Using Ensemble OCT-Derived Features beyond Intensity Features for Enhanced Stargardt Atrophy Prediction with Deep...

Area of Science:

• Ophthalmology
• Medical Imaging
• Artificial Intelligence

Background:

• Stargardt disease is a leading cause of juvenile macular dystrophy.
• Spectral-domain optical coherence tomography (SD-OCT) is crucial for monitoring retinal changes.
• Current methods often rely on mean intensity features for atrophy prediction.

Purpose of the Study:

• To develop an enhanced prediction model for Stargardt disease atrophy.
• To explore the utility of advanced OCT-derived features beyond mean intensity.
• To improve the accuracy and reliability of Stargardt disease progression assessment.

Main Methods:

• An ensemble deep learning neural network was employed.
• Advanced OCT-derived features were utilized to augment data.
• Predictions were made for six-month and twelve-month intervals.

Main Results:

• The proposed model achieved a median Dice coefficient of 0.830 (6-month) and 0.828 (12-month).
• This represents a significant improvement over models using only mean intensity features (0.744 and 0.762, respectively).
• Performance varied based on the retinal layers analyzed.

Conclusions:

• Advanced OCT features enhance Stargardt disease atrophy prediction.
• The deep learning model shows promise for assessing prognosis and quantifying disease progression.
• Further research into layer-specific features could refine predictive accuracy.