Detail publikace

Optical Disc Segmentation Using Fully Convolutional Neural Network in Retina Images

Originální název

Optical Disc Segmentation Using Fully Convolutional Neural Network in Retina Images

Anglický název

Optical Disc Segmentation Using Fully Convolutional Neural Network in Retina Images

Jazyk

en

Originální abstrakt

This paper focuses on optic disc segmentation, which is one of the main steps in glaucoma diagnostics. A novel method, based on semantic, pixel-wise segmentation using the fully convolutional network is applied to the RIM-ONE dataset. This approach is advantageous because no additional preprocessing or postprocessing is needed. Moreover, results are promising, reaching mean IOU at about 0.7 and thus can compete with state of the art methods. The only disadvantage lays in the need of training dataset of sufficient size.

Anglický abstrakt

This paper focuses on optic disc segmentation, which is one of the main steps in glaucoma diagnostics. A novel method, based on semantic, pixel-wise segmentation using the fully convolutional network is applied to the RIM-ONE dataset. This approach is advantageous because no additional preprocessing or postprocessing is needed. Moreover, results are promising, reaching mean IOU at about 0.7 and thus can compete with state of the art methods. The only disadvantage lays in the need of training dataset of sufficient size.

BibTex


@inproceedings{BUT149748,
  author="Branislav {Hesko} and Radim {Kolář} and Vratislav {Harabiš}",
  title="Optical Disc Segmentation Using Fully Convolutional Neural Network in Retina Images",
  annote="This paper focuses on optic disc segmentation, which is one of the main steps in glaucoma diagnostics. A novel method, based on semantic, pixel-wise segmentation using the fully convolutional network is applied to the RIM-ONE dataset. This approach is advantageous because no additional preprocessing or postprocessing is needed. Moreover, results are promising, reaching mean IOU at about 0.7 and thus can compete with state of the art methods. The only disadvantage lays in the need of training dataset of sufficient size.",
  address="Vysoké učení technické v Brně, Fakulta elektrotechniky a komunikačních technologií",
  booktitle="Proceedings of IEEE Student Branch Conference Blansko 2018",
  chapter="149748",
  edition="2018",
  howpublished="online",
  institution="Vysoké učení technické v Brně, Fakulta elektrotechniky a komunikačních technologií",
  year="2018",
  month="september",
  pages="16--20",
  publisher="Vysoké učení technické v Brně, Fakulta elektrotechniky a komunikačních technologií",
  type="conference paper"
}