Search

This paper proposes a novel method for acquiring an intestinal direction vector (IDV) based on a single static wireless endoscopic image. The IDV can be used for navigation of wireless capsule endoscopy, and for multicamera microball system, this information can help to select cameras to capture images of the region of interest. Our proposal is based on lumen detection, which involves Bayer-format downsample, adaptive threshold segmentation, and radial texture detection. Then, an IDV calculation method with only one single static image by modeling the small intestine and image capture process is put forward. The performance of the proposed method is verified with experiments based on real digestive tract images and the microball demo system. The lumen detection method achieves 95.5% precision and 98.1% sensitivity. The experimental results in pig intestine show that the error of IDV is limited by a sphere with center (−0.00629, 0.00097, 0.00061) and radius 0.085. [ABSTRACT FROM PUBLISHER]

A system for epileptic seizure detection in electroencephalography (EEG) is described in this paper. One of the challenges is to distinguish rhythmic discharges from nonstationary patterns occurring during seizures. The proposed approach is based on an adaptive and localized time–frequency representation of EEG signals by means of rational functions. The corresponding rational discrete short-time Fourier transform (DSTFT) is a novel feature extraction technique for epileptic EEG data. A multilayer perceptron classifier is fed by the coefficients of the rational DSTFT in order to separate seizure epochs from seizure-free epochs. The effectiveness of the proposed method is compared with several state-of-art feature extraction algorithms used in offline epileptic seizure detection. The results of the comparative evaluations show that the proposed method outperforms competing techniques in terms of classification accuracy. In addition, it provides a compact representation of EEG time-series. [ABSTRACT FROM PUBLISHER]

This paper proposes a new computer-aided method for the skin lesion classification applicable to both melanocytic skin lesions (MSLs) and nonmelanocytic skin lesions (NoMSLs). The computer-aided skin lesion classification has drawn attention as an aid for detection of skin cancers. Several researchers have developed methods to distinguish between melanoma and nevus, which are both categorized as MSL. However, most of these studies did not focus on NoMSLs such as basal cell carcinoma (BCC), the most common skin cancer and seborrheic keratosis (SK) despite their high incidence rates. It is preferable to deal with these NoMSLs as well as MSLs especially for the potential users who are not enough capable of diagnosing pigmented skin lesions on their own such as dermatologists in training and physicians with different expertise. We developed a new method to distinguish among melanomas, nevi, BCCs, and SKs. Our method calculates 828 candidate features grouped into three categories: color, subregion, and texture. We introduced two types of classification models: a layered model that uses a task decomposition strategy and flat models to serve as performance baselines. We tested our methods on 964 dermoscopy images: 105 melanomas, 692 nevi, 69 BCCs, and 98 SKs. The layered model outperformed the flat models, achieving detection rates of 90.48%, 82.51%, 82.61%, and 80.61% for melanomas, nevi, BCCs, and SKs, respectively. We also identified specific features effective for the classification task including irregularity of color distribution. The results show promise for enhancing the capability of the computer-aided skin lesion classification. [ABSTRACT FROM PUBLISHER]

High performance of the brain–computer interface (BCI) needs efficient algorithms to extract discriminative features from raw electroencephalography (EEG) signals. In this paper, we present a novel scheme to extract spatial spectral features for the motor imagery-based BCI. The learning task is formulated by maximizing the mutual information between spatial spectral features (MMISS) and class labels, by which a unique objective function directly related to Bayes classification error is optimized. The spatial spectral features are assumed to follow a parametric Gaussian distribution, which has been validated by the normal distribution Mardia's test, and under this assumption the estimation of mutual information is derived. We propose a gradient based alternative and iterative learning algorithm to optimize the cost function and derive the spatial and spectral filters simultaneously. The experimental results on dataset IVa of BCI competition III and dataset IIa of BCI competition IV show that the proposed MMISS is able to efficiently extract discriminative features from motor imagery-based EEG signals to enhance the classification accuracy compared to other existing algorithms. [ABSTRACT FROM PUBLISHER]

Mild traumatic brain injury (mTBI) appears as low contrast lesions in magnetic resonance (MR) imaging. Standard automated detection approaches cannot detect the subtle changes caused by the lesions. The use of context has become integral for the detection of low contrast objects in images. Context is any information that can be used for object detection but is not directly due to the physical appearance of an object in an image. In this paper, new low-level static and dynamic context features are proposed and integrated into a discriminative voxel-level classifier to improve the detection of mTBI lesions. Visual features, including multiple texture measures, are used to give an initial estimate of a lesion. From the initial estimate novel proximity and directional distance, contextual features are calculated and used as features for another classifier. This feature takes advantage of spatial information given by the initial lesion estimate using only the visual features. Dynamic context is captured by the proposed posterior marginal edge distance context feature, which measures the distance from a hard estimate of the lesion at a previous time point. The approach is validated on a temporal mTBI rat model dataset and shown to have improved dice score and convergence compared to other state-of-the-art approaches. Analysis of feature importance and versatility of the approach on other datasets are also provided. [ABSTRACT FROM PUBLISHER]

This paper presents novel methods for classification of fetal heart rate (FHR) signals into categories that are meaningful for clinical implementation. They are based on generative models (GMs) and Bayesian theory. Instead of using scalar features that summarize information obtained from long-duration data, the models allow for explicit use of feature sequences derived from local patterns of FHR evolution. We compare our methods with a deterministic expert system for classification and with a support vector machine approach that relies on system-identification and heart rate variability features. We tested the classifiers on 83 retrospectively collected FHR records, with the gold-standard true diagnosis defined using umbilical cord pH values. We found that our methods consistently performed as well as or better than these, suggesting that the use of GMs and the Bayesian paradigm can bring significant improvement to automatic FHR classification approaches. [ABSTRACT FROM PUBLISHER]

