Detail publikace

Efficient Convolutional Neural Network Based Optic Disc Analysis Using Digital Fundus Images

JOSHI, R. DUTTA, M. SIKORA, P. KIAC, M.

Originální název

Efficient Convolutional Neural Network Based Optic Disc Analysis Using Digital Fundus Images

Anglický název

Efficient Convolutional Neural Network Based Optic Disc Analysis Using Digital Fundus Images

Jazyk

en

Originální abstrakt

Glaucoma is a disease of optic disk which is a worldwide problem and is one of the major and critical causes of the irreversible loss of the vision. It is the second most significant vision problem after cataract. In glaucoma, the sight loss is due to the permanent damage of the optic nerve which is the result of a chronic elevation of intraocular pressure. Most of the developed methods are costly and time consuming. Imaging of fundus for screening is the most common glaucoma detection technique. In this paper, an automatic and fast deep learning based method for glaucoma detection is presented. Digital fundus images of different resolution from three eye dataset are augmented by rotating and scaling at different values. The automatic setup can process the fundus image in 0.2 seconds. The proposed method is tested on the 1220 fundus images and achieved an average accuracy of 93.698%, sensitivity of 89.054% and specificity of 95.848%. Also, the results signify the proposed method is resolution independent, rotation and scaling invariant, and have rapid glaucoma screening time.

Anglický abstrakt

Glaucoma is a disease of optic disk which is a worldwide problem and is one of the major and critical causes of the irreversible loss of the vision. It is the second most significant vision problem after cataract. In glaucoma, the sight loss is due to the permanent damage of the optic nerve which is the result of a chronic elevation of intraocular pressure. Most of the developed methods are costly and time consuming. Imaging of fundus for screening is the most common glaucoma detection technique. In this paper, an automatic and fast deep learning based method for glaucoma detection is presented. Digital fundus images of different resolution from three eye dataset are augmented by rotating and scaling at different values. The automatic setup can process the fundus image in 0.2 seconds. The proposed method is tested on the 1220 fundus images and achieved an average accuracy of 93.698%, sensitivity of 89.054% and specificity of 95.848%. Also, the results signify the proposed method is resolution independent, rotation and scaling invariant, and have rapid glaucoma screening time.

Dokumenty

BibTex


@inproceedings{BUT164733,
  author="Rakesh Chandra {Joshi} and Malay Kishore {Dutta} and Pavel {Sikora} and Martin {Kiac}",
  title="Efficient Convolutional Neural Network Based Optic Disc Analysis Using Digital Fundus Images",
  annote="Glaucoma is a disease of optic disk which is a worldwide problem and is one of the major and critical causes of the irreversible loss of the vision. It is the second most significant vision problem after cataract. In glaucoma, the sight loss is due to the permanent damage of the optic nerve which is the result of a chronic elevation of intraocular pressure. Most of the developed methods are costly and time consuming. Imaging of fundus for screening is the most common glaucoma detection technique. In this paper, an automatic and fast deep learning based method for glaucoma detection is presented. Digital fundus images of different resolution from three eye dataset are augmented by rotating and scaling at different values. The automatic setup can process the fundus image in 0.2 seconds. The proposed method is tested on the 1220 fundus images and achieved an average accuracy of 93.698%, sensitivity of 89.054% and specificity of 95.848%. Also, the results signify the proposed method is resolution independent, rotation and scaling invariant, and have rapid glaucoma screening time.",
  address="IEEE",
  booktitle="Proceedings of the 2020 43rd International Conference on Telecommunications and Signal Processing (TSP)",
  chapter="164733",
  doi="10.1109/TSP49548.2020.9163560",
  howpublished="online",
  institution="IEEE",
  year="2020",
  month="august",
  pages="533--536",
  publisher="IEEE",
  type="conference paper"
}