Vasiliki E. Kosmidou

Sign Language Recognition Using Intrinsic-Mode Sample Entropy on sEMG and Accelerometer Data

Sign language forms a communication channel among the deaf; however, automated gesture recognition could further expand their communication with the hearers. In this work, data from five-channel surface electromyogram and 3-D accelerometer from the signers dominant hand were analyzed using intrinsic-mode entropy (IMEn) for the automated recognition of Greek sign language (GSL) isolated signs. Discriminant analysis was used to identify the effective scales of the intrinsic-mode functions and the window length for the calculation of the IMEn that contributes to the efficient classification of the GSL signs. Experimental results from the IMEn analysis applied to GSL signs corresponding to 60-word lexicon repeated ten times by three native signers have shown more than 93% mean classification accuracy using IMEn as the only source of the classification feature set. This provides a promising bed-set toward the automated GSL gesture recognition.

Evaluation of surface EMG features for the recognition of American Sign Language gestures

In this work, analysis of the surface electromyogram (sEMG) signal is proposed for the recognition of American sign language (ASL) gestures. To this purpose, sixteen features are extracted from the sEMG signal acquired from the users forearm, and evaluated by the Mahalanobis distance criterion. Discriminant analysis is used to reduce the number of features used in the classification of the signed ASL gestures. The proposed features are tested against noise resulting in a further reduced set of features, which are evaluated for their discriminant ability. The classification results reveal that 97.7% of the inspected ASL gestures were correctly recognized using sEMG-based features, providing a promising solution to the automatic ASL gesture recognition problem

Brain source localization of MMN, P300 and N400: Aging and gender differences

The localization of neuronal generators during an ERP study, using a high-density electroencephalogram (HD-EEG) equipment was made on three Evoked Related Potential (ERP) components, i.e., the Mismatch Negativity (MMN), the P300 and the N400. Furthermore, the ERP characteristics, their field distribution and the area of their maximum field intensity were extracted and compared between young and elderly, as well as between females and males. A two tone oddball experiment was conducted, involving 27 young adults and 18 elderly, healthy and right handed, and HD-EEG data were acquired. These data were then subjected to auditory ERPs extraction and thorough statistical analysis. The derived experimental results revealed significant age-related differences to both the latencies and the amplitudes of the MMN and the P300 and the topographic distribution of the HD-EEG amplitudes. Additionally, a shift in the maximum intensities from frontal to temporal lobe with aging appeared in the case of the P300, whereas no effect was observed for the MMN component. No statistical significant differences (p>0.05) due to age was found in N400 characteristics. Finally, gender-related differences were significant in the response time of the subjects, finding males response faster. The level and the location of the maximum intensity of sources also differed between genders, especially in young subjects. These findings justify the enhanced potential of HD-EEG data to accurately reflect the age and gender dependencies at the three components of simple auditory ERPs and pave the way for the investigation of neurodegenerative pathologies, such as the Alzheimers disease.

Electroencephalogram and Alzheimer’s Disease: Clinical and Research Approaches

Alzheimers disease (AD) is a neurodegenerative disorder that is characterized by cognitive deficits, problems in activities of daily living, and behavioral disturbances. Electroencephalogram (EEG) has been demonstrated as a reliable tool in dementia research and diagnosis. The application of EEG in AD has a wide range of interest. EEG contributes to the differential diagnosis and the prognosis of the disease progression. Additionally such recordings can add important information related to the drug effectiveness. This review is prepared to form a knowledge platform for the project entitled “Cognitive Signal Processing Lab,” which is in progress in Information Technology Institute in Thessaloniki. The team tried to focus on the main research fields of AD via EEG and recent published studies.

Enhanced Sign Language Recognition Using Weighted Intrinsic-Mode Entropy and Signer's Level of Deafness

Sign language (SL) forms an important communication canal for the deaf. In this paper, enhanced SL recognition, by relating the individual way of signing with the signers level of deafness (LoD) through a novel hybrid adaptive weighting (HAW) process applied to surface electromyogram and 3-D accelerometer data, is proposed. Using a LoD-driven genetic algorithm, HAW optimally weights the intrinsic modes of the acquired signals, preparing them for sample entropy (SampEn) estimation that follows. The resulting feature set, namely, weighted intrinsic-mode entropy (IMEn) (wIMEn), aims at increasing the SL-sign-classification accuracy alone or boosted by signer identification and/or signers LoD-based group identification. The wIMEn was compared with three other feature sets, i.e., time frequency, SampEn, and IMEn, regarding their discrimination ability (both among signers and SL signs). Data from the dominant hand of nine subjects with various LoD were analyzed for the classification of 61 Greek SL (GSL) signs. Experimental results have shown that the introduced wIMEn feature set exhibited higher performance compared to others, both in signer identification and signers LoD-based group identification and in GSL sign classification. The findings suggest that LoD could be considered in the construction of a signer-independent SL-classification system toward the enhancement of its performance.

