A Deep Learning Based Automatic Severity Detector for Diabetic Retinopathy

Rawan T. Al Saad, Somaya Al-Maadeed, Md Abdullah Al Mamun, Sabri Boughorbel

Research output: Chapter in Book/Report/Conference proceedingConference contribution

Abstract

Automated Diabetic Retinopathy (DR) screening methods with high accuracy have the strong potential to assist doctors in evaluating more patients and quickly routing those who need help to a specialist. In this work, we used Deep Convolutional Neural Network architecture to diagnosing DR from digital fundus images and accurately classifying its severity. We train this network using a graphics processor unit (GPU) on the publicly available Kaggle dataset. We used Theano, Lasagne, and cuDNN libraries on two Amazon EC2 p2.xlarge instances and demonstrated impressive results, particularly for a high-level classification task. On the dataset of 30,262 training images and 4864 testing images, our model achieves an accuracy of 72%. Our experimental results showed that increasing the batch size does not necessarily speed up the convergence of the gradient computations. Also, it demonstrated that the number and size of fully connected layers do not have a significant impact on the performance of the model.

Original languageEnglish
Title of host publicationMachine Learning and Data Mining in Pattern Recognition - 14th International Conference, MLDM 2018, Proceedings
PublisherSpringer Verlag
Pages64-76
Number of pages13
ISBN (Print)9783319961354
DOIs
Publication statusPublished - 1 Jan 2018
Event14th International Conference on Machine Learning and Data Mining in Pattern Recognition, MLDM 2018 - New York, United States
Duration: 15 Jul 201819 Jul 2018

Publication series

NameLecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)
Volume10934 LNAI
ISSN (Print)0302-9743
ISSN (Electronic)1611-3349

Other

Other14th International Conference on Machine Learning and Data Mining in Pattern Recognition, MLDM 2018
CountryUnited States
CityNew York
Period15/7/1819/7/18

Fingerprint

Detector
Detectors
Graphics Processors
Image Model
Network Architecture
Network architecture
Digital Image
Batch
Screening
High Accuracy
Speedup
Routing
Neural Networks
Gradient
Neural networks
Testing
Unit
Experimental Results
Learning
Deep learning

Keywords

  • Convolutional Neural Networks
  • Deep learning
  • Diabetic retinopathy
  • Medical imaging

ASJC Scopus subject areas

  • Theoretical Computer Science
  • Computer Science(all)

Cite this

Al Saad, R. T., Al-Maadeed, S., Al Mamun, M. A., & Boughorbel, S. (2018). A Deep Learning Based Automatic Severity Detector for Diabetic Retinopathy. In Machine Learning and Data Mining in Pattern Recognition - 14th International Conference, MLDM 2018, Proceedings (pp. 64-76). (Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics); Vol. 10934 LNAI). Springer Verlag. https://doi.org/10.1007/978-3-319-96136-1_6

A Deep Learning Based Automatic Severity Detector for Diabetic Retinopathy. / Al Saad, Rawan T.; Al-Maadeed, Somaya; Al Mamun, Md Abdullah; Boughorbel, Sabri.

Machine Learning and Data Mining in Pattern Recognition - 14th International Conference, MLDM 2018, Proceedings. Springer Verlag, 2018. p. 64-76 (Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics); Vol. 10934 LNAI).

Research output: Chapter in Book/Report/Conference proceedingConference contribution

