Scholarlink

<p>Anaemia presents a significant health concern globally due to its association with a deficiency in red blood cells or haemoglobin (Hb) levels. However, traditional screening methods are invasive and costly, posing challenges, particularly in resource-constrained environments. As a result, there’s an increasing need for affordable, accurate, portable, non-invasive solutions. This research integrates modern computational approaches with the conventional visual inspection of pallor in the palm to develop a non-invasive method for anaemia screening. The system works by applying and then releasing pressure on the palm to induce pallor, subsequently tracking the color changes over time. A time-domain analysis is performed to identify patterns that correlate with blood haemoglobin (Hb) concentration. The features extracted from this process serve as inputs to deep learning models. Specifically, we evaluate the performance of three recurrent neural network architectures-namely, standard Recurrent Neural Networks (RNNs), Long Short-Term Memory (LSTM), and Bidirectional LSTM (Bi-LSTM)-in combination with Convolutional Neural Networks (CNNs) for accurate Hb level prediction. We have evaluated these prediction models using two different train-test-split scenarios: an 80:20 split and a 70:30 split. Out of these three proposed models, the model using Bi-LSTM exhibits superior performance, achieving a mean RMSE value of 0.44&#xa0;g/dL and classification accuracy of 95.45% in the 80:20 train-test-split scenario, with clinically tested Hb levels serving as the gold standard. In contrast, the 70:30 split ensures consistent performance across all three recurrent networks, offering competitive results compared to the 80:20 split.</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

Estimating Haemoglobin Levels Non-Invasively Through Analysis of Palm Pallor Using Deep Learning Networks

  • 期刊论文

  • 发表日期：

    2026年3月10日

查看全文
  • Sumana Naskar,
  • Abhishek Kesarwani,
  • Sunanda Das,
  • Dakshina Ranjan Kisku,
  • Mamata Dalui

摘要

Anaemia presents a significant health concern globally due to its association with a deficiency in red blood cells or haemoglobin (Hb) levels. However, traditional screening methods are invasive and costly, posing challenges, particularly in resource-constrained environments. As a result, there’s an increasing need for affordable, accurate, portable, non-invasive solutions. This research integrates modern computational approaches with the conventional visual inspection of pallor in the palm to develop a non-invasive method for anaemia screening. The system works by applying and then releasing pressure on the palm to induce pallor, subsequently tracking the color changes over time. A time-domain analysis is performed to identify patterns that correlate with blood haemoglobin (Hb) concentration. The features extracted from this process serve as inputs to deep learning models. Specifically, we evaluate the performance of three recurrent neural network architectures-namely, standard Recurrent Neural Networks (RNNs), Long Short-Term Memory (LSTM), and Bidirectional LSTM (Bi-LSTM)-in combination with Convolutional Neural Networks (CNNs) for accurate Hb level prediction. We have evaluated these prediction models using two different train-test-split scenarios: an 80:20 split and a 70:30 split. Out of these three proposed models, the model using Bi-LSTM exhibits superior performance, achieving a mean RMSE value of 0.44 g/dL and classification accuracy of 95.45% in the 80:20 train-test-split scenario, with clinically tested Hb levels serving as the gold standard. In contrast, the 70:30 split ensures consistent performance across all three recurrent networks, offering competitive results compared to the 80:20 split.