Yalin Zheng

Automated Vessel Segmentation Using Infinite Perimeter Active Contour Model with Hybrid Region Information with Application to Retinal Images

Automated detection of blood vessel structures is becoming of crucial interest for better management of vascular disease. In this paper, we propose a new infinite active contour model that uses hybrid region information of the image to approach this problem. More specifically, an infinite perimeter regularizer, provided by using L2 Lebesgue measure of the γ-neighborhood of boundaries, allows for better detection of small oscillatory (branching) structures than the traditional models based on the length of a features boundaries (i.e., H1 Hausdorff measure). Moreover, for better general segmentation performance, the proposed model takes the advantage of using different types of region information, such as the combination of intensity information and local phase based enhancement map. The local phase based enhancement map is used for its superiority in preserving vessel edges while the given image intensity information will guarantee a correct features segmentation. We evaluate the performance of the proposed model by applying it to three public retinal image datasets (two datasets of color fundus photography and one fluorescein angiography dataset). The proposed model outperforms its competitors when compared with other widely used unsupervised and supervised methods. For example, the sensitivity (0.742), specificity (0.982) and accuracy (0.954) achieved on the DRIVE dataset are very close to those of the second observers annotations.

Automated segmentation of lumbar vertebrae in digital videofluoroscopic images

Low back pain is a significant problem in the industrialized world. Diagnosis of the underlying causes can be extremely difficult. Since mechanical factors often play an important role, it can be helpful to study the motion of the spine. Digital videofluoroscopy has been developed for this study and it can provide image sequences with many frames, but which often suffer due to noise, exacerbated by the very low radiation dosage. Thus, determining vertebra position within the image sequence presents a considerable challenge. There have been many studies on vertebral image extraction, but problems of repeatability, occlusion and out-of-plane motion persist. In this paper, we show how the Hough transform (HT) can be used to solve these problems. Here, Fourier descriptors were used to describe the vertebral body shape. This description was incorporated within our HT algorithm from which we can obtain affine transform parameters, i.e., scale, rotation and center position. The method has been applied to images of a calibration model and to images from two sequences of moving human lumbar spines. The results show promise and potential for object extraction from poor quality images and that models of spinal movement can indeed be derived for clinical application.

Automated Segmentation of Foveal Avascular Zone in Fundus Fluorescein Angiography

PURPOSE. To describe and evaluate the performance of a computerized automated segmentation technique for use in quantification of the foveal avascular zone (FAZ). METHODS. A computerized technique for automated segmentation of the FAZ using images from fundus fluorescein angiography (FFA) was applied to 26 transit-phase images obtained from patients with various grades of diabetic retinopathy. The area containing the FAZ zone was first extracted from the original image and smoothed by a Gaussian kernel (sigma = 1.5). An initializing contour was manually placed inside the FAZ of the smoothed image and iteratively moved by the segmentation program toward the FAZ boundary. Five tests with different initializing curves were run on each of 26 images to assess reproducibility. The accuracy of the program was also validated by comparing results obtained by the program with the FAZ boundaries manually delineated by medical retina specialists. Interobserver performance was then evaluated by comparing delineations from two of the experts. RESULTS. One-way analysis of variance indicated that the disparities between different tests were not statistically significant, signifying excellent reproducibility for the computer program. There was a statistically significant linear correlation between the results obtained by automation and manual delineations by experts. CONCLUSIONS. This automated segmentation program can produce highly reproducible results that are comparable to those made by clinical experts. It has the potential to assist in the detection and management of foveal ischemia and to be integrated into automated grading systems.

Retinal Vessel Segmentation: An Efficient Graph Cut Approach with Retinex and Local Phase

Our application concerns the automated detection of vessels in retinal images to improve understanding of the disease mechanism, diagnosis and treatment of retinal and a number of systemic diseases. We propose a new framework for segmenting retinal vasculatures with much improved accuracy and efficiency. The proposed framework consists of three technical components: Retinex-based image inhomogeneity correction, local phase-based vessel enhancement and graph cut-based active contour segmentation. These procedures are applied in the following order. Underpinned by the Retinex theory, the inhomogeneity correction step aims to address challenges presented by the image intensity inhomogeneities, and the relatively low contrast of thin vessels compared to the background. The local phase enhancement technique is employed to enhance vessels for its superiority in preserving the vessel edges. The graph cut-based active contour method is used for its efficiency and effectiveness in segmenting the vessels from the enhanced images using the local phase filter. We have demonstrated its performance by applying it to four public retinal image datasets (3 datasets of color fundus photography and 1 of fluorescein angiography). Statistical analysis demonstrates that each component of the framework can provide the level of performance expected. The proposed framework is compared with widely used unsupervised and supervised methods, showing that the overall framework outperforms its competitors. For example, the achieved sensitivity (0:744), specificity (0:978) and accuracy (0:953) for the DRIVE dataset are very close to those of the manual annotations obtained by the second observer.

