LOCTseg: A lightweight fully convolutional network for end-to-end optical coherence tomography segmentation.

Area of Science:

• Ophthalmology
• Medical Imaging
• Computer Vision
• Artificial Intelligence

Background:

• Optical coherence tomography (OCT) is crucial for high-resolution retinal imaging in clinical practice.
• Manual segmentation of retinal structures in OCT images is time-consuming and requires specialized expertise.
• Variability in retinal structures presents a significant challenge for automated OCT image analysis.

Purpose of the Study:

• To develop a novel, end-to-end automatic solution for semantic segmentation of OCT images.
• To introduce LOCTSeg, a lightweight fully convolutional network (FCN) designed for efficient and accurate OCT b-scan segmentation.
• To evaluate the performance of LOCTSeg against state-of-the-art methods on public and private OCT datasets.

Main Methods:

• A novel lightweight fully convolutional network (FCN) architecture, termed LOCTSeg, was developed for end-to-end semantic segmentation.
• LOCTSeg was trained and evaluated on two public datasets (AROI and HCMS) and one private dataset (ERM).
• Performance was assessed using the Dice score, comparing LOCTSeg against existing segmentation techniques.

Main Results:

• LOCTSeg achieved improved Dice scores on the AROI dataset (69% to 73%) and HCMS dataset (91% to 92%).
• The model demonstrated superior performance on ERM segmentation, outperforming other lightweight FCNs by 4% to 15%.
• LOCTSeg offers competitive inference speed without compromising segmentation accuracy.

Conclusions:

• LOCTSeg provides an effective and efficient solution for automatic semantic segmentation of OCT images.
• The proposed FCN architecture significantly advances the state-of-the-art in OCT image analysis.
• LOCTSeg holds promise for improving diagnostic marker segmentation in clinical ophthalmology.