Assessment of frame-averaging algorithms in OCT image analysis

Background and Objective: To evaluate frame registration and averaging algorithms for optical coherence tomography. Patients and Methods: Twenty normal and 20 glaucomatous eyes were imaged. Objective differences were measured by comparing noise variance, spread of edge, and contrast-to-noise ratio. Subjective image quality was also evaluated. Results: Two frame-averaging algorithms (FA400 and FA407) had better noise variance and contrast-to-noise ratio but worse spread of edge than did single frames (P < .01). Both algorithms provided better subjective assessments of structure boundaries than did single images (P < .001). FA407 had significantly lower spread of edge and better internal limiting membrane visualization than FA400. Conclusion: Frame averaging significantly suppressed speckle noise and increased the visibility of retinal structures, but imperfect image registration caused edge blurring that could be detected by the spread of edge parameter. In frame-averaging algorithms, higher contrast-to-noise ratio and lower noise variance indicated better noise suppression, but spread of edge was most sensitive in comparing edge preservation between algorithms.