Segmentation of Preretinal Space in Optical Coherence Tomography Images Using Deep Neural Networks.

Area of Science:

• Ophthalmology
• Medical Imaging
• Artificial Intelligence

Background:

• Accurate segmentation of the preretinal area in optical coherence tomography (OCT) is crucial for diagnosing vitreomacular diseases.
• Current segmentation solutions in scientific or commercial OCT imaging lack the ability to diagnose vitreomacular traction changes.

Purpose of the Study:

• To propose an efficient semantic segmentation method for the preretinal area between the inner limiting membrane (ILM) and posterior cortical vitreous (PCV).
• To evaluate the effectiveness of various neural networks and compare them with graph theory techniques.
• To demonstrate the potential for diagnosing vitreomacular traction changes using the proposed segmentation solution.

Main Methods:

• Utilized a database of 3D OCT scans from the Optovue RTVue XR Avanti device.
• Tested semantic segmentation using UNet, Attention UNet, ReLayNet, and LFUNet neural networks.
• Assessed effectiveness with the Dice coefficient and compared with graph theory techniques, incorporating relative distance maps and varying kernel sizes.

Main Results:

• Achieved high segmentation effectiveness, reaching up to 96.35% for the posterior cortical vitreous (PCV).
• Demonstrated that larger kernel sizes in convolutional layers can enhance segmentation quality.
• Showcased the superiority of the proposed neural network approach over traditional graph theory techniques.

Conclusions:

• The proposed neural network-based segmentation method offers an efficient and accurate solution for the preretinal area.
• The technique shows significant potential for clinical application in diagnosing vitreomacular traction changes.
• This advancement addresses a current gap in scientific and commercial OCT imaging solutions.