Automated artefacts detection for OCT-Angiography images using deep learning

Abstract

Optical Coherence Tomography Angiography (OCT-Angiography) is a recent non-invasive imaging technique which enables visualization of microvasculature in the eye. There is increasing interest in the use of OCT-Angiography for disease studies and diagnosis. However, interpretation of OCT-Angiography can be affected by localized artefacts which only degrades image quality in a focal region of the image. This study presents a Defect Detection System (DDS), capable of automatic identification of artefacts in an OCT-Angiography image. Three convolutional neural network (CNN) architectures (VGG-16, VGG-19, ResNet-50) from the ImageNet classification were used to train the automated classifier using transfer learning. Results show that VGG-19 obtained the highest accuracy of 99.52% compared to the other networks. The results are promising for the use of DDS for automated OCT-Angiography image quality assessment.