Neuroscience informatics最新文献_第3页

Evaluating the effect of point-sampling on univariate point and interval forecasting of cerebral physiologic signals using ARIMA modeling in acute traumatic neural injury 用ARIMA模型评价点采样对急性创伤性神经损伤脑生理信号单变量点区间预测的效果

Neuroscience informatics

Pub Date : 2026-03-01 Epub Date: 2025-12-17 DOI: 10.1016/j.neuri.2025.100248

Nuray Vakitbilir , Kevin Y. Stein , Tobias Bergmann , Noah Silvaggio , Amanjyot Singh Sainbhi , Abrar Islam , Logan Froese , Rakibul Hasan , Mansoor Hayat , Marcel Aries , Frederick A. Zeiler

High-resolution physiological signals, such as intracranial pressure (ICP) and regional cerebral oxygen saturation (rSO₂), are critical for managing traumatic brain injury (TBI) by enabling continuous monitoring of cerebral autoregulation and vascular reactivity. These signals provide essential insights into brain perfusion dynamics, supporting timely clinical interventions. However, the high temporal resolution of these data introduces challenges in real-time use, integration into predictive models, and computational efficiency. Consequently, resolution reduction techniques are essential for simplifying the data while retaining critical features necessary for accurate prediction and modeling. Using the Multi-omic Analytics and Integrative Neuroinformatics in the HUman Brain (MAIN-HUB) Lab database, high-frequency cerebral physiologic dataset, we aimed to evaluate the effects of point-sampling resolution reduction on point and interval predictions using the autoregressive integrated moving average (ARIMA) model for both raw signals and derived indices. Temporal resolution was reduced by selecting the first value within non-overlapping intervals, ranging from 1-min (min) to 12-h windows. A total of 125 patient data was analyzed across various temporal resolutions. The results indicated that ARIMA models performed well at higher resolutions (e.g., 1-min), capturing short-term physiological dynamics with lower errors. However, as resolution decreased, errors and variability increased, particularly for signals like mean arterial pressure (MAP) and cerebral perfusion pressure (CPP), which exhibit rapid or complex physiological changes. The findings underscore the need to carefully balance temporal resolution, model performance, and computational efficiency, especially when dealing with high-frequency physiological data in clinical settings.

高分辨率的生理信号，如颅内压（ICP）和区域脑氧饱和度（rSO2），通过连续监测大脑自动调节和血管反应性，对治疗创伤性脑损伤（TBI）至关重要。这些信号提供了脑灌注动力学的基本见解，支持及时的临床干预。然而，这些数据的高时间分辨率在实时使用、集成到预测模型和计算效率方面带来了挑战。因此，分辨率降低技术对于简化数据，同时保留准确预测和建模所需的关键特征是必不可少的。利用人脑多组学分析和综合神经信息学（MAIN-HUB）实验室数据库、高频脑生理学数据集，我们旨在利用原始信号和衍生指数的自回归综合移动平均（ARIMA）模型，评估点采样分辨率降低对点和区间预测的影响。通过选择非重叠间隔内的第一个值来降低时间分辨率，范围从1分钟（min）到12小时窗口。在不同的时间分辨率下，共分析了125例患者的数据。结果表明，ARIMA模型在高分辨率（例如1分钟）下表现良好，以较低的误差捕获短期生理动力学。然而，随着分辨率的降低，误差和变异性增加，特别是对于平均动脉压（MAP）和脑灌注压（CPP）等信号，它们表现出快速或复杂的生理变化。研究结果强调了仔细平衡时间分辨率、模型性能和计算效率的必要性，特别是在临床环境中处理高频生理数据时。

{"title":"Evaluating the effect of point-sampling on univariate point and interval forecasting of cerebral physiologic signals using ARIMA modeling in acute traumatic neural injury","authors":"Nuray Vakitbilir , Kevin Y. Stein , Tobias Bergmann , Noah Silvaggio , Amanjyot Singh Sainbhi , Abrar Islam , Logan Froese , Rakibul Hasan , Mansoor Hayat , Marcel Aries , Frederick A. Zeiler","doi":"10.1016/j.neuri.2025.100248","DOIUrl":"10.1016/j.neuri.2025.100248","url":null,"abstract":"<div><div>High-resolution physiological signals, such as intracranial pressure (ICP) and regional cerebral oxygen saturation (rSO<sub>2</sub>), are critical for managing traumatic brain injury (TBI) by enabling continuous monitoring of cerebral autoregulation and vascular reactivity. These signals provide essential insights into brain perfusion dynamics, supporting timely clinical interventions. However, the high temporal resolution of these data introduces challenges in real-time use, integration into predictive models, and computational efficiency. Consequently, resolution reduction techniques are essential for simplifying the data while retaining critical features necessary for accurate prediction and modeling. Using the Multi-omic Analytics and Integrative Neuroinformatics in the HUman Brain (MAIN-HUB) Lab database, high-frequency cerebral physiologic dataset, we aimed to evaluate the effects of point-sampling resolution reduction on point and interval predictions using the autoregressive integrated moving average (ARIMA) model for both raw signals and derived indices. Temporal resolution was reduced by selecting the first value within non-overlapping intervals, ranging from 1-min (min) to 12-h windows. A total of 125 patient data was analyzed across various temporal resolutions. The results indicated that ARIMA models performed well at higher resolutions (e.g., 1-min), capturing short-term physiological dynamics with lower errors. However, as resolution decreased, errors and variability increased, particularly for signals like mean arterial pressure (MAP) and cerebral perfusion pressure (CPP), which exhibit rapid or complex physiological changes. The findings underscore the need to carefully balance temporal resolution, model performance, and computational efficiency, especially when dealing with high-frequency physiological data in clinical settings.</div></div>","PeriodicalId":74295,"journal":{"name":"Neuroscience informatics","volume":"6 1","pages":"Article 100248"},"PeriodicalIF":0.0,"publicationDate":"2026-03-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145790330","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

引用次数: 0

Letter to editor regarding “A systematic review and meta-analysis on the diagnostic accuracy of artificial intelligence and computer-aided diagnosis of lumbar prolapsed intervertebral disc” 关于“人工智能和计算机辅助诊断腰椎间盘突出症诊断准确性的系统评价和荟萃分析”的致编辑信

