Hongying Lilian Tang

Histological image retrieval based on semantic content analysis

The demand for automatic recognition and retrieval of medical images for screening, reference, and management is increasing. We present an intelligent content-based image retrieval system called I-Browse, which integrates both iconic and semantic content for histological image analysis. The I-Browse system combines low-level image processing technology with high-level semantic analysis of medical image content through different processing modules in the proposed system architecture. Similarity measures are proposed and their performance is evaluated. Furthermore, as a byproduct of semantic analysis, I-Browse allows textual annotations to be generated for unknown images. As an image browser, apart from retrieving images by image example, it also supports query by natural language.

An Automated Detection System for Microaneurysms That Is Effective across Different Racial Groups

Patients without diabetic retinopathy (DR) represent a large proportion of the caseload seen by the DR screening service so reliable recognition of the absence of DR in digital fundus images (DFIs) is a prime focus of automated DR screening research. We investigate the use of a novel automated DR detection algorithm to assess retinal DFIs for absence of DR. A retrospective, masked, and controlled image-based study was undertaken. 17,850 DFIs of patients from six different countries were assessed for DR by the automated system and by human graders. The systems performance was compared across DFIs from the different countries/racial groups. The sensitivities for detection of DR by the automated system were Kenya 92.8%, Botswana 90.1%, Norway 93.5%, Mongolia 91.3%, China 91.9%, and UK 90.1%. The specificities were Kenya 82.7%, Botswana 83.2%, Norway 81.3%, Mongolia 82.5%, China 83.0%, and UK 79%. There was little variability in the calculated sensitivities and specificities across the six different countries involved in the study. These data suggest the possible scalability of an automated DR detection platform that enables rapid identification of patients without DR across a wide range of races.

Localizing Microaneurysms in Fundus Images Through Singular Spectrum Analysis

Goal: Reliable recognition of microaneurysms (MAs) is an essential task when developing an automated analysis system for diabetic retinopathy (DR) detection. In this study, we propose an integrated approach for automated MA detection with high accuracy. Methods: Candidate objects are first located by applying a dark object filtering process. Their cross-section profiles along multiple directions are processed through singular spectrum analysis. The correlation coefficient between each processed profile and a typical MA profile is measured and used as a scale factor to adjust the shape of the candidate profile. This is to increase the difference in their profiles between true MAs and other non-MA candidates. A set of statistical features of those profiles is then extracted for a K-nearest neighbor classifier. Results: Experiments show that by applying this process, MAs can be separated well from the retinal background, the most common interfering objects and artifacts. Conclusion: The results have demonstrated the robustness of the approach when testing on large scale datasets with clinically acceptable sensitivity and specificity. Significance: The approach proposed in the evaluated system has great potential when used in an automated DR screening tool or for large scale eye epidemiology studies.

Measurement of optical cup-to-disc ratio in fundus images for glaucoma screening

Glaucoma is the second leading cause of blindness. There is no cure for glaucoma yet and early detection is critical to avoid total loss of vision. Cup to disc ratio (CDR) is commonly used by ophthalmologists to diagnose glaucoma. The purpose of this work is to develop an automatic approach to measure the cup to disc ratio (CDR) for glaucoma screening. In this paper, superpixels clustering algorithm; simple linear iterative clustering (SLIC) and a feed-forward neural network classifier have been utilized. A set of superpixels features are extracted and then used for training the classifier. To detect the optic disc and cup boundaries, the classifier is used to classify the superpixels in the region of interest. Then morphological operations and elliptical estimation approach are applied for the final cup and disc boundary detection and segmentation. The CDR is calculated based on these segmentations. Experiments show that by training the non-linear classifier on a set of efficient features, the optic disc and cup can be correctly estimated even in a low contrast images. The results have also shown that effectiveness of the approach with 92% sensitivity and 88% specificity.

Real-time search-free multiple license plate recognition via likelihood estimation of saliency

A search-free car license plate localization method based on 3-D Bayesian saliency estimation has been developed for the first time.This method uses a 3-D object tracking algorithm based on Bayesian methods to estimate the 3-D salient regions by exploiting the motion information in the videos.The algorithm is fast and robust to variation in the environment and is suitable for localizing multiple plates.The tracking approach exhibits acceptable accuracy for different noise and artefact levels.The localization method uses adaptive size in closing to improve the accuracy. Display Omitted In this paper, we propose a novel search-free localization method based on 3-D Bayesian saliency estimation. This method uses a new 3-D object tracking algorithm which includes: object detection, shadow detection and removal, and object recognition based on Bayesian methods. The algorithm is tested over three image datasets with different levels of complexities, and the results are compared with those of benchmark methods in terms of speed and accuracy. Unlike most search-based license-plate extraction methods, our proposed 3-D Bayesian saliency algorithm has lower execution time (less than 60źms), more accuracy, and it is a search-free algorithm which works in noisy backgrounds.

