Scholarlink

<p>The management of Alzheimer’s disease (AD) and successful clinical intervention depend on an early and precise diagnosis. However, existing Deep Learning (DL) models often struggle with limited accuracy, high error rates and insufficient interpretability. To address these limitations, this research proposes an interpretable DL model called Interpretable Temporal-Spatial Graph Attention Network with Hyperbolic Sine Optimizer (IT-SGAN-HSO). The model integrates multimodal neuroimaging data, including MRI &amp; PET scans, and enhances performance through a multistage pipeline. Pre-processing is carried out using a Hybrid Recursive Reversible Box Filter-based Fast Adaptive Bilateral Filter (H2RBF-FABF) to improve image quality. Segmentation is performed using DeepLabV3 combined with Steerable Transformers (DLV3-STr), while feature extraction and classification utilizes a Hamiltonian Quantum Generative Adversarial Temporal-Spatial Graph Attention Network (HQGATSGAN). The Hyperbolic Sine Optimizer (HSO) further optimizes model training by accelerating convergence and reducing computational complexity. The model was evaluated on two benchmark datasets: OASIS and ADNI. The method achieved an average accuracy of 98.7%, with precision, recall, specificity &amp; <i>F</i>1-score all exceeding 98%. It also demonstrated significantly lower error rates (0.1%) and faster inference time (0.2&#xa0;s) compared to methods. These results highlight the proposed model’s potential for early and reliable AD diagnosis and progression monitoring in clinical settings.</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

Interpretable Temporal-Spatial Graph Attention Network with Hyperbolic Sine Optimizer Algorithm for Alzheimer’s Disease Diagnosis Through Multiscale Feature Modeling

  • 期刊论文

  • 发表日期：

    2025年8月12日

查看全文
  • Biswajit Brahma,
  • Gorli L Aruna Kumari,
  • Bhawani Sankar Panigrahi,
  • Sanjay Kumar Sen,
  • Susanta Kumar Sahoo

摘要

The management of Alzheimer’s disease (AD) and successful clinical intervention depend on an early and precise diagnosis. However, existing Deep Learning (DL) models often struggle with limited accuracy, high error rates and insufficient interpretability. To address these limitations, this research proposes an interpretable DL model called Interpretable Temporal-Spatial Graph Attention Network with Hyperbolic Sine Optimizer (IT-SGAN-HSO). The model integrates multimodal neuroimaging data, including MRI & PET scans, and enhances performance through a multistage pipeline. Pre-processing is carried out using a Hybrid Recursive Reversible Box Filter-based Fast Adaptive Bilateral Filter (H2RBF-FABF) to improve image quality. Segmentation is performed using DeepLabV3 combined with Steerable Transformers (DLV3-STr), while feature extraction and classification utilizes a Hamiltonian Quantum Generative Adversarial Temporal-Spatial Graph Attention Network (HQGATSGAN). The Hyperbolic Sine Optimizer (HSO) further optimizes model training by accelerating convergence and reducing computational complexity. The model was evaluated on two benchmark datasets: OASIS and ADNI. The method achieved an average accuracy of 98.7%, with precision, recall, specificity & F1-score all exceeding 98%. It also demonstrated significantly lower error rates (0.1%) and faster inference time (0.2 s) compared to methods. These results highlight the proposed model’s potential for early and reliable AD diagnosis and progression monitoring in clinical settings.