Wireless accelerometry is feasible in acute monitoring of upper limb motor recovery after ischemic stroke.

BACKGROUND: Clinical deterioration in the acute stage of ischemic stroke powerfully predicts outcome and may serve as a marker for urgent intervention. However, accurate monitoring of acute stroke patients is hampered by the lack of validated continuous monitoring devices. We sought to assess the use of wireless accelerometry in this setting, hypothesizing that stroke patients would have a greater difference in movement between upper limbs than controls and that the magnitude of correlation between upper limb movements would be negatively associated with the National Institutes of Health Stroke Scale (NIHSS) score. METHODS: In this pilot study, 20 patients with acute ischemic stroke and unilateral upper limb weakness and 10 controls were recruited from a comprehensive stroke centre. All subjects were fitted with two 3-axis accelerometers and underwent 24 h of continuous accelerometry recording of upper limb movements and repeat NIHSS assessments. The intra-class correlation coefficient (ICC), assessing the similarity (or otherwise) of spontaneous movements in each arm was calculated. The association between NIHSS (total and motor subset scores) and the magnitude of ICC was estimated by Spearman's rank correlation, receiver-operating characteristic curve analysis was performed and the optimal diagnostic threshold value of ICC was calculated. RESULTS: The magnitude of the ICC was significantly associated with the baseline NIHSS score (p = 0.02) and non-significantly associated with the baseline NIHSS motor score (p = 0.08). At the optimal diagnostic threshold of ICC magnitude = 0.7, wireless accelerometry distinguished patients from controls with a sensitivity of 0.95, a specificity of 0.6 and a diagnostic odds ratio of 28.5. CONCLUSIONS: The wireless accelerometry system successfully detects a motor deficit in the setting of acute ischemic stroke, accurately differentiating patients from controls, and correlates well with the baseline NIHSS score. Its use is feasible in the acute stroke setting. Overall, it shows promise as a diagnostic tool to continuously monitor acute stroke patients but requires validation in a larger trial.

Time-frequency mapping of the rhythmic limb movements distinguishes convulsive epileptic from psychogenic nonepileptic seizures.

PURPOSE: A definite diagnosis of psychogenic nonepileptic seizures (PNES) usually requires in-patient video-electroencephalography (EEG) monitoring. Previous research has shown that convulsive psychogenic nonepileptic seizures (PNES) demonstrate a characteristic pattern of rhythmic movement artifact on the EEG. Herein we sought to examine the potential for time-frequency mapping of data from a movement-recording device (accelerometer) worn on the wrist as a diagnostic tool to differentiate between convulsive epileptic seizures and PNES. METHODS: Time-frequency mapping was performed on accelerometer traces obtained during 56 convulsive seizure-like events from 35 patients recorded during in-patient video-EEG monitoring. Twenty-six patients had PNES, eight had epileptic seizures, and one had both seizure types. The time-frequency maps were derived from fast Fourier transformations to determine the dominant frequency for sequential 2.56-s blocks for the course of each event. KEY FINDINGS: The coefficient of variation (CoV) of limb movement frequency for the PNES events was less than for the epileptic seizure events (median, 17.18% vs. 52.23%; p < 0.001). A blinded review of the time-frequency maps by an epileptologist was accurate in differentiating between the event types, that is, 38 (92.7%) of 41 and 6 (75%) of 8 nonepileptic and epileptic seizures, respectively, were diagnosed correctly, with seven events classified as "nondiagnostic." Using a CoV cutoff score of 32% resulted in similar classification accuracy, with 42 (93%) of 45 PNES and 10 (91%) of 11 epileptic seizure events correctly diagnosed. SIGNIFICANCE: Time-frequency analysis of data from a wristband movement monitor could be utilized as a diagnostic tool to differentiate between epileptic and nonepileptic convulsive seizure-like events.

Motor recovery monitoring using acceleration measurements in post acute stroke patients.

BACKGROUND: Stroke is one of the major causes of morbidity and mortality. Its recovery and treatment depends on close clinical monitoring by a clinician especially during the first few hours after the onset of stroke. Patients who do not exhibit early motor recovery post thrombolysis may benefit from more aggressive treatment. METHOD: A novel approach for monitoring stroke during the first few hours after the onset of stroke using a wireless accelerometer based motor activity monitoring system is developed. It monitors the motor activity by measuring the acceleration of the arms in three axes. In the presented proof of concept study, the measured acceleration data is transferred wirelessly using iMote2 platform to the base station that is equipped with an online algorithm capable of calculating an index equivalent to the National Institute of Health Stroke Score (NIHSS) motor index. The system is developed by collecting data from 15 patients. RESULTS: We have successfully demonstrated an end-to-end stroke monitoring system reporting an accuracy of calculating stroke index of more than 80%, highest Cohen's overall agreement of 0.91 (with excellent κ coefficient of 0.76). CONCLUSION: A wireless accelerometer based 'hot stroke' monitoring system is developed to monitor the motor recovery in acute-stroke patients. It has been shown to monitor stroke patients continuously, which has not been possible so far with high reliability.

