Deep Semi-Supervised and Self-Supervised Learning for Diabetic Retinopathy Detection

Jose Arrieta; Oscar J. Perdomo; Fabio A. González

doi:10.1117/12.2669723

Deep Semi-Supervised and Self-Supervised Learning for Diabetic Retinopathy Detection

Jose Arrieta
, Oscar J. Perdomo
, Fabio A. González

Research output: Chapter in Book/Inform › Conference contribution

6 Scopus citations

Abstract

Diabetic retinopathy (DR) is one of the leading causes of blindness in the working-age population of developed countries, caused by a side effect of diabetes that reduces the blood supply to the retina. Deep neural networks have been widely used in automated systems for DR classification on eye fundus images. However, these models need a large number of annotated images. In the medical domain, annotations from experts are costly, tedious, and time-consuming; as a result, a limited number of annotated images are available. This paper presents a semi-supervised method that leverages unlabeled images and labeled ones to train a model that detects diabetic retinopathy. The proposed method uses unsupervised pretraining via self-supervised learning followed by supervised fine-tuning with a small set of labeled images and knowledge distillation to increase the performance in a classification task. This method was evaluated on the EyePACS test and Messidor-2 dataset achieving 0.94 and 0.89 AUC respectively using only 2% of EyePACS train labeled images.

Original language	English (US)
Title of host publication	18th International Symposium on Medical Information Processing and Analysis
Editors	Jorge Brieva, Pamela Guevara, Natasha Lepore, Marius G. Linguraru, Leticia Rittner, Eduardo Romero Castro
Publisher	SPIE
ISBN (Electronic)	978-151-066-254-4
DOIs	https://doi.org/10.1117/12.2669723
State	Published - 2023
Event	18th International Symposium on Medical Information Processing and Analysis - Valparaiso, Chile Duration: Nov 9 2022 → Nov 11 2022

Publication series

Name	Proceedings of SPIE - The International Society for Optical Engineering
Volume	12567
ISSN (Print)	0277-786X
ISSN (Electronic)	1996-756X

Conference

Conference	18th International Symposium on Medical Information Processing and Analysis
Country/Territory	Chile
City	Valparaiso
Period	11/9/22 → 11/11/22

UN SDGs

This output contributes to the following UN Sustainable Development Goals (SDGs)

SDG 3 Good Health and Well-being

All Science Journal Classification (ASJC) codes

Electronic, Optical and Magnetic Materials
Condensed Matter Physics
Computer Science Applications
Applied Mathematics
Electrical and Electronic Engineering

Access to Document

10.1117/12.2669723

Cite this

Arrieta, J., Perdomo, O. J., & González, F. A. (2023). Deep Semi-Supervised and Self-Supervised Learning for Diabetic Retinopathy Detection. In J. Brieva, P. Guevara, N. Lepore, M. G. Linguraru, L. Rittner, & E. R. Castro (Eds.), 18th International Symposium on Medical Information Processing and Analysis Article 125671J (Proceedings of SPIE - The International Society for Optical Engineering; Vol. 12567). SPIE. https://doi.org/10.1117/12.2669723

Arrieta, Jose ; Perdomo, Oscar J. ; González, Fabio A. / Deep Semi-Supervised and Self-Supervised Learning for Diabetic Retinopathy Detection. 18th International Symposium on Medical Information Processing and Analysis. editor / Jorge Brieva ; Pamela Guevara ; Natasha Lepore ; Marius G. Linguraru ; Leticia Rittner ; Eduardo Romero Castro. SPIE, 2023. (Proceedings of SPIE - The International Society for Optical Engineering).

@inproceedings{ce4f0b8ca5e64217926a4acae74cd20f,

title = "Deep Semi-Supervised and Self-Supervised Learning for Diabetic Retinopathy Detection",

abstract = "Diabetic retinopathy (DR) is one of the leading causes of blindness in the working-age population of developed countries, caused by a side effect of diabetes that reduces the blood supply to the retina. Deep neural networks have been widely used in automated systems for DR classification on eye fundus images. However, these models need a large number of annotated images. In the medical domain, annotations from experts are costly, tedious, and time-consuming; as a result, a limited number of annotated images are available. This paper presents a semi-supervised method that leverages unlabeled images and labeled ones to train a model that detects diabetic retinopathy. The proposed method uses unsupervised pretraining via self-supervised learning followed by supervised fine-tuning with a small set of labeled images and knowledge distillation to increase the performance in a classification task. This method was evaluated on the EyePACS test and Messidor-2 dataset achieving 0.94 and 0.89 AUC respectively using only 2\% of EyePACS train labeled images.",

author = "Jose Arrieta and Perdomo, \{Oscar J.\} and Gonz{\'a}lez, \{Fabio A.\}",

note = "Publisher Copyright: {\textcopyright} 2023 SPIE.; 18th International Symposium on Medical Information Processing and Analysis ; Conference date: 09-11-2022 Through 11-11-2022",