Neuroscience informatics

Pub Date : 2025-12-01 Epub Date: 2025-09-30 DOI: 10.1016/j.neuri.2025.100235

Estanislao Arana

引用次数: 0

Functional MRI in hypertension – A systematic review of brain connectivity, regional activity, and cognitive impairment 高血压的功能性MRI——对脑连通性、区域活动和认知障碍的系统回顾

Neuroscience informatics

Pub Date : 2025-12-01 Epub Date: 2025-09-09 DOI: 10.1016/j.neuri.2025.100233

Sathya Sabina Muthu , Suresh Sukumar , Rajagopal Kadavigere , Shivashankar K.N. , K. Vaishali , Ramesh Babu M.G. , Hari Prakash Palaniswamy , Abhimanyu Pradhan , Winniecia Dkhar , Nitika C. Panakkal , Sneha Ravichandran , Dilip Shettigar , Poovitha Shruthi Paramashiva

Hypertension is increasingly recognized as a key contributor to cognitive decline and brain structure and function alterations. Functional Magnetic Resonance Imaging (fMRI) provides a non-invasive means to detect early disruptions in neural networks before clinical symptoms of cognitive impairment emerge. This systematic review explored the application of fMRI in assessing brain functional changes and cognitive performance in individuals with hypertension. A comprehensive search of electronic databases identified eight relevant studies, most of which employed resting-state fMRI techniques. Findings majorly demonstrated that hypertension is associated with altered connectivity within key neural networks, including the default mode network, frontoparietal network, and salience network. Additional observations included reduced regional homogeneity and changes in low-frequency fluctuations. These neural alterations were linked to decreased memory, executive function, and attention. While the findings support the potential of fMRI as an early biomarker for hypertension-related cognitive impairment, the evidence remains limited by the small number of studies and geographic concentration. Nonetheless, fMRI holds promise for clinical application in identifying individuals at risk and guiding timely interventions. Additional longitudinal studies with broader geographic representation are necessary to confirm these insights and facilitate the integration of fMRI into the routine evaluation and management of hypertension-related brain alterations.

高血压越来越被认为是认知能力下降和大脑结构和功能改变的关键因素。功能磁共振成像（fMRI）提供了一种非侵入性手段，在认知障碍的临床症状出现之前检测神经网络的早期中断。本系统综述探讨了功能磁共振成像在高血压患者脑功能变化和认知表现评估中的应用。通过对电子数据库的全面搜索，确定了8项相关研究，其中大多数采用了静息状态功能磁共振成像技术。研究结果主要表明，高血压与关键神经网络的连通性改变有关，包括默认模式网络、额顶叶网络和显著性网络。其他观察结果包括区域均匀性降低和低频波动的变化。这些神经变化与记忆力、执行功能和注意力下降有关。虽然这些发现支持fMRI作为高血压相关认知障碍的早期生物标志物的潜力，但由于研究数量少和地理集中，证据仍然有限。尽管如此，功能磁共振成像仍有望在临床应用中识别有风险的个体并指导及时干预。需要更多具有更广泛地理代表性的纵向研究来证实这些见解，并促进将功能磁共振成像纳入高血压相关脑改变的常规评估和管理。

{"title":"Functional MRI in hypertension – A systematic review of brain connectivity, regional activity, and cognitive impairment","authors":"Sathya Sabina Muthu , Suresh Sukumar , Rajagopal Kadavigere , Shivashankar K.N. , K. Vaishali , Ramesh Babu M.G. , Hari Prakash Palaniswamy , Abhimanyu Pradhan , Winniecia Dkhar , Nitika C. Panakkal , Sneha Ravichandran , Dilip Shettigar , Poovitha Shruthi Paramashiva","doi":"10.1016/j.neuri.2025.100233","DOIUrl":"10.1016/j.neuri.2025.100233","url":null,"abstract":"<div><div>Hypertension is increasingly recognized as a key contributor to cognitive decline and brain structure and function alterations. Functional Magnetic Resonance Imaging (fMRI) provides a non-invasive means to detect early disruptions in neural networks before clinical symptoms of cognitive impairment emerge. This systematic review explored the application of fMRI in assessing brain functional changes and cognitive performance in individuals with hypertension. A comprehensive search of electronic databases identified eight relevant studies, most of which employed resting-state fMRI techniques. Findings majorly demonstrated that hypertension is associated with altered connectivity within key neural networks, including the default mode network, frontoparietal network, and salience network. Additional observations included reduced regional homogeneity and changes in low-frequency fluctuations. These neural alterations were linked to decreased memory, executive function, and attention. While the findings support the potential of fMRI as an early biomarker for hypertension-related cognitive impairment, the evidence remains limited by the small number of studies and geographic concentration. Nonetheless, fMRI holds promise for clinical application in identifying individuals at risk and guiding timely interventions. Additional longitudinal studies with broader geographic representation are necessary to confirm these insights and facilitate the integration of fMRI into the routine evaluation and management of hypertension-related brain alterations.</div></div>","PeriodicalId":74295,"journal":{"name":"Neuroscience informatics","volume":"5 4","pages":"Article 100233"},"PeriodicalIF":0.0,"publicationDate":"2025-12-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145060025","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

引用次数: 0

Enhancing neuromolecular imaging classification in low-data regimes with generative machine learning: A case study in HDAC PET/MR imaging of alcohol use disorder 利用生成式机器学习增强低数据体制下的神经分子成像分类：酒精使用障碍的HDAC PET/MR成像案例研究

Neuroscience informatics

Pub Date : 2025-12-01 Epub Date: 2025-08-14 DOI: 10.1016/j.neuri.2025.100225

Tyler N. Meyer , Olga Andreeva , Roger D. Weiss , Wei Ding , Iris Shen , Changning Wang , Ping Chen , Tewodros Mulugeta Dagnew

Introduction

Positron Emission Tomography (PET) is a vital modality for investigating brain related disorders. However, data scarcity especially for novel molecular targets like neuroepigenetic enzymes combined with difficult-to-recruit patient populations limits the development of machine learning (ML) models. Our primary objective is to enhance single-subject classification of neuromolecular imaging data and facilitate biomarker discovery. We demonstrate our approach using histone deacetylase (HDAC) PET/MR imaging in Alcohol Use Disorder (AUD).

