Bing Nan Li

Speech Emotion Recognition Using Fourier Parameters

Recently, studies have been performed on harmony features for speech emotion recognition. It is found in our study that the first- and second-order differences of harmony features also play an important role in speech emotion recognition. Therefore, we propose a new Fourier parameter model using the perceptual content of voice quality and the first- and second-order differences for speaker-independent speech emotion recognition. Experimental results show that the proposed Fourier parameter (FP) features are effective in identifying various emotional states in speech signals. They improve the recognition rates over the methods using Mel frequency cepstral coefficient (MFCC) features by 16.2, 6.8 and 16.6 points on the German database (EMODB), Chinese language database (CASIA) and Chinese elderly emotion database (EESDB). In particular, when combining FP with MFCC, the recognition rates can be further improved on the aforementioned databases by 17.5, 10 and 10.5 points, respectively.

Evaluation of robust wave image processing methods for magnetic resonance elastography

Magnetic resonance elastography (MRE) is a promising modality for in vivo quantification and visualization of soft tissue elasticity. It involves three stages of processes for (1) external excitation, (2) wave imaging and (3) elasticity reconstruction. One of the important issues to be addressed in MRE is wave image processing and enhancement. In this study we approach it from three different ways including phase unwrapping, directional filtering and noise suppression. The relevant solutions were addressed briefly. Some of them were implemented and evaluated on both simulated and experimental MRE datasets. The results confirm that wave image enhancement is indispensable before carrying out MRE elasticity reconstruction.

Quantitative analysis of patients with celiac disease by video capsule endoscopy: A deep learning method☆

BACKGROUNDnCeliac disease is one of the most common diseases in the world. Capsule endoscopy is an alternative way to visualize the entire small intestine without invasiveness to the patient. It is useful to characterize celiac disease, but hours are need to manually analyze the retrospective data of a single patient. Computer-aided quantitative analysis by a deep learning method helps in alleviating the workload during analysis of the retrospective videos.nnnMETHODnCapsule endoscopy clips from 6 celiac disease patients and 5 controls were preprocessed for training. The frames with a large field of opaque extraluminal fluid or air bubbles were removed automatically by using a pre-selection algorithm. Then the frames were cropped and the intensity was corrected prior to frame rotation in the proposed new method. The GoogLeNet is trained with these frames. Then, the clips of capsule endoscopy from 5 additional celiac disease patients and 5 additional control patients are used for testing. The trained GoogLeNet was able to distinguish the frames from capsule endoscopy clips of celiac disease patients vs controls. Quantitative measurement with evaluation of the confidence was developed to assess the severity level of pathology in the subjects.nnnRESULTSnRelying on the evaluation confidence, the GoogLeNet achieved 100% sensitivity and specificity for the testing set. The t-test confirmed the evaluation confidence is significant to distinguish celiac disease patients from controls. Furthermore, it is found that the evaluation confidence may also relate to the severity level of small bowel mucosal lesions.nnnCONCLUSIONSnA deep convolutional neural network was established for quantitative measurement of the existence and degree of pathology throughout the small intestine, which may improve computer-aided clinical techniques to assess mucosal atrophy and other etiologies in real-time with videocapsule endoscopy.

MREJ: MRE elasticity reconstruction on ImageJ

Magnetic resonance elastography (MRE) is a promising method for health evaluation and disease diagnosis. It makes use of elastic waves as a virtual probe to quantify soft tissue elasticity. The wave actuator, imaging modality and elasticity interpreter are all essential components for an MRE system. Efforts have been made to develop more effective actuating mechanisms, imaging protocols and reconstructing algorithms. However, translating MRE wave images into soft tissue elasticity is a nontrivial issue for health professionals. This study contributes an open-source platform - MREJ - for MRE image processing and elasticity reconstruction. It is established on the widespread image-processing program ImageJ. Two algorithms for elasticity reconstruction were implemented with spatiotemporal directional filtering. The usability of the method is shown through virtual palpation on different phantoms and patients. Based on the results, we conclude that MREJ offers the MRE community a convenient and well-functioning program for image processing and elasticity interpretation.

An integrative framework for 3D cobb angle measurement on CT images

OBJECTIVEnMeasuring the Cobb angle on computed tomography (CT) images remains a challenging but requisite task for clinical diagnoses of scoliosis. Traditionally, clinical practitioners resort to manual demarcation, but this approach is inefficient and subjective. Most of the existing computerized algorithms are two-dimensional (2D) and incapable of multi-angle calibration.nnnMETHODSnA novel integrative framework based on curvature features and geometric constraints is proposed to measure three-dimensional (3D)Cobb angles on CT images. This framework enables Cobb angle estimation in stereo and accomplishes the synchronous computation of the Cobb angle in three imaging planes. The whole system was quantitatively evaluated on 22 spine models obtained from the clinic.nnnRESULTSnThe results demonstrate that the integrative framework performs well in clinical Lenke classification and outperforms both the traditional manual method and the 2D digital method as evidenced by high intra-observer and inter-observer reliability (ICC>0.94, SEM 0.9°-1.2° for intra-observer, ICC>0.94, SEM 0.8°-1.2° for inter-observer). This 3D framework is also robust across different models (SE<3°).nnnCONCLUSIONSnThe new integrative framework is able to measure the Cobb angles in three imaging planes simultaneously and is therefore clinically advantageous.

