Search

This paper presents a voice activity detector (VAD) for automatic switching between a noise classifier and a speech enhancer as part of the signal processing pipeline of hearing aid devices. The developed VAD consists of a computationally efficient feature extractor and a random forest classifier. Previously used signal features as well as two newly introduced signal features are extracted and fed into the classifier to perform automatic switching. This switching approach is compared to two popular VADs. The results obtained indicate the introduced approach outperforms these existing approaches in terms of both detection rate and processing time.

In this paper, the development of a speech processing pipeline on smartphones for hearing aid devices (HADs) is presented. This pipeline is used for noise suppression and speech enhancement (SE) to improve speech quality and intelligibility. The proposed method is implemented to run in real-time on Android smartphones. The results of the testing conducted indicate that the proposed method suppresses the noise and improves the perceptual quality of speech in terms of three objective measures of perceptual evaluation of speech quality (PESQ), noise attenuation level (NAL), and the coherent speech intelligibility index (CSII).

Information theoretic-based similarity measures, in particular mutual information, are widely used for intermodal/intersubject 3-D brain image registration. However, conventional mutual information does not consider spatial dependency between adjacent voxels in images, thus reducing its efficacy as a similarity measure in image registration. This paper first presents a review of the existing attempts to incorporate spatial dependency into the computation of mutual information (MI). Then, a recently introduced spatially dependent similarity measure, named spatial MI, is extended to 3-D brain image registration. This extension also eliminates its artifact for translational misregistration. Finally, the effectiveness of the proposed 3-D spatial MI as a similarity measure is compared with three existing MI measures by applying controlled levels of noise degradation to 3-D simulated brain images.

This paper presents an improved environment-adaptive noise suppression solution for the cochlear implants speech processing pipeline. This improvement is achieved by using a multi-band data-driven approach in place of a previously developed single-band data-driven approach. Seven commonly encountered noisy environments of street, car, restaurant, mall, bus, pub and train are considered to quantify the improvement. The results obtained indicate about 10% improvement in speech quality measures.

This paper presents a medication adherence monitoring system for pill bottles based on a wearable inertial sensor. Signal templates corresponding to the two actions of twist-cap and hand-to-mouth are created using a camera-assisted training phase. The act of pill intake is then identified by performing a moving window dynamic time warping in real-time between signal templates and the signals acquired by the wearable inertial sensor. The outcomes of the experimentations carried out indicate that the developed medical adherence monitoring system identifies the act of pill intake with a high degree of accuracy.

This paper presents a home-based Senior Fitness Test (SFT) measurement system by using an inertial sensor and a depth camera in a collaborative way. The depth camera is used to monitor the correct pose of a subject for a fitness test and any deviation from the correct pose while the inertial sensor is used to measure the number of a fitness test action performed by the subject within the time duration specified by the fitness protocol. The results indicate that this collaborative approach leads to high success rates in providing the SFT measurements under realistic conditions.

This paper presents the real-time implementation of an adaptive speech processing pipeline for cochlear implants on the smartphone platform. The pipeline is capable of real-time classification of background noise environment and automated tuning of a noise suppression component based upon the detected background noise environment. This pipeline was previously implemented on the FDA-approved PDA platform for cochlear implant studies. The paper discusses the steps taken to achieve the real-time implementation of the pipeline on the smartphone platform. In addition, it includes the real-time timing as well as the noise suppression results when the entire pipeline was run on the smartphone platform.

In this paper, the properties of a new class of causal Markov random fields, named symmetric Markov mesh random field, are initially discussed. It is shown that the symmetric Markov mesh random fields from the upper corners are equivalent to the symmetric Markov mesh random fields from the lower corners. Based on this new random field, a symmetric, corner-independent, and isotropic image model is then derived which incorporates the dependency of a pixel on all its neighbors. The introduced image model comprises the product of several local 1D density and 2D joint density functions of pixels in an image thus making it computationally tractable and practically feasible by allowing the use of histogram and joint histogram approximations to estimate the model parameters. An image restoration application is also presented to confirm the effectiveness of the model developed. The experimental results demonstrate that this new model provides an improved tool for image modeling purposes compared to the conventional Markov random field models.