Methods

We propose Catalysis Training pipeline, a framework that augments real imaging data with high-quality synthetic data generated by a Wasserstein Conditional Generative Adversarial Network (WCGAN). Using [¹¹C]Martinostat PET/MR imaging, we extracted 1-D standardized uptake value ratio (SUVR) tabular features representing HDAC enzyme expression density across eight cingulate subregions. These were used to train and test ML classifiers, including Support Vector Machine (SVM), XGBoost, and Random Forest, under leave-one-out cross-validation.

Results

Integrating synthetic data in the training process improved classification accuracy significantly: +26% for XGBoost and Random Forest (from 59% to 85%), and +18% for SVM (from 70% to 88%). Synthetic samples improved model generalizability. Key hemispheric and subregional cingulate HDAC patterns were also identified as potential biomarkers.

Conclusion

Our results demonstrate that generative AI can help overcome data scarcity in low-data regime neuroimaging applications. Catalysis Training provides a scalable strategy to enhance ML-driven biomarker discovery and disease classification, especially for rare or difficult-to-study disorders like AUD. Clinically, cingulate HDAC expression measured by [¹¹C]Martinostat PET/MR shows promise as an objective biomarker for AUD, complementing DSM-based diagnosis and informing novel treatment strategies.

正电子发射断层扫描（PET）是研究脑相关疾病的重要方式。然而，数据稀缺，尤其是神经表观遗传酶等新分子靶点的数据稀缺，加上难以招募的患者群体，限制了机器学习（ML）模型的发展。我们的主要目标是加强神经分子成像数据的单学科分类，促进生物标志物的发现。我们在酒精使用障碍（AUD）中使用组蛋白去乙酰化酶（HDAC） PET/MR成像证明了我们的方法。方法我们提出了催化训练管道，这是一个利用Wasserstein条件生成对抗网络（WCGAN）生成的高质量合成数据增强真实成像数据的框架。使用[11C]Martinostat PET/MR成像，我们提取了代表8个扣带亚区HDAC酶表达密度的1-D标准化摄取值比（SUVR）表格特征。这些被用来训练和测试ML分类器，包括支持向量机（SVM）、XGBoost和随机森林，在留一交叉验证下。结果在训练过程中集成合成数据显著提高了分类准确率：XGBoost和Random Forest的分类准确率为+26%（从59%提高到85%），SVM的分类准确率为+18%（从70%提高到88%）。合成样本提高了模型的泛化能力。关键的半球和分区域扣带HDAC模式也被确定为潜在的生物标志物。我们的研究结果表明，生成式人工智能可以帮助克服低数据机制神经成像应用中的数据稀缺性。Catalysis Training提供了一种可扩展的策略，以增强机器学习驱动的生物标志物发现和疾病分类，特别是对于罕见或难以研究的疾病，如AUD。在临床上，[11C]Martinostat PET/MR测量的扣带HDAC表达有望作为AUD的客观生物标志物，补充基于dsm的诊断并为新的治疗策略提供信息。

{"title":"Enhancing neuromolecular imaging classification in low-data regimes with generative machine learning: A case study in HDAC PET/MR imaging of alcohol use disorder","authors":"Tyler N. Meyer , Olga Andreeva , Roger D. Weiss , Wei Ding , Iris Shen , Changning Wang , Ping Chen , Tewodros Mulugeta Dagnew","doi":"10.1016/j.neuri.2025.100225","DOIUrl":"10.1016/j.neuri.2025.100225","url":null,"abstract":"<div><h3>Introduction</h3><div>Positron Emission Tomography (PET) is a vital modality for investigating brain related disorders. However, data scarcity especially for novel molecular targets like neuroepigenetic enzymes combined with difficult-to-recruit patient populations limits the development of machine learning (ML) models. Our primary objective is to enhance single-subject classification of neuromolecular imaging data and facilitate biomarker discovery. We demonstrate our approach using histone deacetylase (HDAC) PET/MR imaging in Alcohol Use Disorder (AUD).</div></div><div><h3>Methods</h3><div>We propose <em>Catalysis Training pipeline</em>, a framework that augments real imaging data with high-quality synthetic data generated by a Wasserstein Conditional Generative Adversarial Network (WCGAN). Using [<sup>11</sup>C]Martinostat PET/MR imaging, we extracted 1-D standardized uptake value ratio (SUVR) tabular features representing HDAC enzyme expression density across eight cingulate subregions. These were used to train and test ML classifiers, including Support Vector Machine (SVM), XGBoost, and Random Forest, under leave-one-out cross-validation.</div></div><div><h3>Results</h3><div>Integrating synthetic data in the training process improved classification accuracy significantly: +26% for XGBoost and Random Forest (from 59% to 85%), and +18% for SVM (from 70% to 88%). Synthetic samples improved model generalizability. Key hemispheric and subregional cingulate HDAC patterns were also identified as potential biomarkers.</div></div><div><h3>Conclusion</h3><div>Our results demonstrate that generative AI can help overcome data scarcity in low-data regime neuroimaging applications. Catalysis Training provides a scalable strategy to enhance ML-driven biomarker discovery and disease classification, especially for rare or difficult-to-study disorders like AUD. Clinically, cingulate HDAC expression measured by [<sup>11</sup>C]Martinostat PET/MR shows promise as an objective biomarker for AUD, complementing DSM-based diagnosis and informing novel treatment strategies.</div></div>","PeriodicalId":74295,"journal":{"name":"Neuroscience informatics","volume":"5 4","pages":"Article 100225"},"PeriodicalIF":0.0,"publicationDate":"2025-12-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"144886494","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

引用次数: 0

Deep learning for fetal brain imaging: A systematic review and framework towards privacy-preserving neurodevelopmental informatics 胎儿脑成像的深度学习：对隐私保护神经发育信息学的系统回顾和框架

