Cataract is one of the leading causes of blindness worldwide. A computer-aided approach to assess nuclear cataract automatically and objectively is proposed in this paper. An enhanced Active Shape Model (ASM) is investigated to extract robust lens contour from slit-lamp images. The mean intensity in the lens area, the color information on the central posterior subcapsular reflex, and the profile on the visual axis are selected as the features for grading. A Support Vector Machine (SVM) scheme is proposed to grade nuclear cataract automatically. The proposed approach has been tested using the lens images from Singapore National Eye Centre. The mean error between the automatic grading and grader's decimal grading is 0.38. Statistical analysis shows that 97.8% of the automatic grades are within one grade difference to human grader's integer grades. Experimental results indicate that the proposed automatic grading approach is promising in facilitating nuclear cataract diagnosis.

>Entry in SPIE Digital Library<

 

In recent years more and more computer aided diagnosis (CAD) systems are being used routinely in hospitals. Image-based knowledge discovery plays important roles in many CAD applications, which have great potential to be integrated into the next-generation picture archiving and communication systems (PACS). Robust medical image segmentation tools are essentials for such discovery in many CAD applications. In this paper we present a platform with necessary tools for performance benchmarking for algorithms of liver segmentation and volume estimation used for liver transplantation planning. It includes an abdominal computer tomography (CT) image database (DB), annotation tools, a ground truth DB, and performance measure protocols. The proposed architecture is generic and can be used for other organs and imaging modalities. In the current study, approximately 70 sets of abdominal CT images with normal livers have been collected and a user-friendly annotation tool is developed to generate ground truth data for a variety of organs, including 2D contours of liver, two kidneys, spleen, aorta and spinal canal. Abdominal organ segmentation algorithms using 2D atlases and 3D probabilistic atlases can be evaluated on the platform. Preliminary benchmark results from the liver segmentation algorithms which make use of statistical knowledge extracted from the abdominal CT image DB are also reported. We target to increase the CT scans to about 300 sets in the near future and plan to make the DBs built available to medical imaging research community for performance benchmarking of liver segmentation algorithms.

>Entry in SPIE Digital Library<

 

Cataract is one of the leading causes of blindness worldwide. A computer-aided approach to assess nuclear cataract automatically and objectively is proposed in this paper. An enhanced Active Shape Model (ASM) is investigated to extract robust lens contour from slit-lamp images. The mean intensity in the lens area, the color information on the central posterior subcapsular reflex, and the profile on the visual axis are selected as the features for grading. A Support Vector Machine (SVM) scheme is proposed to grade nuclear cataract automatically. The proposed approach has been tested using the lens images from Singapore National Eye Centre. The mean error between the automatic grading and grader's decimal grading is 0.38. Statistical analysis shows that 97.8% of the automatic grades are within one grade difference to human grader's integer grades. Experimental results indicate that the proposed automatic grading approach is promising in facilitating nuclear cataract diagnosis.

>Entry in SPIE Digital Library<

 

Medical image segmentation is a challenging process due to possible image over-segmentation and under-segmentation (leaking). The CALM medical image segmentation system is constructed with an innovative scheme that cascades threshold level-set and region-growing segmentation algorithms using Union and Intersection set operators. These set operators help to balance the over-segmentation rate and under-segmentation rate of the system respectively. While adjusting the curvature scalar parameter in the threshold level-set algorithm, we observe that the abrupt change in the size of the segmented areas coincides with the occurrences of possible leaking. Instead of randomly choose a value or use the system default curvature scalar values, this observation prompts us to use the following formula in CALM to automatically decide the optimal curvature values gamma to prevent the occurrence of leaking : delta2S/deltagamma2 >= M, where S is the size of the segmented area and M is a large positive number. Motivated for potential applications in organ transplant and analysis, the CALM system is tested on the segmentation of the kidney regions from the Magnetic Resonance images taken from the National University Hospital of Singapore. Due to the nature of MR imaging, low-contrast, weak edges and overlapping regions of adjacent organs at kidney boundaries are frequently seen in the datasets and hence kidney segmentation is prone to leaking. The kidney segmentation accuracy rate achieved by CALM is 22% better compared with those achieved by the component algorithms or the system without leaking detection mechanism. CALM is easy-to-implement and can be applied to many applications besides kidney segmentation.

>Entry in SPIE Digital Library<

 

Objective quantification of lens images is essential for cataract assessment and treatment. In this paper, bottom-up and top-down strategies are combined to detect the lens contour from the slit-lamp images. The center of the lens is localized by horizontal and vertical intensity profile clustering and the lens contour is estimated by fitting an ellipse. A modified active shape model (ASM) is further applied to detect the contour of the lens. The average intensity inside the lens is employed as the indicator of nuclear opacity. The relationship between our automated nuclear cataract assessment and the clinical grading is analyzed. The preliminary study of forty images shows that the difference between automatic grading and clinical grading is acceptable.

>Entry in IEEExplore<

 

Widely used in the evaluation of retrieval systems, the pooling method collects top ranked images from submitted retrieval systems resulting in possibly a very large pool of images. Inevitably, the pool may contain outliers. Human experts then manually annotate the relevance of them to create a ground truth for evaluation. Studies show that this annotation is time-consuming, tedious and inconsistent. To reduce human workload, this paper introduces an automatic method to detect outliers. Different from traditional detection methods using unsupervised techniques only, we utilize both supervised and unsupervised techniques sequentially as both positive and negative examples are (partially) available in this context. Specifically, support vector machines (SVMs) and fuzzy c-means clustering are used to predict data relevance and "outlierness". Performance improvements using our method after outlier removal have been validated on the medical image retrieval task in ImageCLEF 2004.