A computationally efficient speech enhancement pipeline in noisy environments based on a single-processor implementation is developed for utilization in bilateral cochlear implant systems. A two-channel joint objective function is defined and a closed form solution is obtained based on the weighted-Euclidean distortion measure. The computational efficiency and no need for synchronization aspects of this pipeline make it a suitable solution for real-time deployment. A speech quality measure is used to show its effectiveness in six different noisy environments as compared to a similar one-channel enhancement pipeline when using two separate processors or when using independent sequential processing.

This paper presents an improved noise classification in environment-adaptive speech processing pipelines of cochlear implants. This improvement is achieved by using a dual-microphone and by using a computationally efficient feature-level combination approach to achieve real-time operation. A new measure named Suppression Advantage is also defined in order to quantify the noise suppression improvement of an entire pipeline due to noise classification. The noise classification and suppression improvement results are presented for four commonly encountered noise environments.

The extraction of regions-of-interest (ROIs) in hyperspectral images of breast cancer specimens is currently carried out manually or by visual inspection. In order to address the labor-intensive and time-consuming process of the manual extraction of ROIs in hyperspectral images, an algorithm is developed in this paper to automate the extraction process. This is achieved by using a contrast module and a homogeneity module to duplicate the same manual or visual steps that an expert goes through in order to extract ROIs. The success of the automated process is determined by comparing the classification rates of the automated approach with the manual approach in terms of the ability to separate cancer cases from normal cases.

This paper presents the real-time implementation of an environment-adaptive noise suppression algorithm on an FDA-approved PDA platform for cochlear implant studies. This added capability involves identifying the background noise environment in real-time and adapting a data-driven noise suppression approach to that noise environment on-the-fly. Various software optimization steps are taken in order to achieve a real-time throughput on the PDA platform involving both the speech decomposition and the adaptive noise suppression components. Real-time timing results and a quantitative measure of noise suppression are presented.

This paper presents a statistical image modeling approach based on Markov random field to synthesize cervical tissue second harmonic generation (SHG) images. Binary images representing fiber and pore areas of the cervix tissue are first obtained from SHG images using an image processing pipeline consisting of noise removal, contrast enhancement and optimal thresholding. These binary images are modeled using a Markov random field whose parameters are estimated via the least squares method. The parameters are then used to synthesize fiber and pore areas of cervical tissue in the form of binary images. The effectiveness of the synthesis is demonstrated by reporting the classification outcome for two classes of cervical SHG images collected from mice at two different stages of normal pregnancy. The developed synthesis allows generation of realistic fiber and pore area binary images for cervical tissue studies.

Rapid and efficient imaging of the brain to monitor brain activity and neural connectivity is performed through functional MRI and diffusion tensor imaging (DTI) using the Echo-planar imaging (EPI) sequence. An entire volume of the brain is imaged by EPI in a few seconds through the measurement of all k-space lines within one repetition time. However, this makes the sequence extremely sensitive to imperfections of magnetic field. In particular, the error caused by susceptibility induced magnetic field inhomogeneity accumulates over the duration of phase encoding, which in turn results in severe geometric distortion (warping) in EPI scans. EPI distortion correction through unwarping can be performed by field map based or image based techniques. However, due to the lack of ground truth it has been difficult to compare and validate different approaches. In this paper we propose a hybrid field map guided constrained deformable registration approach and compare it to field map based and image based unwarping approaches through a novel in-vivo validation framework which is based on the acquisition and alignment of EPI scans with different phase encoding directions. The quantitative evaluation results show that our hybrid approach of field map guided deformable registration to an undistorted T2-weighted image outperforms the other approaches.

This paper presents a wavelet-based speech coding strategy for cochlear implants. In addition, it describes the real-time implementation of this strategy on a personal digital assistant (PDA) platform. Three wavelet packet decomposition tree structures are considered and their performance in terms of computational complexity, spectral leakage, fixed-point accuracy, and real-time processing are compared to other commonly used strategies in cochlear implants. A real-time mechanism is introduced for updating the wavelet coefficients recursively. It is shown that the proposed strategy achieves higher analysis rates than the existing strategies while being able to run in real time on a PDA platform. In addition, it is shown that this strategy leads to a lower amount of spectral leakage. The PDA implementation is made interactive to allow users to easily manipulate the parameters involved and study their effects.