Neuroscience informatics

Pub Date : 2025-12-01 Epub Date: 2025-10-28 DOI: 10.1016/j.neuri.2025.100241

Sayma Alam Suha, Rifat Shahriyar

Fetal neurodevelopment is a complex process of neural growth during pregnancy, where early detection of abnormalities is vital, and deep learning offers promising techniques for this purpose. The objective of this systematic review is to investigate deep learning applications in fetal neurodevelopment, aiming to synthesize cutting-edge research, examine methodologies, identify research gaps, and propose a federated learning framework. Following PRISMA 2020 guidelines, 55 peer-reviewed articles were selected from an initial 900 records across major databases and additional sources where each article was examined through six specific data extraction criteria. Peer-reviewed articles from 2005 to 2025, specifically those exploring automated deep learning for fetal neurodevelopment using clinical images were included, while non-deep learning analyses were excluded. Risk of bias was qualitatively assessed based on design, data diversity, validation, and reporting. Key scopes of the studies included brain segmentation and regionalization (50.91%), structural measurement (12.73%), image reconstruction, enhancement and synthesis (21.82%) and predictive modeling and clinical classification (14.55%) which also distinguishes between tasks involving pixel-level analysis and image-level predictions. The 55 included studies used diverse datasets (753 to 433,000 images) as well as synthetic image data in some recent works covering wide-ranging gestational ages, mainly using MRI and ultrasound images. The systematic analysis explicitly categorizes each study by task type, applied methodology (U-Net variants, transformer-based models, CNNs, implicit neural representations), and corresponding evaluation metrics—segmentation (DSC, IoU, HD95), classification (Accuracy, Precision, AUC), regression (MAE, RMSE, R²), and reconstruction (PSNR, SSIM), facilitating standardized performance comparisons and establishing clear benchmarks for future research in automated fetal brain imaging. Significant gaps that were identified include inadequate data diversity, privacy measures, limited clinical interpretability and validity of AI models, and insufficient integration of multimodal data. To address these challenges, a unified framework is proposed that integrates multimodal data fusion, explainable artificial intelligence (XAI) paradigms, and federated learning architectures complemented by synthetic data generation techniques to ensure robust privacy preservation in real-world application. This work was not specifically funded, and the review was not registered.

胎儿神经发育是怀孕期间神经生长的一个复杂过程，早期发现异常是至关重要的，深度学习为这一目的提供了有前途的技术。本系统综述的目的是研究深度学习在胎儿神经发育中的应用，旨在综合前沿研究，检查方法，确定研究空白，并提出一个联合学习框架。根据PRISMA 2020指南，从主要数据库和其他来源的初始900条记录中选择了55篇同行评议的文章，其中每篇文章都通过六个特定的数据提取标准进行了检查。2005年至2025年的同行评审文章，特别是那些利用临床图像探索胎儿神经发育自动化深度学习的文章被纳入其中，而非深度学习分析被排除。偏倚风险根据设计、数据多样性、验证和报告进行定性评估。主要研究范围包括脑分割和区域化（50.91%）、结构测量（12.73%）、图像重建、增强和合成（21.82%）以及预测建模和临床分类（14.55%），并区分涉及像素级分析和图像级预测的任务。这55项纳入的研究使用了不同的数据集（753至43.3万张图像），以及一些近期作品中的合成图像数据，涵盖了广泛的胎龄，主要使用MRI和超声图像。系统分析根据任务类型、应用方法（U-Net变量、基于变压器的模型、cnn、隐式神经表征）和相应的评估指标(分割（DSC、IoU、HD95）、分类（准确度、精度、AUC）、回归（MAE、RMSE、R2）和重建（PSNR、SSIM）对每项研究进行了明确的分类，促进了标准化的性能比较，并为未来的自动化胎儿脑成像研究建立了明确的基准。发现的重大差距包括数据多样性不足、隐私措施、人工智能模型的临床可解释性和有效性有限，以及多模式数据整合不足。为了应对这些挑战，提出了一个统一的框架，该框架集成了多模态数据融合、可解释的人工智能（XAI）范式和联合学习架构，辅以合成数据生成技术，以确保在实际应用中健壮的隐私保护。这项工作没有特别资助，评论也没有注册。

{"title":"Deep learning for fetal brain imaging: A systematic review and framework towards privacy-preserving neurodevelopmental informatics","authors":"Sayma Alam Suha, Rifat Shahriyar","doi":"10.1016/j.neuri.2025.100241","DOIUrl":"10.1016/j.neuri.2025.100241","url":null,"abstract":"<div><div>Fetal neurodevelopment is a complex process of neural growth during pregnancy, where early detection of abnormalities is vital, and deep learning offers promising techniques for this purpose. The objective of this systematic review is to investigate deep learning applications in fetal neurodevelopment, aiming to synthesize cutting-edge research, examine methodologies, identify research gaps, and propose a federated learning framework. Following PRISMA 2020 guidelines, 55 peer-reviewed articles were selected from an initial 900 records across major databases and additional sources where each article was examined through six specific data extraction criteria. Peer-reviewed articles from 2005 to 2025, specifically those exploring automated deep learning for fetal neurodevelopment using clinical images were included, while non-deep learning analyses were excluded. Risk of bias was qualitatively assessed based on design, data diversity, validation, and reporting. Key scopes of the studies included brain segmentation and regionalization (50.91%), structural measurement (12.73%), image reconstruction, enhancement and synthesis (21.82%) and predictive modeling and clinical classification (14.55%) which also distinguishes between tasks involving pixel-level analysis and image-level predictions. The 55 included studies used diverse datasets (753 to 433,000 images) as well as synthetic image data in some recent works covering wide-ranging gestational ages, mainly using MRI and ultrasound images. The systematic analysis explicitly categorizes each study by task type, applied methodology (U-Net variants, transformer-based models, CNNs, implicit neural representations), and corresponding evaluation metrics—segmentation (DSC, IoU, HD95), classification (Accuracy, Precision, AUC), regression (MAE, RMSE, R<sup>2</sup>), and reconstruction (PSNR, SSIM), facilitating standardized performance comparisons and establishing clear benchmarks for future research in automated fetal brain imaging. Significant gaps that were identified include inadequate data diversity, privacy measures, limited clinical interpretability and validity of AI models, and insufficient integration of multimodal data. To address these challenges, a unified framework is proposed that integrates multimodal data fusion, explainable artificial intelligence (XAI) paradigms, and federated learning architectures complemented by synthetic data generation techniques to ensure robust privacy preservation in real-world application. This work was not specifically funded, and the review was not registered.</div></div>","PeriodicalId":74295,"journal":{"name":"Neuroscience informatics","volume":"5 4","pages":"Article 100241"},"PeriodicalIF":0.0,"publicationDate":"2025-12-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145465619","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

引用次数: 0

Smart web interface for student mental health prediction using machine learning with blockchain technology 使用区块链技术的机器学习进行学生心理健康预测的智能网络界面

