A Hybrid System for Automatic Identification of Corneal Layers on In Vivo Confocal Microscopy Images

Area of Science:

• Ophthalmology
• Medical Imaging
• Artificial Intelligence

Background:

• Accurate identification of corneal layers is crucial for diagnosing corneal lesions using in vivo confocal microscopy (IVCM).
• Current methods for corneal layer identification can be subjective and time-consuming.

Purpose of the Study:

• To develop a reliable automated system for identifying corneal layers from IVCM images.
• To improve the accuracy and efficiency of corneal lesion assessment.

Main Methods:

• A hybrid classification system was developed using convolutional neural networks and K-nearest neighbors.
• Two fusion strategies, weighted voting and LightGBM, were employed to combine base classifier results.
• Confidence stratification was used to identify potential model errors.

Main Results:

• Both hybrid systems significantly outperformed individual base classifiers.
• The weighted voting hybrid system achieved a weighted F1 score of 0.9111.
• Confidence stratification successfully identified over half of the misclassified samples.

Conclusions:

• The proposed hybrid approach effectively integrates scanning depth and pixel data for accurate corneal layer identification in IVCM images.
• The confidence stratification method aids in detecting system misclassifications.
• This work provides a foundation for automated corneal layer identification in IVCM imaging.