It has been reported that concentrations of several biomarkers in diabetics’ breath show significant difference from those in healthy people's breath. Concentrations of some biomarkers are also correlated with the blood glucose levels (BGLs) of diabetics. Therefore, it is possible to screen for diabetes and predict BGLs by analyzing one's breath. In this paper, we describe the design of a novel breath analysis system for this purpose. The system uses carefully selected chemical sensors to detect biomarkers in breath. Common interferential factors, including humidity and the ratio of alveolar air in breath, are compensated or handled in the algorithm. Considering the intersubject variance of the components in breath, we build subject-specific prediction models to improve the accuracy of BGL prediction. A total of 295 breath samples from healthy subjects and 279 samples from diabetic subjects were collected to evaluate the performance of the system. The sensitivity and specificity of diabetes screening are 91.51% and 90.77%, respectively. The mean relative absolute error for BGL prediction is 21.7%. Experiments show that the system is effective and that the strategies adopted in the system can improve its accuracy. The system potentially provides a noninvasive and convenient method for diabetes screening and BGL monitoring as an adjunct to the standard criteria. [ABSTRACT FROM PUBLISHER]

Content-based image retrieval systems for \3-D medical datasets still largely rely on \2-D image-based features extracted from a few representative slices of the image stack. Most \2 -D features that are currently used in the literature not only model a \3-D tumor incompletely but are also highly expensive in terms of computation time, especially for high-resolution datasets. Radiologist-specified semantic labels are sometimes used along with image-based \2-D features to improve the retrieval performance. Since radiological labels show large interuser variability, are often unstructured, and require user interaction, their use as lesion characterizing features is highly subjective, tedious, and slow. In this paper, we propose a \3-D image-based spatiotemporal feature extraction framework for fast content-based retrieval of focal liver lesions. All the features are computer generated and are extracted from four-phase abdominal CT images. Retrieval performance and query processing times for the proposed framework is evaluated on a database of $44$ hepatic lesions comprising of five pathological types. Bull’s eye percentage score above $85\%$ is achieved for three out of the five lesion pathologies and for $98\%$ of query lesions, at least one same type of lesion is ranked among the top two retrieved results. Experiments show that the proposed system’s query processing is more than $20$ times faster than other already published systems that use $2$-D features. With fast computation time and high retrieval accuracy, the proposed system has the potential to be used as an assistant to radiologists for routine hepatic tumor diagnosis. [ABSTRACT FROM PUBLISHER]

By overlaying preoperative radiological 3-D models onto the intraoperative laparoscopic video, augmented-reality (AR) displays promise to increase surgeons’ visual awareness of high-risk surgical targets (e.g., the location of a tumor). Existing AR surgical systems lack in robustness and accuracy because of the many challenges in endoscopic imagery, such as frequent changes in illumination, rapid camera motions, prolonged organ occlusions, and tissue deformations. The frequent occurrence of these events can cause the loss of image (anchor) points, and thus, the loss of the AR display after a few frames. In this paper, we present the design of a new AR system that represents a first step toward long term and accurate augmented surgical display for monocular (calibrated and uncalibrated) endoscopic videos. Our system uses correspondence-search methods, and a new weighted sliding-window registration approach, to automatically and accurately recover the overlay by predicting the image locations of a high number of anchor points that were lost after a sudden image change. The effectiveness of the proposed system in maintaining a long term (over 2 min) and accurate (less than 1 mm) augmentation has been documented over a set of real partial-nephrectomy laparascopic videos. [ABSTRACT FROM PUBLISHER]

This study addresses the classification problem of the HEp-2 cell using indirect immunofluorescent (IIF) image analysis, which can indicate the presence of autoimmune diseases by finding antibodies in the patient serum. Generally, the method used for IIF analysis remains subjective, and depends too heavily on the experience and expertise of the physician. Recently, studies have shown that it is possible to identify the cell patterns using IIF image analysis and machine learning techniques. However, it still has large gap in recognition rates to the physical experts’ one. This paper explores an approach in which the discriminative features of HEp-2 cell images in IIF are extracted and then, the patterns of the HEp-2 cell are identified using machine learning techniques. Motivated by the progress in the research field of computer vision, as a result of which small local pixel pattern distributions can now be highly discriminative, the proposed strategy employs a parametric probability process to model local image patches (textons: microstructures in the cell image) and extract the higher-order statistics of the model parameters for the image description. The proposed strategy can adaptively characterize the microtexton space of HEp-2 cell images as a generative probability model, and discover the parameters that yield a better fitting of the training space, which would lead to a more discriminant representation for the cell image. The simple linear support vector machine is used for cell pattern identification because of its low computational cost, in particular for large-scale datasets. Experiments using the open HEp-2 cell dataset used in the ICIP2013 contest validate that the proposed strategy can achieve a much better performance than the widely used local binary pattern (LBP) histogram and its extensions, rotation invariant co-occurrence LBP, and pairwise rotation invariant co-occurrence LBP, and that the achieved recognition error rate is even very significantly below the observed intralaboratory variability. [ABSTRACT FROM PUBLISHER]