Neuroscience informatics

Pub Date : 2025-12-01 Epub Date: 2025-09-30 DOI: 10.1016/j.neuri.2025.100236

Mishu Deb Nath , Md. Khabir Uddin Ahamed , Omayer Ahmed , Tanvir Ahmed , Sujit Roy , Mohammed Nasir Uddin

Student mental health is becoming a growing global concern, with more students facing psychological stress, anxiety, and related disorders. These mental health challenges often develop gradually and, if ignored, can negatively affect a student's academic performance and personal life. Early detection is essential, but high costs, limited resources, and time constraints often hinder it. The study proposes a machine learning-based approach to predict and assess student mental health, addressing this problem. Using rich psychological and behavioral data, the system can identify early signs of mental distress. An extensive evaluation of 12 machine learning models identified the top six performers. Logistic regression, Decision Tree, Extra Tree, Adaboost, Gradient Boosting, and XGBoost. Among these, the fine-tuned Random Forest algorithm achieved the highest performance, with an impressive accuracy of 95.6%. To ensure practical implementation, a Streamlit-based application was developed. This application enables educators and mental health professionals to perform real-time analysis and receive predictions in a clear and user-friendly format. The study incorporates blockchain technology to ensure the secure handling of sensitive data. Data collected through the Web interface, such as responses to mental health questionnaires, is securely stored using blockchain technology. This integrated system offers a reliable and scalable solution for monitoring and supporting student mental health.

随着越来越多的学生面临心理压力、焦虑和相关障碍，学生心理健康正在成为一个日益受到全球关注的问题。这些心理健康挑战往往是逐渐发展的，如果忽视，可能会对学生的学习成绩和个人生活产生负面影响。早期检测至关重要，但高昂的成本、有限的资源和时间限制往往会阻碍检测。该研究提出了一种基于机器学习的方法来预测和评估学生的心理健康，以解决这一问题。利用丰富的心理和行为数据，该系统可以识别出精神困扰的早期迹象。对12个机器学习模型的广泛评估确定了表现最好的6个模型。逻辑回归，决策树，额外树，Adaboost，梯度增强和XGBoost。其中，经过微调的Random Forest算法取得了最高的性能，准确率达到了惊人的95.6%。为了确保实际实现，开发了一个基于streamlite的应用程序。该应用程序使教育工作者和心理健康专业人员能够以清晰和用户友好的格式进行实时分析和接收预测。该研究采用区块链技术来确保敏感数据的安全处理。通过Web界面收集的数据，如对心理健康问卷的回答，使用区块链技术安全地存储。这个综合系统为监测和支持学生的心理健康提供了可靠和可扩展的解决方案。

{"title":"Smart web interface for student mental health prediction using machine learning with blockchain technology","authors":"Mishu Deb Nath , Md. Khabir Uddin Ahamed , Omayer Ahmed , Tanvir Ahmed , Sujit Roy , Mohammed Nasir Uddin","doi":"10.1016/j.neuri.2025.100236","DOIUrl":"10.1016/j.neuri.2025.100236","url":null,"abstract":"<div><div>Student mental health is becoming a growing global concern, with more students facing psychological stress, anxiety, and related disorders. These mental health challenges often develop gradually and, if ignored, can negatively affect a student's academic performance and personal life. Early detection is essential, but high costs, limited resources, and time constraints often hinder it. The study proposes a machine learning-based approach to predict and assess student mental health, addressing this problem. Using rich psychological and behavioral data, the system can identify early signs of mental distress. An extensive evaluation of 12 machine learning models identified the top six performers. Logistic regression, Decision Tree, Extra Tree, Adaboost, Gradient Boosting, and XGBoost. Among these, the fine-tuned Random Forest algorithm achieved the highest performance, with an impressive accuracy of 95.6%. To ensure practical implementation, a Streamlit-based application was developed. This application enables educators and mental health professionals to perform real-time analysis and receive predictions in a clear and user-friendly format. The study incorporates blockchain technology to ensure the secure handling of sensitive data. Data collected through the Web interface, such as responses to mental health questionnaires, is securely stored using blockchain technology. This integrated system offers a reliable and scalable solution for monitoring and supporting student mental health.</div></div>","PeriodicalId":74295,"journal":{"name":"Neuroscience informatics","volume":"5 4","pages":"Article 100236"},"PeriodicalIF":0.0,"publicationDate":"2025-12-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145220108","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

引用次数: 0

Morphometric characterization of early- and late-onset Parkinson's disease: An ROI-based study of classification and correlation 早发和晚发帕金森病的形态计量学特征：基于roi的分类和相关性研究

Neuroscience informatics

Pub Date : 2025-12-01 Epub Date: 2025-08-22 DOI: 10.1016/j.neuri.2025.100228

Sadhana Kumari , Bharti Rana , Shefali Chaudhary , Roopa Rajan , S. Senthil Kumaran , Achal Kumar Srivastava , Leve Joseph Devarajan

Introduction

Parkinson's disease (PD) is associated with progressive neurodegeneration, particularly involving cortico-basal ganglia-thalamo-cortical circuits that underlie motor and cognitive functions. We investigated the morphological brain features derived from structural MRI to differentiate early (EOPD) and late-onset PD (LOPD) from age-related healthy controls.

Methods

3D T1-weighted MRI was acquired in 114 subjects (27 EOPD, 32 YHC, 28 LOPD, and 27 OHC). Gray matter volume (GMV), white matter volumes (WMV), fractal dimension (FD), gyrification index (GI), and cortical thickness (CT) were extracted using CAT12 software. Three tasks, (i) identification of statistically significant regions, (ii) automatic diagnosis using machine learning using individual and combined features, and (iii) correlation study were performed to quantify the relationship between morphological features and clinical variables.

Results

EOPD exhibited a reduction in GMV and cortical complexity in frontal, parietal and temporal lobes compared to YHC. We achieved the highest classification accuracy of 89.06% using FD and CT for EOPD vs YHC, 90.91% using GMV, WMV and FD for LOPD vs OHC and 89.29% using WMV and FD for EOPD vs LOPD after data augmentation for class balancing. EOPD revealed a negative correlation of GMV with UPDRS II (in medial frontal cortex, precuneus and supplementary motor cortex), FD with UPDRS III in pericalcarine; GI and UPDRS II in transverse temporal and pars opercularis; CT with UPDRS III in superior frontal regions.