Using sample entropy for automated sign language recognition on sEMG and accelerometer data

Communication using sign language (SL) provides alternative means for information transmission among the deaf. Automated gesture recognition involved in SL, however, could further expand this communication channel to the world of hearers. In this study, data from five-channel surface electromyogram and three-dimensional accelerometer from signers’ dominant hand were subjected to a feature extraction process. The latter consisted of sample entropy (SampEn)-based analysis, whereas time-frequency feature (TFF) analysis was also performed as a baseline method for the automated recognition of 60-word lexicon Greek SL (GSL) isolated signs. Experimental results have shown a 66 and 92% mean classification accuracy threshold using TFF and SampEn, respectively. These results justify the superiority of SampEn against conventional methods, such as TFF, to provide with high recognition hit-ratios, combined with feature vector dimension reduction, toward a fast and reliable automated GSL gesture recognition.

Brain source localization of MMN and P300 ERPs in mild cognitive impairment and Alzheimer's disease: a high-density EEG approach

Alzheimers disease is the most common neurodegenerative disease of Western societies, suggesting the need for early diagnosis, even in preclinical stages. In this vein, the localization of neuronal generators of event-related potential (ERP) components, that is, the mismatch negativity and the P300, based on high-density electroencephalogram data, was explored as a means to enhance their sensitivity as markers of preclinical Alzheimers disease (AD). A 2-tone oddball experiment was conducted, involving 21 healthy elderly, 21 mild cognitive impairment, and 21 mild AD patients, while high-density electroencephalogram data were recorded. The results revealed longer latencies of both mismatch negativity and P300 and slower and far less accurate responses as neurodegeneration progressed. Standardized low-resolution electromagnetic tomography revealed that source differences between healthy and mild cognitive impairment and healthy and AD patients for both ERP components were present in the same Brodmann area independently of the ERP and the stage of cognitive decline. This finding indicates an early change of source activation related to cognitive performance and may be used to improve the diagnostic and prognostic value of ERPs.

Cognitive MMN and P300 in mild cognitive impairment and Alzheimer's disease: A high density EEG-3D vector field tomography approach.

Precise preclinical detection of dementia for effective treatment and stage monitoring is of great importance. Miscellaneous types of biomarkers, e.g., biochemical, genetic, neuroimaging, and physiological, have been proposed to diagnose Alzheimers disease (AD), the usual suspect behind manifested cognitive decline, and mild cognitive impairment (MCI), a neuropathology prior to AD that does not affect cognitive functions. Event related potential (ERP) methods constitute a non-invasive, inexpensive means of analysis and have been proposed as sensitive biomarkers of cognitive impairment; besides, various ERP components are strongly linked with working memory, attention, sensory processing and motor responses. In this study, an auditory oddball task is employed, to acquire high density electroencephalograhy recordings from healthy elderly controls, MCI and AD patients. The mismatch negativity (MMN) and P300 ERP components are then extracted and their relationship with neurodegeneration is examined. Then, the neural activation at these components is reconstructed using the 3D vector field tomography (3D-VFT) inverse solution. The results reveal a decline of both ERPs amplitude, and a statistically significant prolongation of their latency as cognitive impairment advances. For the MMN, higher brain activation is usually localized in the inferior frontal and superior temporal gyri in the controls. However, in AD, parietal sites exhibit strong activity. Stronger P300 generators are mostly found in the frontal lobe for the controls, but in AD they often shift to the temporal lobe. Reduction in inferior frontal source strength and the switch of the maximum intensity area to parietal and superior temporal sites suggest that these areas, especially the former, are of particular significance when neurodegenerative disorders are investigated. The modulation of MMN and P300 can serve to produce biomarkers of dementia and its progression, and brain imaging can further contribute to the diagnostic efficiency of ERPs.

Intrinsic mode entropy: An enhanced classification means for automated Greek Sign Language gesture recognition

Sign language forms a communication channel among the deaf; however, automated gesture recognition could further expand their communication with the hearers. In this work, data from three-dimensional accelerometer and five-channel surface electromyogram of the users dominant forearm are analyzed using intrinsic mode entropy (IMEn) for the automated recognition of Greek Sign Language (GSL) gestures. IMEn was estimated for various window lengths and evaluated by the Mahalanobis distance criterion. Discriminant analysis was used to identify the effective scales of the intrinsic mode functions and the window length for the calculation of the IMEn that contributes to the correct classification of the GSL gestures. Experimental results from the IMEn analysis of GSL gestures corresponding to ten words have shown 100% classification accuracy using IMEn as the only classification feature. This provides a promising bed-set towards the automated GSL gesture recognition.

Age Effect in Human Brain Responses to Emotion Arousing Images: The EEG 3D-Vector Field Tomography Modeling Approach

Understanding of the brain responses to emotional stimulation remains a great challenge. Studies on the aging effect in neural activation report controversial results. In this paper, pictures of two classes of facial affect, i.e., anger and fear, were presented to young and elderly participants. High-density 256-channel EEG data were recorded and an innovative methodology was used to map the activated brain state at the N170 event-related potential component. The methodology, namely 3D Vector Field Tomography, reconstructs the electrostatic field within the head volume and requires no prior modeling of the individuals brain. Results showed that the elderly exhibited greater N170 amplitudes, while age-based differences were also observed in the topographic distribution of the EEG recordings at the N170 component. The brain activation analysis was performed by adopting a set of regions of interest. Results on the maximum activation area appeared to be emotion-specific; the anger emotional conditions induced the maximum activation in the inferior frontal gyrus, while fear activated more the superior temporal gyrus. The approach used here shows the potential of the proposed computational model to reveal the age effect on the brain activation upon emotion arousing images, which could be further transferred to the design of assistive clinical applications.