An improved strabismus screening method with combination of meta-learning and image processing under data scarcity

Area of Science:

• Ophthalmology
• Computer Vision
• Machine Learning

Background:

• Strabismus screening is crucial for early intervention.
• Limited availability of normal and strabismic images poses a challenge for accurate classification.
• Existing meta-learning methods alone may not achieve optimal performance due to data scarcity.

Purpose of the Study:

• To develop and evaluate a novel method for strabismus screening that combines meta-learning with image processing techniques.
• To improve classification accuracy in strabismus detection, particularly under conditions of limited data.
• To enhance the performance of meta-learning by integrating supplementary image features.

Main Methods:

• A meta-learning approach was pre-trained on a public dataset to extract distinctive image features.
• Image processing methods were employed to extract supplementary features from eye regions (iris position, corneal light reflex).
• Principal Component Analysis (PCA) reduced feature dimensionality for integration, followed by Support Vector Machine (SVM) classification.

Main Results:

• The proposed method achieved a classification accuracy of 0.805, with a sensitivity of 0.768 and specificity of 0.842.
• Meta-learning alone yielded a classification accuracy of 0.709, with a sensitivity of 0.740 and specificity of 0.678.
• The combined approach demonstrated a significant improvement in classification accuracy over meta-learning alone.

Conclusions:

• The developed strabismus screening method shows promising results, achieving higher accuracy than meta-learning alone.
• Integrating meta-learning with image processing effectively addresses the challenge of data scarcity in strabismus detection.
• This hybrid approach offers a viable solution for improving the accuracy and reliability of automated strabismus screening.