Statistical framework for validation without ground truth of choroidal thickness changes detection.

Area of Science:

• Ophthalmology
• Medical Imaging
• Biomedical Engineering

Background:

• Monitoring choroidal thickness changes is crucial for understanding and treating ocular diseases like myopia.
• Evaluating the performance of automated detection algorithms is challenging due to the lack of a definitive ground truth.
• Existing methods relying on averaged expert segmentations are susceptible to inter-expert variability.

Purpose of the Study:

• To present a novel validation approach for automated ocular disease detection that is independent of a ground truth.
• To establish a validation method based on the consensus between algorithms and human experts.
• To evaluate a registration-based method for choroidal thickness analysis in young myopic subjects.

Main Methods:

• Developed a novel validation approach comparing joint agreement between multiple experts and an algorithm.
• Utilized an index to measure the common agreement for validation.
• Conducted an observational study on 90 OCT 3D-volume datasets from healthy subjects (ages 8-18) in myopic Asian urban regions.
• Adapted a state-of-the-art evaluation method involving paired t-tests after excluding individual expert results.

Main Results:

• The novel validation approach demonstrated effectiveness in assessing algorithm performance without a ground truth.
• Comparison between the novel method, expert consensus, and adapted state-of-the-art method provided insights into detection accuracy.
• Results from the registration-based method were evaluated on the specific dataset.

Conclusions:

• The proposed ground-truth-independent validation method offers a robust alternative for evaluating ocular disease detection algorithms.
• This approach mitigates limitations associated with inter-expert variability in ground truth generation.
• The study validates the utility of automated choroidal thickness monitoring in pediatric myopia research.