Classification of sleep apnea types using wavelet packet analysis of short-term ECG signals.

OBJECTIVE: Obstructive sleep apnea (OSA) causes a pause in airflow with reduced breathing effort. In contrast, central sleep apnea (CSA) event is not accompanied with breathing effort. The aim of this study is to differentiate CSA and OSA events using wavelet packet analysis and support vector machines of ECG signals over 5 s period. METHODS: Eight level wavelet packet analysis was performed on each 5 s clip using Daubechies (DB3) mother wavelet and for comparison discrete wavelet analysis was performed using Symlet (SYM3) wavelets. The choice of wavelet basis function was based on a grid search using Daubechies, Symlet and biorthogonal wavelets with decomposition levels varying between 2 and 5. Support vector machine is used for two-class classification. Out of 29 overnight polysomnographic studies, 23 of them were used in the training phase and 6 patients were used for independent testing. RESULTS: The proposed algorithm is shown to perform better in classifying CSA and OSA with wavelet packet features (accuracy-91%, sensitivity-88.14% and specificity-91.11%) than with the traditional wavelet decomposition based features (accuracy-83.79%, sensitivity-89.18% and specificity-83.59%). The independent test resulted in overall classification accuracy, sensitivity and specificity of 91.08, 91.02 and 91.09% respectively using wavelet packet analysis. CONCLUSIONS: The classification result indicates the possibility of non-invasively classifying CSA and OSA events based on shorter segments of ECG signals.

Surface Potential Simulation and Electrode Design for in-Ear EEG Measurement.

The need for everyday-real-time EEG sensing has resulted in the development of minimalistic and discreet wearable EEG measuring devices. A front runner in this race is in-ear worn device. However, the position of the ear as well as its restricted accessibility poses certain challenges in the design of such devices - from the type of materials used to the placement of electrodes. These choices are important as they will determine the quality of the EEG signal available and in turn the accuracy of the application. We therefore create a simulation model of the human ear, allowing us to understand the impact of these choices on our design of an In-Ear EEG wearable. We first study the signal acquisition characteristics of a proposed gold-plated electrode against two other state-of-the-art electrode materials for in-ear EEG data collection. We further validate this electrode choice by fabricating a personalized silicone-based earpiece and collecting in-situ EEG data. This work explores the properties of using gold plated electrodes in capturing in-ear EEG signals enabling unobtrusive collection of the brain physiology data in real world setting.

Forward Model and Deep Learning Based Iterative Deconvolution for Robust Dynamic CT Perfusion.

Perfusion maps obtained from low-dose computed tomography (CT) data suffer from poor signal to noise ratio. To enhance the quality of the perfusion maps, several works rely on denoising the low-dose CT (LD-CT) images followed by conventional regularized deconvolution. Recent works employ deep neural networks (DNN) for learning a direct mapping between the noisy and the clean perfusion maps ignoring the convolution-based forward model. DNN-based methods are not robust to practical variations in the data that are seen in real-world applications such as stroke. In this work, we propose an iterative framework that combines the perfusion forward model with a DNN-based regularizer to obtain perfusion maps directly from the LD-CT dynamic data. To improve the robustness of the DNN, we leverage the anatomical information from the contrast-enhanced LD-CT images to learn the mapping between low-dose and standard-dose perfusion maps. Through empirical experiments, we show that our model is robust both qualitatively and quantitatively to practical perturbations in the data.

Anatomical Landmark Detection using Deep Appearance-Context Network.

Accurate identification of anatomical landmarks is a crucial step in medical image analysis. While deep neural networks have shown impressive performance on computer vision tasks, they rely on a large amount of data, which is often not available. In this work, we propose an attention-driven end-to-end deep learning architecture, which learns the local appearance and global context separately that helps in stable training under limited data. The experiments conducted demonstrate the effectiveness of the proposed approach with impressive results in localizing landmarks when evaluated on cephalometric and spine X-ray image data. The predicted landmarks are further utilized in biomedical applications to demonstrate the impact.

