Scholarlink

<p>The early and accurate diagnosis of acute myeloid leukemia (AML) is essential for adopting effective therapeutic procedures that are highly significant in ensuring the survival of patients. However, present deep learning procedures such as CNNs suffer from resolution dependence and require fixed-sized input images. This resizing might distort the morphological traits or eliminate certain features in blood smear images, causing a possible lowering of diagnostic accuracy. To overcome this drawback, proposed a Modified AlexNet be used with a Spatial Pyramid Pooling (SPP) layer to empower the robust classification of AML subtypes from variable-sized blood smear images without resizing. Since the SPP layer pools the convolutional output into a fixed-length feature vector before the fully connected layer regardless of input size, it removes the restriction on resolution and provides cell morphology preservation. In the proposed network scheme, a pre-trained AlexNet weight is used for transfer learning and fine-tuned on an AML dataset to speed up convergence and attain classification accuracy. The proposed model is compared with baseline models such as CNN, SVM, and Random Forest, and shows superior performance with an accuracy of 95.6%, precision of 94.3%, and ROC-AUC of 97%. Moreover, the model shows consistent and confident classification across multiple AML subtypes, with 96% highest F1-score for the PML_RARA AML subtype. The findings indicate that Modified AlexNet with SPP is a reliable, interpretable, and accurate solution for AML detection, with the possibility of broader applications in medical image analysis.</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 Novel Modified AlexNet with Spatial Pyramid Pooling for Robust Detection of Acute Myeloid Leukemia from Blood Smear Images

  • 期刊论文

  • 发表日期：

    2025年12月27日

查看全文
  • S. Shobha Christila Sobanasingh Devapaul,
  • S. Saravanasundaram Sakthivel,
  • N. P. Ananthamoorthy Nalligoundenpalayam Periyasamy

摘要

The early and accurate diagnosis of acute myeloid leukemia (AML) is essential for adopting effective therapeutic procedures that are highly significant in ensuring the survival of patients. However, present deep learning procedures such as CNNs suffer from resolution dependence and require fixed-sized input images. This resizing might distort the morphological traits or eliminate certain features in blood smear images, causing a possible lowering of diagnostic accuracy. To overcome this drawback, proposed a Modified AlexNet be used with a Spatial Pyramid Pooling (SPP) layer to empower the robust classification of AML subtypes from variable-sized blood smear images without resizing. Since the SPP layer pools the convolutional output into a fixed-length feature vector before the fully connected layer regardless of input size, it removes the restriction on resolution and provides cell morphology preservation. In the proposed network scheme, a pre-trained AlexNet weight is used for transfer learning and fine-tuned on an AML dataset to speed up convergence and attain classification accuracy. The proposed model is compared with baseline models such as CNN, SVM, and Random Forest, and shows superior performance with an accuracy of 95.6%, precision of 94.3%, and ROC-AUC of 97%. Moreover, the model shows consistent and confident classification across multiple AML subtypes, with 96% highest F1-score for the PML_RARA AML subtype. The findings indicate that Modified AlexNet with SPP is a reliable, interpretable, and accurate solution for AML detection, with the possibility of broader applications in medical image analysis.