Automated Quantification of Lens Cortex and Nuclear Opacity Based on Swept-Source Anterior Segment Optical Coherence...

Area of Science:

• Ophthalmology
• Medical Imaging
• Artificial Intelligence

Background:

• Cataract surgery candidates require accurate lens opacity assessment.
• Current quantification methods can be subjective and time-consuming.
• Advancements in imaging and AI offer potential for automated analysis.

Purpose of the Study:

• To develop and validate an automated method for quantifying lens cortex and nuclear opacity.
• Utilize swept-source anterior segment optical coherence tomography (AS-OCT) for image acquisition.
• Employ artificial intelligence (AI) for precise opacity measurement.

Main Methods:

• Trained two AI segmentation models based on the nnUNet framework.
• Quantified lens cortex and nucleus opacity using swept-source AS-OCT images from 504 participants.
• Validated AI model performance against manual measurements by ophthalmologists.

Main Results:

• AI models achieved high segmentation accuracy (MIoU of 0.959 for cortex, 0.928 for nucleus).
• Excellent agreement was observed in opacity quantification (ICC of 0.9933 for cortex, 0.9939 for nucleus).
• AI performance was comparable to manual assessments.

Conclusions:

• The AI model accurately and objectively quantifies lens cortex and nucleus opacity from AS-OCT images.
• This automated method provides a more precise, objective, and rapid quantification solution.
• The developed AI tool is suitable for both clinical practice and research applications.