Generation of Data for Training Retinal Image Segmentation Models

Document Type

Conference Article

Publication Title

Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)

Abstract

Biomedical image segmentation requires pixel-wise labelling which is extremely time-consuming and the availability of additional training data is highly beneficial for training Deep Learning models. In addition to using Classical Image augmentation, generative adversarial networks have been used to augment the training data. This work is an investigation of the usefulness of generated medical images for Deep Learning segmentation models. An attempt has been made to create a computer-generated retinal image segmentation dataset using various state-of-the-art image generative deep learning models and use that dataset to train a supervised image segmentation model. Our experiments demonstrate that the generated data can be used successfully for medical segmentation tasks and improves the model’s performance over using classical augmentations.

First Page

483

Last Page

491

DOI

10.1007/978-3-031-45170-6_50

Publication Date

1-1-2023

This document is currently not available here.

Share

COinS