Fluid-SegNet: Multi-dimensional loss-driven Y-Net with dilated convolutions for OCT B-scan fluid segmentation

Area of Science:

• Ophthalmology
• Medical Imaging
• Artificial Intelligence

Background:

• Optical Coherence Tomography (OCT) is essential for retinal imaging in ophthalmology.
• Accurate segmentation of fluid regions in OCT B-scans is critical for disease diagnosis and treatment planning.

Purpose of the Study:

• To introduce Fluid-SegNet, a novel deep learning framework for accurate fluid region segmentation in OCT B-scans.
• To address challenges in segmenting fine details, small regions, and heterogeneous fluid areas.

Main Methods:

• Development of Fluid-SegNet, a deep learning-based segmentation framework.
• Evaluation of Fluid-SegNet on three public datasets: UMN, AROI, and OIMHS.

Main Results:

• Fluid-SegNet achieved mean Dice scores of 0.8725 (UMN), 0.6967 (AROI), and 0.8020 (OIMHS).
• The model demonstrated effectiveness and robustness in segmenting fluid regions across diverse datasets and imaging conditions.
• Fluid-SegNet successfully addressed challenges of fine detail delineation and fluid region heterogeneity.

Conclusions:

• Fluid-SegNet enhances the accuracy of fluid region segmentation in OCT B-scans.
• The proposed method offers a robust solution for automated retinal disease diagnosis and visual acuity prediction.