Class-wise performance on the test set.

Alzheimer’s disease (AD) is a neurodegenerative illness causing a significant decrease in cognitive function, and early, accurate diagnosis is of great therapeutic and diagnostic value. Currently, there is promising potential for applying various types of artificial intelligence techniques, such as enhanced models of deep learning, for classifying Alzheimer’s disease. Therefore, this study proposes an Outline of deep learning to classify Alzheimer’s disease with segmentation using the Multi-Layer U-Net and a hybrid classification approach combining multi-scale EfficientNet with SVM. The proposed methodology consists of a four-phase process: (1) Whole brain segmentation, (2) Gray matter segmentation using multi-layer U-Net segmentation, (3) Feature extraction using Multi-Scale Efficient Net with SVM for classification, and (4) XAI (explainable AI) techniques by integrating Saliency Map Quantitative Analysis for increased clinical trustworthiness and model interpretability. It is found that the experiment results provide promising classification performance for three classes – Alzheimer’s Disease (AD), Mild Cognitive Impairment (MCI) and Cognitively Normal (CN) with an overall accuracy of 97.78% ± 0.54%, precision of 97.18% ± 1.14% (AD), 97.78% ± 0.29% (CN) and 97.03% ± 1.10% (MCI), recall of 97.90% ± 0.77% (AD), 97.49% ± 1.34% (CN) and 97.25% ± 0.99% (MCI), and F1 score of 97.74% ± 0.63% (AD), 97.78% ± 0.79% (CN), and 97.54% ± 0.69%(MCI). The results obtained underscore the elegance of the proposed approach in correctly classifying Alzheimer’s disease stages. Future work will evaluate the model on publicly accessible Alzheimer’s disease MRI datasets and incorporate advanced XAI techniques for increased interpretability and diagnostic reliability. The work focuses on Human Health.

阿尔茨海默病（Alzheimer’s Disease, AD）是一类会引发认知功能显著衰退的神经退行性疾病，早期精准诊断具备极高的诊疗价值。当前，各类人工智能技术（如优化后的深度学习模型）在阿尔茨海默病分类任务中展现出可观的应用潜力。为此，本研究提出一种面向阿尔茨海默病分类的深度学习框架，该框架采用多层U-Net（Multi-Layer U-Net）完成图像分割，并结合多尺度高效神经网络（Multi-Scale EfficientNet）与支持向量机（Support Vector Machine, SVM）构建混合分类方案。所提方法包含四阶段流程：（1）全脑分割；（2）采用多层U-Net完成灰质分割；（3）基于多尺度高效神经网络提取特征，并结合支持向量机完成分类；（4）通过整合显著性图定量分析技术实现可解释人工智能（Explainable AI, XAI），以提升临床可信度与模型可解释性。实验结果显示，本方法在三类任务（阿尔茨海默病AD、轻度认知障碍（Mild Cognitive Impairment, MCI）与认知正常（Cognitively Normal, CN））上均展现出优异的分类性能：整体准确率为97.78%±0.54%；精确率分别为97.18%±1.14%（AD）、97.78%±0.29%（CN）与97.03%±1.10%（MCI）；召回率分别为97.90%±0.77%（AD）、97.49%±1.34%（CN）与97.25%±0.99%（MCI）；F1值分别为97.74%±0.63%（AD）、97.78%±0.79%（CN）与97.54%±0.69%（MCI）。所得结果凸显了所提方法在阿尔茨海默病病程阶段精准分类中的优越性。未来研究将基于公开可用的阿尔茨海默病磁共振成像（Magnetic Resonance Imaging, MRI）数据集对本模型进行验证，并引入更先进的可解释人工智能技术以进一步提升模型可解释性与诊断可靠性。本研究聚焦人类健康领域。