Scholarlink

Anaemia is a global public health concern affecting children and pregnant women, occurring from the reduction in red blood cells and reducing the oxygen transport. Traditional anaemia detection methods involve costly, time consuming and painful blood draws from veins. This study focuses on non-invasive alternatives, emphasizing cost effectiveness and efficiency crucial in resource-constrained environments. With billions of individuals being affected globally, early diagnosis is imperative to prevent associated disorders. By exploring accessible methods, our research aims to enhance timely identification and intervention for improved global public health. This research leverages a public dataset named EYES-DEFY-ANAEMIA, comprising 218 eye conjunctiva photos from Italian and Indian patients. Convolution bases of pre-trained Convolutional Neural Network (CNN) models were used for feature extraction in conjunction with conventional machine learning (ML) models as classifiers. Notably, the optimal performance was achieved by using the MobileNetV2 convolutional base as the feature extractor in conjunction with a K-Nearest neighbour classifier. The results demonstrated an impressive recall of 0.957, coupled with an F1 score of 0.880 and accuracy of 0.857. This approach showcases the efficacy of combining deep learning based feature extraction with traditional ML models for robust classification in the context of eye conjunctiva images, specifically addressing the challenges of anaemia detection.

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

Anaemia Prediction from Conjunctiva Images: Leveraging Pre-trained CNNs for Feature Extraction and Traditional ML for Classification

  • 会议论文

  • 发表日期：

    2025年

查看全文
  • Sayan Paul,
  • Bholanath Roy,
  • Vijay Bhaskar Semwal

摘要

Anaemia is a global public health concern affecting children and pregnant women, occurring from the reduction in red blood cells and reducing the oxygen transport. Traditional anaemia detection methods involve costly, time consuming and painful blood draws from veins. This study focuses on non-invasive alternatives, emphasizing cost effectiveness and efficiency crucial in resource-constrained environments. With billions of individuals being affected globally, early diagnosis is imperative to prevent associated disorders. By exploring accessible methods, our research aims to enhance timely identification and intervention for improved global public health. This research leverages a public dataset named EYES-DEFY-ANAEMIA, comprising 218 eye conjunctiva photos from Italian and Indian patients. Convolution bases of pre-trained Convolutional Neural Network (CNN) models were used for feature extraction in conjunction with conventional machine learning (ML) models as classifiers. Notably, the optimal performance was achieved by using the MobileNetV2 convolutional base as the feature extractor in conjunction with a K-Nearest neighbour classifier. The results demonstrated an impressive recall of 0.957, coupled with an F1 score of 0.880 and accuracy of 0.857. This approach showcases the efficacy of combining deep learning based feature extraction with traditional ML models for robust classification in the context of eye conjunctiva images, specifically addressing the challenges of anaemia detection.