Incorporating patch subspace model in Mumford-Shah type active contours.

In this paper, we propose a unified energy minimization model for segmentation of non-smooth image structures, e.g., textures, based on Mumford-Shah functional and linear patch model. We consider that image patches of a non-smooth image structure can be modeled by a patch subspace, and image patches of different non-smooth image structures belong to different patch subspaces, which leads to a computational framework for segmentation of non-smooth image structures. Motivated by the Mumford-Shah model, we show that this segmentation framework is equivalent to minimizing a piecewise linear patch reconstruction energy. We also prove that the error of segmentation is bounded by the error of the linear patch reconstruction, meaning that improving the linear patch reconstruction for each region leads to reduction of the segmentation error. In addition, we derive an algorithm for the linear patch reconstruction with proven global optimality and linear rate of convergence. The segmentation in our method is achieved by minimizing a single energy functional without requiring predefined features. Hence, compared with the previous methods that require predefined texture features, our method can be more suitable for handling general textures in unsupervised segmentation. As a by-product, our method also produces a dictionary of optimized orthonormal descriptors for each segmented region. We mainly evaluate our method on the Brodatz textures. The experiments validate our theoretical claims and show the clear superior performance of our methods over other related methods for segmentation of the textures.

Region-based nasopharyngeal carcinoma lesion segmentation from MRI using clustering- and classification-based methods with learning.

In clinical diagnosis of nasopharyngeal carcinoma (NPC) lesion, clinicians are often required to delineate boundaries of NPC on a number of tumor-bearing magnetic resonance images, which is a tedious and time-consuming procedure highly depending on expertise and experience of clinicians. Computer-aided tumor segmentation methods (either contour-based or region-based) are necessary to alleviate clinicians' workload. For contour-based methods, a minimal user interaction to draw an initial contour inside or outside the tumor lesion for further curve evolution to match the tumor boundary is preferred, but parameters within most of these methods require manual adjustment, which is technically burdensome for clinicians without specific knowledge. Therefore, segmentation methods with a minimal user interaction as well as automatic parameters adjustment are often favored in clinical practice. In this paper, two region-based methods with parameters learning are introduced for NPC segmentation. Two hundred fifty-three MRI slices containing NPC lesion are utilized for evaluating the performance of the two methods, as well as being compared with other similar region-based tumor segmentation methods. Experimental results demonstrate the superiority of adopting learning in the two introduced methods. Also, they achieve comparable segmentation performance from a statistical point of view.

A computer assisted method for nuclear cataract grading from slit-lamp images using ranking.

In clinical diagnosis, a grade indicating the severity of nuclear cataract is often manually assigned by a trained ophthalmologist to a patient after comparing the lens' opacity severity in his/her slit-lamp images with a set of standard photos. This grading scheme is often subjective and time-consuming. In this paper, a novel computer-aided diagnosis method via ranking is proposed to facilitate nuclear cataract grading following conventional clinical decision-making process. The grade of nuclear cataract in a slit-lamp image is predicted using its neighboring labeled images in a ranked image list, which is achieved using a learned ranking function. This ranking function is learned via direct optimization on a newly proposed approximation to a ranking evaluation measure. Our proposed method has been evaluated by a large dataset composed of 1000 different cases, which are collected from an ongoing clinical population-based study. Both experimental results and comparison with several existing methods demonstrate the benefit of grading via ranking by our proposed method.

Towards automatic detection of age-related macular degeneration in retinal fundus images.

Age-related macular degeneration (AMD) is a leading cause of blindness worldwide. The disease is highly associated with age, and becoming increasingly prevalent in our aging societies. Drusen is a pathological feature that is well-associated with AMD. In this paper, we present a method of detecting drusen in retinal fundus images. The method first determines the location of the macula, which is used as a landmark for a clinical drusen grading overlay. Subsequently, regions of drusen are identified though a maximal region-based pixel intensity approach via RGB and HSV channels. Methods of reducing the effect of retinal and choroidal vessels are also described. The system is tested on a sample set of 16 fundus images from a clinical study, with half having drusen. Experiments on the results show a sensitivity and specificity of 0.75 on the test image set.

