Scholarlink

Alzheimer’s disease is a neurodegenerative disorder characterized by irreversible brain cell damage, leading to progressive cognitive decline and memory impairment. Although currently incurable, early detection is crucial for effective management and slowing disease progression. This study evaluates the performance of deep learning models using transfer learning on the Alzheimer’s Disease Neuroimaging Initiative (ADNI) dataset to classify MRI scans into four categories: Cognitively Normal (CN), Early Mild Cognitive Impairment (EMCI), Late Mild Cognitive Impairment (LMCI), and Alzheimer’s Disease (AD). We compared six pre-trained models (VGG16, VGG19, ResNet50, ResNet152, MobileNetV2, and EfficientNetB7) using two data-splitting approaches: the Late-slice-level and Early-subject-level split. ResNet50 showed the highest performance, achieving an accuracy of 93.47% with data leakage and 78.62% without leakage. We also integrated the Kolmogorov Arnold Network (KAN) with ResNet50, achieving 91.37% and 74.63% accuracies with and without data leakage, respectively, although this integration did not surpass the traditional ResNet50 model. The study further examined model performance using 2D brain slices from different anatomical views (Axial, Coronal, Sagittal) and varying slice counts (5, 10, 15). Model consistency has been assessed through 3-way, binary, and one-vs-rest classification tasks. Moreover, Grad-CAM visualization has been utilized to interpret the model’s understanding of cognitive decline, providing insights into its decision-making process.

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Assessing CNN Models for Multi-stage Alzheimer’s Disease Classification with Data Splitting Techniques

  • 会议论文

  • 发表日期：

    2026年

查看全文
  • Tanush Mohindra,
  • Anshul Sharma,
  • Vandana Bharti,
  • Rishav Singh,
  • Sanjay Kumar Singh

摘要

Alzheimer’s disease is a neurodegenerative disorder characterized by irreversible brain cell damage, leading to progressive cognitive decline and memory impairment. Although currently incurable, early detection is crucial for effective management and slowing disease progression. This study evaluates the performance of deep learning models using transfer learning on the Alzheimer’s Disease Neuroimaging Initiative (ADNI) dataset to classify MRI scans into four categories: Cognitively Normal (CN), Early Mild Cognitive Impairment (EMCI), Late Mild Cognitive Impairment (LMCI), and Alzheimer’s Disease (AD). We compared six pre-trained models (VGG16, VGG19, ResNet50, ResNet152, MobileNetV2, and EfficientNetB7) using two data-splitting approaches: the Late-slice-level and Early-subject-level split. ResNet50 showed the highest performance, achieving an accuracy of 93.47% with data leakage and 78.62% without leakage. We also integrated the Kolmogorov Arnold Network (KAN) with ResNet50, achieving 91.37% and 74.63% accuracies with and without data leakage, respectively, although this integration did not surpass the traditional ResNet50 model. The study further examined model performance using 2D brain slices from different anatomical views (Axial, Coronal, Sagittal) and varying slice counts (5, 10, 15). Model consistency has been assessed through 3-way, binary, and one-vs-rest classification tasks. Moreover, Grad-CAM visualization has been utilized to interpret the model’s understanding of cognitive decline, providing insights into its decision-making process.