Al Saad, RT, Al-Maadeed, S, Al Mamun, MA & Boughorbel, S 2018, A Deep Learning Based Automatic Severity Detector for Diabetic Retinopathy. in Machine Learning and Data Mining in Pattern Recognition - 14th International Conference, MLDM 2018, Proceedings. Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics), vol. 10934 LNAI, Springer Verlag, pp. 64-76, 14th International Conference on Machine Learning and Data Mining in Pattern Recognition, MLDM 2018, New York, United States, 15/7/18. https://doi.org/10.1007/978-3-319-96136-1_6
Al Saad RT, Al-Maadeed S, Al Mamun MA, Boughorbel S. A Deep Learning Based Automatic Severity Detector for Diabetic Retinopathy. In Machine Learning and Data Mining in Pattern Recognition - 14th International Conference, MLDM 2018, Proceedings. Springer Verlag. 2018. p. 64-76. (Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)). https://doi.org/10.1007/978-3-319-96136-1_6
Al Saad, Rawan T. ; Al-Maadeed, Somaya ; Al Mamun, Md Abdullah ; Boughorbel, Sabri. / A Deep Learning Based Automatic Severity Detector for Diabetic Retinopathy. Machine Learning and Data Mining in Pattern Recognition - 14th International Conference, MLDM 2018, Proceedings. Springer Verlag, 2018. pp. 64-76 (Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)).
@inproceedings{7bb3945018b740cf952855650281129d,
title = "A Deep Learning Based Automatic Severity Detector for Diabetic Retinopathy",
abstract = "Automated Diabetic Retinopathy (DR) screening methods with high accuracy have the strong potential to assist doctors in evaluating more patients and quickly routing those who need help to a specialist. In this work, we used Deep Convolutional Neural Network architecture to diagnosing DR from digital fundus images and accurately classifying its severity. We train this network using a graphics processor unit (GPU) on the publicly available Kaggle dataset. We used Theano, Lasagne, and cuDNN libraries on two Amazon EC2 p2.xlarge instances and demonstrated impressive results, particularly for a high-level classification task. On the dataset of 30,262 training images and 4864 testing images, our model achieves an accuracy of 72{\%}. Our experimental results showed that increasing the batch size does not necessarily speed up the convergence of the gradient computations. Also, it demonstrated that the number and size of fully connected layers do not have a significant impact on the performance of the model.",
keywords = "Convolutional Neural Networks, Deep learning, Diabetic retinopathy, Medical imaging",
author = "{Al Saad}, {Rawan T.} and Somaya Al-Maadeed and {Al Mamun}, {Md Abdullah} and Sabri Boughorbel",
year = "2018",
month = "1",
day = "1",
doi = "10.1007/978-3-319-96136-1_6",
language = "English",
isbn = "9783319961354",
series = "Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)",
publisher = "Springer Verlag",
pages = "64--76",
booktitle = "Machine Learning and Data Mining in Pattern Recognition - 14th International Conference, MLDM 2018, Proceedings",

}

TY - GEN

T1 - A Deep Learning Based Automatic Severity Detector for Diabetic Retinopathy

AU - Al Saad, Rawan T.

AU - Al-Maadeed, Somaya

AU - Al Mamun, Md Abdullah

AU - Boughorbel, Sabri

PY - 2018/1/1

Y1 - 2018/1/1

N2 - Automated Diabetic Retinopathy (DR) screening methods with high accuracy have the strong potential to assist doctors in evaluating more patients and quickly routing those who need help to a specialist. In this work, we used Deep Convolutional Neural Network architecture to diagnosing DR from digital fundus images and accurately classifying its severity. We train this network using a graphics processor unit (GPU) on the publicly available Kaggle dataset. We used Theano, Lasagne, and cuDNN libraries on two Amazon EC2 p2.xlarge instances and demonstrated impressive results, particularly for a high-level classification task. On the dataset of 30,262 training images and 4864 testing images, our model achieves an accuracy of 72%. Our experimental results showed that increasing the batch size does not necessarily speed up the convergence of the gradient computations. Also, it demonstrated that the number and size of fully connected layers do not have a significant impact on the performance of the model.

AB - Automated Diabetic Retinopathy (DR) screening methods with high accuracy have the strong potential to assist doctors in evaluating more patients and quickly routing those who need help to a specialist. In this work, we used Deep Convolutional Neural Network architecture to diagnosing DR from digital fundus images and accurately classifying its severity. We train this network using a graphics processor unit (GPU) on the publicly available Kaggle dataset. We used Theano, Lasagne, and cuDNN libraries on two Amazon EC2 p2.xlarge instances and demonstrated impressive results, particularly for a high-level classification task. On the dataset of 30,262 training images and 4864 testing images, our model achieves an accuracy of 72%. Our experimental results showed that increasing the batch size does not necessarily speed up the convergence of the gradient computations. Also, it demonstrated that the number and size of fully connected layers do not have a significant impact on the performance of the model.

KW - Convolutional Neural Networks

KW - Deep learning

KW - Diabetic retinopathy

KW - Medical imaging

UR - http://www.scopus.com/inward/record.url?scp=85050563641&partnerID=8YFLogxK

UR - http://www.scopus.com/inward/citedby.url?scp=85050563641&partnerID=8YFLogxK

U2 - 10.1007/978-3-319-96136-1_6

DO - 10.1007/978-3-319-96136-1_6

M3 - Conference contribution

SN - 9783319961354

T3 - Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)

SP - 64

EP - 76

BT - Machine Learning and Data Mining in Pattern Recognition - 14th International Conference, MLDM 2018, Proceedings

PB - Springer Verlag

ER -