Open-source deep learning-based automatic segmentation of mouse Schlemm's canal in optical coherence tomography...

Area of Science:

• Ophthalmology
• Medical Imaging
• Artificial Intelligence

Background:

• Accurate segmentation of the Schlemm's canal (SC) is crucial for understanding intraocular pressure and diseases like glaucoma.
• Current methods for SC segmentation in optical coherence tomography (OCT) images can be time-consuming and subjective.
• Automated tools are needed to accelerate research into SC morphology and treatment efficacy.

Purpose of the Study:

• To develop a deep learning-based automatic approach for segmenting the Schlemm's canal (SC) lumen in OCT scans of living mouse eyes.
• To create free, open-source software implementing this novel segmentation method.
• To evaluate the performance of the automated method against manual segmentation by an expert grader.

Main Methods:

• A novel convolutional neural network (CNN) utilizing a U-Net architecture with late fusion, multi-scale input, dilated residual blocks, and attention-gating was developed.
• The CNN model was trained and validated using 163 pairs of intensity and speckle variance (SV) OCT B-scans from 32 living mouse eyes.
• Performance was quantified using the Dice Similarity Coefficient (DSC) and compared to manual segmentation results.

Main Results:

• The proposed deep learning model achieved a mean DSC of 0.694 ± 0.256 and a median DSC of 0.791.
• Manual segmentation by a second expert grader yielded a mean DSC of 0.713 ± 0.209 and a median DSC of 0.763.
• The automated method's performance was comparable to that of a human expert grader.

Conclusions:

• This study presents the first automatic method for SC lumen segmentation in OCT images of living mouse eyes.
• The developed deep learning software provides a reliable and efficient tool for SC segmentation.
• This open-source software is expected to accelerate research on glaucoma and the efficacy of new ocular pressure-lowering drugs.