DenseNet-121 with Dual Attention for ARMD Classification from Fundus Images

In recent years, deep learning and Computer vision have been mainly used in ophthalmology. In this study, we present a DenseNet-121 model with dual attention for the classification of Age-related Macular degeneration (ARMD) and Normal eyes with the help of fundus images. The convolutional block attention module highlights relevant channel and spatial features that DenseNet-121 has extracted. Further, the channel recalibration module uses edge information and statistical features with spatial dimensions to enhance the feature representation. A dataset cohort that contains a total of 890 images from the publically available dataset on Kaggle, namely ARMD_DR_DN_TSLN has been used for the experiment. Our model has shown higher results than the other existing models. An ablation study has also been conducted to check each component’s effectiveness.