Shadow Artifact Detection in OCT Angiography via ML
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Optical coherence tomography angiography (OCTA) systems face challenges in accurately distinguishing shadow artifacts caused by vitreous floaters or pupil vignetting from actual perfusion defects, leading to unreliable vessel density measurements due to signal blockage, especially in areas with severe signal loss.
Innovation Solution
A machine-learning algorithm is applied to detect shadow artifacts by analyzing OCT and OCTA datasets, using features such as reflectance-adjusted thresholding, local flow index, and standard deviation to differentiate between shadowed and non-shadowed areas, allowing for the suppression of shadow artifacts and improvement of signal quality.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If traditional OCTA signal quality assessment is used, then overall scan quality can be evaluated, but localized shadow artifacts cannot be distinguished from actual perfusion defects
Solution Approach 1:
The patent segments the OCTA signal into multiple independent components: flow signal, reflectance signal, and shadow artifact signal. By processing each component separately and combining them through weighted fusion, the system can identify and exclude shadow artifacts while preserving genuine perfusion defect information, thereby improving measurement precision and reliability simultaneously
Solution Approach 2:
The patent introduces reflectance signal as an intermediary component to detect shadow artifacts. The reflectance signal serves as a mediator that identifies areas affected by vitreous floaters or media opacities, allowing the system to differentiate between shadow artifacts and true perfusion defects without discarding the entire scan
2Reliability
If entire scans are discarded due to shadow artifacts, then measurement reliability is maintained, but diagnostic information is lost
Solution Approach 1:
The patent extracts shadow artifact components from the OCTA signal using reflectance-adjusted thresholding and signal component separation. By removing only the shadow artifact portions and retaining the valid flow signal components, the system maintains measurement reliability while preserving diagnostic information from non-shadowed areas
Solution Approach 2:
The patent applies local quality assessment by evaluating signal characteristics at each spatial location independently. This allows the system to identify and exclude only the specific regions affected by shadow artifacts while maintaining high-quality measurements from unaffected regions, preventing unnecessary loss of diagnostic information
3Difficulty of detecting and measuring
If reflectance-based shadow detection is used, then shadow areas can be identified, but atrophied and cystic areas may be misclassified
Solution Approach 1:
The patent merges multiple signal components (flow signal, reflectance signal, and their correlation) to create a comprehensive shadow detection criterion. By combining these independent measurements with weighted fusion, the system achieves robust shadow detection that correctly distinguishes shadow artifacts from atrophied and cystic areas, improving measurement precision while maintaining ease of detection
Data Source
AI summary
Disclosed herein are methods and systems for automated detection of shadow artifacts in optical coherence tomography (OCT) and/or OCT angiography (OCTA). The shadow detection includes applying a machine-learning algorithm to the OCT dataset and the OCTA dataset to detect one or more shadow artifacts in the sample. The machine-learning algorithm is trained with first training data from first training samples that include manufactured shadows and no perfusion defects and second training data from second training samples that include perfusion defects and no manufactured shadows. The shadow artifacts in the OCTA dataset and/or OCT dataset may be suppressed to generate a shadow-suppressed OCTA dataset and/or a shadow-suppressed OCT dataset, respectively. Other embodiments may be described and claimed.