Abstract
Diabetic retinopathy (DR) detection techniques is a biometric modality that deserves systematic review and analysis of the connected algorithms for further improvement. The ophthalmologist uses retinal fundus images for the early detection of DR by segmenting the images. There are several segmentation algorithms reported as earlier. This chapter presents a comprehensive review of the methodology associated with retinal blood vessel extraction presented to date. The vessel segmentation techniques are divided into four main categories depending on their underlying methodology as pattern recognition, vessel tracking, model based, and hybrid approaches. A few of these methods are further classified into subsections. Finally, a comparative analysis of a few of the DR detection techniques will be presented here based on their merits, demerits, and other parameters like sensitivity, specificity, and accuracy and provide detailed information about its significance, present status, limitations, and future scope.
TopDiabetic Retinopathy Detection Technique Analysis
The automatic extraction process of the retinal blood vessel is a necessary step before the decision of the ophthalmologist to perform early screening and accomplish the treatment of DR disease. Different Researchers use different methods to extract blood vessels from fundus images. These vessel extraction techniques are divided into four main categories depending on their underlying methodology as pattern recognition, vessel tracking, model based and hybrid approaches. Few of these classifications are further classified into subsections.
Figure 1. Categorization of retinal vessel extraction technique
Key Terms in this Chapter
Vessel Tracking Method: Vessel tracking methods are used to segment the retinal vessels by tracking the vessels. The algorithms usually work with a set of reliable seed points and use them to track the retinal vasculature on the basis of the texture of local intensity information.
Matched Filtering Method: Matched filtering methods primarily use filters like Gaussian filters or their variation to check the gray distributions and shape of retinal vessels and collect corresponding responses. The presence of the required feature is recognized by using this filtering response.
Model-Based Method: In model-based methods, the explicit vessel models are used to extract the retinal vessel map. The method mainly consists of vector fields, the Hermite model, region growing, active contour, level set, and other methods. The method is classified into two types: (1) deformable models and (2) vessel profile models.
Hybrid Approaches: Hybrid techniques are a combination of two or more computational techniques which provide more advantage to detect components than any other individual technique. It helps to improve data analysis. The hybrid technique helps qualitative research to be effective.
Mathematical Morphology-Based Method: Morphological processing technique is basically combined with other vessel properties, to obtain vessel-like patterns from retinal fundus images. Objective image and Structuring Element (SE) are the main two parts of this technique. This technique offers the identification and segmentation of images by processing and analyzing SE in a binary image.
Neural Network-Based Method: Neural network consists of interconnected nodes which work like neurons of the human brain. Basically, it is a set of the algorithm, which is constructed to identify hidden patterns. It interprets data by using machine perception or labeling raw input.
Multiscale-Based Method: The multi-scale method represents the image at multiple levels/scales. It performs the segmentation of the retinal image on the basis of the variation of the image resolution. This process produced a scale space for the extraction of various structures from the input retinal images.