This paper presents an image processing system to distinguish a lipopolysaccharide (LPS) infection model of preterm labor from normal mouse pregnancy using Second Harmonic Generation (SHG) images of mouse cervix. Two classes of SHG images are considered: images from mice in which premature birth was caused by intrauterine LPS administration and images from normal pregnant mice. A wide collection of image texture features consisting of co-occurrence matrix-based, granulometry-based and wavelet-based are examined. The results obtained indicate that the combination of co-occurrence-based and granulometry-based textures features provides the most effective texture set for separating these two classes of images.

Active Noise Control (ANC) of fMRI acoustic noise using the conventional Filtered-X LMS (FXLMS) approach results in poor cancelation performance and slow convergence due to its broadband nature and the need for high order adaptive filters. High order adaptive filters are needed to effectively model the long acoustic impulse responses. Existing methods to improve the performance of FXLMS based broadband ANC systems are either computationally expensive or need elaborate implementation. In this paper we show a practical method to enhance the performance of FXLMS based algorithms, by deriving a crude estimate of the causalWiener filter and initializing the adaptive filter with the estimated Wiener filter. We observe that very fast convergence to the global minimum can be achieved along with huge gains in the noise cancelation performance. We call this method Wiener initialized FXLMS (WI-FXLMS).We show the effectiveness of the proposed approach for the active noise control of functional MRI acoustic noise and several other realistic noise sources.

Cochlear implant patients often complain about their difficulty in understanding speech in noisy environments. Currently a fixed noise suppression algorithm is used in cochlear implants regardless of the characteristics of the speech or noise environment. Access to an intelligent mechanism to determine the noise environment on-the-fly in order to automatically switch between different noise suppression algorithms in real-time can enhance patients experience with cochlear implants. In this paper, we report the first prototype system implementing such a real-time switching mechanism for automatic selection between two noise suppression algorithms designed for two commonly encountered noisy environments. The results obtained indicate the feasibility of this on-the-fly switching for actual deployment in cochlear implants.

Shannon entropy is a powerful tool in image analysis, but its reliable computation from image data faces an inherent dimensionality problem that calls for a low-dimensional and closed form model for the pixel value distributions. The most promising such models are Markovian, however, the conventional Markov random field is hampered by noncausality and its causal versions are also not free of difficulties. For example, the Markov mesh random field has its own limitations due to the strong diagonal dependency in its local neighboring system. A new model, named quadrilateral Markov random field (QMRF) is introduced in this paper in order to overcome these limitations. A property of QMRF with neighboring size of 2 is then used to decompose an image prior into a product of 2-D joint pdfs in which they are estimated using a joint histogram under the homogeneity assumption. In addition, the paper includes an extension of the introduced method to the computation of image spatial mutual information. Comparisons on synthesized images as well as two applications with real images are presented to motivate the developments in this paper and demonstrate the advantages in the performance of the introduced method over the existing ones.

The present paper describes the design of two stimulators (bench-top and portable) which can be used for animal studies in cochlear implants. The bench-top stimulator is controlled by a high-speed digital output board manufactured by National Instruments and is electrically isolated. The portable stimulator is controlled by a personal digital assistant (PDA) and is based on a custom interface board that communicates with the signal processor in the PDA through the secure digital IO (SDIO) slot. Both stimulators can provide 8 charge-balanced, bipolar channels of pulsatile and analog-like electrical stimulation, delivered simultaneously, interleaved or using a combination of both modes. Flexibility is provided into the construction of arbitrary, but charge-balanced, pulse shapes, which can be either symmetric or asymmetric.

This paper presents a set of validation procedures for nonrigid registration of functional EPI to anatomical MRI brain images. Although various registration techniques have been developed and validated for high-resolution anatomical MRI images, due to a lack of quantitative and qualitative validation procedures, the use of nonrigid registration between functional EPI and anatomical MRI images has not yet been deployed in neuroimaging studies. In this paper, the performance of a robust formulation of a nonrigid registration technique is evaluated in a quantitative manner based on simulated data and is further evaluated in a quantitative and qualitative manner based on in vivo data as compared to the commonly used rigid and affine registration techniques in the neuroimaging software packages. The nonrigid registration technique is formulated as a second-order constrained optimization problem using a free-form deformation model and mutual information similarity measure. Bound constraints, resolution level and cross-validation issues have been discussed to show the degree of accuracy and effectiveness of the nonrigid registration technique. The analyses performed reveal that the nonrigid approach provides a more accurate registration, in particular when the functional regions of interest lie in regions distorted by susceptibility artifacts.

