Related Experiment Video
Updated: May 5, 2026
12:22
Multimodal Volumetric Retinal Imaging by Oblique Scanning Laser Ophthalmoscopy oSLO and Optical Coherence Tomography OCT
Published on: August 4, 2018
8.9K
Pointmaps to Practice: 3D Multi-view Ocular Lesion Mapping
Vasileios Alevizos1,2, George A Papakostas3
1Department of Learning, Informatics, Management and Ethics, Karolinska Institutet, 171 77, Stockholm, Sweden. vasileios.alevizos@pm.me.
Journal of Imaging Informatics in Medicine
|October 24, 2025
Summary
This study introduces a novel method for 3D ocular imaging, enabling precise volumetric mapping of uveal melanoma. The technique uses DUSt3R and Logarithmic Positional Partition Interval Encoding (LPPIE) for accurate and memory-efficient 3D reconstruction.
Area of Science:
- Ophthalmology and Medical Imaging
- Computer Vision and Graphics
Background:
- Accurate 3D volumetric mapping is crucial for guiding uveal melanoma therapy, as 2D imaging lacks the necessary topological detail for precise boundary delineation.
- Existing methods may struggle with the complex geometries and optical properties of the eye, necessitating advanced reconstruction techniques.
Purpose of the Study:
- To develop and evaluate a novel pipeline for accurate, memory-efficient 3D ocular volumetry using single-camera input.
- To enable precise boundary delineation and volumetric mapping of ocular pathologies, including uveal melanoma.
Main Methods:
- Employed DUSt3R for correspondence estimation with self-calibrated poses and dense pointmaps, refined via intrinsic reprojection loss.
- Utilized Logarithmic Positional Partition Interval Encoding (LPPIE) for depth data, pointmaps, and camera parameters to minimize memory usage.
- Evaluated the pipeline on various ocular images (melanoma, nevus, melanosis, pterygium, phantoms) using metrics like completeness, MAE, RMSE, and rotational/translational errors.
Main Results:
- Achieved completeness (C) of approximately 0.43 for melanoma cases, with rotation mismatch around 2°, RMSE of 0.021, and translation error of 3.9 cm.
- Simpler morphologies yielded higher completeness (up to 0.61) with stable MAE (0.005-0.008) and high accuracy (δ<1.25 ≈ 0.98-0.99).
- LPPIE significantly reduced memory footprint with minimal texture degradation; coherent meshes were generated under modest computational load.
Conclusions:
- The proposed method offers a viable approach for accessible ocular volumetry using portable hardware and minimal calibration.
- The pipeline demonstrates potential for clinical application in guiding uveal melanoma therapy through accurate 3D reconstruction.
- Further refinements could enhance robustness for irregular surfaces, specular reflections, and motion artifacts, improving clinical utility.
