Scholarlink

<p>Alzheimer’s disease (AD) is a degenerative neurological condition that severely affects cognitive abilities, highlighting the importance of early and precise diagnosis for timely intervention. Deep learning (DL) models, especially pre-trained convolutional neural networks, have demonstrated effectiveness in classifying diseases using medical imaging. However, optimizing these models remains a challenge for improving classification accuracy. This paper proposes CENRBO-DenseNet121-AD, an optimized DL model for Alzheimer’s disease classification, using MRI scans from the ADNI dataset. Our framework uses an Improved Newton–Raphson-based optimizer based on a Chaotic strategy and Enhanced solution quality ESQ (CENRBO) to improve hyperparameter tuning and convergence efficiency. CENRBO represents a first-of-its-kind hybrid metaheuristic optimizer that unifies Newton–Raphson convergence with adaptive chaotic diversification and convergence-triggered ESQ refinement in a self-regulating framework specifically designed for hyperparameter optimization of deep CNNs in medical imaging. Experimental evaluations demonstrated that CENRBO-DenseNet121-AD achieved a classification accuracy of 99.89%, outperforming manually optimized DenseNet121 and state-of-the-art deep learning models. The proposed model was also compared with pre-trained architectures and traditional metaheuristic-based optimization methods, confirming its superior performance in terms of sensitivity, specificity, precision, and F1-score. Additionally, the effectiveness of CENRBO was validated on complex global optimization tasks using CEC’2022 benchmark functions, further proving its robustness and adaptability. To assess model generalizability, additional experiments were conducted on the OASIS-3 dataset, confirming consistent performance across independent cohorts. Moreover, Grad-CAM visualizations were integrated to highlight discriminative brain regions, enhancing interpretability and clinical trust. These results highlight the potential of CENRBO-DenseNet121-AD as a highly accurate and reliable model for Alzheimer’s disease classification, setting a new benchmark in medical imaging-based diagnostic research through algorithmic innovation, ablation-validated optimization, and clinical-grade performance.</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

A deep learning approach to Alzheimer’s disease classification using an improved Newton–Raphson optimizer

  • 期刊论文

  • 发表日期：

    2026年5月15日

查看全文
  • Marwa M. Emam,
  • Amina Salhi,
  • F. M. Aldosari,
  • Rayan Alshamrani,
  • Ashrf Althbiti,
  • Atef Ismail

摘要

Alzheimer’s disease (AD) is a degenerative neurological condition that severely affects cognitive abilities, highlighting the importance of early and precise diagnosis for timely intervention. Deep learning (DL) models, especially pre-trained convolutional neural networks, have demonstrated effectiveness in classifying diseases using medical imaging. However, optimizing these models remains a challenge for improving classification accuracy. This paper proposes CENRBO-DenseNet121-AD, an optimized DL model for Alzheimer’s disease classification, using MRI scans from the ADNI dataset. Our framework uses an Improved Newton–Raphson-based optimizer based on a Chaotic strategy and Enhanced solution quality ESQ (CENRBO) to improve hyperparameter tuning and convergence efficiency. CENRBO represents a first-of-its-kind hybrid metaheuristic optimizer that unifies Newton–Raphson convergence with adaptive chaotic diversification and convergence-triggered ESQ refinement in a self-regulating framework specifically designed for hyperparameter optimization of deep CNNs in medical imaging. Experimental evaluations demonstrated that CENRBO-DenseNet121-AD achieved a classification accuracy of 99.89%, outperforming manually optimized DenseNet121 and state-of-the-art deep learning models. The proposed model was also compared with pre-trained architectures and traditional metaheuristic-based optimization methods, confirming its superior performance in terms of sensitivity, specificity, precision, and F1-score. Additionally, the effectiveness of CENRBO was validated on complex global optimization tasks using CEC’2022 benchmark functions, further proving its robustness and adaptability. To assess model generalizability, additional experiments were conducted on the OASIS-3 dataset, confirming consistent performance across independent cohorts. Moreover, Grad-CAM visualizations were integrated to highlight discriminative brain regions, enhancing interpretability and clinical trust. These results highlight the potential of CENRBO-DenseNet121-AD as a highly accurate and reliable model for Alzheimer’s disease classification, setting a new benchmark in medical imaging-based diagnostic research through algorithmic innovation, ablation-validated optimization, and clinical-grade performance.