Single feature spatio-temporal architecture for EEG Based cognitive load assessment.

The study of electroencephalography (EEG) data for cognitive load analysis plays an important role in identification of stress-inducing tasks. This can be useful in applications such as optimal work allocation, increasing efficiency in the workplace and ensuring safety in difficult work environments. In order for such systems to be realistically deployable, easy acquisition and processing of the data on a wearable device is imperative. Current techniques primarily perform offline processing to analyse a multi-channel EEG to make a post facto assessment. This work focusses on building a new deep learning architecture that performs a single feature based spatio-temporal analysis of EEG data. This is achieved by creating a brain topographic map based on a single feature followed by spatio-temporal analysis using the developed network architecture. Data from two cognitive load experiments on the Physionet EEGMAT dataset were used to validate the performance. The network achieves an accuracy of 98.3% which is better than similar state-of-the-art approaches. Moreover, the proposed approach facilitates analysis of the spatial propagation of a signal, which is not possible through conventional EEG signal representations.

Analyzing temporal variability of standard descriptors of Poincaré plots.

The Poincaré map is a visual technique to recognize the hidden correlation patterns of a time series signal. The standard descriptors of the Poincaré map are used to quantify the plot that measures the gross variability of the time series data. However, the problem lies in capturing temporal information of the plot quantitatively. In this article, we propose a new formulation for calculating the standard descriptors SD1 and SD2 from localized measures SD1^(w) and SD2^(w). To justify the importance of the temporal measure, SD1^(w), SD2^(w) are calculated for the 2 case studies (normal sinus rhythm [NSR] vs congestive heart failure and NSR vs arrhythmia) and are compared with the performance using the overall measures (SD1, SD2). Using overall SD1, receiver operating characteristic areas of 0.72 and 0.86 were obtained for NSR vs congestive heart failure and NSR vs arrhythmia, and using the proposed method resulted in 0.82 and 0.89. Because we have shown that the overall SD1 and SD2 are functions of the respective localized measures SD1^(w) and SD2^(w), we can conclude that use of localized measure provides equal or higher performance in pathology detection compared with the overall SD1 or SD2.

Surgical tool segmentation and localization using spatio-temporal deep network.

Laparoscopic cholecystectomy surgery is a minimally invasive surgery to remove the gallbladder, where surgical instruments are inserted through small incisions in the abdomen with the help of a laparoscope. Identification of tool presence and precise segmentation of tools from the video is very important in understanding the quality of the surgery and training budding surgeons. Precise segmentation of tools is required to track the tools during real-time surgeries. In this paper, a new pixel-wise instance segmentation algorithm is proposed, which segments and localizes the surgical tool using spatio-temporal deep network. The performance of the proposed has been compared with the state-of-the-art image-based instance segmentation method using the Cholec80 dataset. It is also compared with methods in the literature using frame-level presence detection and spatial detection with good results.

Complex correlation measure: a novel descriptor for Poincaré plot.

BACKGROUND: Poincaré plot is one of the important techniques used for visually representing the heart rate variability. It is valuable due to its ability to display nonlinear aspects of the data sequence. However, the problem lies in capturing temporal information of the plot quantitatively. The standard descriptors used in quantifying the Poincaré plot (SD1, SD2) measure the gross variability of the time series data. Determination of advanced methods for capturing temporal properties pose a significant challenge. In this paper, we propose a novel descriptor "Complex Correlation Measure (CCM)" to quantify the temporal aspect of the Poincaré plot. In contrast to SD1 and SD2, the CCM incorporates point-to-point variation of the signal. METHODS: First, we have derived expressions for CCM. Then the sensitivity of descriptors has been shown by measuring all descriptors before and after surrogation of the signal. For each case study, lag-1 Poincaré plots were constructed for three groups of subjects (Arrhythmia, Congestive Heart Failure (CHF) and those with Normal Sinus Rhythm (NSR)), and the new measure CCM was computed along with SD1 and SD2. ANOVA analysis distribution was used to define the level of significance of mean and variance of SD1, SD2 and CCM for different groups of subjects. RESULTS: CCM is defined based on the autocorrelation at different lags of the time series, hence giving an in depth measurement of the correlation structure of the Poincaré plot. A surrogate analysis was performed, and the sensitivity of the proposed descriptor was found to be higher as compared to the standard descriptors. Two case studies were conducted for recognizing arrhythmia and congestive heart failure (CHF) subjects from those with NSR, using the Physionet database and demonstrated the usefulness of the proposed descriptors in biomedical applications. CCM was found to be a more significant (p = 6.28E-18) parameter than SD1 and SD2 in discriminating arrhythmia from NSR subjects. In case of assessing CHF subjects also against NSR, CCM was again found to be the most significant (p = 9.07E-14). CONCLUSION: Hence, CCM can be used as an additional Poincaré plot descriptor to detect pathology.