Epitomized summarization of wireless capsule endoscopic videos for efficient visualization.

A video recording of an examination by Wireless Capsule Endoscopy (WCE) may typically contain more than 55,000 video frames, which makes the manual visual screening by an experienced gastroenterologist a highly time-consuming task. In this paper, we propose a novel method of epitomized summarization of WCE videos for efficient visualization to a gastroenterologist. For each short sequence of a WCE video, an epitomized frame is generated. New constraints are introduced into the epitome formulation to achieve the necessary visual quality for manual examination, and an EM algorithm for learning the epitome is derived. First, the local context weights are introduced to generate the epitomized frame. The epitomized frame preserves the appearance of all the input patches from the frames of the short sequence. Furthermore, by introducing spatial distributions for semantic interpretation of image patches in our epitome formulation, we show that it also provides a framework to facilitate the semantic description of visual features to generate organized visual summarization of WCE video, where the patches in different positions correspond to different semantic information. Our experiments on real WCE videos show that, using epitomized summarization, the number of frames have to be examined by the gastroenterologist can be reduced to less than one-tenth of the original frames in the video.

A computer-aided diagnosis system of nuclear cataract via ranking.

A novel computer-aided diagnosis system of nuclear cataract via ranking is firstly proposed in this paper. The grade of nuclear cataract in a slit-lamp image is predicted based on its neighboring labeled images in a ranked images list, which is achieved using an optimal ranking function. A new ranking evaluation measure is proposed for learning the optimal ranking function via direct optimization. Our system has been tested by a large dataset composed of 1000 slit-lamp images from 1000 different cases. Both experimental results and comparison with several state-of-the-art methods indicate the superiority of our system.

A kernel-induced space selection approach to model selection in KLDA.

Model selection in kernel linear discriminant analysis (KLDA) refers to the selection of appropriate parameters of a kernel function and the regularizer. By following the principle of maximum information preservation, this paper formulates the model selection problem as a problem of selecting an optimal kernel-induced space in which different classes are maximally separated from each other. A scatter-matrix-based criterion is developed to measure the "goodness" of a kernel-induced space, and the kernel parameters are tuned by maximizing this criterion. This criterion is computationally efficient and is differentiable with respect to the kernel parameters. Compared with the leave-one-out (LOO) or k-fold cross validation (CV), the proposed approach can achieve a faster model selection, especially when the number of training samples is large or when many kernel parameters need to be tuned. To tune the regularization parameter in the KLDA, our criterion is used together with the method proposed by Saadi (2004). Experiments on benchmark data sets verify the effectiveness of this model selection approach.

Two criteria for model selection in multiclass support vector machines.

Practical applications call for efficient model selection criteria for multiclass support vector machine (SVM) classification. To solve this problem, this paper develops two model selection criteria by combining or redefining the radius-margin bound used in binary SVMs. The combination is justified by linking the test error rate of a multiclass SVM with that of a set of binary SVMs. The redefinition, which is relatively heuristic, is inspired by the conceptual relationship between the radius-margin bound and the class separability measure. Hence, the two criteria are developed from the perspective of model selection rather than a generalization of the radius-margin bound for multiclass SVMs. As demonstrated by extensive experimental study, the minimization of these two criteria achieves good model selection on most data sets. Compared with the k-fold cross validation which is often regarded as a benchmark, these two criteria give rise to comparable performance with much less computational overhead, particularly when a large number of model parameters are to be optimized.

Semi-supervised nasopharyngeal carcinoma lesion extraction from magnetic resonance images using online spectral clustering with a learned metric.

In this paper, we consider the extraction of nasopharyngeal carcinoma lesion from MR images as a region segmentation problem. We propose a semi-supervised segmentation approach to segment the lesion in two steps. First, a metric is learned in a supervised fashion, which maximizes the separation between two groups of pixels (tumor or non-tumor) with minimal user interaction. Second, the learned metric is used to complete extraction of tumor region in an unsupervised fashion. Several experiments were conducted to evaluate the performance of similar methods with learned metrics for grouping or classifying pixels to form the tumor region. It is observed that the spectral clustering-based method performs well and the performance is comparable or marginally better than the discriminative SVM-based method.

