Comprehensive assessment of the anterior segment in refraction corrected OCT based on multitask learning

Area of Science:

• Ophthalmology
• Medical Imaging
• Artificial Intelligence

Background:

• Anterior segment diseases are a major cause of irreversible blindness.
• Current diagnostic methods lack comprehensive analysis of all key anterior segment structures.
• Accurate segmentation and quantification are crucial for clinical diagnosis and treatment.

Purpose of the Study:

• To develop a fully automated multitask deep learning method for simultaneous segmentation, quantification, and landmark detection in AS-OCT images.
• To propose a refraction correction method to address geometric distortions in AS-OCT imaging.
• To facilitate clinical evaluation of anterior segment diseases and surgical outcomes, such as cataract surgery.

Main Methods:

• A multitask deep learning network was designed for simultaneous segmentation and landmark detection.
• The network analyzes major anterior segment structures (iris, lens, cornea) and implants (ICL, IOL), and detects landmarks (scleral spur, iris root).
• A refraction correction algorithm was integrated to improve geometric accuracy of AS-OCT images, using 1251 images from 180 patients.

Main Results:

• The proposed deep learning network outperformed state-of-the-art methods in segmentation and landmark detection.
• High agreement was observed between manual and automated measurements of clinical parameters (anterior chamber, pupil, iris, ICL, IOL).
• The method demonstrated its utility in facilitating clinical evaluation, exemplified by cataract surgery assessment.

Conclusions:

• The developed automated multitask deep learning method offers a robust solution for comprehensive analysis of AS-OCT images.
• This approach enhances the accuracy and efficiency of diagnosing anterior segment diseases and evaluating surgical interventions.
• The proposed technique holds significant potential for advancing clinical practice in ophthalmology.