文献速递神经MR

超越复杂架构:用于脑MRI稳健阿尔茨海默病分类的简化CNN流程

Beyond complex architectures: a streamlined CNN pipeline for robust Alzheimer's disease classification from brain MRI.
PubMed 文献 · Neuroradiology · 2026-07-04

本研究采用卷积神经网络(CNN)深度学习流程,基于OASIS和ADNI两个公开数据集(ADNI含21,324张MRI图像,OASIS含6,400张MRI图像),区分阿尔茨海默病、轻度认知障碍(MCI)与健康对照。模型在ADNI图像上准确率达99.67%,在OASIS图像上达99.06%,作者希望该技术能辅助医生预测和检测阿尔茨海默病。

要点

  • CNN流程在ADNI和OASIS数据集上分别达到99.67%和99.06%的分类准确率。
  • 研究使用公开数据集,按阿尔茨海默病、MCI、健康对照三类进行MRI图像分类。
  • 该方法为辅助阿尔茨海默病早期检测提供了简化深度学习流程的参考。

核心结论

  • 简化CNN流程在ADNI和OASIS脑MRI数据集上分别取得99.67%和99.06%的准确率,可有效分类阿尔茨海默病、MCI与健康对照。

研究设计与样本

  • 研究为基于深度学习的分类研究,采用卷积神经网络(CNN)流程。
  • 数据来自OASIS和ADNI两个公开数据集。
  • ADNI数据包含21,324张MRI图像:MCI患者7,572张、健康对照5,904张、阿尔茨海默病患者7,848张;OASIS数据包含6,400张MRI图像。
  • 参考标准为数据集原有的阿尔茨海默病、MCI、健康对照分类标签。

诊断标准与关键数值

  • CNN模型在ADNI图像上的分类准确率为99.67%。
  • CNN模型在OASIS图像上的分类准确率为99.06%。

临床应用与阅片要点

  • 该简化CNN流程可用于脑MRI的阿尔茨海默病、MCI与健康对照三分类辅助分析。
  • 模型在公开数据集上准确率高,但临床落地前需在本地数据上验证泛化能力。

局限性

  • 原文未报告外部独立验证、多中心数据或前瞻性临床验证。
  • 数据集类别标签作为参考标准,未说明影像学诊断细节与标注一致性。

以上解读依据论文摘要(全文未开放获取或许可不允许翻译转载)。

英文原文(摘要)· 点击展开

BACKGROUND AND PURPOSE: Alzheimer's disease, a common type of dementia, gradually steals memories and impacts daily life as brain cells deteriorate. We explored how Artificial Intelligence (AI) could help spot early signs of Alzheimer's using MRI brain scans.

MATERIALS AND METHODS: Our study focused on a deep learning approach, specifically a convolutional neural network (CNN), to distinguish between Alzheimer's, Mild Cognitive Impairment (MCI), and healthy individuals. We used two well-known datasets, OASIS and ADNI, for this work. After carefully preparing the ADNI data (which included 21,324 MRI images: 7,572 from MCI patients, 5,904 from healthy controls, and 7,848 from Alzheimer's patients) and the OASIS data (6,400 MRI images), our model performed exceptionally well.

RESULTS: The CNN Model achieved an accuracy of 99.67% with the ADNI images and 99.06% with the OASIS images. These encouraging results, which stand up well against other studies, show that our method can effectively analyze large amounts of data and accurately classify Alzheimer's.

CONCLUSIONS: Our main hope is that this kind of technology can give doctors and caregivers better tools to predict and detect the disease, ultimately saving time, reducing costs, and helping those affected by Alzheimer's.

来源:Neuroradiology,查看原文。版权归原作者所有。

引用

  1. Jabli MA, Mourad M. Beyond complex architectures: a streamlined CNN pipeline for robust Alzheimer's disease classification from brain MRI. Neuroradiology. 2026. PMID: 42400646. doi:10.1007/s00234-026-04088-4

供执业医师工作参考,不替代临床判断;数值与分级以指南原文为准。