Adaptive feature squeeze network for nuclear cataract classification in AS-OCT image.

Area of Science:

• Ophthalmology
• Medical Imaging
• Artificial Intelligence

Background:

• Nuclear cataract (NC) is a prevalent age-related condition impacting vision and quality of life.
• Cataract surgery effectively restores vision, but accurate NC severity classification is crucial.
• Anterior segment optical coherence tomography (AS-OCT) offers noninvasive imaging of the lens nucleus.

Purpose of the Study:

• To develop and evaluate an automated system for classifying nuclear cataract severity using AS-OCT images.
• To introduce a novel Convolutional Neural Network (CNN) framework, Adaptive Feature Squeeze Network (AFSNet), for this task.
• To investigate the efficacy of focusing CNN analysis on the lens nucleus region for NC classification.

Main Methods:

• Proposed a novel CNN framework, Adaptive Feature Squeeze Network (AFSNet), incorporating an adaptive feature squeeze module.
• The AFSNet module dynamically adjusts local and global feature importance.
• Conducted experiments on both a clinical AS-OCT dataset and a public OCT dataset, utilizing Class Activation Mapping (CAM) for interpretability.

Main Results:

• AFSNet achieved superior performance in classifying NC severity compared to strong baselines and state-of-the-art methods.
• CNN models demonstrated better classification accuracy when focusing on the nucleus region versus the entire lens.
• CAM technique successfully localized discriminative regions, enhancing model interpretability.

Conclusions:

• The proposed AFSNet framework provides an effective and automated solution for NC severity classification from AS-OCT images.
• Focusing CNN analysis on the lens nucleus region improves classification performance.
• The study highlights the potential of AI in ophthalmology for objective disease assessment and improved patient care.