Description
Flutter TensorFlow Lite Artificial Intelligence Application Detection of Diabetic Retinopathy
DR Scanner mobile app was developed with Flutter. In terms of code structure, it was written with the principle of clean code.
The interface is simple and very easy to edit.
It can be run on both Android and iOS platforms.
It has an application structure that can be installed on Google Play Store, App Store and Huawei App Gallery platforms.
It can be easily run offline as there are no options such as database and remote connection.
Developed with a modern, simple user-friendly interface using the advantages of Flutter, DRScanner makes it easier for you to detect Diabetic Retinopathy. And in seconds!
All the steps that need to be done in detail are included in the documentation.
It should not be thought of as just an eye scanning application.
If the Tensorflow model is changed, it can also be used to detect other diseases.
High glucose level disrupts the structure of the retinal layer in the eyes and causes diabetic retinopathy which is characterized with new pathologic blood vessels in the eyes. Although diabetic retinopathy is not clear at the beginning of the disease, it is the most common problem in people who have diabetes, and causes blindness or cloudy vision if it is not diagnosed at the beginning of the disease. For early diagnosis of diabetic retinopathy, regular fundus controls and examination of the edema in the vessels of the retina are made periodically by ophthalmologists.
With in the scope of this study, it is made possible to provide the early diagnosis and the level of diabetic retinopathy by using deep learning, image processing methods and convolutional neural networks of the retina. In order to provide ease and rapid of diagnosis of the diabetic retinopathy in daily life, the diagnosis protocol has been turned into a mobile application. With the mobile application, both the diagnosis and more regular results of the diabetic retinopathy can be obtained easily and practically.
Dataset Information
| Tag Number | | Tag |
| 0 | Patient |
| 1 | Healthy |
Steps Taken
- All images are cropped and resized using the resize script and pre-processing script.
- Images without retinopathy were projected using the rotation script; Images with retinopathy were reflected and rotated 90, 120, 180 and 270 degrees.
- After rotating and reflecting with and without retinopathy, the class imbalance has been resolved and detected several thousand images have retinopathy.
- In total, there are 5000 images processed by the neural network.
- All images were converted to NumPy Arrays using the conversion script. NumPy Arrays combined images and tags in an array and send the images to CNN.
- The model was created by using the TensorFlow and Keras libraries. For CNN, encoding was done by using anaconda as IDE and Jupyter Notepad within anaconda.
- The pictures are tagged and parsed the pictures used to train them in two different sequences according to the labelling.
- The pictures were then brought to a fixed size (255*255) by grayscale method .
- The images are then passed through CNN and are called learning.
- The trained model can be saved and then tested with pictures.
There are no reviews yet.