服这些困难，我们引入了一种称为SC-PPO的新型混合任务调度框架，该框架将谱聚类与深度强化学习算法近端策略优化（PPO）相结合。首先利用谱聚类技术对结构相似的任务进行聚类，从而简化任务调度问题，为决策提供更高层次的抽象。然后使用PPO代理根据任务特征、虚拟机（VM）状态、可靠性值和资源可用性来调度集群任务。PPO代理使用多目标奖励函数进行训练，该函数平衡了完工时间、能量、任务可靠性和信任感知VM选择。模拟实验在谷歌Cloud Jobs （GoCJ）数据集上的不同雾云模拟环境中进行。提出的SC-PPO方法与三种代表性基线进行了比较：可靠性改进鲸鱼优化算法（RIWOA）、深度Q-Network （DQN）和优势行动者-评论家（A2C）算法。所获得的结果表明，所提出的SC-PPO方法优于基线，在完工时间方面提高了20%以上，能耗更低，可靠性得分更高，并且在处理大规模工作负载时提高了可伸缩性。

{"title":"SC-PPO: Spectral clustering-guided Proximal Policy Optimization for distributed workflow scheduling in cloud–fog computing","authors":"Sambeet Patro , Sangram Keshari Swain , S. Sudheer Mangalampalli","doi":"10.1016/j.eij.2026.100920","DOIUrl":"10.1016/j.eij.2026.100920","url":null,"abstract":"<div><div>In fog–cloud computing, efficient task scheduling is crucial to meet the performance requirements of modern applications such as smart healthcare, intelligent transportation, industrial automation. It needs to process large-scale, latency-sensitive,dependency-rich tasks, which can be modeled as workflow-directed acyclic graphs (DAGs). Existing task scheduling mechanisms face difficulties in managing conflicting objectives such as makespan, energy, fault tolerance. To overcome these difficulties, we introduce a novel hybrid task scheduling framework called SC-PPO, which integrates spectral clustering with Proximal Policy Optimization (PPO), a deep reinforcement learning algorithm. The spectral clustering technique is first used to cluster structurally similar tasks, thereby simplifying task scheduling problem and obtaining a higher-level abstraction for decision-making. A PPO agent is then employed to schedule the clustered tasks based on task characteristics, virtual machine (VM) status, reliability values, resource availability. The PPO agent is trained using a multi-objective reward function that balances makespan, energy, task reliability, and trust-aware VM selection. Simulation experiments were conducted in a diverse fog–cloud simulation environment on the Google Cloud Jobs (GoCJ) dataset. The proposed SC-PPO approach was compared with three representative baselines: the Reliability-Improved Whale Optimization Algorithm (RIWOA), Deep Q-Network (DQN), and Advantage Actor–Critic (A2C) algorithm. The results obtained indicate that the proposed SC-PPO approach outperforms the baselines with more than a 20% improvement in makespan, lower energy consumption, higher reliability scores, and improved scalability for handling large-scale workloads.</div></div>","PeriodicalId":56010,"journal":{"name":"Egyptian Informatics Journal","volume":"33 ","pages":"Article 100920"},"PeriodicalIF":4.3,"publicationDate":"2026-03-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"147396551","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":3,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

引用次数: 0

KDLog: a selective knowledge distillation approach for sequential log anomaly detection KDLog：用于顺序日志异常检测的选择性知识蒸馏方法

IF 4.3 3区计算机科学 Q1 COMPUTER SCIENCE, ARTIFICIAL INTELLIGENCE

Egyptian Informatics Journal

Pub Date : 2026-03-01 Epub Date: 2025-12-27 DOI: 10.1016/j.eij.2025.100879

Hailong Cheng , Shi Ying , Xiaoyu Duan , Wanli Yuan

Log anomaly detection is a critical task for ensuring the reliability of complex systems. However, existing methods often suffer from poor adaptability and substantial retraining overhead as log data evolve. This paper introduces a novel framework called KDLog, a knowledge-distillation-based approach that enables accurate and efficient log anomaly detection in dynamic environments. KDLog employs a two-stage selective-distillation mechanism, in which a lightweight student model is trained using the high-confidence outputs generated by a teacher model, effectively preventing negative knowledge transfer. Compared with state-of-the-art methods, KDLog improves overall accuracy by 4.5%, F1-score by 4.3%, and recall by 3.3% on average across real-world datasets (HDFS and BGL). Moreover, it reduces model update time by 60–78% and achieves a smaller model size, by up to 50%, compared with deep learning baselines such as DeepLog and LogAnomaly. Statistical significance tests confirm the robustness of these improvements. Unlike prior methods, KDLog also demonstrates strong resilience to unseen log patterns, with less than a 4% performance drop under simulated log-template drift. These gains make KDLog a scalable and practical solution for real-time anomaly detection, effectively bridging the gap between high-performance learning and operational efficiency in production environments.

