Texture Attention Network for Diabetic Retinopathy Classification

Diabetic Retinopathy (DR) is a disease caused by high level of glucose in retina vessels. This malicious put millions people around the world at risk for vision loss each year. Being life-threatening disease, early diagnosis can be an effective step treatment and prevention loss. To automate process, computer-aided methods are not only useful detecting diabetic signatures but also provide information regarding grade optometrist to determine appropriate treatment. Several deep classification models proposed literature solve retinopathy task, however, these usually lack incorporate attention mechanism better encode semantic dependency highlight most important region boosting model performance. overcome limitations, we propose style content recalibration inside neural network adaptively scale informative regions classification. In our method, input image passes through encoder module both high-level features. Next, utilizing separation mechanism, decompose representational space into (e.g., texture features) contextual representation. The takes representation applies high-pass filter while spatial normalization uses convolutional operation more detect signs. Once modules applied features, fusion combines features form normalized decoding path. decoder performs grading healthy, non-healthy tasks. Our experiment on APTOS Kaggle dataset (accuracy 0.85) demonstrates significant improvement compared work. fact reveals applicability method real-world scenario.