An evolutionary approach for determining Hidden Markov Model for medical image analysis

Hidden Markov Model (HMM) is a technique highly capable of modelling the structure of an observation sequence. In this paper, HMM is used to provide the contextual information for detecting clinical signs present in diabetic retinopathy screen images. However, there is a need to determine a feature set that best represents the complexity of the data as well as determine an optimal HMM. This paper addresses these problems by automatically selecting the best feature set while evolving the structure and obtaining the parameters of a Hidden Markov Model. This novel algorithm not only selects the best feature set, but also identifies the topology of the HMM, the optimal number of states, as well as the initial transition probabilities.

Automated detection of diabetic retinopathy in three European populations

Objective: Currently 1/12 of the world’s population has diabetes mellitus (DM), many are or will be screened by having retinal images taken. This current study aims to compare the DAPHNE software’s ability to detect DR in three different European populations compared to human grading carried out at the Moorfields Eye Hospital Reading Centre (MEHRC). Participants: Retinal images were taken from participants of the HAPIEE study (Lithuania, n=1014), the PAMDI study (Italy, n=882) and the MARS study (Germany, n=909). Methods: All anonymized images were graded by human graders at MEHRC for the presence of DR. Independently, and without any knowledge of the human grader’s results, the DAPHNE software analysed the images and divided the participants into DR and no-DR groups. Main outcome measures: The primary outcomes were sensitivity, specificity, positive predictive value (PPV) and negative predictive value (NPV) of the DAPHNE software with regards to the identification of DR or no-DR on retinal images as compared to the human grader as reference standard. Results: A total of 2805 participants were enrolled from the three study sites. The sensitivity of the DAPHNE software was above 93% in all three studies specificity was above 80%, the PPV was above 28% and the NPV was not below 98.8% in any of the studies. The DAPHNE software did not miss any vision-threatening DR. The areas under the curve (AUC) for all three studies were above 0.96. DAPHNE reduced manual human workload by 70% but had a total false positive rate of 63%. Conclusions: The DAPHNE software showed to be reliable to detect DR on three different European populations, using three different imaging settings. Further testing is required to see scalability, performance on live DR screening systems and on camera settings different to these studies.

Search-free license plate localization based on saliency and local variance estimation

In recent years, the performance and accuracy of automatic license plate number recognition (ALPR) systems have greatly improved, however the increasing number of applications for such systems have made ALPR research more challenging than ever. The inherent computational complexity of search dependent algorithms remains a major problem for current ALPR systems. This paper proposes a novel search-free method of localization based on the estimation of saliency and local variance. Gabor functions are then used to validate the choice of candidate license plate. The algorithm was applied to three image datasets with different levels of complexity and the results compared with a number of benchmark methods, particularly in terms of speed. The proposed method outperforms the state of the art methods and can be used for real time applications.

Optimal and simultaneous designs of Hermitian transforms and masks for reducing intraclass separations of feature vectors for anomaly detection of diabetic retinopathy images

This paper proposes a novel methodology for the optimal and simultaneous designs of both Hermitian transforms and masks for reducing the intraclass separations of feature vectors for anomaly detection of diabetic retinopathy images. Each class of training images associates with a Hermitian transform, a mask and a known represented feature vector. The optimal and simultaneous designs of both the Hermitian transforms and the masks are formulated as least squares optimization problems subject to the Hermitian constraints. Since the optimal mask of each class of training images is dependent on the corresponding optimal Hermitian transform, only the Hermitian transforms are required to be designed. Nevertheless, the Hermitian transform design problems are optimization problems with highly nonlinear objective functions subject to the complex valued quadratic Hermitian constraints. This kind of optimization problems is very difficult to solve. To address the difficulty, this paper proposes a singular value decomposition approach for deriving a condition on the solutions of the optimization problems as well as an iterative approach for solving the optimization problems. Since the matrices characterizing the discrete Fourier transform, discrete cosine transform and discrete fractional Fourier transform are Hermitian, the Hermitian transforms designed by our proposed approach are more general than existing transforms. After both the Hermitian transforms and the masks for all classes of training images are designed, they are applied to test images. The test images will assign to the classes where the Euclidean 2-norms of the differences between the processed feature vectors of the test images and the corresponding represented feature vectors are minimum. Computer numerical simulation results show that the proposed methodology for the optimal and simultaneous designs of both the Hermitian transforms and the masks is very efficient and effective. The proposed technique is also very efficient and effective for reducing the intraclass separations of feature vectors for anomaly detection of diabetic retinopathy images.

Clinical Content Detection for Medical Image Retrieval

Content-based image retrieval (CBIR) is the most widely used method for searching large-scale medical image collections; however this approach is not suitable for high-level applications as human experts are accustomed to manage medical images based on their clinical features rather than primitive features. Automatic detection of clinical features in a large-scale image database and realization of image retrieval by clinical content are still open issues. This paper presents a Markov random field (MRF) based model for clinical content detection. Multiple classifiers are applied to recognize a wide range of clinical features in a large-scale histological image database, and they are further combined to generate more reliable and robust estimation. Spatial contexts will cooperate with local estimations in the MRF based model to make a decision based on global consistency. The detected clinical features will provide a basis for image retrieval. Experiments have been carried out in a large-scale histological image database with promising results