During my internship at “Achilles Resolute Private Limited” (https://www.achillesresolute.com/), I was design various classification and segmentation CNN models using Python Keras, Opencv, Sklearn etc. libraries.
I used Deep learning-based segmentation algorithms to segment out below 40 days of baby's retinopathy images and also calculate the tortuosity index to find the severity of the diseases. Challenge in this work is that mask binary dataset is not given. so we have to collect some datasets from different fundus datasets and train the model. Also, baby's retinopathy images are not very clear as fundus dataset, so image processing techniques also used to fix these issues.
This repository is an extend of that work. I add some other important features in this repo regarding Fundus images.
Blood vessels damaged from diabetic retinopathy can cause vision loss in two ways:Fragile, abnormal blood vessels can develop and leak blood into the center of the eye, blurring vision. This is proliferative retinopathy and is the fourth and most advanced stage of the disease.
Fluid can leak into the center of the macula, the part of the eye where sharp, straight-ahead vision occurs. The fluid makes the macula swell, blurring vision. This condition is called macular edema. It can occur at any stage of diabetic retinopathy, although it is more likely to occur as the disease progresses. About half of the people with proliferative retinopathy also have macular edema.
At first, you will see a few specks of blood, or spots, “floating” in your vision. If spots occur, see your eye care professional as soon as possible. You may need treatment before more serious bleeding occurs. Hemorrhages tend to happen more than once, often during sleep.
In this part of the project I use odir dataset and aptos dataset. At first using odir dataset I detect which disease present then using aptos dataset decide diabetic retinopathy diesses grading or sevierity.
odir dataset labels are as follows:
- normal (N)
- diabetes (D)
- glaucoma (G)
- cataract (C)
- AMD (A),
- hypertension (H)
- myopia (M)
- other diseases/abnormalities (O)
aptos dataset labels are as follows:
- No DR
- Mild
- Moderate
- Severe
- Proliferative DR
Tortuosity is a property of a curve being tortuous (twisted; having many turns). There have been several attempts to quantify this property. Tortuosity is commonly used to describe diffusion and fluid flow in porous media, such as soils and snow.
For this part of the project I use [DRIVE] dataset.
You can see from histogram plot that there is some serious channel issue. Try to fix it first with image processing
For segmentation here I use UNET.
React
Typescript
Django
Laravel
Facebook
Microsoft
Google
Alibaba
D3
Tencent