This tutorial explains the evolving approaches on deep learning (DL) modeling and their dependence on statistically comprehensive datasets as input in various brain scan neuroimages. Powerful visual modalities, e.g., magnetic resonance images and positron emission tomography, can show neural changes during Alzheimer’s disease (AD) development. Computer vision’s recent success has lent impetus to numerous DL modeling publications reporting accuracy above 90%, using AD NeuroImage (ADNI) datasets. However, several limitations exist when using DL for AD image interpretation. Due to the lack of a comprehensive dataset and medical images’ complexity, there is little to no clinical value in such DL approaches. Furthermore, many of the published research results in the field are not comparable in experimenting with the ADNI datasets without well-accepted evaluation criteria. This tutorial describes the fundamentals and gaps in applying DL methodology over ADNI datasets.

中文翻译：

---

阿尔茨海默氏病深度学习研究简介

本教程介绍了深度学习（DL）建模的不断发展的方法，以及它们对各种大脑扫描神经图像中作为输入的统计综合数据集的依赖性。强大的视觉模式，例如磁共振图像和正电子发射断层扫描，可以显示阿尔茨海默氏病（AD）发展过程中的神经变化。计算机视觉的最近成功为使用AD NeuroImage（ADNI）数据集的许多DL建模出版物提供了90％以上的准确性的动力。但是，使用DL进行AD图像解释时存在一些限制。由于缺乏全面的数据集和医学图像的复杂性，这种DL方法几乎没有临床价值。此外，在没有公认的评估标准的情况下，使用ADNI数据集进行实验时，该领域中许多已发表的研究结果无法与之相比。本教程描述了将DL方法应用于ADNI数据集的基础知识和差距。