Accurate segmentation of different brain tissues is of much importance in magnetic resonance imaging. This paper presents a comparison of the existing segmentation algorithms that are deployed in the neuroimaging community as part of two widely used software packages. The results obtained in this comparison can be used to select the appropriate segmentation algorithm for the neuroimaging application of interest. In addition to the entire brain area, a comparison is carried out for the subcortical region of the brain in terms of its gray matter composition.

Magnetic resonance field map images are normally used in characterizing the magnetic field inhomogeneity for distortion correction in Echo-Planar Imaging (EPI) and accurate localization in functional MRI (fMRI). In this paper, the computation and applications of an average field map image template is investigated based on real field maps. The introduced methodology and the obtained field map image templates may be used in EPI and fMRI image analysis, distortion correction, registration, and functional localization when high-resolution field map images are not available for individual datasets. The introduced methodology involves three stages of pre-processing, registration, and spatial normalization. The analysis and results presented in this paper show the impact and usefulness of the investigated methodology in several applications.

Functional localization is a concept which involves the application of a sequence of geometrical and statistical image processing operations in order to define the location of brain activity or to produce functional/parametric maps with respect to the brain structure or anatomy. Considering that functional brain images do not normally convey detailed structural information and, thus, do not present an anatomically specific localization of functional activity, various image registration techniques are introduced in the literature for the purpose of mapping functional activity into an anatomical image or a brain atlas. The problems addressed by these techniques differ depending on the application and the type of analysis, i.e., single-subject versus group analysis. Functional to anatomical brain image registration is the core part of functional localization in most applications and is accompanied by intersubject and subject-to-atlas registration for group analysis studies. Cortical surface registration and automatic brain labeling are some of the other tools towards establishing a fully automatic functional localization procedure. While several previous survey papers have reviewed and classified general-purpose medical image registration techniques, this paper provides an overview of brain functional localization along with a survey and classification of the image registration techniques related to this problem.

Data obtained from 102 skin lesions in vivo by spectroscopic oblique-incidence reflectometry were analyzed. The participating physicians initially divided the skin lesions into two visually distinguishable groups based on the lesions' melanocytic conditions. Group 1 consisted of the following two cancerous and benign subgroups: (1) basal cell carcinomas and squamous cell carcinomas and (2) benign actinic keratoses, seborrheic keratoses, and warts. Group 2 consisted of (1) dysplastic nevi and (2) benign common nevi. For each group, a bootstrap-based Bayes classifier was designed to separate the benign from the dysplastic or cancerous tissues. A genetic algorithm was then used to obtain the most effective combination of spatiospectral features for each classifier. The classifiers, tested with prospective blind studies, reached statistical accuracies of 100% and 95% for groups 1 and 2, respectively. Properties that related to cell-nuclear size, to the concentration of oxyhemoglobin, and to the concentration of deoxyhemoglobin as well as the derived concentration of total hemoglobin and oxygen saturation were defined to explain the origins of the classification outcomes.

This paper presents a photo-identification algorithm using Zernike moment invariants embedded in a subspace optimal for pattern identification. Fisher discriminants are used and the invariants are projected onto the subspace spanned by the Fisher basis vectors. The technique has been applied to photo-identification of gray whales (Eschrichtius robustus) using their field images. White patches (blotches) appearing on a gray whale's left and right flukes constitute unique identifying features and have been used here for individual identification. The fluke area is extracted from a fluke image via the live-wire edge detection algorithm, followed by optimal thresholding of the fluke area to obtain the blotches. Zernike moment invariants are then calculated for the blotches and projected onto the subspace spanned by Fisher basis vectors. These invariants are used as the feature vector representing a database image. During matching, the database images are ranked depending on the degree of similarity between a query and database feature vectors. The results show that the use of this algorithm leads to a significant reduction in the amount of manual search that is normally done by marine biologists.

This paper discusses a real-time digital signal processor (DSP)-based hierarchical neural network classifier capable of classifying both analog and digital modulation signals. A high-performance DSP processor, namely the TMS320C6701, is utilized to implement different kinds of classifiers including a hierarchical neural network classifier. A total of 31 statistical signal features are extracted and used to classify 11 modulation signals plus white noise. The modulation signals include CW, AM, FM, SSB, FSK2, FSK4, PSK2, PSK4, OOK, QAM16, and QAM32. A classification hierarchy is introduced and the genetic algorithm is employed to obtain the most effective set of features at each level of the hierarchy. The classification results and the number of operations on the DSP processor indicate the effectiveness of the introduced hierarchical neural network classifier in terms of both classification rate and processing time.