A hybrid approach for nucleus stain separation in histopathological images.

Difficulties in automation of histology image analysis are caused due to varying stain colors in the histology slides and the interaction of stains. Incorrect stain separation results in incorrect nucleus segmentation. A new hybrid algorithm has been proposed combining de-staining and wedge separation algorithms, which provides better stain separation and maintains color integrity of the input image. The proposed algorithm is tested on 36 histopathological images covering varying tissues and compared with popular methods in the area with excellent results in high nuclei density category.

Cloud solution for histopathological image analysis using region of interest based compression.

Recent technological gains have led to the adoption of innovative cloud based solutions in medical imaging field. Once the medical image is acquired, it can be viewed, modified, annotated and shared on many devices. This advancement is mainly due to the introduction of Cloud computing in medical domain. Tissue pathology images are complex and are normally collected at different focal lengths using a microscope. The single whole slide image contains many multi resolution images stored in a pyramidal structure with the highest resolution image at the base and the smallest thumbnail image at the top of the pyramid. Highest resolution image will be used for tissue pathology diagnosis and analysis. Transferring and storing such huge images is a big challenge. Compression is a very useful and effective technique to reduce the size of these images. As pathology images are used for diagnosis, no information can be lost during compression (lossless compression). A novel method of extracting the tissue region and applying lossless compression on this region and lossy compression on the empty regions has been proposed in this paper. The resulting compression ratio along with lossless compression on tissue region is in acceptable range allowing efficient storage and transmission to and from the Cloud.

Protein topology classification using two-stage support vector machines.

The determination of the first 3-D model of a protein from its sequence alone is a non-trivial problem. The first 3-D model is the key to the molecular replacement method of solving phase problem in x-ray crystallography. If the sequence identity is more than 30%, homology modelling can be used to determine the correct topology (as defined by CATH) or fold (as defined by SCOP). If the sequence identity is less than 25%, however, the task is very challenging. In this paper we address the topology classification of proteins with sequence identity of less than 25%. The input information to the system is amino acid sequence, the predicted secondary structure and the predicted real value relative solvent accessibility. A two stage support vector machine (SVM) approach is proposed for classifying the sequences to three different structural classes (alpha, beta, alpha+beta) in the first stage and 39 topologies in the second stage. The method is evaluated using a newly curated dataset from CATH with maximum pairwise sequence identity less than 25%. An impressive overall accuracy of 87.44% and 83.15% is reported for class and topology prediction, respectively. In the class prediction stage, a sensitivity of 0.77 and a specificity of 0.91 is obtained. Data file, SVM implementation (SVMHEAVY) and result files can be downloaded from http://www.ee.unimelb.edu.au/ISSNIP/downloads/.

Crowd Event Detection on Optical Flow Manifolds.

Analyzing crowd events in a video is key to understanding the behavioral characteristics of people (humans). Detecting crowd events in videos is challenging because of articulated human movements and occlusions. The aim of this paper is to detect the events in a probabilistic framework for automatically interpreting the visual crowd behavior. In this paper, crowd event detection and classification in optical flow manifolds (OFMs) are addressed. A new algorithm to detect walking and running events has been proposed, which uses optical flow vector lengths in OFMs. Furthermore, a new algorithm to detect merging and splitting events has been proposed, which uses Riemannian connections in the optical flow bundle (OFB). The longest vector from the OFB provides a key feature for distinguishing walking and running events. Using a Riemannian connection, the optical flow vectors are parallel transported to localize the crowd groups. The geodesic lengths among the groups provide a criterion for merging and splitting events. Dispersion and evacuation events are jointly modeled from the walking/running and merging/splitting events. Our results show that the proposed approach delivers a comparable model to detect crowd events. Using the performance evaluation of tracking and surveillance 2009 dataset, the proposed method is shown to produce the best results in merging, splitting, and dispersion events, and comparable results in walking, running, and evacuation events when compared with other methods.

Automatic Detection and Classification of Convulsive Psychogenic Nonepileptic Seizures Using a Wearable Device.

