FLANN based Exudates Extraction

Motivation

As a continuation to the Glaucoma detection project we have now removed the Blood vessels from the fundus image. So the next step is to remove the Hard exudates which obstruct the ONH detection.

ONH is the main marker to detect Glaucoma, but the extraction of ONH from fundus images is not accurate due to the presence of Exudates and other fundus pathogens. So in this study Blood vessels segmentation is the first step. I have applied Active contours(snakes) for this purpose.

We have developed a Classification-based algorithm to extract Exudates because these approaches proved to give better performance and also have a huge scope for the implementation of the existing evolutionary algorithms. The Classification-based methods can be broadly divided into three different steps.

1. Candidate region extraction
2. Feature extraction
3. Classification

1. Candidate region extraction

This is the first step in the extraction process where we extract all the pixels that are probable to be exudate pixels. All the yellowish objects in the image are coarsely separated, these are the Hard Exudates candidate regions which are processed further to classify. In this study, we have employed the Luminosity and contrast Normalization due to its better estimation of exudate pixels. Other methods are:

1. Luminosity and contrast normalization.
2. K-means clustering
3. Fuzzy C-means Clustering
4. LoG Transformation on different Intensity Bands
5. Stationery Wavelet Transform

2. Feature Extraction

Feature selection for solving a classification based problems depends on the discriminatory power of features. Most of the traditional feature selection methods are classifier-dependent. On the other hand, for a medical image analysis a classifier independent feature analysis is more beneficiary in terms of robustness and scalability. The ideal features are defined by their ability to gather information concerning the structure of the data rather than serving the requirements of a particular classifier. Logistic regression based feature selection method is most commonly used for obtaining classifier- independent features. These are the following 13 features selected based on their higher discriminatory.

1. Mean of blue channel intensity inside the region.
2. Mean of green channel intensity inside the region.
3. Standard deviation of the red channel inside the region.
4. Standard deviation of the blue channel inside the region
5. Mean of green channel intensity around the region.
6. Mean of blue channel intensity around the region
7. Region centroid in blue channel: x
8. Region centroid in blue channel: y
9. Color difference of the Red channel
10. Color difference of the green channel
11. Color difference of the blue channel
12. Region compactness
13. Homogeneity

3. Classification

FLANN model, first proposed by Pao is a single layered neural network with single neuron at the output. The architecture of the FLANN contains less computational load and high convergence rate than those of traditional neural networks due to its single layered structure. FLANN model can capture the non-linear relationship between the inputs and the output unlike the Multiple Regression which can capture only linear relationship between them. Below figure shows the simplified block diagram of the FLANN model. The elements of the input pattern vector applied to the FLANN model are the feature vectors obtained from the candidate regions and the output of the model is the 1 if the region is an exudate otherwise 0. The conventional non-linear functional expansions employed in the model are Trigonometric, Legendre & Chebyshev. In this study, only trigonometric expansion is employed as it is observed experimentally that trigonometric expansion provides better performance when compared to others.