Measurement of central subfield thickness based on depth learning

Abstract

Central subfield thickness (CST) can assist in the diagnosis of many diseases, which can be observed through OCT images. This paper proposes a new deep learning framework for measuring CST. In this paper, the original OCT image is segmented based on U-Net, and a classification task is introduced here to determine whether the original image is taken from the center of the eye, so as to improve the segmentation effect of the center of the retina. The CST value of the segmented image is calculated through a double tower regression model, which is composed of the reduced dimension self-attention model and ResNet splicing. Through experimental verification, the regression accuracy of this framework is about 8% higher than that of other models.