Artifacts are not rare edge cases. Motion, low signal, shadowing, segmentation and projection can change retinal images and quantitative outputs. The safest interpretation pairs the processed image with raw structural context and a deliberate quality check.
- Automated segmentation fails more often when pathology distorts normal retinal anatomy.
- OCTA adds artifact classes that can imitate or erase vascular patterns, including projection and flow-threshold effects.
- A device quality score is useful but cannot replace visual inspection of the scan and segmentation boundaries.
Every retinal image is constructed
Retinal imaging feels direct because the outputs are visual. But OCT, OCTA and widefield images are created through acquisition and processing choices: optics collect a signal, software reconstructs it, algorithms align frames, boundaries are segmented and display settings transform measurements into something readable. An artifact is any feature introduced or altered by that pipeline rather than by the underlying biology. Some are obvious. Others can look convincingly anatomical.
Motion: the eye never truly holds still
Microsaccades, drift, blinks and fixation changes can distort scans. Structural OCT may show discontinuities or duplicated features. OCTA is particularly vulnerable because it deliberately interprets change between repeated scans as motion from blood. Eye tracking and registration reduce motion artifacts, but they do not eliminate them. In a large glaucoma OCTA dataset, poor-quality images were associated with multiple artifact types, reinforcing that quality control is a routine part of analysis rather than a research nicety.
Segmentation: when software draws the wrong boundary
Many OCT outputs depend on algorithms tracing layer boundaries. Disease breaks the assumptions those algorithms were trained on. Fluid, epiretinal membranes, atrophy, pigment-epithelium detachments and other distortions can pull boundaries into the wrong place. In a study of 149 eyes with chorioretinal disease plus healthy controls, segmentation error was far more common in diseased eyes; neovascular AMD was among the most problematic groups. A thickness map generated from a wrong boundary can look precise while measuring the wrong anatomy.
Projection: vessels appearing where they are not
OCTA can display superficial-vessel signal in deeper slabs because moving blood casts dynamic shadows onto structures beneath it. Projection-removal algorithms attempt to suppress that effect, but correction is not perfect. A vessel-like pattern in a deep slab may therefore represent true deep flow, residual projection or a mixture. This is one reason experienced readers compare en-face angiograms with cross-sectional flow overlays rather than interpreting isolated slabs.
Shadowing and low signal: absence can be artificial
Cataract, vitreous opacity, hemorrhage, exudate, pigmentation and poor alignment can reduce the light reaching or returning from the retina. Low signal can erase detail and can make OCTA look nonperfused where flow is simply not detectable. A 2024 study of eyes with diabetic macular edema found artifacts in roughly one third of scans, including segmentation, motion, projection and low-signal artifacts. The exact rate depends on device and population, but the principle is stable: pathology increases the chance that the image pipeline is challenged.
Widefield distortion: flattening a sphere
Ultra-widefield images map a curved retinal surface onto a flat display. Peripheral anatomy can appear stretched or compressed, and eyelids, eyelashes and steering can influence the visible field. Stereographic projection can improve spatial measurement, but a wide image should not be interpreted like an undistorted map. Comparing lesion area across devices or projections without accounting for geometry can produce misleading conclusions.
Quantitative outputs inherit the artifact
Vessel density, retinal thickness, nonperfusion area and layer volumes feel objective because they are numbers. Yet those numbers are downstream of image quality and segmentation. If the scan is miscentered, the slab is wrong or a vessel is suppressed by shadow, the metric faithfully quantifies the artifact. This is why reproducible imaging studies describe acquisition criteria, exclusions and correction procedures rather than reporting a number without context.
A practical quality-control habit
Before trusting a retinal image, ask four questions: Was the signal adequate? Was the scan centered and free of major motion? Are segmentation boundaries anatomically plausible? Does the processed map agree with the underlying B-scan or source image? When something surprising appears, seek confirmation in adjacent slices or another modality. The goal is not skepticism for its own sake. It is calibrated trust—the ability to know when a beautiful image is also a reliable measurement.
Automation can hide the point of failure
Modern software makes imaging feel seamless by correcting motion, centering scans and drawing boundaries automatically. The better the interface, the easier it is to forget that an algorithm made those choices. Automated correction can sometimes create a plausible-looking output even when the source data were weak. Keeping access to raw or minimally processed views is therefore valuable, especially when a quantitative result contradicts the clinical picture.
Artifacts can be systematic, not random
A systematic artifact is more dangerous than occasional noise because it can bias an entire dataset in the same direction. For example, a segmentation algorithm may fail consistently in eyes with a particular pathology or anatomical feature. If those failures are not recognized, a research model can learn the artifact or a clinical metric can appear to differ between groups for technical rather than biological reasons. Quality control is therefore part of evidence quality, not merely image aesthetics.
Artifact prevalence and appearance depend on device, protocol, disease and software version. This article provides a framework, not a complete device-specific artifact atlas.
Sources & original records
We prioritize primary records, clinical-trial registries, peer-reviewed literature and authoritative institutions. Manufacturer material is labeled when used to describe a product or company position.
- Prevalences of segmentation errors and motion artifacts in OCT-angiography differ among retinal diseasesPubMed · Clinical artifact study · PMID 29982897 · DOI 10.1007/s00417-018-4053-2
- Clinical Features Related to OCT Angiography Artifacts in Patients with Diabetic Macular EdemaPubMed · Clinical artifact study · PMID 38447922
- OCT Angiography Artifacts in GlaucomaPubMed · Large imaging-quality study · PMID 33819524