日志异常检测是保证复杂系统可靠性的一项重要任务。然而，随着日志数据的发展，现有方法的适应性差，并且需要大量的再训练开销。本文介绍了一种名为KDLog的新框架，这是一种基于知识提取的方法，可以在动态环境中准确有效地检测日志异常。KDLog采用两阶段选择蒸馏机制，其中使用教师模型生成的高置信度输出训练轻量级学生模型，有效防止负知识转移。与最先进的方法相比，KDLog在真实数据集（HDFS和BGL）上的总体准确率提高了4.5%，f1分数提高了4.3%，召回率平均提高了3.3%。此外，与DeepLog和LogAnomaly等深度学习基线相比，它将模型更新时间减少了60-78%，模型大小减少了50%。统计显著性检验证实了这些改进的稳健性。与以前的方法不同，KDLog还显示出对未见日志模式的强大弹性，在模拟日志模板漂移下性能下降不到4%。这些优点使KDLog成为实时异常检测的可扩展实用解决方案，有效地弥合了生产环境中高性能学习和操作效率之间的差距。

{"title":"KDLog: a selective knowledge distillation approach for sequential log anomaly detection","authors":"Hailong Cheng , Shi Ying , Xiaoyu Duan , Wanli Yuan","doi":"10.1016/j.eij.2025.100879","DOIUrl":"10.1016/j.eij.2025.100879","url":null,"abstract":"<div><div>Log anomaly detection is a critical task for ensuring the reliability of complex systems. However, existing methods often suffer from poor adaptability and substantial retraining overhead as log data evolve. This paper introduces a novel framework called KDLog, a knowledge-distillation-based approach that enables accurate and efficient log anomaly detection in dynamic environments. KDLog employs a two-stage selective-distillation mechanism, in which a lightweight student model is trained using the high-confidence outputs generated by a teacher model, effectively preventing negative knowledge transfer. Compared with state-of-the-art methods, KDLog improves overall accuracy by 4.5%, F1-score by 4.3%, and recall by 3.3% on average across real-world datasets (HDFS and BGL). Moreover, it reduces model update time by 60–78% and achieves a smaller model size, by up to 50%, compared with deep learning baselines such as DeepLog and LogAnomaly. Statistical significance tests confirm the robustness of these improvements. Unlike prior methods, KDLog also demonstrates strong resilience to unseen log patterns, with less than a 4% performance drop under simulated log-template drift. These gains make KDLog a scalable and practical solution for real-time anomaly detection, effectively bridging the gap between high-performance learning and operational efficiency in production environments.</div></div>","PeriodicalId":56010,"journal":{"name":"Egyptian Informatics Journal","volume":"33 ","pages":"Article 100879"},"PeriodicalIF":4.3,"publicationDate":"2026-03-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145884767","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":3,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

引用次数: 0

Adaptive acoustic feedback control in aphasia Therapy: A Graph-Based learning approach for Unintended resonance suppression in Mandarin (Chinese)-Speaking aphasic patients 失语症治疗中的自适应声反馈控制：基于图的学习方法对华语失语症患者非预期共振抑制

IF 4.3 3区计算机科学 Q1 COMPUTER SCIENCE, ARTIFICIAL INTELLIGENCE

Egyptian Informatics Journal

Pub Date : 2026-03-01 Epub Date: 2026-02-28 DOI: 10.1016/j.eij.2026.100908

Jiayi Zhang , Nuria Serrano , Francisco Hernando-Gallego , Mohammad Khishe

Aphasia therapy for Mandarin-speaking patients presents distinct challenges due to the language’s tonal characteristics and the presence of unforeseen vocal resonance, which reduces intelligibility and distorts tone contours. Current automatic speech feedback systems face challenges managing such distortions, especially in real-time and customized clinical contexts. This paper develops a novel framework, named graph-based adaptive acoustic feedback control (GA-AFC), that integrates graph neural networks (GNNs) with reinforcement learning (RL) to model and suppress articulation-resonance mismatches in aphasic speech in a dynamic manner. Unlike black-box automatic speech recognition (ASR) and traditional autoregressive models, GA-AFC constructs an articulation-resonance graph based on acoustic features such as harmonicity, pitch, energy, and Mel-frequency cepstral coefficients (MFCCs). The system utilizes GNN encoders to capture phoneme-tonal transitions and employs an RL policy to adapt acoustic feedback in real-time. Experimental evaluations on three benchmark Mandarin datasets, i.e., Common Voice (Mandarin), AISHELL-1, and HKUST, demonstrate that GA-AFC achieves substantial improvements in both fluency enhancement and recognition accuracy. In the context of aphasic speech, the model achieves an average word error reduction (WER) of 17.2% relative to Wav2Vec 2.0 and 30.1% relative to DeepSpeech, alongside a 14.8% improvement in tone classification accuracy on the HKUST corpus. Regarding resonance suppression, GA-AFC logs a spectral deviation of baseline systems by 28.6%, achieving a MOS score of 4.4 (±0.3) in subjective listening tests, which surpasses all comparative models. Moreover, the system demonstrates rapid convergence, with adaptation times of less than 20 s and feedback latencies of under 140 ms, making it suitable for real-time clinical use. The findings indicate that GA-AFC provides a responsive, adaptable, and clinically applicable framework for customizable speech feedback in Mandarin aphasia therapy, proposing a novel approach to tone- and resonance-sensitive neural interventions in speech rehabilitation.

