Scholarlink

<p>The increasing prevalence of neurological disorders poses significant challenges for healthcare systems and requires early, accurate diagnosis for effective treatment and management. Although deep learning shows great potential, a comprehensive comparison of modern architectures across different data types and modalities remains limited. This study aims to address that gap by systematically reviewing recent advances in deep learning-based diagnosis of neurological disorders. The review focuses on Deep Convolutional Neural Networks (DCNNs), lightweight CNNs, and transformer-based models, employing the PRISMA methodology to identify and analyze 133 peer-reviewed studies. Of these, 95 studies focus on six major DCNN architectures—AlexNet, U-Net, ResNet, CapsNet, DenseNet, and EfficientNet—while 16 examine lightweight CNNs, and 22 investigate transformer models. DCNN architectures are used for diagnosing various neurological disorders. Conversely, lightweight CNNs mainly utilize neuroimaging and EEG data to detect brain tumors, Alzheimer’s disease, autism, and epilepsy. Meanwhile, transformer-based architectures target conditions such as Alzheimer’s, schizophrenia, stroke, epilepsy, and brain tumors, using neuroimaging, EEG, and behavioral data, with limited research on other disorders. Neuroimaging is the most common data type, featured in 73 DCNN studies, followed by EEG data in 18 studies, with limited use of facial, speech, behavioral, and genetic data. The study also provides a detailed analysis of preprocessing techniques for neuroimaging and signal data, fusion strategies, and post-processing methods essential for diagnosis—areas often overlooked in previous reviews. The findings of this study highlight the dominance of neuroimaging and suggest opportunities to diversify data types, fusion strategies, and underexplored models.</p>

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

Deep Convolutional Neural Networks in Neurological Disorders Diagnosis: Comprehensive Review of Cutting-Edge Architectures, Challenges, and Future Directions

  • 期刊论文

  • 发表日期：

    2025年11月28日

查看全文
  • Hoda Zamani,
  • Zahra Asghari Varzaneh,
  • Mohammad H. Nadimi-Shahraki

摘要

The increasing prevalence of neurological disorders poses significant challenges for healthcare systems and requires early, accurate diagnosis for effective treatment and management. Although deep learning shows great potential, a comprehensive comparison of modern architectures across different data types and modalities remains limited. This study aims to address that gap by systematically reviewing recent advances in deep learning-based diagnosis of neurological disorders. The review focuses on Deep Convolutional Neural Networks (DCNNs), lightweight CNNs, and transformer-based models, employing the PRISMA methodology to identify and analyze 133 peer-reviewed studies. Of these, 95 studies focus on six major DCNN architectures—AlexNet, U-Net, ResNet, CapsNet, DenseNet, and EfficientNet—while 16 examine lightweight CNNs, and 22 investigate transformer models. DCNN architectures are used for diagnosing various neurological disorders. Conversely, lightweight CNNs mainly utilize neuroimaging and EEG data to detect brain tumors, Alzheimer’s disease, autism, and epilepsy. Meanwhile, transformer-based architectures target conditions such as Alzheimer’s, schizophrenia, stroke, epilepsy, and brain tumors, using neuroimaging, EEG, and behavioral data, with limited research on other disorders. Neuroimaging is the most common data type, featured in 73 DCNN studies, followed by EEG data in 18 studies, with limited use of facial, speech, behavioral, and genetic data. The study also provides a detailed analysis of preprocessing techniques for neuroimaging and signal data, fusion strategies, and post-processing methods essential for diagnosis—areas often overlooked in previous reviews. The findings of this study highlight the dominance of neuroimaging and suggest opportunities to diversify data types, fusion strategies, and underexplored models.