Related Experiment Video
Updated: Jun 19, 2026
08:39
Longitudinal Micro-Computed Tomography Image Analysis for User-Defined Region of Interest in Critical-Sized Bone Defects
Published on: June 24, 2025
Automated anatomically guided quality assessment for OCTA via multi-region analysis and statistical calibration
Enrui Zhang1, Hengyi Yuan1, Lei Zhang1,2
1School of Information Science and Engineering, Yanshan University, Qinhuangdao 066004, China.
Biomedical Optics Express
|June 18, 2026
Summary
A new framework improves optical coherence tomography angiography (OCTA) quality control by analyzing specific regions, ensuring accurate vascular measurements. This method enhances automated data filtering for AI and device calibration.
Area of Science:
- Ophthalmology
- Medical Imaging
- Artificial Intelligence
Background:
- Reliable optical coherence tomography angiography (OCTA) necessitates standardized quality control for accurate vascular quantification.
- Current OCTA quality assessment methods often rely on subjective grading or global metrics, neglecting region-specific signal characteristics.
- Existing approaches fail to account for the varying artifact sensitivity and signal formation across different anatomical regions within OCTA images.
Purpose of the Study:
- To develop an anatomically informed OCTA quality assessment framework integrating multi-region segmentation and semi-supervised learning.
- To enable region-specific quality evaluation by segmenting images into large vessels, capillary networks, and the foveal avascular zone (FAZ).
- To establish a robust quality control system for OCTA data, supporting AI training and device calibration.
Main Methods:
- Utilized the Segment Anything Model for partitioning OCTA images into distinct anatomical regions (large vessels, capillaries, FAZ).
- Extracted physically interpretable metrics such as vessel edge sharpness, contrast-to-noise ratio, and signal-to-noise ratio to create a decorrelation-grounded feature space.
- Implemented a distribution-calibrated semi-supervised learning strategy with a modulation factor of 0.9 for stable grading with limited annotations.
Main Results:
- Achieved an average Dice coefficient of 81.21% for region segmentation on public and clinical datasets.
- Demonstrated a grading accuracy of 90.0% with a Cohen kappa of 0.864, indicating high reliability.
- Validated the framework's effectiveness on the OCTA-500 dataset and an independent clinical dataset.
Conclusions:
- The proposed framework transforms global heuristic scoring into anatomically resolved, measurement-consistent evaluation for OCTA.
- It supports automated data filtering crucial for AI training pipelines by providing standardized quality control.
- The framework can be integrated into OCT acquisition workflows for device-level performance calibration and consistent quality assurance.
