Вы находитесь на странице: 1из 5

DRIVE: Digital Retinal Images For Vessel Extraction

Abdul Mugeesh, ABES Engineering College, Rohit Sharma, MIET Jammu, Kanishk Nama, Thapar
University ,Dr. Suneet Kumar Gupta.


Abstract— ​Retinal Images and it’s information is tools that can segment the blood vessels from it’s
used to diagnose the possible retinal diseases such as retinal images for various studies.
diabetic retinopathy. However, due to non linear In this study, we are using Dense Block U-net
radiance and variance in modalities, the dissimilarity model which is inspired from Dense net and Retina
between blood vessels and background is very small. net. The database is provided by DRIVE database.
This problem embeds a need to perform some
The photographs for the DRIVE database were
operations to segment blood vessels from it’s retinal
images and improves the segmentation accuracy.In this obtained from a diabetic retinopathy screening
study, we propose a modified convolutional neural program in The Netherlands. The screening
network (CNN) architecture termed Fully Dense UNet population consisted of 400 diabetic subjects between
(FD-UNet) for segmentation of retinal blood vessels 25-90 years of age. Forty photographs have been
from their respective images . randomly selected, 33 do not show any sign of
diabetic retinopathy and 7 show signs of mild early
Index Terms​—Image reconstruction, image diabetic retinopathy. Each image has been JPEG
augmentation, D​ ense U-net, Retinal blood vessels, compressed. The set of 40 images has been divided
biomedical imaging into a training and a test set, both containing 20
images. The dataset consists of training set and
ground truth set and the resulting images are
I. INTRODUCTION compared with ground truth for loss evaluation and
training.
F​eature segmentation is an important problem in
image processing when there is contrast between the
background and the feature. With the help of
II. Related work:
Computer vision we can perform segmentation by Image segmentation is basically the segmentation of
taking this problem as a binary classification some specfic features from an image, here we are
problem. The growing application of Computer working on blood vessel segmentation. The purpose
vision has enable the bio-medical science community of doing image segmentation is to have proper vision
to optimize diagnostic process and visualisation. over blood vessels for doing diagnostic
Ophthalmology has benifited greatly from it which retinopathy. Olaf Ronneberger, Philipp Fischer, and
until few months ago depends heavily on visually Thomas Brox have done previous work by publishing
oriented tasks. the study that shows different results on leaverging
The Retina is the innermost part of the eye. In recent U-net model, their model has achieved an accuracy of
years, much research has been made into developing 92.13% consisting of 23 convolutional layers.They
an automatic have also used data augmentation to teach the
diagnostic system to cope up with diseases like network the desired invariance and robustness
Diabetic retinopathy. Retinal images analysis, it’s properties. A model called Retina U-net was also
segmentation and inspection of it’s features like developed by Orobix on top of U-net structure and
length, width, tortuosity serves as significant got accuracy of 97.90%.However,we modify the
indicators for analysing the presence of retinopathic conv2d layer to dense block and reduced the number
diseases. of layers from 23 to 10 and finally got better result.
The manual grading of these images, low contrast This method however do not achieve very good
and analysing is time consuming and requires alot of results when performing on larger epochs and on
mental concentration. Hence, the need for software images with a high number of exudates.
III METHODOLOGY

3.1 Dataset: 3.3 Convolution 2D

The DRIVE database has been established to enable This convo2D operation is extensively used in the
comparative studies on segmentation of blood vessels field of digital image processing wherein the 2D
in retinal images. The photographs for the DRIVE matrix representing the image will be convolved with
database were obtained from a diabetic retinopathy a comparatively smaller matrix called 2D kernel. The
screening program in The Netherlands. The screening U-net model uses Convolution 2D operation for the
population consisted of 400 diabetic subjects between processing of image for extraction of features. In our
25-90 years of age. Forty photographs have been model, we are using 10 convolution 2D layers with the
randomly selected, 33 do not show any sign of diabetic filter size varying as 32x32, 64x64,128x128,256x256
retinopathy and 7 show signs of mild early diabetic and 512x512.
retinopathy. Each image has been JPEG compressed.
The set of 40 images has been divided into a training 3.4 Batch Normalisation
and a test set, both containing 20 images.
 
We are processing the images in batches. As we know
3.2 MODEL: different features have different characteristics.That’s
  why we are performing batch normalisation to speed
The proposed segmentation method is summarized in up the process. In our model, this step is implemented
Fig. 1. It consists of a sequence of steps namely, because we are processing the input images on
Convolution 2D layer, batch normalisation Max batches. After applying the convolution 2d function,
pooling, image generation and morphological we normalizes the output of a previous activation layer
processing for enhancing features. Each of these steps by subtracting the batch mean and dividing by the
is detailed in the following section batch standard deviation.

