Scholarlink

Alzheimer’s disease (AD) is a progressive and irreversible neurological illness that destroys thinking, memory power and being a common disease in the world making difficult task to identify in the early stages. Deep learning models are intensively used for early detection of AD through which brain disorder progression can be reduced. Though a lot of work is carried out for detection, it is very essential to classify the AD stages with significantly good performance. Hence there is a need for AD detection and classification system with effective better high detection accuracy. In this study, we used 51,864 images for AD dataset and all the images are pre-processed to remove skull brains in axial view. We proposed transfer learning of two pre-trained models such as SqueezeNet and ResNet-50 for feature extraction and classification. In our experiment, we considered 8600 images for testing and 43,264 images for training. We achieved 99.1% and 99.8% accuracy in SqueezeNet and ResNet-50 respectively on augmented dataset.

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

Alzheimer Disease Classification using Transfer Learning of Pre-trained Model

  • 章节

  • 发表日期：

    2026年

查看全文
  • U. Chiraag,
  • Soumyalatha Naveen

摘要

Alzheimer’s disease (AD) is a progressive and irreversible neurological illness that destroys thinking, memory power and being a common disease in the world making difficult task to identify in the early stages. Deep learning models are intensively used for early detection of AD through which brain disorder progression can be reduced. Though a lot of work is carried out for detection, it is very essential to classify the AD stages with significantly good performance. Hence there is a need for AD detection and classification system with effective better high detection accuracy. In this study, we used 51,864 images for AD dataset and all the images are pre-processed to remove skull brains in axial view. We proposed transfer learning of two pre-trained models such as SqueezeNet and ResNet-50 for feature extraction and classification. In our experiment, we considered 8600 images for testing and 43,264 images for training. We achieved 99.1% and 99.8% accuracy in SqueezeNet and ResNet-50 respectively on augmented dataset.