Conclusion

Distinct morphometric changes were observed in patients with EOPD and LOPD in comparison with HC, suggesting the utility of morphological measures in early diagnosis of PD.

帕金森病（PD）与进行性神经退行性变有关，特别是涉及运动和认知功能基础的皮质-基底神经节-丘脑-皮质回路。我们研究了来自结构MRI的脑形态学特征，以区分早期（EOPD）和晚发性PD （LOPD）与年龄相关的健康对照。方法对114例患者（EOPD 27例，YHC 32例，LOPD 28例，OHC 27例）进行3d t1加权MRI检查。采用CAT12软件提取脑灰质体积（GMV）、白质体积（WMV）、分形维数（FD）、旋转指数（GI）、皮质厚度（CT）。三个任务，(i)识别统计显著区域，（ii）使用机器学习使用单个和组合特征进行自动诊断，以及（iii）进行相关性研究，以量化形态学特征与临床变量之间的关系。结果与YHC相比，opd表现出额叶、顶叶和颞叶的GMV和皮质复杂性降低。使用FD和CT对EOPD与YHC的分类准确率为89.06%，使用GMV、WMV和FD对LOPD与OHC的分类准确率为90.91%，使用WMV和FD对EOPD与LOPD的分类准确率为89.29%。EOPD显示GMV与UPDRS II（内侧额叶皮质、楔前叶和辅助运动皮质）呈负相关，FD与UPDRS III呈负相关；颞部和包部的GI和UPDRS II；额上区UPDRS III CT检查。结论与HC相比，EOPD和LOPD患者的形态学变化明显，提示形态学检测在PD早期诊断中的应用。

{"title":"Morphometric characterization of early- and late-onset Parkinson's disease: An ROI-based study of classification and correlation","authors":"Sadhana Kumari , Bharti Rana , Shefali Chaudhary , Roopa Rajan , S. Senthil Kumaran , Achal Kumar Srivastava , Leve Joseph Devarajan","doi":"10.1016/j.neuri.2025.100228","DOIUrl":"10.1016/j.neuri.2025.100228","url":null,"abstract":"<div><h3>Introduction</h3><div>Parkinson's disease (PD) is associated with progressive neurodegeneration, particularly involving cortico-basal ganglia-thalamo-cortical circuits that underlie motor and cognitive functions. We investigated the morphological brain features derived from structural MRI to differentiate early (EOPD) and late-onset PD (LOPD) from age-related healthy controls.</div></div><div><h3>Methods</h3><div>3D T1-weighted MRI was acquired in 114 subjects (27 EOPD, 32 YHC, 28 LOPD, and 27 OHC). Gray matter volume (GMV), white matter volumes (WMV), fractal dimension (FD), gyrification index (GI), and cortical thickness (CT) were extracted using CAT12 software. Three tasks, (i) identification of statistically significant regions, (ii) automatic diagnosis using machine learning using individual and combined features, and (iii) correlation study were performed to quantify the relationship between morphological features and clinical variables.</div></div><div><h3>Results</h3><div>EOPD exhibited a reduction in GMV and cortical complexity in frontal, parietal and temporal lobes compared to YHC. We achieved the highest classification accuracy of 89.06% using FD and CT for EOPD vs YHC, 90.91% using GMV, WMV and FD for LOPD vs OHC and 89.29% using WMV and FD for EOPD vs LOPD after data augmentation for class balancing. EOPD revealed a negative correlation of GMV with UPDRS II (in medial frontal cortex, precuneus and supplementary motor cortex), FD with UPDRS III in pericalcarine; GI and UPDRS II in transverse temporal and pars opercularis; CT with UPDRS III in superior frontal regions.</div></div><div><h3>Conclusion</h3><div>Distinct morphometric changes were observed in patients with EOPD and LOPD in comparison with HC, suggesting the utility of morphological measures in early diagnosis of PD.</div></div>","PeriodicalId":74295,"journal":{"name":"Neuroscience informatics","volume":"5 4","pages":"Article 100228"},"PeriodicalIF":0.0,"publicationDate":"2025-12-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"144908997","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

引用次数: 0

Power-to-power cross-frequency coupling as a novel approach for temporal lobe seizure detection and analysis 功率-功率交叉频率耦合作为一种新的颞叶癫痫检测和分析方法

Neuroscience informatics

Pub Date : 2025-12-01 Epub Date: 2025-10-28 DOI: 10.1016/j.neuri.2025.100240

Bar Lehmann, Andrei V. Medvedev

Objective

Power-to-power cross-frequency coupling (CFC) is a novel method to index the dynamic spatio-temporal interactions between brain rhythms, including high frequency oscillations (HFOs). This research evaluates this promising method's capacity for seizure detection with intracranial EEG. Seizures can be conceptualized as composites of different electrographic patterns including (1) spike, (2) ripple-on-spike, and (3) ripple-on-oscillation. This study also performs a basic CFC analysis of each of these components which has potential to further the understanding of epileptogenic processes.

Methods

In this study, deep learning networks including Stacked Sparse Autoencoder (SSAE) and Long Short Term Memory (LSTM) are trained to detect seizures and help characterize CFC patterns for these three common seizure components. The analysis uses intracranial EEG (iEEG) records from the ieeg.org (Mayo Clinic files) database. Temporal Lobe Epilepsy (TLE) seizures (

n = 120

) from 26 patients were analyzed along with segments of background activity. Power-to-power coupling was calculated between all frequencies 1–250 Hz pairwise using the EEGLAB toolbox. CFC matrices of seizure and background activity were used as training or testing inputs to the autoencoder.

Results

The trained network was able to recognize background and seizure segments (not used in training) with a sensitivity of 90.2%, specificity of 96.8% and overall accuracy of 93.4%. The three seizure components (spike, ripple-on-spike, ripple-on-oscillation) were also observed to have unique CFC signatures.

Conclusions

The results provide evidence both for (1) the relevance of power-to-power coupling (PPC) for TLE seizure detection in iEEG, as well as (2) there existing unique PPC signatures of three common seizure components.