3.5 Max pooling 2D

Max pooling is a sample-based discretization process.


The objective is to down-sample an input
representation reducing its dimensionality and
allowing for assumptions to be made about features
contained in the sub-regions binned. We have used
Max pooling with every convolution 2D layer as a 2x2
max pooling operation with stride 2 for
downsampling. At each downsampling step we double
the number of feature channels.
3.6 DenseBlock:

Fig 3.6.1: A 5-layer dense block with a growth rate of R =


4. Every layer takes all feature maps of all preceding
layers.
Fig 4.1 U-net Model on utilising the augmentation and
In DenseNet, whereas traditional convolutional Relu exhibits the best performance with patch size:
96x96 and 10 conv layers. We got PR curve of 71.99%.
networks with L layers have L connections—one
between each layer and its subsequent layer—our
network has ( L × (L + 1)) ÷ 2 direct connections. In
our model, we have used denseNet because it alleviate
the vanishing-gradient problem, strengthen feature
propagation, encourage feature reuse, and substantially
reduce the number of parameters For the feature of
segmentation, recent studies shows that dense
convolutional networks proves to be best.

3.7 Training

Our model is trained on 40 images which is 20 images


on training and 20 images validation on NVIDIA tesla
GPU. The information about training.
Number of Epochs: 100
Batch size: 20
Patch size: 96x96
Max filter : 512x512.
Fig 4.2 The ROC curve value is 0.9818 which depicts that
our model is 98.18% capable of distinguishing between
classes correctly.
IV. Experimental Results.

We empirically demonstrate Dense U-Net’s


effectiveness on several hyper parameters and
compare our all results , and also with other
achitectures and implementations. After training on
DRIVE dataset, our results are as follows.
Fig 4.5 The Resulting image is the final segmented image
of the given image obtained from model.
​ Table 4.1​ ​AUC ROC curve of Different models

V. Conclusion:

The performance of this segmentation model for the


retinal image for each subtask is comparable with the
retina-Unet@github but not able to beat
VesselNet@github whose highest accuracy is 98.41%
where as our best result got was 98.18%. As a future
work, we will consider doing hyper-parameter tuning
Fig 4.3 The colorful image is the original image and it has and try different convolutional 2D layers. Second,
been cropped around the FOV(45 degree) Attention based Unet and DeepLab-v3+ are also worth
to try.

VI. Acknowledgement
The authors are supported in part by the respected
Mentor Dr. Suneet Kumar gupta and Bennett
University for the opportunity to research.At an
individual level, Rohit Sharma is supported by his
college MIET Jammu, Abdul Mugeesh is supported by
ABES college where as Kanishk Nama is supported by
Thapar university.
Fig 4.4 Ground Truth image is the manual segmentation
of the given image by the experienced experts to compare
the segmentation results of the model.
VII. References:
[1] C. Cortes, X. Gonzalvo, V. Kuznetsov, M. Mohri,
and S. Yang. Adanet: Adaptive structural learning of
artificial neural networks. arXiv preprint
arXiv:1607.01097, 2016
[2] G. Huang, Zhuang liu and Laurens maaten:
Densely Connected Convolutional Networks.
arXiv:1608.06993, 2018
[3] Preethi, M., and R. Vanithamani. "Review of
retinal blood vessel detection methods for automated
diagnosis of Diabetic Retinopathy." International
Conference on Advances in Engineering, Science and
Management IEEE, 2012:262-265.
[4] Fu, Huazhu, et al. "Retinal vessel segmentation via
deep learning network and fully-connected conditional
random fields." IEEE, International Symposium on
Biomedical Imaging IEEE, 2016:698-701.
[5] Liskowski et al., "Segmenting Retinal Blood
Vessels with Deep Neural Networks", IEEE
Transactions on Medical Imaging, vol. PP, no. 99, pp.
1-1, 2016.
[6] Qiaoliang et al., "A Cross-Modality Learning
Approach for Vessel Segmentation in Retinal Images",
IEEE Transactions on Medical Imaging, vol. 35, no. 1,
pp. 109-118, 2016.
[7]Melinscak et al., "Retinal vessel segmentation
using deep neural networks", In Proceedings of the
10th International Conference on Computer Vision
Theory and Applications (VISIGRAPP 2015), (2015),
pp. 577–582.
[8] A. Krizhevsky, I. Sutskever, and G. E. Hinton.
Imagenet classification with deep convolutional neural
networks. In NIPS, 2012. 3, 7