Reinforcement-based leveraging transfer learning for multiclass optical coherence tomography images classification.

Area of Science:

• Ophthalmology and Medical Imaging

Background:

• Accurate diagnosis of retinal diseases like Diabetic Macular Edema (DME) and Age-related Macular Degeneration (AMD) is critical for preventing vision loss.
• Optical Coherence Tomography (OCT) is vital for identifying these conditions, particularly AMD due to its increasing prevalence.

Purpose of the Study:

• To introduce a novel Reinforcement-Based Leveraging Transfer Learning (RBLTL) framework for enhanced OCT image classification.
• To improve the accuracy and generalization of automated diagnosis for retinal diseases.

Main Methods:

• Integration of reinforcement Q-learning with transfer learning using pre-trained models (InceptionV3, DenseNet201, InceptionResNetV2).
• Dynamic hyperparameter optimization within the RBLTL framework to mitigate overfitting and boost performance.

Main Results:

• Achieved high testing accuracies of 98.75%, 98.90%, and 99.20% in multiclass OCT image classification across three scenarios.
• Demonstrated the framework's effectiveness in categorizing OCT images for DME and AMD.

Conclusions:

• The RBLTL framework offers a reliable and versatile approach for automated medical image classification.
• This method has significant implications for improving clinical diagnostics of retinal diseases.