目的功率-功率交叉频率耦合（CFC）是一种用于表征脑节律（包括高频振荡）之间动态时空相互作用的新方法。本研究评估了该方法在颅内脑电图检测癫痫发作中的应用前景。癫痫发作可以被定义为不同电图模式的复合，包括(1)尖峰，(2)尖峰上的波纹，(3)振荡上的波纹。本研究还对这些成分进行了基本的氯氟化碳分析，这有可能进一步了解癫痫发生过程。方法在本研究中，深度学习网络包括堆叠稀疏自编码器（SSAE）和长短期记忆（LSTM）进行训练，以检测癫痫发作，并帮助表征这三种常见癫痫发作成分的CFC模式。分析使用来自ieeg.org（梅奥诊所档案）数据库的颅内脑电图（iEEG）记录。对26例颞叶癫痫（TLE）发作（n=120）及其背景活动片段进行分析。使用EEGLAB工具箱两两计算所有频率1-250 Hz之间的功率-功率耦合。癫痫发作和背景活动的CFC矩阵被用作自动编码器的训练或测试输入。结果训练后的神经网络能够识别背景和未用于训练的癫痫片段，灵敏度为90.2%，特异性为96.8%，总体准确率为93.4%。三种发作成分（尖峰，尖峰上的波纹，振荡上的波纹）也被观察到具有独特的CFC特征。结论电-电耦合（power-to-power coupling， PPC）与eeg中TLE发作检测的相关性，以及三种常见发作成分存在独特的PPC特征。

{"title":"Power-to-power cross-frequency coupling as a novel approach for temporal lobe seizure detection and analysis","authors":"Bar Lehmann, Andrei V. Medvedev","doi":"10.1016/j.neuri.2025.100240","DOIUrl":"10.1016/j.neuri.2025.100240","url":null,"abstract":"<div><h3>Objective</h3><div>Power-to-power cross-frequency coupling (CFC) is a novel method to index the dynamic spatio-temporal interactions between brain rhythms, including high frequency oscillations (HFOs). This research evaluates this promising method's capacity for seizure detection with intracranial EEG. Seizures can be conceptualized as composites of different electrographic patterns including (1) spike, (2) ripple-on-spike, and (3) ripple-on-oscillation. This study also performs a basic CFC analysis of each of these components which has potential to further the understanding of epileptogenic processes.</div></div><div><h3>Methods</h3><div>In this study, deep learning networks including Stacked Sparse Autoencoder (SSAE) and Long Short Term Memory (LSTM) are trained to detect seizures and help characterize CFC patterns for these three common seizure components. The analysis uses intracranial EEG (iEEG) records from the ieeg.org (Mayo Clinic files) database. Temporal Lobe Epilepsy (TLE) seizures (<span><math><mi>n</mi><mo>=</mo><mn>120</mn></math></span>) from 26 patients were analyzed along with segments of background activity. Power-to-power coupling was calculated between all frequencies 1–250 Hz pairwise using the EEGLAB toolbox. CFC matrices of seizure and background activity were used as training or testing inputs to the autoencoder.</div></div><div><h3>Results</h3><div>The trained network was able to recognize background and seizure segments (not used in training) with a sensitivity of 90.2%, specificity of 96.8% and overall accuracy of 93.4%. The three seizure components (spike, ripple-on-spike, ripple-on-oscillation) were also observed to have unique CFC signatures.</div></div><div><h3>Conclusions</h3><div>The results provide evidence both for (1) the relevance of power-to-power coupling (PPC) for TLE seizure detection in iEEG, as well as (2) there existing unique PPC signatures of three common seizure components.</div></div>","PeriodicalId":74295,"journal":{"name":"Neuroscience informatics","volume":"5 4","pages":"Article 100240"},"PeriodicalIF":0.0,"publicationDate":"2025-12-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145465624","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

引用次数: 0

Entity-augmented neuroscience knowledge retrieval using ontology and semantic understanding capability of LLM 基于LLM本体和语义理解能力的实体增强神经科学知识检索

Neuroscience informatics

Pub Date : 2025-12-01 Epub Date: 2025-10-28 DOI: 10.1016/j.neuri.2025.100237

Pralaypati Ta , Sriram Venkatesaperumal , Keerthi Ram , Mohanasankar Sivaprakasam

Neuroscience research publications encompass a vast wealth of knowledge. Accurately retrieving existing information and discovering new insights from this extensive literature is essential for advancing the field. However, when knowledge is dispersed across multiple sources, current state-of-the-art retrieval methods often struggle to extract the necessary information. A knowledge graph (KG) can integrate and link knowledge from multiple sources. However, existing methods for constructing KGs in neuroscience often rely on labeled data and require domain expertise. Acquiring large-scale, labeled data for a specialized area like neuroscience presents significant challenges. This work proposes novel methods for constructing KG from unlabeled large-scale neuroscience research corpus utilizing large language models (LLM), neuroscience ontology, and text embeddings. We analyze the semantic relevance of neuroscience text segments identified by LLM for building the knowledge graph. We also introduce an entity-augmented information retrieval algorithm to extract knowledge from the KG. Several experiments were conducted to evaluate the proposed approaches. The results demonstrate that our methods significantly enhance knowledge discovery from the unlabeled neuroscience research corpus. The performance of the proposed entity and relation extraction method is comparable to the existing supervised method. It achieves an F1 score of 0.84 for entity extraction from the unlabeled data. The knowledge obtained from the KG improves answers to over 52% of neuroscience questions from the PubMedQA dataset and questions generated using selected neuroscience entities.

神经科学研究出版物包含了大量的知识。准确地检索现有信息，并从这些广泛的文献中发现新的见解，对于推进该领域至关重要。然而，当知识分散在多个来源时，当前最先进的检索方法往往难以提取必要的信息。知识图（KG）可以整合和链接来自多个来源的知识。然而，现有的神经科学中构建kg的方法往往依赖于标记数据，需要领域的专业知识。为像神经科学这样的专业领域获取大规模的、有标签的数据是一项重大挑战。这项工作提出了利用大型语言模型（LLM）、神经科学本体和文本嵌入从未标记的大规模神经科学研究语料库中构建KG的新方法。我们分析由LLM识别的神经科学文本片段的语义相关性，构建知识图谱。我们还引入了一种实体增强信息检索算法来从知识库中提取知识。进行了几个实验来评估所提出的方法。结果表明，我们的方法显著提高了未标记神经科学研究语料库的知识发现。所提出的实体和关系提取方法的性能与现有的监督方法相当。对于从未标记数据中提取实体，它达到了0.84的F1分数。从KG获得的知识提高了来自PubMedQA数据集和使用选定的神经科学实体生成的问题的52%以上的神经科学问题的答案。