We discuss the use of a noninvasive in vivo optical technique, diffuse reflectance spectroscopic imaging with oblique incidence, to distinguish between benign and cancer-prone skin lesions. Various image features were examined to classify the images from lesions into benign and cancerous categories. Two groups of lesions were processed separately: Group 1 includes keratoses, warts versus carcinomas; and group 2 includes common nevi versus dysplastic nevi. A region search algorithm was developed to extract both one- and two-dimensional spectral information. A bootstrap-based Bayes classifier was used for classification. A computer-assisted tool was then devised to act as an electronic second opinion to the dermatologist. Our approach generated only one false-positive misclassification out of 23 cases collected for group 1 and two misclassifications out of 34 cases collected for group 2 under the worst estimation condition.

This paper presents a syntactic/semantic string representation scheme as well as a string matching method as part of a computer-assisted system to identify dolphins from photographs of their dorsal fins. A low-level string representation is constructed from the curvature function of a dolphin's fin trailing edge, consisting of positive and negative curvature primitives. A high-level string representation is then built over the low-level string via merging appropriate groupings of primitives in order to have a less sensitive representation to curvature fluctuations or noise. A family of syntactic/semantic distance measures between two strings is introduced. A composite distance measure is then defined and used as a dissimilarity measure for database search, highlighting both the syntax (structure or sequence) and semantic (attribute or feature) differences. The syntax consists of an ordered sequence of significant protrusions and intrusions on the edge, while the semantics consist of seven attributes extracted from the edge and its curvature function. The matching results are reported for a database of 624 images corresponding to 164 individual dolphins. The identification results indicate that the developed string matching method performs better than the previous matching methods including dorsal ratio, curvature, and curve matching. The developed computer-assisted system can help marine mammalogists in their identification of dolphins, since it allows them to examine only a handful of candidate images instead of the currently used manual searching of the entire database.

Marine biologists use a measurement called the "Dorsal Ratio" in the process of manual identification of bottlenose dolphins. The dorsal ratio denotes the relative distances of the two largest notches from the tip on the dorsal fin. The manual computation of this ratio is time consuming, labor intensive, and user dependent. This paper presents a computer-assisted system to extract the dorsal ratio for use in identification of individual animals. The first component of the system consists of active contour modeling where the trailing edge of the dorsal fin is detected. This is followed by a curvature module to find the characteristic fin points: tip and two most prominent notches. Curvature smoothing is performed at various smoothing scales, and wavelet coefficients are utilized to select an appropriate smoothing scale. The dorsal ratio is then computed from the curvature function at the appropriate smoothing scale. The system was tested using 296 digitized images of dolphins, representing 94 individual dolphins. The results obtained indicate that the computer extracted dorsal ratio can be used in place of the manually extracted dorsal ratio as part of the manual identification process.

We have investigated the possibility of using diffuse reflectance polarimetry to detect changes caused by different molecular compounds and concentrations in tissue-simulating phantoms. The effects of glucose, B-alanine and l-lysine at different concentrations in turbid media have been investigated separately. This approach is based on the effect of optical properties on the polarization state of light. The results show that this method has potential for determining changes in molecular concentrations in highly scattering biological media from polarization images.

An automatic, neural network-based approach was applied to segment normal brain compartments and lesions on MR images. Two supervised networks, backpropagation (BPN) and counterpropagation, and two unsupervised networks, Kohonen learning vector quantizer and analog adaptive resonance theory, were trained on registered T2-weighted and proton density images. The classes of interest were background, gray matter, white matter, cerebrospinal fluid, macrocystic encephalomalacia, gliosis, and "unknown." A comprehensive feature vector was chosen to discriminate these classes. The BPN combined with feature conditioning, multiple discriminant analysis followed by Hotelling transform, produced the most accurate and consistent classification results. Classification of normal brain compartments were generally in agreement with expert interpretation of the images. Macrocystic encephalomalacia and gliosis were recognized and, except around the periphery, classified in agreement with the clinician's report used to train the neural network.