Automatic segmentation of hyperreflective foci in OCT images

Area of Science:

• Ophthalmology
• Medical Imaging
• Artificial Intelligence

Background:

• Age-related Macular Degeneration (AMD) is a primary cause of vision loss.
• Tracking Hyperreflective Foci (HF) in Optical Coherence Tomography (OCT) images is crucial for AMD patient management.
• Automated analysis of OCT images can improve diagnostic accuracy and treatment monitoring.

Purpose of the Study:

• To develop a deep learning framework for the automatic segmentation of HF in OCT images.
• To assess the performance of various neural networks in HF segmentation.
• To establish a foundation for a clinical decision support system for AMD.

Main Methods:

• Collection and annotation of OCT images.
• Image preprocessing and feature extraction.
• Training and evaluation of Conventional, Deep, and Convolutional Neural Networks for HF segmentation.

Main Results:

• The developed neural network systems achieved high Dice Coefficient values for HF segmentation.
• Segmentation accuracy was comparable to, and often exceeded, manual annotations by multiple physicians (above 95% similarity).
• Performance was validated on clinical data excluded from the training set.

Conclusions:

• Neural networks provide accurate HF segmentation in OCT images.
• The achieved accuracy supports the integration of this technology into clinical practice.
• This research enables the development of a decision support system for everyday clinical use in AMD management.