INTEGRATING AN ATTENTION MECHANISM INTO A CONVOLUTIONAL NEURAL NETWORK TO IMPROVE OPHTHALMIC DISEASE CLASSIFICATION
Egor N. Volkov, Alexey Nikolaevich Averkin, Sergei A. Yarushev · SOFT MEASUREMENTS AND COMPUTING · 2024
The article considers an approach to improving the classification of diseases based on optical coherence tomography (OCT) images using an attention mechanism integrated into the architecture of a convolutional neural network. An overview and limitations of existing methods of processing OCT images are given. Public datasets are described. The architectural features of the proposed model are shown in detail, including the design of attention blocks and their integration into the convolutional network model. An experimental comparison of the proposed model with traditional convolutional neural networks without an attention mechanism (VGG16, EfficientNetB0) demonstrated improved results on the test dataset (miavgF1 – 0,987, miavgP – 0,998, miavgR – 0,978).