Incorporating temporal information into level set functional for robust ventricular boundary detection from echocardiographic image sequence.

Echocardiographic images often suffer from dropouts that lead to loss of signals on the ventricular boundary and cause the level set curve used to detect the boundary leaking out from the gaps on the boundary. In this paper, a novel method that incorporates temporal information into the level set functional is proposed to solve the leakage problem encountered when detecting the heart wall boundary from the echocardiographic image sequence. The ventricular boundary is quantitatively partitioned and classified into strong and weak segments. The weak segments are considered to be weakened by dropouts and there is low confidence on the presence of boundary. Temporal information from neighboring frames is exploited as a regularizer into the level set equation. Hence, the original boundary information in the weak segments can be reconstructed and the curve leakage problem can be remedied. Experimental results demonstrate the advantages of the proposed method for the intended task.

Incorporating temporal information into active contour method for detecting heart wall boundary from echocardiographic image sequence.

A new method which incorporates temporal information into the active contour function is proposed to solve the dropout and speckle noise problems encountered when detecting the inner heart wall boundary from echocardiographic image sequence. The ventricular boundary is considered to be composed of strong and weak segments. The weak segments are interfered by image degradations in ultrasound, and they are too weak to constrain the curve evolution. Temporal information is incorporated into the external energy of the active contour function to recover the missing boundary and strengthen the weak segments. Experimental results demonstrate the advantages of the proposed method for the intended task.

Extraction of metastatic lymph nodes from MR images using two deformable model-based approaches.

We presented and evaluated two deformable model-based approaches, region plus contour deformation (RPCD), and level sets to extract metastatic cervical nodal lesions from pretreatment T2-weighted magnetic resonance images. The RPCD method first uses a region deformation to achieve a rough boundary of the target node from a manually drawn initial contour, based on signal statistics. After that, an active contour deformation is employed to drive the rough boundary to the real node-normal tissue interface. Differently, the level sets move a manually drawn initial contour toward the desired nodal boundary under the control of the evolvement speed function, which is influenced by image gradient force. The two methods were tested by extracting 33 metastatic cervical nodes from 18 nasopharyngeal carcinoma patients. Experiments on a basis of pixel matching to reference standard showed that RPCD and level sets achieved averaged percentage matching at 82-84% and 87-88%, respectively. In addition, both methods had significantly lower interoperator variances than the manual tracing method. It was suggested these two methods could be useful tools for the evaluation of metastatic nodal volume as an indicator of classification and treatment response, or be alternatives for the delineation of metastatic nodal lesions in radiation treatment planning.

The relationship between nasopharyngeal carcinoma tumor volume and TNM T-classification: a quantitative analysis.

Recent findings show that tumor volume is a significant prognostic factor for the treatment of nasopharyngeal carcinoma (NPC). The inclusion of tumor volume as an additional prognostic factor in the UICC TNM classification system was suggested; however, how tumor volume could possibly be incorporated is still unexplored. In this paper, we report a quantitative analysis on the relationship between NPC tumor volume and T-classification, using the data from 206 NPC patients. By T-classification and semi-automatic tumor volume measurement, the difference in tumor volumes among the various TNM T-classification groups was examined. In addition, a statistics-based analysis scheme, which used the T-classification as the "gold standard", was proposed to classify NPC tumors into volume-based groups to explore the possible links. The results show that NPC tumor volume has positive correlation with advancing T-classification groups and significant difference existed in the distribution of T-classification among various volume-based groups (P < 0.001). By the proposed statistical scheme, tumor volume could be included as an additional prognostic factor in the TNM framework, following validation studies.

Correlation between MR imaging-derived nasopharyngeal carcinoma tumor volume and TNM system.

