An Open-Source Deep Learning Algorithm for Efficient and Fully Automatic Analysis of the Choroid in Optical Coherence...

Area of Science:

• Ophthalmology
• Medical Imaging
• Artificial Intelligence

Background:

• Accurate choroid segmentation in Optical Coherence Tomography (OCT) is crucial for systemic disease research.
• Existing methods are often semi-automatic, requiring manual intervention and introducing subjectivity.
• There is a need for open-source, fully automatic algorithms for efficient choroidal analysis.

Purpose of the Study:

• To develop DeepGPET, an open-source, fully automatic deep learning algorithm for choroid segmentation in OCT data.
• To evaluate the performance and efficiency of DeepGPET compared to a validated semi-automatic method.

Main Methods:

• A U-Net model with a MobileNetV3 backbone was finetuned on a dataset of 715 OCT B-scans.
• Ground-truth segmentations were generated using the semi-automatic Gaussian Process Edge Tracing (GPET) method.
• Performance was assessed using segmentation agreement metrics, choroidal thickness/area measurements, and qualitative clinical evaluation.

Main Results:

• DeepGPET demonstrated excellent agreement with GPET (AUC=0.9994, Dice=0.9664).
• It significantly reduced processing time from 34.49s (GPET) to 1.25s per image.
• Clinical ophthalmologist evaluation showed comparable qualitative performance to GPET.

Conclusions:

• DeepGPET provides a highly accurate and efficient solution for automatic choroid segmentation.
• Its open-source nature and reduced processing time facilitate large-scale choroidal measurements.
• The algorithm can be deployed clinically without a trained operator, enhancing objectivity and accessibility.