Interpretable Automated Detection and Measurement of Dome-Shaped Macula on Optical Coherence Tomography

Purpose

To develop an interpretable automated pipeline for dome-shaped macula (DSM) detection and measurement from optical coherence tomography (OCT) B-scans.

Methods

We analyzed one eye from each of 300 children aged 4–18 years in the Shanghai Child and Adolescent Large-scale Eye Study High Myopia (SCALE-HM) cohort (50 DSM-positive; 100 test eyes). A U-Net pretrained on OCTA-500 and fine-tuned on the development set segmented the retinal pigment epithelium (RPE) boundary. A geometric module located the apex, fitted a tangent-line baseline, and computed perpendicular height in physical coordinates.

An eye was DSM-positive if any gradable B-scan had height ≥ 50 µm. Two masked graders evaluated the test set; a third adjudicated disagreements.

Results

The area under the receiver operating characteristic curve was 0.927 (95% confidence interval [CI], 0.850–0.986), sensitivity was 82.1% (95% CI, 64.4%–92.1%), and specificity was 100% (95% CI, 94.9%–100%) against the adjudicated reference. In concordant-positive eyes, automated height agreed with manual measurements (intraclass correlation coefficient, 0.949; mean absolute error, 6.9 µm; bias, − 5.7 µm, versus Doctor A). Mean RPE-boundary error on pediatric test scans was 1.54 µm.

Conclusions

The pipeline may be considered a reliable and reproducible tool for automated DSM detection and height measurement on radial OCT, with each value traceable to land-marks on the source B-scan. External validation across populations, devices, and acquisition protocols is needed.

Translational Relevance

The pipeline provides an objective and reproducible method to detect DSM and quantify dome height, width, and orientation across OCT meridians. Because each measurement can be verified on the source B-scan, it enables standardized DSM assessment.