Scholarlink

<p>A dangerous hematological malignancy, acute lymphoblastic leukemia (ALL) has a survival rate that is drastically affected by how long it takes to diagnose the disease. Though convolutional neural networks (CNNs) have improved medical imaging, the clinical dependability of most previous research is limited due to their reliance on single models, imbalance in the datasets, and absence of statistical validation. This study proposes an ensemble framework integrating pre-trained CNNs (DenseNet-121, ResNet-34) for feature extraction with machine learning classifiers—Support Vector Machine (SVM), K-Nearest Neighbor (KNN), Random Forest (RF), AdaBoost, and Backpropagation Network (BPN). Experiments on the C-NMC leukemia dataset (10,661 images) show that the ensemble achieves 92.5% accuracy and 93.1% F1-score, outperforming DenseNet-121 and ResNet-34 by 5.6% and 6.3%, respectively. The model also records the highest AUC (0.975) across classifiers. Statistical tests (<i>t</i>-test, Wilcoxon) confirm that the improvements are significant (<i>p</i> &lt; 0.05). The proposed method demonstrates practical potential as an automated clinical decision-support tool, reducing manual interpretation errors and expediting diagnosis. By combining CNN-based deep features with ensemble machine learning, the framework improves robustness, sensitivity, and applicability in real-world hematology workflows.</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

Efficient convolutional neural networks for acute lymphoblastic leukaemia prediction in computer vision

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2025年12月16日

查看全文
  • S. B. Mohan,
  • S. Sathya,
  • S. Rajalakshmi,
  • G. Gurumoorthy,
  • Rajkumar Sivanraju

摘要

A dangerous hematological malignancy, acute lymphoblastic leukemia (ALL) has a survival rate that is drastically affected by how long it takes to diagnose the disease. Though convolutional neural networks (CNNs) have improved medical imaging, the clinical dependability of most previous research is limited due to their reliance on single models, imbalance in the datasets, and absence of statistical validation. This study proposes an ensemble framework integrating pre-trained CNNs (DenseNet-121, ResNet-34) for feature extraction with machine learning classifiers—Support Vector Machine (SVM), K-Nearest Neighbor (KNN), Random Forest (RF), AdaBoost, and Backpropagation Network (BPN). Experiments on the C-NMC leukemia dataset (10,661 images) show that the ensemble achieves 92.5% accuracy and 93.1% F1-score, outperforming DenseNet-121 and ResNet-34 by 5.6% and 6.3%, respectively. The model also records the highest AUC (0.975) across classifiers. Statistical tests (t-test, Wilcoxon) confirm that the improvements are significant (p < 0.05). The proposed method demonstrates practical potential as an automated clinical decision-support tool, reducing manual interpretation errors and expediting diagnosis. By combining CNN-based deep features with ensemble machine learning, the framework improves robustness, sensitivity, and applicability in real-world hematology workflows.