>Entry in IEEExplore<

 

In this paper we report our work using visual feature fusion for the tasks of medical image retrieval and annotation in the benchmark of ImageCLEF 2005. In the retrieval task, we use visual features without text information, having no relevance feedback. Both local and global features in terms of both structural and statistical nature are captured. We first identify visually similar images manually and form templates for each query topic. A pre-filtering process is utilized for a coarse retrieval. In the fine retrieval, two similarity measuring channels with different visual features are used in parallel and then combined in the decision level to produce a final score for image ranking. Our approach is evaluated over all 25 query topics with each containing example image(s) and topic textual statements. Over 50,000 images we achieved a mean average precision of 14.6%, as one of the best performed runs. In the annotation task, visual features are fused in an early stage by concatenation with normalization. We use support vector machines (SVM) with RBF kernels for the classification. Our approach is trained over a 9,000 image training set and tested over the given test set with 1000 images and on 57 classes with a correct classification rate of about 80%.

>Entry in Springer<

 

In this paper, we present an efficient and convenient relevance feedback (RF) by using a semi-supervised kernel-specified k-means clustering (SKKC) technique. SKKC is used to cluster the retrieval results so that RF can be conducted on the cluster level. Compared with traditional RF conducted on the point/single-image level, the new RF will facilitate the RF selection and reduce user's efforts on it. Furthermore, the proposed approach enables an accumulated learning ability by recording and learning from the history of users' RFs. The new RF is applied in a content-based medical image retrieval (CBMIR) system. Experimental results on ImageCLEF database of around 9,000 images have shown that the proposed new RF is able to improve effectiveness and efficiency of CBMIR

>Entry in IEEExplore<

 

In this paper, a multi-class classification system is developed for medical images. We have mainly explored ways to use different image features, and compared two classifiers: Principle Component Analysis (PCA) and Supporting Vector Machines (SVM) with RBF (radial basis functions) kernels. Experimental results showed that SVM with a combination of the middle-level blob feature and low-level features (down-scaled images and their texture maps) achieved the highest recognition accuracy. Using the 9000 given training images from ImageCLEF05, our proposed method has achieved a recognition rate of 88.9% in a simulation experiment. And according to the evaluation result from the ImageCLEF05 organizer, our method has achieved a recognition rate of 82% over its 1000 testing images.

>Entry in IEEExplore<

 

We propose a novel framework for content-based image retrieval with multiple parallel retrieval engines (MultiPRE) to achieve higher retrieval performance. Visual features, including both low-level features, such as color, texture and region features, and middle-level structure features, such as blob representation of objects are used to capture geometrical and statistical characteristics of images. Both clustering analysis and discrimination analysis are used as similarity measures in multiple retrieval engines, which are based on~principal component analysis (PCA) and support vector machines (SVM), respectively. Finally outputs of these engines are fused to determine ranking lists of retrieved images for given retrieval topics. The proposed framework has been evaluated based on the 26 image query topics over the CasImage database~with over 9000 medical images~used in ImageCLEF 2004, an international research effort for content-based image retrieval performance benchmark. Experiments show that the proposed framework achieved significantly better performance in terms of both the mean and the variance of average precision than the best run reported in ImageCLEF2004.

>Entry in the ACM Digital Library<

 

Medical image retrieval is still mainly a research domain with a large variety of applications and techniques. With the ImageCLEF 2004 benchmark, an evaluation framework has been created that includes a database, query topics and ground truth data. Eleven systems (with a total of more than 50 runs) compared their performance in various configurations. The results show that there is not any one feature that performs well on all query tasks. Key to successful retrieval is rather the selection of features and feature weights based on a specific set of input features, thus on the query task. In this paper we propose a novel method based on query topic dependent image features (QTDIF) for content-based medical image retrieval. These feature sets are designed to capture both inter-category and intra-category statistical variations to achieve good retrieval performance in terms of recall and precision. We have used Gaussian Mixture Models (GMM) and blob representation to model medical images and construct the proposed novel QTDIF for CBIR. Finally, trained multi-class support vector machines (SVM) are used for image similarity ranking. The proposed methods have been tested over the Casimage database with around 9000 images, for the given 26 image topics, used for imageCLEF 2004. The retrieval performance has been compared with the medGIFT system, which is based on the GNU Image Finding Tool (GIFT). The experimental results show that the proposed QTDIF-based CBIR can provide significantly better performance than systems based general features only.

>Entry in SPIE Digital Libary<

 

In this paper, two discriminative fusion schemes are proposed for automatic image annotation. One is the ensemble-pattern association based fusion and another is the model-based transformation. The fusion approaches are studied and evaluated in a unified framework for AIA based on the text representation of the image content and the MC MFoM learning. The schemes are flexible for fusing diverse visual features and multiple modalities. The discriminative learning can automatically weight the most important features for the classification. We evaluate the fusion schemes based on the Corel and TRECVID 2003 datasets. The experimental results clearly show that the proposed fusion schemes give a significant improvement in term of the mean of F1 as well as the number of the detected concepts.

>Entry in IEEEXplore<