TY - GEN
T1 - A Deep Learning Based Automatic Severity Detector for Diabetic Retinopathy
AU - AlSaad, Rawan
AU - Al-Maadeed, Somaya
AU - Al Mamun, Md Abdullah
AU - Boughorbel, Sabri
N1 - Publisher Copyright:
© 2018, Springer International Publishing AG, part of Springer Nature.
PY - 2018
Y1 - 2018
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 - https://www.scopus.com/pages/publications/85050563641
U2 - 10.1007/978-3-319-96136-1_6
DO - 10.1007/978-3-319-96136-1_6
M3 - Conference contribution
AN - SCOPUS:85050563641
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
A2 - Perner, Petra
PB - Springer Verlag
T2 - 14th International Conference on Machine Learning and Data Mining in Pattern Recognition, MLDM 2018
Y2 - 15 July 2018 through 19 July 2018
ER -