Highly reliable personalized noninvasive hemoglobin estimation by using Vision Transformers and dual fine-tuning

Area of Science:

• Biomedical Engineering
• Artificial Intelligence in Healthcare
• Hematology

Background:

• Anemia affects billions globally, requiring frequent monitoring.
• Traditional blood tests for hemoglobin are invasive and impractical for continuous assessment.
• Noninvasive diagnostic methods are crucial for early anemia detection.

Purpose of the Study:

• To develop the first patient-specific AI system for noninvasive hemoglobin estimation using palpebral conjunctiva images.
• To enhance anemia screening accuracy and accessibility through personalized diagnostics.
• To reduce reliance on conventional blood tests for hematological monitoring.

Main Methods:

• Utilized a vision transformer (ViT) architecture with dual fine-tuning for personalized hemoglobin estimation.
• Collected a dataset of conjunctival images from the same patients over multiple days for individualized calibration.
• Developed a patient-specific model for noninvasive hemoglobin level assessment.

Main Results:

• Achieved a high R-squared value of 0.94, 98% accuracy, and a mean absolute error of 0.25 g/dL.
• Demonstrated performance comparable to laboratory-based hemoglobin tests.
• Attained 100% sensitivity in detecting all anemic cases, eliminating false-negatives.

Conclusions:

• The developed AI system offers a precise, rapid, and accessible method for anemia screening.
• This noninvasive approach can redefine long-term hematological monitoring, especially in resource-limited settings.
• Improved clinical decision-making and reduced healthcare costs are potential benefits.