Related Experiment Video
Updated: Jun 16, 2025
04:48
Application of Deep Learning-Based Medical Image Segmentation via Orbital Computed Tomography
Published on: November 30, 2022
2.7K
Deep Learning-Based Model for Non-invasive Hemoglobin Estimation via Body Parts Images: A Retrospective Analysis and
En-Ting Lin1, Shao-Chi Lu1, An-Sheng Liu1
1Department of Computer Science and Information Engineering, National Taiwan University, CSIE Der Tian Hall No. 1, Sec. 4, Roosevelt Road, Taipei, 106319, Taiwan.
Journal of Imaging Informatics in Medicine
|August 19, 2024
Summary
A new deep learning model, BPANet, uses images from the conjunctiva, palm, and fingernail to predict anemia non-invasively. This AI tool offers a reliable alternative to subjective visual pallor assessments in clinical settings.
Area of Science:
- Medical Imaging
- Artificial Intelligence
- Hematology
Background:
- Anemia affects over a billion people globally, posing a significant health challenge.
- Current diagnostic methods, like hemoglobin measurement, are invasive.
- Physician's visual assessment of pallor is subjective and experience-dependent.
Purpose of the Study:
- To develop a novel, non-invasive deep learning model for anemia prediction.
- To improve anemia detection accuracy by integrating multi-body part imaging.
- To create a reliable tool for real-time anemia assessment in clinical and home settings.
Main Methods:
- A deep learning model, Body-Part-Anemia Network (BPANet), was developed using retrospective data (EYES-DEFY-ANEMIA).
- The model utilizes input images from conjunctiva, palm, and fingernail.
- A fusion attention mechanism and dual loss function were employed to enhance feature learning and handle data imbalance.
Main Results:
- BPANet achieved an accuracy of 0.849 and an F1-score of 0.828 on the retrospective dataset.
- Prospective validation on 101 patients demonstrated prediction accuracy of 0.716 and F1-score of 0.788.
- The model showed excellent performance in non-invasive anemia detection.
Conclusions:
- A novel non-invasive hemoglobin prediction model (BPANet) based on multi-body part imaging was developed and validated.
- BPANet demonstrates potential for accurate and reliable anemia detection.
- The model offers a promising solution for real-time anemia screening in diverse settings.
