Convolutional Neural Networks for Spectroscopic Analysis in Retinal Oximetry

Area of Science:

• Ophthalmology
• Biomedical Engineering
• Medical Imaging

Background:

• Retinal oximetry is crucial for assessing eye hemodynamics and health.
• Current oximetry methods suffer from inconsistent measurements, limiting clinical use.
• Reliable and reproducible oximetry is essential for clinical relevance.

Purpose of the Study:

• To develop an advanced algorithm for accurate multi-wavelength retinal oximetry.
• To overcome limitations of existing oximetry techniques regarding consistency and reproducibility.
• To enable precise measurement of hemoglobin oxygenation in the retina.

Main Methods:

• Developed a novel convolutional neural network (CNN) algorithm for multi-wavelength oximetry.
• The CNN algorithm is calibration-free and does not require prior spectral knowledge.
• Algorithm performance was evaluated against previous techniques and in the presence of spectral variations.

Main Results:

• The CNN algorithm demonstrated significantly improved calculation performance over existing methods.
• Achieved accurate deduction of effective oxygenation (SO₂) and fractional oxygenation.
• The algorithm proved invariable to spectral shifting, enhancing robustness.

Conclusions:

• The developed CNN algorithm offers a reliable and reproducible method for retinal oximetry.
• Accurate and independent measurement of hemoglobin concentrations has high potential for diagnosing and monitoring eye diseases.
• This advancement paves the way for wider clinical adoption of retinal oximetry.