Biomedical Engineering: Applications, Basis and Communications

The work aims at classifying EEG of mild cognitive impairment (MCI) patients from that of normal control (NC) subjects using recurrence quantification analysis (RQA) and a simple visual memory task, which is commonly used in memory clinics. EEG of MCI and NC groups are recorded under three cognitive conditions, resting eyes closed (EC) and two phases of the task, namely, picture viewing (learning phase, PIC) and picture recollection (immediate free recall phase, PICREC). Complexity analysis of EEG is performed using RQA measures, recurrence rate (RR) and entropy (ENTR). Mean values of these measures over electrodes from four cortical regions are used for statistical analysis of group differences, under the different cognitive conditions. In all the cortical regions, the mean RQA RR and ENTR values of MCI group are observed to be higher compared to NC group under the task conditions. Receiver operating characteristics (ROC) analysis is used for assessing the classification efficiency of the RQA-based method applied to EEG of MCI subjects. A fair classification is obtained in all the four cortical regions during the PIC condition using RR and in all regions except frontal, using ENTR. In the PICREC condition, a good classification is obtained in the temporal, parietal and occipital regions and a fair classification is attained in the frontal region using RR. In this condition, the ENTR values provided a fair classification in all the four cortical regions. These RQA measures are used as feature vectors of SVM classifier to further confirm the classification efficiency of the couplets of RQA RR and ENTR. These results indicate RQA method can efficiently classify MCI EEG based on complexity levels using the simple immediate free recall task.

Objective: Parkinson’s Disease (PD) is a neurodegenerative disease that is categorized by tremor, rigidity, and bradykinesia. Currently, there is no standard method to diagnose patients with PD. One of the common symptoms of PD is gait disorders which are caused by rigid muscles. Gait disorders may start some years before disease diagnosis. Therefore, better understanding of the gait signal can lead to early diagnosis of PD.

Methods: Computer-aided system has been useful in early detection of PD symptoms. In the present study, gait disturbances have received attention as potential biomarkers for early diagnosing of PD. Time and frequency analysis of gait signals together can provide more useful information. Wavelet-based features were extracted from stride, swing and double support time signals of healthy subjects and PD patients. These signals were decomposed into five levels using “sym4” wavelet. Mean and standard deviation (SD) of the absolute values of the approximation and detailed coefficients at each level were computed. Then final features were picked accordingly to obtain the best result for the classification.

Results: Support Vector Machine (SVM) was employed for classification of patients and healthy people. The classifier performance was measured based on accuracy, sensitivity and specificity. The classifier performance is obtained with 93.3% accuracy employing linear kernel.

Conclusions: The proposed system can be employed as a Decision Support Systems (DSSs) for early diagnosing of PD. Presenting DSSs can be employed to screen suspected cases of PD disease for further evaluation. Studying large number of patients and healthy subjects may lead to more precise study on PD. Also, it seems that using other different classifiers, along with our features, can reduce the diagnosis error.

According to WHO, 2.6 million babies die during pregnancy. Good monitoring during the prenatal period could provide a significant reduction of this mortality rate. This is possible by detection and extraction of the fetal electrocardiogram (FECG). Extraction of that information is complex due to other noise coming from the mother and within the fetus that drowns out the fetal heart signal. However, new technology and improved filtering technique have provided ways to more accurately and efficiently gather various electrical components regarding fetal heart condition. In this paper, we propose a new source separation filtering method exploiting linear and nonlinear filtering techniques. Our method is a non-invasive extraction technique, where the source signal is the cardiac electrical signal acquired by non-invasive electrodes to facilitate the collection of signals and reduce the cost of the acquisition system; it differs from other existing methods in minimizing the number of input signals and the simplicity of its implementation. The fetal heart signal is drowned out by the maternal electrocardiogram (MECG). The problem that arises is the exact knowledge of the MECG signal affecting the chosen measuring electrode, since the MECG is dependent on the position of the electrode and the type of tissue that goes through. Therefore, its knowledge can be made only by a mathematical estimation. A DWT decomposition with adaptive thresholding based on an LMS filter is applied to extract the fetal signal. So first we extract the QRS complex of the FECG and detect the fetal heart rate (FHR).

This work decodes two-class motor imagery (MI) based on four main processing steps: (i) Raw electroencephalographic (EEG) signal is decomposed to single trials and spatial filters are estimated for each trial by common spatial filtering (CSP) method; (ii) features are extracted by taking the log transformation (normal distribution) of the spatially filtered EEG signal; (iii) optimal channel selection algorithm is proposed to reduce the number of EEG channels, such approach is regarded as key technological advantage in the implementation of brain–computer interface (BCI) to reduce the system processing time; (iv) finally, support vector machine (SVM) is employed to discriminate two classes of left and right hand MI. Two variations of SVM were proposed: polynomial function kernel and radial-based function RBF kernel. The results revealed that CSP succeeded in removing the strong correlation bound between the EEG samples by maximizing the variance of class 2 samples while minimizing the variance of class 1 samples. The channel selection algorithm achieved its goal to reduce the data dimension by selecting two channels out of three having the lowest variance entropies of 0.239 and 0.261 for channel 1 and channel 2, respectively. The features vector was divided into 80% train and 20% test with five-fold cross validation. The classification performance of SVM-polynomial kernel was 87.86% while it is 95.72% for SVM-RBF kernel as average accuracy of five-folds for both. Thus SVM-RBF is superior to SVM-Poly in the proposed framework.

