Novel methods for automatic analysis on vascular
PhD thesis, University of Nottingham.
Imaging has become an essential component in many fields of medical and clinical research. The availability of medical imaging provides an excellent opportunity for monitoring and diagnosis of diseases. Automatic analysis of medical images can help with these processes. To date, automatic medical image analysis is still a difficult job in the field of computer vision because of factors such as noise and intensity inhomogeneity, which affect the performance of an image analysis framework.
There are many types of medical images available and the analysis of each of these image types requires its own specially designed methods. Blood vessels are implicated in a large number of diseases, including diabetic retinopathy, cardiovascular diseases, and Alzheimer’s disease, which cause damage to blood vessels. Because of this, blood vessel analysis is of great importance in medical diagnostic applications. In addition, a type of glial cell called the microglia have the similar appearance to blood vessels, but have a damaging role in a range of brain disorders. Previous research has found that the development of blood vessels and the presence of microglia are strongly correlated. Thus, the analysis of microglial images can also help provide a valuable tool to gain insight into the biology of blood vessels in health and disease.
The primary objective of this Ph.D. thesis is to develop an automatic and reliable image analysis framework for analysing microvasculature data. A typical medical image analysis framework often includes several steps before diagnosing the disease of a patient. Firstly, the representation of an image is typically converted into a form that is easier to analyse, in a process known as segmentation.Traditional vessel segmentation methods often fail for images that contain high– density microvessels and background noise. The errors occur in the form of missing, undesired broken or incorrectly merged vessels, eventually leading to poor segmentation results. A method for both 2D and 3D vessel segmentation that is suitable for segmenting microvessels has therefore been proposed here. The method incorporates the advantages of a line filter and a Hessian–based vessel filter to enhance vessels, and this has shown to be reliable for noisy and inhomogeneous images. For evaluating the proposed method, three synthetic vessel-like models were created, each with different structural properties that are characteristic of the true vasculature, and varies levels of Gaussian noise were applied to these images to simulate noise present in real MRI data. Experimental results have shown that the proposed method produces good vessel segmentation on synthetic vessel images, with significant ability to deal with high noise levels. For example, the method produced 94.46%, 95.29% and 92.48% of segmentation accuracies on branch, spiral and column models with -80dB noise. In comparison, the line filter and Hessian–based filter failed to provide valid segmentation. Furthermore, the method has been applied to three real brain datasets: Micro-CT rat, monkey and MRI human brains. Since there is no ground truth available for the brain models to obtain quantitative analysis, selected results using the line filter, Hessian–based vessel filter, and the proposed method are compared, to demonstrate the strength of the proposed method. As shown by the testing, the proposed method performed very well and handled the noise better.
After segmentation, essential information can be extracted from the segmented structure for classification. This is known as feature extraction. Traditional vessel analysis methods, such as counting the number of branches or measuring the curvature and diameter of individual vessel, are unsuitable for the analysing microvasculature. Previous research used fractal analysis for calculating fractal dimensions of blood vessels, but at present, there has been no systematic research extracting discriminant features from the structure for classification. A novel quantitative analysis method, based on multifractal analysis, is performed on the segmented structures and support vector machine for classifying the structure based on their multifractal features is then introduced. Three publicly available retinal vascular image databases (STARE, MESSIDOR, DIARETDB) are used for the experiments. The proposed methods have produced accuracies of 85.5%, 77% and 81.2% for classification of healthy and diabetic retinal vasculatures on each of the datasets respectively. Experiments show that classification with multiple fractal features produces better rates compared with methods using a single fractal dimension value. Also, experiments demonstrate that classification accuracy can be affected by the accuracy of vessel segmentation algorithms.
The use of multifractal analysis for feature extraction can also be applied to analysing microglia morphology and for classification of microglial activation states. However, there is little prior research on the segmentation of microglial images. Common image segmentation methods are not suitable for the microglial histology images used, so an automated method capable of accurately and efficiently segmenting the microglia from histology images is introduced. The method includes the use of the fast split Bregman algorithm for image denoising and segmentation. A set of selected images were used for validation of the proposed automated technique against manual analysis. Experiments show that the proposed segmentation method has produced similar results to those obtained by the manual method, the method also overcomes the common problems that were described in previous research due to the inhomogeneity of the images. The classification method is then applied to a selected microglial cells dataset, and have produced accuracies of 94.27% for classifying microglial cells in four different activation states. Results have shown that the method is scalable for analysing large microglia datasets.
Thesis (University of Nottingham only)
||blood vessel, microglial cell, image segmentation, image enhancement, classification, fractal analysis
||Q Science > QA Mathematics > QA 75 Electronic computers. Computer science
R Medicine > RC Internal medicine
||UK Campuses > Faculty of Science > School of Computer Science
||20 Jul 2016 14:00
||01 Jan 2017 07:36
Actions (Archive Staff Only)