Scholarlink

<p>Manual microscopic analysis of blood smears for malaria detection is labor-intensive, time-consuming, and prone to human error. To address these limitations, this study proposes a self-supervised learning approach using RotNet, a rotation-based model, to differentiate between healthy and malaria-infected cells. RotNet learns image representations by predicting the rotation angle of input images, eliminating the need for manually labeled data. This method uses synthetic rotation labels to train the model, significantly reducing dependency on human annotations. The core contribution of this work lies in using embeddings and weights from a rotation-based pretext task on unlabeled images to improve classification accuracy in a downstream convolutional neural network (CNN) task. Key highlights include: (1) an efficient and accurate deep learning solution for malaria detection, (2) a novel rotation-based self-supervised strategy, and (3) a practical approach to tackle the scarcity of labeled medical data. The proposed method was tested on the National Institutes of Health (NIH) malaria dataset, achieving a remarkable AUC of 99.8%, surpassing previously reported results. Moreover, similar accuracy was attained using only 10% of the labeled data compared to fully supervised training, highlighting the effectiveness of self-supervised pretraining. Overall, RotNet demonstrates high potential as a robust, scalable, and efficient tool for automated malaria diagnosis.</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

Optimized deep CNN with rotation-driven features for malaria parasite detection

  • 期刊论文

  • 发表日期：

    2025年9月8日

查看全文
  • Sudhakar Kumar,
  • Sunil K. Singh,
  • Gopal Mengi,
  • Animesh Singh,
  • Arun Kumar Dubey,
  • Brij B. Gupta,
  • Wadee Alhalabi,
  • Varsha Arya,
  • Nadia Nedjah

摘要

Manual microscopic analysis of blood smears for malaria detection is labor-intensive, time-consuming, and prone to human error. To address these limitations, this study proposes a self-supervised learning approach using RotNet, a rotation-based model, to differentiate between healthy and malaria-infected cells. RotNet learns image representations by predicting the rotation angle of input images, eliminating the need for manually labeled data. This method uses synthetic rotation labels to train the model, significantly reducing dependency on human annotations. The core contribution of this work lies in using embeddings and weights from a rotation-based pretext task on unlabeled images to improve classification accuracy in a downstream convolutional neural network (CNN) task. Key highlights include: (1) an efficient and accurate deep learning solution for malaria detection, (2) a novel rotation-based self-supervised strategy, and (3) a practical approach to tackle the scarcity of labeled medical data. The proposed method was tested on the National Institutes of Health (NIH) malaria dataset, achieving a remarkable AUC of 99.8%, surpassing previously reported results. Moreover, similar accuracy was attained using only 10% of the labeled data compared to fully supervised training, highlighting the effectiveness of self-supervised pretraining. Overall, RotNet demonstrates high potential as a robust, scalable, and efficient tool for automated malaria diagnosis.