Scholarlink

<p>Acute myeloid leukaemia a major cause of morbidity and mortality in any part of the world, with the burden greatly increasing in sub-Saharan Africa. This study developed a Convolutional neural network for the detection and classification of AML. Data for this study were collected from an online data repository that consists of peripheral blood smears selected from 100 patients diagnosed with different subtypes of AML at the Laboratory of Leukaemia Diagnostics at Munich University Hospital, and smears from 100 patients found to exhibit no morphological features of hematological malignancies in the same time frame. The study adopted a ResNet50 model architecture as the base model (a convolutional neural network) and was modified with a custom classification head that reduces the 2048-dimensional feature vector to 1024 before the final 10-class mapping, combined with specific freezing strategies tailored for the Single Cell Morphological Dataset of Leukocytes. The feature selection and model formulation were carried out using a machine-learning toolkit in Python. The Modified ResNet50 model was compared with other state-of-the-art architectures like VGG16 and AlexNet, with our Proposed modified ResNet50 model having the best performance with an accuracy of 91%. The results highlighted the effectiveness of the CNN model and demonstrated its proficiency in identifying key morphological classes. The findings underscore the potential of CNN-based tools to assist clinicians in AML diagnosis, particularly in enhancing speed, accuracy, and consistency in resource-constrained 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

AML Diagnosis via Transfer Learning: A Modified ResNet50 Approach to Leukocyte Classification

  • 期刊论文

  • 发表日期：

    2026年7月8日

查看全文
  • Oladayo Remilekun Falola,
  • Muritala Abiola Asafa,
  • Peter Adebayo Idowu

摘要

Acute myeloid leukaemia a major cause of morbidity and mortality in any part of the world, with the burden greatly increasing in sub-Saharan Africa. This study developed a Convolutional neural network for the detection and classification of AML. Data for this study were collected from an online data repository that consists of peripheral blood smears selected from 100 patients diagnosed with different subtypes of AML at the Laboratory of Leukaemia Diagnostics at Munich University Hospital, and smears from 100 patients found to exhibit no morphological features of hematological malignancies in the same time frame. The study adopted a ResNet50 model architecture as the base model (a convolutional neural network) and was modified with a custom classification head that reduces the 2048-dimensional feature vector to 1024 before the final 10-class mapping, combined with specific freezing strategies tailored for the Single Cell Morphological Dataset of Leukocytes. The feature selection and model formulation were carried out using a machine-learning toolkit in Python. The Modified ResNet50 model was compared with other state-of-the-art architectures like VGG16 and AlexNet, with our Proposed modified ResNet50 model having the best performance with an accuracy of 91%. The results highlighted the effectiveness of the CNN model and demonstrated its proficiency in identifying key morphological classes. The findings underscore the potential of CNN-based tools to assist clinicians in AML diagnosis, particularly in enhancing speed, accuracy, and consistency in resource-constrained settings.