Automated Montaging of Ultrasound Scans for Determination of Ocular Globe Shape

Introduction: The global increase in myopia highlights the urgent need for reliable and automated methods to monitor structural changes in the eye, especially in the peripheral regions beyond the central fovea, where conventional imaging is limited. This study develops and validates an automated pipeline that enables comprehensive assessment of posterior eye shape from extended-range B-scan ultrasonography, addressing the limitations of current partial coherence interferometry and manual alignment approaches.Methods: Extended-range B-scan ultrasound images were acquired at three gaze angles to capture peripheral retinal dimensions. The method utilizes the Segment Anything Model (SAM) for interactive and accurate boundary segmentation, followed by translation-based alignment to merge the triplet of images into a unified posterior eye wall representation. Multi-set averaging across repeated acquisitions was further evaluated for enhancing reproducibility. SAM results were compared to those from manual merging of images. Results: Averaging multiple frames before merging significantly reduced the standard deviation of measurements and improved consistency compared with merging a single image. Using this automated triplet-merging approach, reproducibility yielded mean absolute improvements of 31.46 microns (30.12%) in the horizontal view and 29.04 microns (39.53%) in the vertical view relative to manual merging, indicating greater reproducibility and stability across acquisitions.Conclusion:The proposed operator-independent pipeline provides a scalable, consistent assessment of peripheral eye morphology. By enabling automated quantification of retinal curvature and globe shape, this approach supports more precise monitoring of myopia progression and enhances the reliability of evaluating treatment efficacy in both clinical trials and routine practice.