Insu Song

Content-based classification of breath sound with enhanced features

Since breath sound (BS) contains important indicators of respiratory health and disease, analysis and detection of BS has become an important topic, with diagnostic and assessment of treatment capabilities. In this paper, the identification and classification of respiratory disorders based on the enhanced perceptual and cepstral feature set (PerCepD) is proposed. The hybrid PerCepD feature can capture the time-frequency characteristics of BS very well. Thus, it is very effective for the exploration and classification of normal and pathological BS related data. The classification models based on support vector machine (SVM) and artificial neural network (ANN) have been adopted to achieve automatic detection from BS data. The high detection accuracy results validate the performance of the proposed feature sets and classification model. The experimental results also demonstrate that the high accuracy of the pathological BS data can provide reliable diagnostic suggestions for breath disorders, such as flu, pneumonia and bronchitis.

Fast action recognition using negative space features

Due to the number of potential applications and their inherent complexity, automatic capture and analysis of actions have become an active research area. In this paper, an implicit method for recognizing actions in a video is proposed. Existing implicit methods work on the regions of subjects, but our proposed system works on the surrounding regions, called negative spaces, of the subjects. Extracting features from negative spaces facilitates the system to extract simple, yet effective features for describing actions. These negative-space based features are robust to deformed actions, such as complex boundary variations, partial occlusions, non-rigid deformations and small shadows. Unlike other implicit methods, our method does not require dimensionality reduction, thereby significantly improving the processing time. Further, we propose a new method to detect cycles of different actions automatically. In the proposed system, first, the input image sequence is background segmented and shadows are eliminated from the segmented images. Next, motion based features are computed for the sequence. Then, the negative space based description of each pose is obtained and the action descriptor is formed by combining the pose descriptors. Nearest Neighbor classifier is applied to recognize the action of the input sequence. The proposed system was evaluated on both publically available action datasets and a new fish action dataset for comparison, and showed improvement in both its accuracy and processing time. Moreover, the proposed system showed very good accuracy for corrupted image sequences, particularly in the case of noisy segmentation, and lower frame rate. Further, it has achieved highest accuracy with lowest processing time compared with the state-of-art methods.

A health social network recommender system

People with chronic health conditions require support beyond normal health care systems. Social networking has shown great potential to provide the needed support. Because of the privacy and security issues of health information systems, it is often difficult to find patients who can support each other in the community. We propose a social-networking framework for patient care, in particular for parents of children with Autism Spectrum Disorders (ASD). In the framework, health service providers facilitate social links between parents using similarities of assessment reports without revealing sensitive information. A machine learning approach was developed to generate explanations of ASD assessments in order to assist clinicians in their assessment. The generated explanations are then used to measure similarities between assessments in order to recommend a community of related parents. For the first time, we report on the accuracy of social linking using an explanation-based similarity measure.

Robust and secure watermarking scheme for breath sound

Due to the development of the Internet, security and intellectual property protection have attracted significant interest in the copyright protection field recently. A novel watermarking scheme for breath sounds, combining lifting wavelet transform (LWT), discrete cosine transform (DCT), singular value decomposition (SVD) and dither modulation (DM) quantization is proposed in this paper as a way to insert encrypted source and identity information in breath sounds while maintaining significant biological signals. In the proposed scheme, LWT is first performed to decompose the signal, and then DCT is applied on the approximate coefficients. SVD is carried out on the LWT-DCT coefficients to derive singular values. DM is adopted to quantize the singular values of each of the LWT-DCT blocks; thus, the watermark extraction is blind by using the DM algorithm. The novelty of our proposed method also includes the introduction of the particle swarm optimization (PSO) technique to optimize the quantization steps for the DM approach. The experimental results demonstrate that the proposed watermarking scheme obtains good robustness against common manipulation attacks and preserves imperceptivity. The performance comparison results verify that our scheme outperforms existing approaches in terms of robustness and imperceptibility.

Optimal watermarking scheme for breath sound

In this paper, a new watermarking scheme for breath sound based on lifting wavelet transform (LWT), discrete cosine transform (DCT), singular value decomposition (SVD) and dither modulation (DM) quantization is proposed to embed encrypted source and identity information, and medical conditions, such as cold and flu symptoms in breath sound while preserving important biological signals for detecting breathing patterns and breathing rates. In the proposed scheme, LWT is first carried out to decompose the signal followed by applying DCT on the approximate coefficients. SVD is then performed on the LWT-DCT coefficients to get the singular values. The novelty of our proposed method includes the introduction of the particle swarm optimization (PSO) technique to optimization the quantization steps of the DM approach too. Simulation results show that our watermarking scheme achieves good robustness against common signal processing attacks and maintains the imperceptivity. The comparison results also show good performance of our scheme.

