Scholarlink

<p>Acute lymphoblastic leukemia (ALL) is a critical hematological malignancy where prompt and accurate diagnosis is essential for effective treatment. This study proposes a novel methodology combining efficient image segmentation with a stacked ensemble of pre-trained convolutional neural networks (CNNs) and XGBoost (FXGBoost) for accurately classifying peripheral blood smear (PBS) images. The used dataset includes 3,256 annotated PBS images from 89 patients covering benign and malignant cases. The proposed model achieves outstanding diagnostic performance, with an accuracy of 99.39% and an F1-score of 0.9963, outperforming several existing deep learning and ensemble methods. Statistical validation using confidence intervals and Tukey honestly significant difference (HSD) testing confirms the results’ significance. Comparative evaluations also include transformer-based models, and inference time per image (ITP) analysis, which is provided to assess computational feasibility. To ensure clinical applicability, we incorporate model interpretability using Gradient-weighted Class Activation Map (Grad-CAM) and address challenges such as class imbalance and overfitting. We acknowledge limitations related to dataset scope and generalizability, and future directions include domain adaptation, explainability enhancement, and real-time deployment. This work contributes a robust and clinically relevant framework for ALL diagnoses, demonstrating how AI-based tools can augment medical decision-making in real-world 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

Fast Detection of Acute Lymphoblastic Leukemia Through Stacked Pre-trained Ensemble Learning and Efficient Segmentation

  • 期刊论文

  • 发表日期：

    2025年7月9日

查看全文
  • Muhammad A. O. Ahmed,
  • Raed Alotaibi,
  • Yasser Abdel Satar,
  • Nora Gaber,
  • Nahla F. Omran,
  • Omar Reyad

摘要

Acute lymphoblastic leukemia (ALL) is a critical hematological malignancy where prompt and accurate diagnosis is essential for effective treatment. This study proposes a novel methodology combining efficient image segmentation with a stacked ensemble of pre-trained convolutional neural networks (CNNs) and XGBoost (FXGBoost) for accurately classifying peripheral blood smear (PBS) images. The used dataset includes 3,256 annotated PBS images from 89 patients covering benign and malignant cases. The proposed model achieves outstanding diagnostic performance, with an accuracy of 99.39% and an F1-score of 0.9963, outperforming several existing deep learning and ensemble methods. Statistical validation using confidence intervals and Tukey honestly significant difference (HSD) testing confirms the results’ significance. Comparative evaluations also include transformer-based models, and inference time per image (ITP) analysis, which is provided to assess computational feasibility. To ensure clinical applicability, we incorporate model interpretability using Gradient-weighted Class Activation Map (Grad-CAM) and address challenges such as class imbalance and overfitting. We acknowledge limitations related to dataset scope and generalizability, and future directions include domain adaptation, explainability enhancement, and real-time deployment. This work contributes a robust and clinically relevant framework for ALL diagnoses, demonstrating how AI-based tools can augment medical decision-making in real-world settings.