Figure 1.

Algorithm workflow. Frames from an AOSLO image sequence were rejected using the conventional definition of an outlier as an observation being at least 3 SDs (σ) away from the mean (μ). They were rejected based on mean pixel intensity (Ī ≤ μ − 3σ or Ī ≥ μ + 3σ), intraframe motion (ΣR² ≥ μ + 3σ), NCC peak height (NCC ≤ μ − 3σ) after motion tracking, and distance to cluster centroid (d ≥ μ + 3σ) after clustering. An additional “strict” rejection was conducted by removing frames with the lowest 20% of NCC peak heights. The frames from each distinct spatial cluster with the lowest ΣR² values were output.