Multi-scale Microaneurysms Segmentation Using Embedding Triplet Loss

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. Applying a patch-wise approach with healthy patches only sampled from healthy patient images gives the ability of learning segmentation even in cases where not all instances of a lesion are annotated in the gold standard images. This approach introduces a 30.29% improvement over the 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. Applying a patch-wise approach with healthy patches only sampled from healthy patient images gives the ability of learning segmentation even in...     »