Multi-scale Microaneurysms Segmentation Using Embedding Triplet Loss
This work addresses diabetic retinopathy detection, a critical medical imaging task, but is incremental as it builds on existing deep learning methods with specific enhancements.
The paper tackled microaneurysms segmentation in fundus images for diabetic retinopathy detection by introducing a two-stage deep learning approach with multi-scale input and embedding triplet loss, achieving a 30.29% relative improvement over a fully convolutional neural network.
Deep learning techniques are recently being used in fundus image analysis and diabetic retinopathy detection. Microaneurysms are an important indicator of diabetic retinopathy progression. We introduce a two-stage deep learning approach for microaneurysms segmentation using multiple scales of the input with selective sampling and embedding triplet loss. The model first segments on two scales and then the segmentations are refined with a classification model. To enhance the discriminative power of the classification model, we incorporate triplet embedding loss with a selective sampling routine. The model is evaluated quantitatively to assess the segmentation performance and qualitatively to analyze the model predictions. This approach introduces a 30.29% relative improvement over the fully convolutional neural network.