Imaging and Evaluation of Corneal Vascularization Using Fluorescein and Indocyanine Green Angiography

PURPOSE To evaluate indocyanine green angiography (ICGA) and fluorescein angiography (FA) in imaging and quantifying corneal neovascularization (CNV). METHODS Patients with CNV were studied using a standardized protocol of color digital photography, FA, and ICGA. Images were graded independently by two observers and assessed for quality, phases of fluorescence, and leakage. Areas of CNV and vasculature geometric properties were analyzed and quantified by an automated program. RESULTS Twenty-three patients with good quality images were included. Mean times to appearance of ICG and fluorescein were 17 and 20 seconds (P = 0.10). Best images for analysis were obtained at 64 seconds for ICGA and 47 seconds for FA. CNV not apparent on color or FA, particularly in the presence of scarring, was well delineated by ICGA. Leakage of ICGA did not occur. Fluorescein leakage from apical CNV images occurred significantly earlier (32 seconds) in patients with CNV of <6-month duration than those of >1-year (50 seconds) duration (P = 0.04). Mean area of CNV and vessel diameter were similar with ICGA (8.79 mm(2), 0.058 mm) or FA (7.74 mm(2), 0.054 mm) but significantly larger than on color (1.94 mm(2), 0.026 mm) images (P < 0.01). Vessel tortuosity was similar on ICGA (1.16), FA (1.17), and color (1.15) (P = 0.27). CONCLUSIONS Combined use of FA and ICGA are valuable tools with which to assess CNV and provide better vessel delineation than can be obtained with only color images. Parameters used to assess CNV, such as leakage, area, diameter, and tortuosity, may be useful measures for evaluating treatment. Videography is useful for detecting early leakage.

Convolutional Neural Networks for Diabetic Retinopathy

Abstract The diagnosis of diabetic retinopathy (DR) through colour fundus images requires experienced clinicians to identify the presence and significance of many small features which, along with a complex grading system, makes this a difficult and time consuming task. In this paper, we propose a CNN approach to diagnosing DR from digital fundus images and accurately classifying its severity. We develop a network with CNN architecture and data augmentation which can identify the intricate features involved in the classification task such as micro-aneurysms, exudate and haemorrhages on the retina and consequently provide a diagnosis automatically and without user input. We train this network using a high-end graphics processor unit (GPU) on the publicly available Kaggle dataset and demonstrate impressive results, particularly for a high-level classification task. On the data set of 80,000 images used our proposed CNN achieves a sensitivity of 95% and an accuracy of 75% on 5,000 validation images.

Lumbar spine visualisation based on kinematic analysis from videofluoroscopic imaging

Low back pain is a significant problem and its cost is enormous to society. However, diagnosis of the underlying causes remains problematic despite extensive study. Reasons for this arise from the deep-rooted situation of the spine and also from its structural complexity. Clinicians have to mentally convert 2-D image information into a 3-D form to gain a better understanding of structural integrity. Therefore, visualisation and animation may be helpful for understanding, diagnosis and for guiding therapy. Some low back pain originates from mechanical disorders, and study of the spine kinematics may provide an insight into the source of the problem. Digital videofluoroscopy was used in this study to provide 2-D image sequences of the spine in motion, but the images often suffer due to noise, exacerbated by the very low radiation dosage. Thus determining vertebrae position within the image sequence presents a considerable challenge. This paper describes a combination of spine kinematic measurements with a solid model of the human lumbar spine for visualisation of spine motion. Since determination of the spine kinematics provides the foundation and vertebral extraction is at the core, this is discussed in detail. Edge detection is a key feature of segmentation and it is shown that phase congruency performs better than most established methods with the rather low-grade image sequences from fluoroscopy. The Hough transform is then applied to determine the positions of vertebrae in each frame of a motion sequence. In the Hough transform, Fourier descriptors are used to represent the vertebral shapes. The results show that the Hough transform is a very promising technique for vertebral extraction from videofluoroscopic images. A dynamic visualisation package has been developed in order to view the moving lumbar spine from any angle and viewpoint. Wire frame models of the vertebrae were built by using CT images from the Visible Human Project and these models are scaled to match the fluoroscopic image data. For animation, the spinal kinematic data from the motion study is incorporated.

