Related Experiment Video
Updated: May 31, 2025
05:35
Author Spotlight: Developing a Point-of-Care Hemoglobin Estimation Method for Anemia Management
Published on: January 19, 2024
708
Noninvasive Anemia Detection and Hemoglobin Estimation from Retinal Images Using Deep Learning: A Scalable Solution
Rehana Khan1, Vinod Maseedupally1, Kaveri A Thakoor2
1School of Optometry and Vision Science, University of New South Wales, Sydney, Australia.
Translational Vision Science & Technology
|January 23, 2025
Summary
This study developed a deep-learning model for noninvasive anemia detection using fundus images. The InceptionV3 model achieved high accuracy in predicting anemia and estimating hemoglobin levels.
Area of Science:
- Ophthalmology
- Medical Imaging
- Artificial Intelligence
Background:
- Anemia diagnosis typically requires invasive blood tests.
- Noninvasive methods for anemia detection are highly desirable.
Purpose of the Study:
- To develop and validate a deep-learning model for noninvasive anemia detection.
- To estimate hemoglobin (Hb) levels using fundus images.
- To identify anemia-related retinal features.
Main Methods:
- A deep-convolutional neural network (InceptionV3) was trained on 2265 participants' fundus images and hematological data.
- The model predicted anemia and estimated Hb levels, with performance evaluated using accuracy, sensitivity, specificity, and ROC curves.
- GradCAM saliency maps and image processing quantified anemia-related retinal features.
Main Results:
- The InceptionV3 model achieved 98% accuracy, 99% sensitivity, and 97% specificity for anemia prediction (AUC=0.98).
- Hb level estimation had a mean absolute error of 0.58 g/dL.
- Anemia was associated with increased retinal vessel tortuosity and reduced vessel density near the optic disc.
Conclusions:
- Deep learning models, like InceptionV3, can accurately detect anemia and estimate Hb levels noninvasively from fundus images.
- Retinal vessel analysis provides quantitative biomarkers for anemia detection.