{"title":"Entity-augmented neuroscience knowledge retrieval using ontology and semantic understanding capability of LLM","authors":"Pralaypati Ta , Sriram Venkatesaperumal , Keerthi Ram , Mohanasankar Sivaprakasam","doi":"10.1016/j.neuri.2025.100237","DOIUrl":"10.1016/j.neuri.2025.100237","url":null,"abstract":"<div><div>Neuroscience research publications encompass a vast wealth of knowledge. Accurately retrieving existing information and discovering new insights from this extensive literature is essential for advancing the field. However, when knowledge is dispersed across multiple sources, current state-of-the-art retrieval methods often struggle to extract the necessary information. A knowledge graph (KG) can integrate and link knowledge from multiple sources. However, existing methods for constructing KGs in neuroscience often rely on labeled data and require domain expertise. Acquiring large-scale, labeled data for a specialized area like neuroscience presents significant challenges. This work proposes novel methods for constructing KG from unlabeled large-scale neuroscience research corpus utilizing large language models (LLM), neuroscience ontology, and text embeddings. We analyze the semantic relevance of neuroscience text segments identified by LLM for building the knowledge graph. We also introduce an entity-augmented information retrieval algorithm to extract knowledge from the KG. Several experiments were conducted to evaluate the proposed approaches. The results demonstrate that our methods significantly enhance knowledge discovery from the unlabeled neuroscience research corpus. The performance of the proposed entity and relation extraction method is comparable to the existing supervised method. It achieves an F1 score of 0.84 for entity extraction from the unlabeled data. The knowledge obtained from the KG improves answers to over 52% of neuroscience questions from the PubMedQA dataset and questions generated using selected neuroscience entities.</div></div>","PeriodicalId":74295,"journal":{"name":"Neuroscience informatics","volume":"5 4","pages":"Article 100237"},"PeriodicalIF":0.0,"publicationDate":"2025-12-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145465738","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

引用次数: 0

Exploring KAN as a next-generation replacement for MLPs in EEG-based seizure detection 探索KAN作为下一代mlp在基于脑电图的癫痫检测中的替代品

Neuroscience informatics

Pub Date : 2025-12-01 Epub Date: 2025-08-14 DOI: 10.1016/j.neuri.2025.100226

Eman Allogmani

Epilepsy is a chronic neurological disorder characterized by recurrent seizures due to abnormal brain activity. Accurate detection of seizures from electroencephalogram (EEG) signals is critical, but it is often challenged by signal noise and class imbalance in real-world data. In this study, we systematically evaluate Kolmogorov–Arnold Networks (KANs)—a recent neural architecture based on the Kolmogorov–Arnold representation theorem—as an alternative to Multi-Layer Perceptrons (MLPs) for EEG-based seizure classification, with a focus on model robustness under noisy conditions. This is the first comprehensive evaluation of KAN's robustness under multiplicative noise in the context of EEG seizure detection. Experiments were conducted using two widely used EEG datasets: the Bonn dataset and the CHB-MIT Scalp EEG dataset. Across multiple network configurations and varying levels of multiplicative noise, we assess performance using F1 Score, AUROC, AUPRC, Sensitivity, and Specificity. Our findings show that KAN achieves more stable performance than MLPs under noisy conditions, particularly in smaller architectures. These results suggest that KAN may offer a robust and generalizable approach for seizure detection in noise-prone clinical settings.

癫痫是一种慢性神经系统疾病，其特征是由于大脑活动异常引起的反复发作。从脑电图（EEG）信号中准确检测癫痫发作是至关重要的，但它经常受到现实数据中信号噪声和类别不平衡的挑战。在本研究中，我们系统地评估了Kolmogorov-Arnold网络（KANs）——一种基于Kolmogorov-Arnold表示定理的最新神经结构——作为多层感知器（mlp）的替代方案，用于基于脑电图的癫痫分类，重点关注了模型在噪声条件下的鲁棒性。这是在脑电图癫痫发作检测的背景下，首次对KAN在乘性噪声下的鲁棒性进行综合评价。实验使用两个广泛使用的脑电数据集：波恩数据集和CHB-MIT头皮脑电数据集。在多个网络配置和不同级别的乘法噪声中，我们使用F1评分、AUROC、AUPRC、灵敏度和特异性来评估性能。我们的研究结果表明，在噪声条件下，特别是在较小的架构中，KAN比mlp实现了更稳定的性能。这些结果表明，KAN可能为易受噪声影响的临床环境中的癫痫发作检测提供了一种强大且可推广的方法。

{"title":"Exploring KAN as a next-generation replacement for MLPs in EEG-based seizure detection","authors":"Eman Allogmani","doi":"10.1016/j.neuri.2025.100226","DOIUrl":"10.1016/j.neuri.2025.100226","url":null,"abstract":"<div><div>Epilepsy is a chronic neurological disorder characterized by recurrent seizures due to abnormal brain activity. Accurate detection of seizures from electroencephalogram (EEG) signals is critical, but it is often challenged by signal noise and class imbalance in real-world data. In this study, we systematically evaluate Kolmogorov–Arnold Networks (KANs)—a recent neural architecture based on the Kolmogorov–Arnold representation theorem—as an alternative to Multi-Layer Perceptrons (MLPs) for EEG-based seizure classification, with a focus on model robustness under noisy conditions. This is the first comprehensive evaluation of KAN's robustness under multiplicative noise in the context of EEG seizure detection. Experiments were conducted using two widely used EEG datasets: the Bonn dataset and the CHB-MIT Scalp EEG dataset. Across multiple network configurations and varying levels of multiplicative noise, we assess performance using F1 Score, AUROC, AUPRC, Sensitivity, and Specificity. Our findings show that KAN achieves more stable performance than MLPs under noisy conditions, particularly in smaller architectures. These results suggest that KAN may offer a robust and generalizable approach for seizure detection in noise-prone clinical settings.</div></div>","PeriodicalId":74295,"journal":{"name":"Neuroscience informatics","volume":"5 4","pages":"Article 100226"},"PeriodicalIF":0.0,"publicationDate":"2025-12-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"144865012","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

引用次数: 0