From Image to Sequence: Exploring Vision Transformers for Optical Coherence Tomography Classification.

Area of Science:

• Ophthalmology
• Medical Imaging
• Artificial Intelligence

Background:

• Optical Coherence Tomography (OCT) is crucial for diagnosing retinal diseases like diabetic macular edema and age-related macular degeneration.
• These conditions pose significant global health challenges, potentially leading to vision loss if not detected early.
• Current OCT image classification methods face challenges due to complex retinal structures and dataset variability.

Purpose of the Study:

• To develop a novel hybrid model for enhanced OCT image classification.
• To address limitations in current OCT analysis methods.
• To improve the accuracy and efficiency of diagnosing critical retinal diseases.

Main Methods:

• Integration of Convolutional Neural Networks (CNNs) for local feature extraction and Vision Transformers (ViT) for long-range pattern recognition.
• Development of a hybrid model leveraging the complementary strengths of CNNs and ViT.
• Application of the hybrid model to OCT image datasets for classification.

Main Results:

• The hybrid model achieved a high accuracy of 99.80% on the OCT2017 dataset.
• The model demonstrated significant parameter efficiency, utilizing only 6.9 million parameters.
• This efficiency surpasses that of larger models like Xception and OpticNet-71.

Conclusions:

• The developed hybrid CNN-ViT model offers a highly accurate and parameter-efficient solution for OCT image classification.
• Its efficiency makes it suitable for clinical settings with limited computational resources.
• This advancement supports rapid and imperative diagnosis of critical retinal diseases.