Cansiz Berke, Ilhan Hamza Osman, Aydin Nizamettin, Serbes Gorkem
Department of Electronics and Communication Engineering, Yildiz Technical University, Istanbul, Türkiye.
Department of Computer Engineering, Yildiz Technical University, Istanbul, Türkiye.
Front Neurosci. 2025 Aug 28;19:1638922. doi: 10.3389/fnins.2025.1638922. eCollection 2025.
Alzheimer's disease has been considered one of the most dangerous neurodegenerative health problems. This disease, which is characterized by memory loss, leads to conditions that adversely affect daily life. Early diagnosis is crucial for effective treatment and is achieved through various imaging technologies. However, these methods are quite costly and their results depend on the expertise of the specialist physician. Therefore, deep learning techniques have recently been utilized as decision support tools for Alzheimer's disease.
In this research, the detection of Alzheimer's disease was investigated using a deep learning model applied to electroencephalography signals, taking advantage of olfactory memory. The dataset comprises three categories: healthy individuals, those with amnestic mild cognitive impairment, and Alzheimer's disease patients. The proposed model integrates three distinct feature types through a transformer-based fusion approach for classification. These feature vectors are derived from the Common Spatial Pattern, Covariance matrix-Tangent Space and a Tunable Q-Factor wavelet coefficient mapping.
The results demonstrated that subject-based classification of rose aroma attained a 93.14% accuracy using EEG-recorded olfactory memory responses.
This output has demonstrated superiority over EEG-based results reported in the literature.
阿尔茨海默病被认为是最危险的神经退行性健康问题之一。这种以记忆丧失为特征的疾病会导致对日常生活产生不利影响的状况。早期诊断对于有效治疗至关重要，可通过各种成像技术实现。然而，这些方法成本相当高，且其结果取决于专科医生的专业知识。因此，深度学习技术最近已被用作阿尔茨海默病的决策支持工具。
在本研究中，利用嗅觉记忆，通过应用于脑电图信号的深度学习模型来研究阿尔茨海默病的检测。数据集包括三类：健康个体、遗忘型轻度认知障碍患者和阿尔茨海默病患者。所提出的模型通过基于变压器的融合方法整合三种不同的特征类型进行分类。这些特征向量来自共同空间模式、协方差矩阵 - 切空间和可调Q因子小波系数映射。
结果表明，使用脑电图记录的嗅觉记忆反应，基于受试者的玫瑰香气分类准确率达到了93.14%。
该结果已证明优于文献中报道的基于脑电图的结果。
