Scholarlink

Malaria, a potentially fatal illness brought on by Plasmodium parasites, requires prompt and precise diagnosis. The conventional diagnostic techniques are laborious and prone to human error. The NIH malaria cell picture dataset is used in this work to investigate deep learning-based malaria categorization. Three CNN models that had already been trained VGG16, ResNet50, and EfficientNetB3 were first assessed. Because VGG16 and ResNet50 performed better than the others, a MultiNet Ensemble model that combines the two architectures was created. The model uses L2 regularization, dropout, batch normalization, and transfer learning with real-time data augmentation (rotation, flipping, shearing, zooming). Using callbacks like EarlyStopping, ReduceLROnPlateau, and ModelCheckpoint, training was optimized. The MultiNet Ensemble outperformed single-model techniques with a 98.33% accuracy rate, excellent precision, recall, and ROC-AUC. This dependable, automated system provides a strong diagnostic tool that is especially appropriate for clinical settings with limited resources.

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

MultiNet Ensemble: A Deep Learning Approach for Accurate Malaria Detection Using Microscopic Images

  • 会议论文

  • 发表日期：

    2025年

查看全文
  • Md. Tofael Ahmed Bhuiyan,
  • Shahriar Manzoor,
  • Khandaker Mohammad Mohi Uddin

摘要

Malaria, a potentially fatal illness brought on by Plasmodium parasites, requires prompt and precise diagnosis. The conventional diagnostic techniques are laborious and prone to human error. The NIH malaria cell picture dataset is used in this work to investigate deep learning-based malaria categorization. Three CNN models that had already been trained VGG16, ResNet50, and EfficientNetB3 were first assessed. Because VGG16 and ResNet50 performed better than the others, a MultiNet Ensemble model that combines the two architectures was created. The model uses L2 regularization, dropout, batch normalization, and transfer learning with real-time data augmentation (rotation, flipping, shearing, zooming). Using callbacks like EarlyStopping, ReduceLROnPlateau, and ModelCheckpoint, training was optimized. The MultiNet Ensemble outperformed single-model techniques with a 98.33% accuracy rate, excellent precision, recall, and ROC-AUC. This dependable, automated system provides a strong diagnostic tool that is especially appropriate for clinical settings with limited resources.