汉语患者的失语症治疗面临着独特的挑战，因为汉语的音调特征和不可预见的声乐共鸣的存在，降低了可理解性，扭曲了音调轮廓。当前的自动语音反馈系统面临着管理这种扭曲的挑战，特别是在实时和定制的临床环境中。本文开发了一种新的框架，称为基于图的自适应声反馈控制（GA-AFC），该框架将图神经网络（gnn）与强化学习（RL）相结合，以动态的方式建模和抑制失语语音中的发音共振不匹配。与黑盒自动语音识别（ASR）和传统的自回归模型不同，GA-AFC基于谐波、音高、能量和mel -频率倒谱系数（MFCCs）等声学特征构建发音共振图。该系统利用GNN编码器捕获音素-音调转换，并采用RL策略实时适应声学反馈。在Common Voice（普通话）、AISHELL-1和HKUST三个基准普通话数据集上的实验评估表明，GA-AFC在流利度增强和识别准确率方面都取得了实质性的进步。在失语语音的情况下，该模型相对于Wav2Vec 2.0实现了17.2%的平均单词错误减少（WER），相对于DeepSpeech实现了30.1%的平均单词错误减少（WER），同时在香港科大语料库上的音调分类精度提高了14.8%。在共振抑制方面，GA-AFC记录的基线系统的频谱偏差为28.6%，主观听力测试的MOS得分为4.4（±0.3），超过所有比较模型。此外，该系统收敛速度快，适应时间小于20秒，反馈延迟小于140毫秒，适合临床实时应用。研究结果表明，GA-AFC为普通话失语症治疗中可定制的语音反馈提供了一个响应性、适应性和临床应用的框架，为语音康复中音调和共振敏感的神经干预提供了一种新的途径。

{"title":"Adaptive acoustic feedback control in aphasia Therapy: A Graph-Based learning approach for Unintended resonance suppression in Mandarin (Chinese)-Speaking aphasic patients","authors":"Jiayi Zhang , Nuria Serrano , Francisco Hernando-Gallego , Mohammad Khishe","doi":"10.1016/j.eij.2026.100908","DOIUrl":"10.1016/j.eij.2026.100908","url":null,"abstract":"<div><div>Aphasia therapy for Mandarin-speaking patients presents distinct challenges due to the language’s tonal characteristics and the presence of unforeseen vocal resonance, which reduces intelligibility and distorts tone contours. Current automatic speech feedback systems face challenges managing such distortions, especially in real-time and customized clinical contexts. This paper develops a novel framework, named graph-based adaptive acoustic feedback control (GA-AFC), that integrates graph neural networks (GNNs) with reinforcement learning (RL) to model and suppress articulation-resonance mismatches in aphasic speech in a dynamic manner. Unlike black-box automatic speech recognition (ASR) and traditional autoregressive models, GA-AFC constructs an articulation-resonance graph based on acoustic features such as harmonicity, pitch, energy, and Mel-frequency cepstral coefficients (MFCCs). The system utilizes GNN encoders to capture phoneme-tonal transitions and employs an RL policy to adapt acoustic feedback in real-time. Experimental evaluations on three benchmark Mandarin datasets, i.e., Common Voice (Mandarin), AISHELL-1, and HKUST, demonstrate that GA-AFC achieves substantial improvements in both fluency enhancement and recognition accuracy. In the context of aphasic speech, the model achieves an average word error reduction (WER) of 17.2% relative to Wav2Vec 2.0 and 30.1% relative to DeepSpeech, alongside a 14.8% improvement in tone classification accuracy on the HKUST corpus. Regarding resonance suppression, GA-AFC logs a spectral deviation of baseline systems by 28.6%, achieving a MOS score of 4.4 (±0.3) in subjective listening tests, which surpasses all comparative models. Moreover, the system demonstrates rapid convergence, with adaptation times of less than 20 s and feedback latencies of under 140 <em>ms</em>, making it suitable for real-time clinical use. The findings indicate that GA-AFC provides a responsive, adaptable, and clinically applicable framework for customizable speech feedback in Mandarin aphasia therapy, proposing a novel approach to tone- and resonance-sensitive neural interventions in speech rehabilitation.</div></div>","PeriodicalId":56010,"journal":{"name":"Egyptian Informatics Journal","volume":"33 ","pages":"Article 100908"},"PeriodicalIF":4.3,"publicationDate":"2026-03-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"147396705","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":3,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

引用次数: 0