Data mining techniques for the screening of age-related macular degeneration

Age related macular degeneration (AMD) is the primary cause of adult blindness. Currently AMD cannot be cured, however early detection does allow the progress of the condition to be inhibited. One of the first symptoms of AMD is the presence of fatty deposits, called drusen, on the retina. The presence of drusen may be identified through the manual inspection/screening of retinal images. This task, however, requires recourse to domain experts and is therefore resource intensive. This paper proposes and compares two data mining techniques to support the automated screening for AMD. The first uses spatial-histograms, that maintain both image colour and spatial information, for the image representation; to which a case based reasoning (CBR) classification technique is applied. The second is founded on a hierarchical decomposition of the image set so that a tree representation is generated. A weighted frequent sub-graph mining technique is then applied to this representation to identify sub-trees that frequently occur across the data set. The identified sub-trees are then encoded in the form of feature vectors to which standard classification techniques can be applied.

Quantifying changes in corneal neovascularization using fluorescein and indocyanine green angiography

PURPOSE To quantify changes in corneal neovascularization in patients with active keratitis after treatment using color imaging, fluorescein angiography (FA), and indocyanine green angiography (ICGA). DESIGN Prospective, interventional case series. METHODS Twelve consecutive patients were studied. A comparison of corneal neovascularization parameters was undertaken before and after resolution of the keratitis. A slit-lamp digital camera acquired images of the neovascularization using color imaging, FA, and ICGA. The best-quality images were selected using a grading system, and the neovascular regions of interest were analyzed using automated in-house software. The parameters of analysis were vessel area, diameter, tortuosity, and FA dye leakage. RESULTS There was a significant reduction in the area of neovascularization after treatment on color imaging (0.78 mm(2); P < .05), FA (2.33 mm(2); P < .01), and ICGA (2.07 mm(2); P < .01). There was also a significant reduction in mean vessel diameter across the region of interest for each patient, more marked on FA (42.74 to 32.52 μm; P < .01) and ICGA (44.77 to 33.29 μm; P < .01) than on color imaging (29.10 to 25.17 μm; P < .01). A significant change in vessel tortuosity was not observed. There was a significant increase in FA dye leakage time (12.41 seconds; P < .05) after treatment. CONCLUSIONS We demonstrate application of an objective method for analyzing changes in corneal neovascularization. The excellent vessel delineation with ICGA even in the presence of stromal scars makes it an ideal agent for measurement of vessel parameters. FA is useful at detecting vessel leakage, and the time to leakage provides a possible measure of vessel staging.

Corneal Angiography for Guiding and Evaluating Fine-Needle Diathermy Treatment of Corneal Neovascularization

PURPOSE To investigate the outcome of selective occlusion of the afferent vessel of corneal neovascular complexes (CoNVs), using angiographically guided fine-needle diathermy (FND). DESIGN Retrospective interventional case series. SUBJECTS Patients with CoNV unresponsive to topical steroid therapy. METHODS Visual acuity, color images, and fluorescein angiography and indocyanine green angiography were measured before and after FND with a minimum of 3 months of follow-up. The number of afferent vessels crossing the limbus, time to fluorescein leakage, area, and geometric properties of the CoNV were determined using an in-house automated program written in numerical computing language (MatLab R14; The MathWorks Inc., Natick, MA). The location of the afferent vessel was identified from the angiographic images and marked at the slit lamp using a needle to make a cut to the depth of the vessel. We then applied FND using an electrolysis needle. MAIN OUTCOME MEASURES Area of CoNV. RESULTS Thirty patients underwent FND for CoNV that had not responded to treatment with topical steroids. The CoNV was associated with previous microbial keratitis (n = 26), intrastromal corneal ring segments (n = 2), ectodermal dysplasia (n = 1), and corneal choristoma (n = 1). Duration of CoNV was >6 months in 23 patients (77%), between 3 and 6 months in 3 patients (10%), and <3 months in 5 patients (13%). The number of afferent vessels per CoNV ranged from 1 to 3, with a mean diameter of 40 μm (standard deviation [SD], 10 μm) and mean time to leakage from apical vessels was 44.22 seconds (minimum, 27.43 seconds; maximum, 63.59 seconds). The number of FND treatments that were required was 1 for 20 patients (66.6%), 2 for 8 patients (26.6%), and 3 for 2 patients (6.6%). After FND, the area of CoNV reduced by 1.80 mm(2) (SD, 1.40 mm(2)), from 2.42 (SD, 1.59) to 0.62 mm(2) (SD, 0.73 mm(2)) up to 12 weeks postoperatively (P < 0.01). CONCLUSIONS The differentiation of afferent and efferent vessels using corneal angiography enables treatment to be selectively applied to the afferent vessels; there are usually 1 to 2 for each CoNV complex.