Toward detection of aliases without string similarity

Entity aliases commonly exist. Accurately detecting these aliases plays a vital role in various applications. In particular, it is critical to detect the aliases that are intentionally hidden from the real identities, such as those of terrorists and frauds. Most existing work does not pay close attention to the aliases that have low/no string similarity to the given entities. In this paper, we propose a classifier that is based on active learning for detecting this type of aliasing. To minimize the cost of pair-wise comparison, a subset-based method is designed to restrict the selection within entity subsets. An active learning classifier is then employed in each entity subset to find the probability of whether a candidate is the alias of a given entity within the subset. After all of the results from the classifier are integrated, a list of aliases is returned for each given entity. For evaluation, we implemented four state-of-the-art methods and compared them with our proposed approach on three datasets. The results clearly demonstrate that this new active learning classifier is superior to those existing methods.

SenseViewer: A unified rendering interface of visual and haptic cues in medical images

We propose a new unified rendering interface - SenseViewer - that allows users to annotate and label visual, auditory and in particular haptic cues to medical images. This new interface may support pre-operation planning, surgical training and medical robot guidance better. Users can touch and feel virtual organs and tissues with haptics. The visual and auditory cues are helpful to define optimal paths for planning and guidance. SenseViewer is one of the earliest dedicated interfaces for rendering visual and haptic cues in medical images. Moreover, it employs magnetic resonance elastography for quantitative soft tissue viscoelasticity that makes haptic cues more realistic. SenseViewer will be further enhanced and integrated into our innovative Image-guided Robot Assisted Training (IRAS) system.

OpenMRE: A Numerical Platform for MRE Study

Magnetic resonance elastography (MRE) offers a noninvasive solution to visualize the mechanical properties of soft tissue, but the study suffers from expensive magnetic resonance scanning. Moreover, translating MRE wave images into soft tissue elasticity is a nontrivial issue for clinical professionals and healthcare practitioners. An interactive system—OpenMRE—is thus developed with the aid of ImageJ for numerical MRE study. It is comprised of two comparatively independent toolkits, namely MREA for simulation and MREP for interpretation. MREA mainly deals with the forward problem of MRE, and provides a numerical platform to determine the propagation and distribution of specially designed elastic wave. It is possible to numerically study some state-of-the-art paradigms including multisource and multifrequency MRE. The resultant wave images are interpretable in MREP that is designed for the inverse problem of MRE. It consists of the algorithms for phase unwrapping, directional filtering, and elasticity reconstruction. In a word, OpenMRE offers the MRE community a convenient and well-functioning system for interactive MRE study.

Liver Function Region Segmentation in Nuclear Medicine Using Mean Shift and Level Set Methods

Single photon emission computed tomography (SPECT) is a popular modality in clinic for liver function evaluation, but it is challenging for computerized segmentation and analysis of SPECT images. Most conventional technologies including mean shift and level set methods are not efficient due to weak contrast and ambiguous boundaries. We propose a new integrative model for liver function region segmentation in nuclear medicine. First, mean shift segmentation is improved by incorporating both space and color information. The enhanced mean shift is able to separate liver dysfunctional regions from their background. Second, the preparatory mean shift segmentation is employed to initialize a Hamilton-Jacobi level set model. It makes level set evolution nearby the objects of interest. Finally, we propose a new object indication function by considering the original nuclear medicine image as well as the relevant mean shift segmentation. This combinational indicator is effective to control level set segmentation. Otherwise, boundary leakage is inevitable. Experimental results on a set of clinical SPECT images confirm the effectiveness and robustness of this new integrative model.

Comprehensive assessment of gait signals using multiple time scale features

It is a challenging problem to detect and analyze gait signals for health evaluation. In this article, we propose a comprehensive assessment method using multiple time scale features to extract gait signal characteristics. Multi-resolution wavelet transform, together with logic regression and correlation analysis, was adapted for statistical analysis. The results show that the primary period and autocorrelation of gait signals vary substantially in three cohorts of people, namely normal young people, healthy old people and those with Parkinsons diseases. Furthermore, it is found that there is a correlation between the periodicity of gait sequences and the degree of Parkinsons diseases. In conclusion, these multiple scale features are very useful for health evaluation.