Algorithm for Detection and Quantification of Hyperreflective Dots on Optical Coherence Tomography in Diabetic...

Area of Science:

• Ophthalmology
• Medical Imaging
• Artificial Intelligence

Background:

• Diabetic macular edema (DME) is a leading cause of vision loss.
• Hyperreflective dots (HRDs) on optical coherence tomography (OCT) are important biomarkers in DME.
• Accurate detection and quantification of HRDs are crucial for DME management.

Purpose of the Study:

• To develop and validate a deep learning algorithm for detecting and quantifying HRDs in OCT images of DME patients.
• To differentiate between hard exudates and small HRDs (potentially activated microglia).

Main Methods:

• Utilized 20 OCT datasets (2,560 b-scans) from DME patients.
• Manual labeling of HRDs by two independent raters.
• Developed a deep learning algorithm trained on rater 1's annotations.
• Employed 4-fold cross-validation for training and validation.
• Evaluated algorithm performance using Dice coefficients, intraclass correlation coefficients (ICCs), and correlation coefficients.

Main Results:

• The algorithm achieved Dice coefficients of 0.70 for total HRDs and 0.72 for hard exudates.
• Stronger correlations (0.95-0.99) and higher ICCs (0.972-0.997) were observed between the algorithm and rater 1 compared to inter-rater agreement.
• The algorithm demonstrated good performance in detecting and quantifying HRDs, including differentiating subtypes.

Conclusions:

• The developed deep learning algorithm is a reliable tool for detecting and quantifying HRDs in OCT images of DME patients.
• The algorithm provides objective and repeatable measurements, valuable for clinical practice and research.
• This technology can enhance the assessment and monitoring of DME.