Cascade spatial and channel-wise multifusion network with criss cross augmentation for corneal segmentation and...

Area of Science:

• Ophthalmology
• Medical Imaging
• Computer Vision

Background:

• Accurate 3D corneal reconstruction from AS-OCT images aids computer-aided diagnosis.
• Existing segmentation methods struggle with low-contrast corneal regions and 180-degree AS-OCT scans.
• There is a need for robust methods for corneal segmentation and 3D reconstruction.

Purpose of the Study:

• To develop an accurate and efficient network for corneal segmentation from AS-OCT images.
• To enable high-quality 3D corneal reconstruction and topographic mapping.
• To address challenges with low-contrast features and non-standard scan angles.

Main Methods:

• Proposed CSCM-CCA-Net incorporating cascade spatial and channel-wise multifusion (CSCM) and criss cross augmentation (CCA).
• CSCM enhances extraction of low-contrast features and contextual dependencies.
• CCA improves shape-preserved feature representation for segmentation accuracy.
• 3D volume reconstruction and topographic thickness map generation via image alignment.
• Transfer function design using intensity and gradient histograms for enhanced visualization.

Main Results:

• CSCM-CCA-Net demonstrated impressive performance in corneal segmentation and 3D reconstruction on the CORNEA benchmark.
• The method achieved superior segmentation accuracy compared to state-of-the-art approaches on challenging fundus datasets (DRIVE, CHASEDB1, FIVES).
• The network effectively handles low-contrast corneal regions and facilitates 3D reconstruction from AS-OCT sequences.

Conclusions:

• CSCM-CCA-Net offers a simple yet efficient solution for accurate corneal segmentation and 3D reconstruction.
• The proposed techniques effectively address limitations of existing medical image segmentation networks.
• The method shows significant potential for advancing computer-aided diagnosis in ophthalmology.