Wound healing is a slow process in diabetic patients due to levels of insulin variations in the body. Thus, we present an automated system to analyze and assess different stages of wound healing process of diabetic patients. The diabetic wound healing stages have been defined into three types, such as the level of tissues present in the wound: The percentage of granulations tissues, Necrotic tissues and Slough tissues present in the diabetic patients. The performance of the proposed method shall be assessed based on the clear accuracy of segmentation of wound region present in the patient body. The Decision Tree-based Segmentation method has yielded a good segmentation accuracy of 98.32% in comparison with the ground truth results of clinical data. Further, the assessment of wound healing stages of proposed method has given a good accuracy of measuring the stages of diabetic patients by measuring the percentage of different types of tissues present in the wound region. Based on the results of classification accuracy of the proposed method, we assess whether the wound is going to heal quickly or not. Thus, we presented an algorithm of Decision Trees for the purpose of segmentation and assessment of wound healing process of diabetic patients.

Due to the advancement in data collection and maintenance strategies, the current clinical databases around the globe are rich in a sense that these contain detailed information not only about the individual’s medical conditions, but also about the environmental features, associated with the individual. Classification within this data could provide new medical insights. Data mining technology has become an attraction for researchers due to its affectivity and efficacy in the field of biomedicine research. Due to the diverse structure of such data sets, only few successful techniques and easy to use softwares, are available in literature. A Bayesian analysis provides a more intuitive statement of probability that hypothesis is true. Bayesian approach uses all available information and can give answers to complex questions more accurately. This means that Bayesian methods include prior information. In Bayesian analysis, no relevant information is excluded as prior represents all the available information apart from data itself. Bayesian techniques are specifically used for decision making. Uncertainty is the main hurdle in making decisions. Due to lack of information about relevant parameters, there is uncertainty about given decision. Bayesian methods measure these uncertainties by using probability. In this study, selected techniques of biostatistical Bayesian inference (the probability based inferencing approach, to identify uncertainty in databases) are discussed. To show the efficiency of a Hybrid technique, its application on two distinct data sets is presented in a novel way.

The application of Brain Computer Interface (BCI) for rehabilitation purpose has gained wide popularity in recent times. BCI for rehabilitation involves detection of brain signals, when the subject performs some sort of Motor Imagery (MI) task, for example, imagination of movement of limbs. Imagination of such movement causes desynchronization of neurons of one part of the brain gets within other parts synchronized. Band power features are best suited for quantification of the synchronization phenomenon. In the present work, extreme learning machine (ELM) and support vector machine (SVM) based classifiers are used to classify the test data. The classifier output is further used to generate control signals for driving a stepper motor, which may be used to drive some neuro-aid application device. In order to achieve a workable model for pragmatic applications, it is necessary to design a robust in nature stepper motor. Open loop analysis, closed loop analysis and performance analysis of motor with possible disturbances are carried out to evaluate the effectiveness of the proposed work. The maximum accuracy using ELM and SVM classifiers are achieved as 90% and 87.78% with a training time of 0.2496s and 3.964s, respectively. In the open loop and closed loop analysis, the desired angular movement (task imagined for rehabilitation) is achieved with an accuracy of 54.14% and 93.4%, respectively. These results suggest that a BCI system can be designed with higher efficiency with the help of MI data.

Cognitive load and emotional states may impact the cognitive learning process. Detection of reliable emotions and cognitive load would benefit the design of the emotional intelligent mobility system for visually impaired peoples (VIPs). Application of learning process using electroencephalography (EEG) offers novel and promising approaches to measure cognitive load and emotional states. EEG is used to identify the physiological index that can lead to detecting cognitive load and emotions which can help to explore the knowledge of learning processes. Basically, EEG is a record of ongoing electrical signals to represent the human brain activity due to external and internal stimuli. Therefore, in this study EEG signals are captured from participants with nine different degrees of sight loss people. EEG signals are then used to measure various cognitive load and emotional states to evaluate cognitive learning process for the VIPs. To support the argument of cognitive learning process, the complexity of the tasks in terms of cognitive load and emotional states are quantified considering diverse factors by extracting features from various well-established metrics such as permutation entropy, event related synchronization/desynchronization, arousal, and valence when VIPs are navigating unfamiliar indoor environments. A classification accuracy of door is 86.67% which is achieved by the proposed model. It has almost 10% of improvement compared to another state-of the-art method who have used same dataset. Moreover, we have achieved 10% and 1% more accuracy in the corridor and open space conditions compared to the existing method. Experimental results also demonstrated that learning process is significantly improved considering wide range of obstacles when they are navigating indoor environments.