Epilepsy is one of the most common neurological disorders and patients suffer from unprovoked seizures. In contrast, psychogenic nonepileptic seizures (PNES) are another class of seizures that are involuntary events not caused by abnormal electrical discharges but are a manifestation of psychological distress. The similarity of these two types of seizures poses diagnostic challenges that often leads in delayed diagnosis of PNES. Further, the diagnosis of PNES involves high-cost hospital admission and monitoring using video-electroencephalogram machines. A wearable device that can monitor the patient in natural setting is a desired solution for diagnosis of convulsive PNES. A wearable device with an accelerometer sensor is proposed as a new solution in the detection and diagnosis of PNES. The seizure detection algorithm and PNES classification algorithm are developed. The developed algorithms are tested on data collected from convulsive epileptic patients. A very high seizure detection rate is achieved with 100% sensitivity and few false alarms. A leave-one-out error of 6.67% is achieved in PNES classification, demonstrating the usefulness of wearable device in the diagnosis of PNES.

Automatic segmentation of the rima glottidis in 4D laryngeal CT scans in Parkinson's disease.

Parkinson's disease (PD) is a progressive, incurable neuro-degenerative disease. Symptoms appear when approximately 70% of mid-brain dopaminergic neurons have died. Temporal analysis of the calculated area of the rima glottidis may give an indication of vocal impairment. In this paper, we present an automatic segmentation algorithm to segment the rima glottidis from 4D CT images using texture features and support vector machines (SVM). Automatic two dimensional region growing is then applied as a post processing step to segment the area accurately. The proposed segmentation algorithm resulted in accurate segmentation and we demonstrate a high correlation between the manually segmented area and automatic segmentation.

Estimation of vocal fold plane in 3D CT images for diagnosis of vocal fold abnormalities.

Vocal folds are the key body structures that are responsible for phonation and regulating air movement into and out of lungs. Various vocal fold disorders may seriously impact the quality of life. When diagnosing vocal fold disorders, CT of the neck is the commonly used imaging method. However, vocal folds do not align with the normal axial plane of a neck and the plane containing vocal cords and arytenoids does vary during phonation. It is therefore important to generate an algorithm for detecting the actual plane containing vocal folds. In this paper, we propose a method to automatically estimate the vocal fold plane using vertebral column and anterior commissure localization. Gray-level thresholding, connected component analysis, rule based segmentation and unsupervised k-means clustering were used in the proposed algorithm. The anterior commissure segmentation method achieved an accuracy of 85%, a good estimate of the expert assessment.

Classification of convulsive psychogenic non-epileptic seizures using muscle transforms obtained from accelerometry signal.

Convulsive psychogenic non-epileptic seizure (PNES) can be characterized as events which mimics epileptic seizures but do not show any characteristic changes on electroencephalogram (EEG). Correct diagnosis requires video-electroencephalography monitoring (VEM) as the diagnosis of PNES is extremely difficult in primary health care. Recent work has demonstrated the usefulness of accelerometry signal taken during a seizure in classification of PNES. In this work, a new direction has been explored to understand the role of different muscles in PNES. This is achieved by modeling the muscle activity of ten different upper limb muscles as a resultant function of accelerometer signal. Using these models, the accelerometer signals recorded from convulsive epileptic patients were transformed into individual muscle components. Based on this, an automated algorithm for classification of convulsive PNES is proposed. The algorithm calculates four wavelet domain features based on signal power, approximate entropy, kurtosis and signal skewness. These features were then used to build a classification model using support vector machines (SVM) classifier. It was found that the transforms corresponding to anterior deltoid and brachioradialis results in good PNES classification accuracy. The algorithm showed a high sensitivity of 93.33% and an overall PNES classification accuracy of 89% with the transform corresponding to anterior deltoid.

Classification of convulsive psychogenic non-epileptic seizures using histogram of oriented motion of accelerometry signals.

A seizure is caused due to sudden surge of electrical activity within the brain. There is another class of seizures called psychogenic non-epileptic seizure (PNES) that mimics epilepsy, but is caused due to underlying psychology. The diagnosis of PNES is done using video-electroencephalography monitoring (VEM), which is a resource intensive process. Recently, accelerometers have been shown to be effective in classification of epileptic and non-epileptic seizures. In this work, we propose a novel feature called histogram of oriented motion (HOOM) extracted from accelerometer signals for classification of convulsive PNES. An automated algorithm based on HOOM is proposed. The algorithm showed a high sensitivity of (93.33%) and an overall accuracy of (80%) in classifying convulsive PNES.