Nested rules in defeasible logic

Defeasible Logic is a rule-based non-monotonic logic with tractable reasoning services. In this paper we extend Defeasible Logic with nested rules. We consider a new Defeasible Logic, called DLns, where we allow one level of nested rules. A nested rule is a rule where the antecedent or the consequent of the rule are rules themselves. The inference conditions for DLns are based on reflection on the inference structures (rules) of the particular theory at hand. Accordingly DLns can be considered an amalgamated reflective system with implicit reflection mechanism. Finally we outline some possible applications of the logic.

Affective Core-Banking Services for Microfinance

In this paper, we present a mobile phone-based banking system, called ACMB (Affective Cashless Mobile Banking), for microfinance. ACMB is designed to provide banking services to the world’s unbanked population of 2.2 billion. To enable interoperability between various microfinance institutions over a heterogeneous network of mobile devices, cell-phone networks and internet services, we define MSDL (Microfinance Service Definition Language) based on WSDL (Web Service Definition Language). MSDL includes a binding for SMS for service queries and utilization. To ensure that the banking service provides an acceptable level of usability and user experience, ACMB incorporates a wellbehaved service interface, which informs the design to create affective banking services based on human emotion models. ACMB was implemented on an Android tablet and evaluated with 147 participants, who performed 804 transactions and exchanged 2,412 SMS messages over a three hour testing period. The results suggest that an ACMB core-banking server on a low-cost mobile device can serve over 15,000 microfinance customers. Therefore, the systems appears to be suitable for most microfinance institutions.

Diagnosis of pneumonia from sounds collected using low cost cell phones

Respiratory diseases, such as pneumonia, cold, flu, and bronchitis, are still the leading causes of child mortality in the world. One solution for alleviating this problem is developing affordable respiratory-health assessment methods using computerized respiratory-sound analysis. This approach has become an active research area due to the recent developments of electronic recording devices, such as electronic stethoscopes. However, all existing methods require specialized equipment, which can be operated only by trained medical personals. We develop a low-cost cell phone-based rapid diagnosis method for respiratory health problems. A total of 367 breath sounds are collected from childrens hospitals in order to develop accurate diagnosis models and evaluation. An extensive analysis is performed on the breath sounds. Statistically significance features are selected for each age group using ANOVA from 1197 acoustic features. The model is evaluated on a binary classification task: pneumonia vs. non-pneumonia. The results showed that the proposed method was able to effectively classify pneumonia even in the presence of environmental noises. The method achieved 91.98% accuracy with 92.06% sensitivity and 90.68% specificity. The results indicate that breath sounds recorded using low-cost mobile devices can be used to detect pneumonia effectively.

Mental Health Informatics

This book introduces approaches that have the potential to transform the daily practice of psychiatrists and psychologists. This includes the asynchronous communication between mental health care providers and clients as well as the automation of assessment and therapy. Speech and language are particularly interesting from the viewpoint of psychological assessment. For instance, depression may change the characteristics of voice in individuals and these changes can be detected by a special form of speech analysis. Computational screening methods that utilize speech and language can detect subtle changes and alert clinicians as well as individuals and caregivers. The use of online technologies in mental health, however, poses ethical problems that will occupy concerned individuals, governments and the wider public for some time. Assuming that these ethical problems can be solved, it should be possible to diagnose and treat mental health disorders online (excluding the use of medication). Speech and language are particularly interesting from the viewpoint of psychological assessment. For instance, depression may change the characteristics of voice in individuals and these changes can be detected by a special form of speech analysis. Computational screening methods that utilize speech and language can detect subtle changes and alert clinicians as well as individuals and caregivers. The use of online technologies in mental health, however, poses ethical problems that will occupy concerned individuals, governments and the wider public for some time. Assuming that these ethical problems can be solved, it should be possible to diagnose and treat mental health disorders online (excluding the use of medication).

Gaussian Hamming Distance

We present new image features for diagnosing general wellbeing states and medical conditions. The new method, called Gaussian Hamming Distance (GHD), generates de-identified features that are highly correlated with general wellbeing states, such as happiness, smoking, and facial palsy. This method allows aid organizations and governments in developing countries to provide affordable medical services. We evaluate the new approach using real face-image data and four classifiers: Naive Bayesian classier, Artificial Neural Network, Decision Tree, and Support Vector Machines (SVM) for predicting general wellbeing states. Its predictive power (over 93 % accuracy) is suitable for providing a variety of online services including recommending useful health information for improving general wellbeing states.