PURPOSE: To measure nasopharyngeal carcinoma tumor volume based on magnetic resonance images using a validated semiautomated measurement methodology and correlate tumor volume with TNM T classification. METHODS AND MATERIALS: The study population consisted of 206 consecutive nasopharyngeal carcinoma patients who had magnetic resonance imaging staging scans. Tumor volume was measured using a semisupervised knowledge-based fuzzy clustering algorithm. Patients were divided into 4 groups according to TNM T classification. The difference in tumor volumes among the various TNM T-classification groups was examined. RESULTS: The mean tumor volume in each T-classification group is as follows: T1, 8.6 mL +/- 5.0 (standard deviation [SD]); T2, 18.1 mL +/- 8.1 (SD); T3, 25.8 mL +/- 14.1 (SD); and T4, 36.2 mL +/- 18.9 (SD). The mean tumor volume increased significantly with advancing T classification (p < 0.0001). Tumor volume in a more advanced T group was significantly larger than that in an adjacent early T group (p < 0.01). CONCLUSION: Validated magnetic resonance imaging-based tumor volume shows positive correlation between tumor volume and advancing T-classification groups. It may be possible to incorporate tumor volume as an additional prognostic factor into the existing TNM system.

Characteristic wave detection in ECG signal using morphological transform.

BACKGROUND: Detection of characteristic waves, such as QRS complex, P wave and T wave, is one of the essential tasks in the cardiovascular arrhythmia recognition from Electrocardiogram (ECG). METHODS: A multiscale morphological derivative (MMD) transform-based singularity detector, is developed for the detection of fiducial points in ECG signal, where these points are related to the characteristic waves such as the QRS complex, P wave and T wave. The MMD detector is constructed by substituting the conventional derivative with a multiscale morphological derivative. RESULTS: We demonstrated through experiments that the Q wave, R peak, S wave, the onsets and offsets of the P wave and T wave could be reliably detected in the multiscale space by the MMD detector. Compared with the results obtained via with wavelet transform-based and adaptive thresholding-based techniques, an overall better performance by the MMD method was observed. CONCLUSION: The developed MMD method exhibits good potentials for automated ECG signal analysis and cardiovascular arrhythmia recognition.

A criterion for optimizing kernel parameters in KBDA for image retrieval.

A criterion is proposed to optimize the kernel parameters in Kernel-based Biased Discriminant Analysis (KBDA) for image retrieval. Kernel parameter optimization is performed by optimizing the kernel space such that the positive images are well clustered while the negative ones are pushed far away from the positives. The proposed criterion measures the goodness of a kernel space, and the optimal kernel parameter set is obtained by maximizing this criterion. Retrieval experiments on two benchmark image databases demonstrate the effectiveness of proposed criterion for KBDA to achieve the best possible performance at the cost of a small fractional computational overhead.

Life-threatening ventricular arrhythmia recognition by nonlinear descriptor.

BACKGROUND: Ventricular tachycardia (VT) and ventricular fibrillation (VF) are ventricular cardiac arrhythmia that could be catastrophic and life threatening. Correct and timely detection of VT or VF can save lives. METHODS: In this paper, a multiscale-based non-linear descriptor, the Hurst index, is proposed to characterize the ECG episode, so that VT and VF can be recognized as different from normal sinus rhythm (NSR) in the descriptor domain. RESULTS: This newly proposed technique was tested using MIT-BIH malignant ventricular arrhythmia database. The relationship between the ECG episode length and the corresponding recognition performance was studied. The experiments demonstrated good performance of the proposed descriptor. An accuracy rate as high as 100% was obtained for VT/VF to be recognized from NSR; for VT and VF to be recognized from each other, the recognition accuracy varies from 84.24% to 100%. In addition, the results were compared favorably against those obtained using Complexity measure. CONCLUSIONS: There is strong potential for using the Hurst index for malignant ventricular arrhythmia recognition in clinical applications.

Texture decomposition by harmonics extraction from higher order statistics.

In this paper, a method of harmonics extraction from Higher Order Statistics (HOS) is developed for texture decomposition. We show that the diagonal slice of the fourth-order cumulants is proportional to the autocorrelation of a related noiseless sinusoidal signal with identical frequencies. We propose to use this fourth-order cumulants slice to estimate a power spectrum from which the harmonic frequencies can be easily extracted. Hence, a texture can be decomposed into deterministic components and indeterministic components as in a unified texture model through a Wold-like decomposition procedure. The simulation and experimental results demonstrated that this method is effective for texture decomposition and it performs better than traditional lower order statistics based decomposition methods.