Publication detail

Simultaneous lesions and optic disc segmentation from ophthalmoscopic images

HESKO, B. et al.

Original Title

Simultaneous lesions and optic disc segmentation from ophthalmoscopic images

English Title

Simultaneous lesions and optic disc segmentation from ophthalmoscopic images

Type

conference paper

Language

en

Original Abstract

In this paper we present a novel approach to retina images segmentation. Simultaneously, 5 classes of objects are segmented including microaneurysms, haemorrhages, hard and soft exudates and optic disc. Segmentation of these eye disease symptoms is not straightforward, segmented objects are small, granular and may not be present in all images. We employ deep learning with fully convolutional methods. For a comparison, two different convolutional networks are used, SegNet and PSPNet. They are based on deep classifiers; therefore, we were able to use pretrained weights and only fine-tune both networks. Results suggest, we have chosen a perspective approach because we reached promising results.

English abstract

In this paper we present a novel approach to retina images segmentation. Simultaneously, 5 classes of objects are segmented including microaneurysms, haemorrhages, hard and soft exudates and optic disc. Segmentation of these eye disease symptoms is not straightforward, segmented objects are small, granular and may not be present in all images. We employ deep learning with fully convolutional methods. For a comparison, two different convolutional networks are used, SegNet and PSPNet. They are based on deep classifiers; therefore, we were able to use pretrained weights and only fine-tune both networks. Results suggest, we have chosen a perspective approach because we reached promising results.

Keywords

deep learning, ophthalmology, retina images, segmentation

Released

05.10.2018

Publisher

Katedra biomedicínskeho inžinierstva a merania

Location

Kočice

ISBN

978-80-8086-271-8

Book

YBERC 2018 International Conference Proceedings

Pages from

1

Pages to

6

Pages count

6

BibTex


@inproceedings{BUT150379,
  author="Branislav {Hesko} and Vratislav {Harabiš}",
  title="Simultaneous lesions and optic disc segmentation from ophthalmoscopic images",
  annote="In this paper we present a novel approach to retina images segmentation. Simultaneously, 5 classes of objects are segmented including microaneurysms, haemorrhages, hard and soft exudates and optic disc. Segmentation of these eye disease symptoms is not straightforward, segmented objects are small, granular and may not be present in all images. We employ deep learning with fully convolutional methods. For a comparison, two different convolutional networks are used, SegNet and PSPNet. They are based on deep classifiers; therefore, we were able to use pretrained weights and only fine-tune both networks. Results suggest, we have chosen a perspective approach because we reached promising results.",
  address="Katedra biomedicínskeho inžinierstva a merania",
  booktitle="YBERC 2018 International Conference Proceedings",
  chapter="150379",
  howpublished="electronic, physical medium",
  institution="Katedra biomedicínskeho inžinierstva a merania",
  year="2018",
  month="october",
  pages="1--6",
  publisher="Katedra biomedicínskeho inžinierstva a merania",
  type="conference paper"
}