year = "2023",

doi = "10.1117/12.2669723",

language = "English (US)",

series = "Proceedings of SPIE - The International Society for Optical Engineering",

publisher = "SPIE",

editor = "Jorge Brieva and Pamela Guevara and Natasha Lepore and Linguraru, \{Marius G.\} and Leticia Rittner and Castro, \{Eduardo Romero\}",

booktitle = "18th International Symposium on Medical Information Processing and Analysis",

address = "United States",

}

Arrieta, J, Perdomo, OJ & González, FA 2023, Deep Semi-Supervised and Self-Supervised Learning for Diabetic Retinopathy Detection. in J Brieva, P Guevara, N Lepore, MG Linguraru, L Rittner & ER Castro (eds), 18th International Symposium on Medical Information Processing and Analysis., 125671J, Proceedings of SPIE - The International Society for Optical Engineering, vol. 12567, SPIE, 18th International Symposium on Medical Information Processing and Analysis, Valparaiso, Chile, 11/9/22. https://doi.org/10.1117/12.2669723

Deep Semi-Supervised and Self-Supervised Learning for Diabetic Retinopathy Detection. / Arrieta, Jose; Perdomo, Oscar J.; González, Fabio A.
18th International Symposium on Medical Information Processing and Analysis. ed. / Jorge Brieva; Pamela Guevara; Natasha Lepore; Marius G. Linguraru; Leticia Rittner; Eduardo Romero Castro. SPIE, 2023. 125671J (Proceedings of SPIE - The International Society for Optical Engineering; Vol. 12567).

Research output: Chapter in Book/Inform › Conference contribution

TY - GEN

T1 - Deep Semi-Supervised and Self-Supervised Learning for Diabetic Retinopathy Detection

AU - Arrieta, Jose

AU - Perdomo, Oscar J.

AU - González, Fabio A.

PY - 2023

Y1 - 2023

N2 - Diabetic retinopathy (DR) is one of the leading causes of blindness in the working-age population of developed countries, caused by a side effect of diabetes that reduces the blood supply to the retina. Deep neural networks have been widely used in automated systems for DR classification on eye fundus images. However, these models need a large number of annotated images. In the medical domain, annotations from experts are costly, tedious, and time-consuming; as a result, a limited number of annotated images are available. This paper presents a semi-supervised method that leverages unlabeled images and labeled ones to train a model that detects diabetic retinopathy. The proposed method uses unsupervised pretraining via self-supervised learning followed by supervised fine-tuning with a small set of labeled images and knowledge distillation to increase the performance in a classification task. This method was evaluated on the EyePACS test and Messidor-2 dataset achieving 0.94 and 0.89 AUC respectively using only 2% of EyePACS train labeled images.

AB - Diabetic retinopathy (DR) is one of the leading causes of blindness in the working-age population of developed countries, caused by a side effect of diabetes that reduces the blood supply to the retina. Deep neural networks have been widely used in automated systems for DR classification on eye fundus images. However, these models need a large number of annotated images. In the medical domain, annotations from experts are costly, tedious, and time-consuming; as a result, a limited number of annotated images are available. This paper presents a semi-supervised method that leverages unlabeled images and labeled ones to train a model that detects diabetic retinopathy. The proposed method uses unsupervised pretraining via self-supervised learning followed by supervised fine-tuning with a small set of labeled images and knowledge distillation to increase the performance in a classification task. This method was evaluated on the EyePACS test and Messidor-2 dataset achieving 0.94 and 0.89 AUC respectively using only 2% of EyePACS train labeled images.

UR - https://www.scopus.com/pages/publications/85159356489

UR - https://www.scopus.com/pages/publications/85159356489#tab=citedBy

U2 - 10.1117/12.2669723

DO - 10.1117/12.2669723

M3 - Conference contribution

AN - SCOPUS:85159356489

T3 - Proceedings of SPIE - The International Society for Optical Engineering

BT - 18th International Symposium on Medical Information Processing and Analysis

A2 - Brieva, Jorge

A2 - Guevara, Pamela

A2 - Lepore, Natasha

A2 - Linguraru, Marius G.

A2 - Rittner, Leticia

A2 - Castro, Eduardo Romero

PB - SPIE

T2 - 18th International Symposium on Medical Information Processing and Analysis

Y2 - 9 November 2022 through 11 November 2022

ER -

Arrieta J, Perdomo OJ, González FA. Deep Semi-Supervised and Self-Supervised Learning for Diabetic Retinopathy Detection. In Brieva J, Guevara P, Lepore N, Linguraru MG, Rittner L, Castro ER, editors, 18th International Symposium on Medical Information Processing and Analysis. SPIE. 2023. 125671J. (Proceedings of SPIE - The International Society for Optical Engineering). doi: 10.1117/12.2669723

Deep Semi-Supervised and Self-Supervised Learning for Diabetic Retinopathy Detection

Abstract

Publication series

Conference

UN SDGs

All Science Journal Classification (ASJC) codes

Access to Document

Other files and links

Fingerprint

Cite this