Dataset: Machine Learning applied to Neuroimaging

Research Context Machine learning and deep learning techniques have become increasingly important for analyzing complex neuroimaging and electrophysiological data in the study of Alzheimer’s disease and mild cognitive impairment. These approaches enable the identification of subtle structural, functional, and signal-based patterns that are difficult to detect using traditional analytical methods, supporting early diagnosis and biomarker discovery. What the Dataset Contains This dataset compiles structured information extracted from peer-reviewed studies that apply machine learning and deep learning methods to neuroimaging and electroencephalography data for Alzheimer’s disease, mild cognitive impairment, and cognitively normal populations. The dataset includes curated metadata describing imaging modalities, machine learning algorithms, sample characteristics, validation strategies, main findings, and reported biomarkers. The included studies span multiple data modalities, including structural magnetic resonance imaging, resting-state functional magnetic resonance imaging, positron emission tomography, and electroencephalography, enabling cross-modal comparison of machine learning performance and biomarker relevance. Notable Insights Machine learning models applied to neuroimaging data can reliably differentiate Alzheimer’s disease from mild cognitive impairment and cognitively normal controls. Functional connectivity alterations, particularly within default mode and frontoparietal networks, are recurrent biomarkers in functional magnetic resonance imaging–based studies. Structural magnetic resonance imaging studies consistently identify hippocampal and temporal lobe atrophy as key discriminative features. Electroencephalography-based deep learning approaches demonstrate promising performance for non-invasive classification across cognitive stages. Model performance varies substantially depending on feature extraction methods and validation strategies, highlighting the importance of methodological transparency. Data Interpretation and Intended Use The dataset is designed for meta-analytical research, methodological comparison of machine learning approaches, and educational purposes. Researchers can use this dataset to: Conduct systematic reviews and meta-analyses of machine learning performance in neuroimaging. Compare biomarkers across imaging modalities. Evaluate the impact of validation strategies on reported classification accuracy. Support reproducibility and secondary analyses in computational neuroscience research. Data Acquisition Summary All data were manually extracted from peer-reviewed publications retrieved from major scientific databases and curated into a standardized structure compatible with Mendeley Data. Each study is documented through structured summary files and a consolidated dataset table, ensuring traceability, consistency, and reproducibility.研究背景 在阿尔茨海默病（Alzheimer’s disease）与轻度认知障碍（mild cognitive impairment, MCI）的相关研究中，机器学习与深度学习技术在复杂神经影像学、电生理数据分析领域的重要性与日俱增。此类方法可识别传统分析手段难以捕捉的细微结构、功能及信号模式，为早期诊断与生物标志物发掘提供支撑。 数据集内容 本数据集整合了从同行评议研究中提取的结构化信息，这些研究将机器学习与深度学习方法应用于阿尔茨海默病、轻度认知障碍及认知正常人群的神经影像学与脑电图（electroencephalography, EEG）数据分析。数据集包含经过整理的元数据，涵盖成像模态、机器学习算法、样本特征、验证策略、核心发现与已报道的生物标志物。 纳入的研究覆盖多种数据模态，包括结构磁共振成像（structural magnetic resonance imaging, sMRI）、静息态功能磁共振成像（resting-state functional magnetic resonance imaging, rs-fMRI）、正电子发射断层扫描（positron emission tomography, PET）及脑电图，支持对机器学习性能与生物标志物相关性开展跨模态对比分析。 核心发现 应用于神经影像学数据的机器学习模型可可靠区分阿尔茨海默病、轻度认知障碍与认知正常对照人群。 基于功能磁共振成像的研究中，功能连接异常（尤其是默认模式网络（default mode network）与额顶网络（frontoparietal network）内的连接改变）是反复被报道的生物标志物。 结构磁共振成像相关研究一致将海马体与颞叶萎缩作为核心判别特征。 基于脑电图的深度学习方法在不同认知阶段的无创分类任务中展现出可观的应用潜力。 模型性能随特征提取方法与验证策略的不同存在显著差异，这凸显了方法学透明性的重要价值。 数据解读与预期用途 本数据集专为荟萃分析研究、机器学习方法的方法学对比及教学场景设计。研究人员可利用本数据集开展以下工作： 1. 针对神经影像学领域的机器学习性能开展系统综述与荟萃分析； 2. 对比不同成像模态下的生物标志物； 3. 评估验证策略对已报道分类准确率的影响； 4. 为计算神经科学研究中的可复现性保障与二次分析提供支持。 数据获取概述 所有数据均从主流科学数据库检索的同行评议文献中手动提取，并整理为适配曼德莱数据平台（Mendeley Data）的标准化结构。每项研究均通过结构化摘要文件与整合数据集表格进行记录，确保数据的可追溯性、一致性与可复现性。