kr-viku / glaucoma-detection

AUTOMATED TYPE CLASSIFICATION OF GLAUCOMA DETECTION USING DEEP LEARNING

GLAUCOMA DETECTION USING DEEP LEARNING

Glaucoma are the leading cause of blindness in the working age population all over the world. Glaucoma through colour fundus images requires experienced clinicians to identify the presence and significance of many small features which, along with complex grading system, makes this difficult and time consuming task. Here we propose a CNN approach to diagnosing Glaucoma from fundus images and accurately classifying its severity. We develop a network with Convolutional Neural Network(CNN) architecture and data augmentation which can identify the intricate features involved in the classification task such as micro aneurysms, exudate and haemorrhages on the retina. We train this network using a high-end graphics processor unit (GPU). An open source Kaggle dataset is used as an input for DRand RIGA dataset is used as an input for Glaucoma. Total number of 25000 images are used for diabetic retinopathy and the testing accuracy for DR is 86%. Total number of 2664 images are used in glaucoma and the testing accuracy for glaucoma is 94%.

The architecture consists of

Data Pre-processing
Data Augmentation
Model trained using CNN
Class Prediction

Datasets

The Dataset for Glaucoma contains 2 classes. A total number of 2664 images are used for glaucoma.

class 0 for Non Glaucoma ( 1488 images ) class 1 for Glaucoma ( 1176 images )

Data Pre-processing:

Conversion of RGB images to Grayscale images. Conversion to grayscale images are done to improvement in testing accuracy.

Data Augmentation :

Rotation, Zooming, Shearing, Horizontal and Vertical Flip are done on images of all classes.
Data augmentation is done for balancing the classes with equal number of images .

Model trained using CNN:

Training the model by using CNN layers.
We have used 3 convolutoinal & max pooling layer, A dropout, Dense and a Fully Connected Layer.

The program runs in Google Colab.

Upload the source file and the datasets to the Google Drive. Open the project.ipynb file using Google Colab web app. Run all the cells in the given order. Output for Glaucoma is a binary classification. Output will be either glaucoma or Not glaucoma. .

glaucoma-detection's Issues

Error in block 5

Here is an error that happened when I execute the GlaucomaTest.ipynb file.
The error message has shown below.
`TypeError Traceback (most recent call last)
in ()
4 epochs = 30,
5 validation_data = test_set,
----> 6 validation_steps = 30/batch_size)

13 frames
/usr/local/lib/python3.7/dist-packages/tensorflow/python/framework/constant_op.py in convert_to_eager_tensor(value, ctx, dtype)
96 dtype = dtypes.as_dtype(dtype).as_datatype_enum
97 ctx.ensure_initialized()
---> 98 return ops.EagerTensor(value, ctx.device_name, dtype)
99
100

TypeError: Cannot convert 0.9375 to EagerTensor of dtype int64`

how can i fix those error thanks a lot

Error

validation_steps = 30/batch_size)
I got error on this line. Kindly guide me how i reslve it.

Saved Model File

Can you share the Gmodel.h5 file?
email- [email protected]

Error occur in block 5 when run the GlaucomaTest.ipynb

Here is an error that happened when I execute the GlaucomaTest.ipynb file.
The error message has shown below.
`TypeError Traceback (most recent call last)
in ()
4 epochs = 30,
5 validation_data = test_set,
----> 6 validation_steps = 30/batch_size)

13 frames
/usr/local/lib/python3.7/dist-packages/tensorflow/python/framework/constant_op.py in convert_to_eager_tensor(value, ctx, dtype)
96 dtype = dtypes.as_dtype(dtype).as_datatype_enum
97 ctx.ensure_initialized()
---> 98 return ops.EagerTensor(value, ctx.device_name, dtype)
99
100

TypeError: Cannot convert 0.9375 to EagerTensor of dtype int64`

Although I adjusted the "validation_steps = 30/batch_size" to "validation_steps = 1 ", the error message was the same.
Can you help me to figure out the bug or how I can solve this problem, thanks a lot!

How did you crop the data?

I have another question about the fundus images.
The images that you uploaded were cropped. But the fundus images that I had were not cropped.
So, can you tell me how what is the methods or any AI model you have used to crop the fundus? Thanks a lot!