Sparse Autoencoder Based Microaneurysm Detection

EasyChair Preprint 9348

Abstract

Microaneurysms (MAs) are the early noticeable lesions in retina, their detection plays a crucial role in the diagnosis of diabetic retinopathy. Here the discriminative features are automatically learned in an unsupervised manner. The stacked sparse autoencoder (SSAE) is effective at learning high-level features from overlapping image patches during training. Using 10-fold cross-validation and fine-tuning yield an improved F-measure of 97.3% and an average area under the ROC curve (AUC) 96.7% obtained. Experimental validation is performed, both quantitative and qualitative, in the public dataset DIARETDB1. The result achieved a better accuracy compared to other methods.

Keyphrases: Micro aneurysm (MA), Stacked Sparse Autoencoder, Support Vector Machine, unsupervised learning

@booklet{EasyChair:9348,
  author    = {P S Bindhya and R Chitra and V S Bibinraj},
  title     = {Sparse Autoencoder Based Microaneurysm Detection},
  howpublished = {EasyChair Preprint 9348},
  year      = {EasyChair, 2022}}