Scholarlink

Early and precise diagnosis of Alzheimer’s Disease is essential to detect early for prompt intervention and effective treatment planning. This study presents a novel optimized CNN-based framework for classifying brain MRI images into four Classes which is imbalance. To address class imbalance, we integrated class weighting and focal loss, ensuring better representation of minority classes while handling hard-to-classify cases. The model achieved remarkable performance, with 99.14% accuracy, precision, recall, F1-score, 99.62% specificity, and 98.59% Cohen’s Kappa, demonstrating robustness across evaluation metrics. Interpretability was enhanced using Local Interpretable Model-agnostic Explanations, providing transparent insights by identifying clinically relevant brain regions influencing predictions. Comparative analysis with state-of-the-art methods, including Siamese CNNs and DenseNet ensembles, highlighted its superiority, while ROC-AUC curves confirmed near-perfect discriminatory capability. This research underscores the potential of combining advanced CNNs with specialized loss functions and Explainable AI techniques for reliable, interpretable, and early AD detection in clinical applications.

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

Integrating XAI with Optimized CNNs: A Novel Approach for Imbalanced Alzheimer’s Disease Classification

  • 会议论文

  • 发表日期：

    2026年

查看全文
  • Soraisam Gobinkumar Singh,
  • Dulumani Das,
  • Utpal Barman

摘要

Early and precise diagnosis of Alzheimer’s Disease is essential to detect early for prompt intervention and effective treatment planning. This study presents a novel optimized CNN-based framework for classifying brain MRI images into four Classes which is imbalance. To address class imbalance, we integrated class weighting and focal loss, ensuring better representation of minority classes while handling hard-to-classify cases. The model achieved remarkable performance, with 99.14% accuracy, precision, recall, F1-score, 99.62% specificity, and 98.59% Cohen’s Kappa, demonstrating robustness across evaluation metrics. Interpretability was enhanced using Local Interpretable Model-agnostic Explanations, providing transparent insights by identifying clinically relevant brain regions influencing predictions. Comparative analysis with state-of-the-art methods, including Siamese CNNs and DenseNet ensembles, highlighted its superiority, while ROC-AUC curves confirmed near-perfect discriminatory capability. This research underscores the potential of combining advanced CNNs with specialized loss functions and Explainable AI techniques for reliable, interpretable, and early AD detection in clinical applications.