Related Experiment Video
Updated: Aug 6, 2026
04:48
Application of Deep Learning-Based Medical Image Segmentation via Orbital Computed Tomography
Published on: November 30, 2022
Data-centric physics-inspired deep learning framework for saturation artifact removal in optical coherence tomography
Jonas Nienhaus1, Thomas Schlegl1, Florian Kapeller1
1Center for Medical Physics and Biomedical Engineering, Medical University of Vienna, Währinger Gürtel 18-20, 1090 Vienna, Austria.
Medical Image Analysis
|July 18, 2026
Summary
We developed a method to recover lost information in Fourier-domain optical coherence tomography (FD-OCT) scans caused by data acquisition saturation. This approach uses a physics-based simulation and a novel multi-input neural network to remove artifacts, improving image quality for clinical use.
Area of Science:
- Medical Imaging
- Optical Engineering
- Computer Vision
Background:
- Saturation in Fourier-domain optical coherence tomography (FD-OCT) imaging causes data loss by clipping spectral interferograms.
- Reconstructed depth profiles exhibit bright axial lines or periodic patterns, masking crucial information and reducing clinical utility.
Purpose of the Study:
- To develop a method for faithfully recovering distorted information from saturated FD-OCT scans.
- To analyze the genesis and appearance of saturation artifacts, including their wavelength dependency.
Main Methods:
- A physics-based simulation model was created to generate realistic saturated interferograms from clean data for neural network training.
- A multi-input, single-output (MISO) network framework was proposed to leverage the wavelength dependency of artifacts.
- The MISO network processes full B-scans alongside images reconstructed from spectral sub-windows.
Main Results:
- Networks trained on simulated data demonstrated successful generalization to real-world artifact removal in ophthalmic anterior segment imaging.
- The method effectively removed artifacts from various sources, including tissue structures and surgical instruments.
- The MISO network showed consistent performance and reliability gains compared to single-input baselines on both swept-source and spectral domain datasets.
Conclusions:
- The proposed approach successfully recovers distorted information in saturated FD-OCT scans.
- The physics-based simulation and MISO network framework enable robust artifact removal.
- This technique enhances the clinical utility of FD-OCT imaging, particularly in ophthalmic applications.
