Serial dual-wavelength illumination for retinal vessel oximetry using a conventional fundus camera: a proof-of-concept study.

Abstract

Purpose: Retinal vessel oximetry requires fundus images at two wavelengths with different hemoglobin absorption characteristics. Simultaneous acquisition using one-chip color cameras may be affected by incomplete spectral separation of detector channels, whereas multi-chip systems such as 3-chip CCD cameras can achieve improved spectral separation. We present a serial dual-wavelength illumination approach integrated into a conventional fundus camera and compare oxygen saturation measurements with those obtained from a commercial retinal oximeter.

Methods: A custom LED illumination unit was fiber-coupled into a fundus camera to provide serial illumination at 548 nm (isosbestic hemoglobin wavelength) and 605 nm (oxygen-sensitive wavelength with differential absorption between oxy- and deoxyhemoglobin), each for 250 ms. Corneal irradiance was 33 mW/cm² at 548 nm and 7 mW/cm² at 605 nm. Sequentially acquired images were registered and analyzed using established retinal oximetry algorithms. Mean arterial and venous oxygen saturation values were compared with measurements from a commercial retinal oximeter in nine participants undergoing routine ophthalmic examination. Linear regression analysis was performed to assess the association between methods, and agreement was additionally evaluated using Bland-Altman analysis.

Results: Mean arterial oxygen saturation was 99.4 ± 5.7% with serial illumination and 100.4 ± 3.0% with standard oximetry. Mean venous oxygen saturation was 71.5 ± 7.0% and 69.5 ± 11.6%, respectively. The mean absolute differences between both techniques were 3.6 ± 2.7% for arteries and 8.4 ± 9.3% for veins. Bland-Altman analysis showed a small arterial bias of 0.94% with limits of agreement from - 7.9% to + 9.8%. For venous measurements, bias was - 1.98% with wider limits of agreement (- 26.9% to + 22.9%). Arterial measurements showed borderline non-significant correlation (R = 0.606, p = 0.084). Venous measurements showed no correlation.

Conclusions: Serial illumination is feasible for retinal vessel oximetry and mitigates detector-level spectral crosstalk associated with simultaneous dual-wavelength imaging using one-chip color cameras. Arterial measurements showed only a small systematic bias compared with the commercial retinal oximeter, whereas venous measurements demonstrated greater variability, indicating the need for further optimization and validation.