Myoelectric control systems—A survey

International Joint Conference on Neural Networks

https://www.repositorio.ufop.br/bitstream/123456789/7011/1/ARTIGO_ECGHeartBeat.pdf

ECG-based heartbeat classification for arrhythmia detection

A novel wavelet sequence based on deep bidirectional LSTM network model for ECG signal classification

User acceptance of myoelectric forearm prostheses is currently low. Awkward control, lack of feedback, and difficult training are cited as primary reasons. Recently, researchers have focused on exploiting the new possibilities offered by advancements in prosthetic technology. Alternatively, researchers could focus on prosthesis acceptance by developing functional requirements based on activities users are likely to perform. In this article, we describe the process of determining such requirements and then the application of these requirements to evaluating the state of the art in myoelectric forearm prosthesis research. As part of a needs assessment, a workshop was organized involving clinicians (representing end users), academics, and engineers. The resulting needs included an increased number of functions, lower reaction and execution times, and intuitiveness of both control and feedback systems. Reviewing the state of the art of research in the main prosthetic subsystems (electromyographic [EMG] sensing, control, and feedback) showed that modern research prototypes only partly fulfill the requirements. We found that focus should be on validating EMG-sensing results with patients, improving simultaneous control of wrist movements and grasps, deriving optimal parameters for force and position feedback, and taking into account the psychophysical aspects of feedback, such as intensity perception and spatial acuity.

/pdf/myoelectric-forearm-prostheses-state-of-the-art-from-a-user-3zwrxoh1nz.pdf

Myoelectric Forearm Prostheses: State of the Art from a User-Centered Perspective

Electrocardiogram (ECG) data compression algorithm is needed that will reduce the amount of data to be transmitted, stored and analyzed, but without losing the clinical information content. An example of application of hierarchical neural network structure is described for compression of ECG signals. Then results of this lossy compression method were compared with two efficient compression methods that are fractal based and wavelet based compressions.

/pdf/hierarchical-neural-network-based-compression-of-ecg-signals-4er8vh0r9k.pdf

Hierarchical neural network based compression of ECG signals

With the developing technology, the constraints of processor speed, memory and input/output of small mobile and wireless devices such as personal digital assistants-PDAs (such as Pocket PCs, Tablet PCs, and smart phones), and sensors, which are the main user devices of pervasive healthcare computing have began to decline. The applicability of the artificial neural networks algorithms that require CPU, memory and I/O intensive operations on small mobile devices is increased in medical area. This paper presents a novel approach for diagnosing diabetes illness using pervasive healthcare computing and artificial neural networks on small mobile and wireless devices.

Diagnosing diabetes illness using pervasive computing and artificial neural networks

The utilization of simulation in surgical resident assessment and training allows quantitation of technical skills in risk-free environments. This transformation of current training paradigms which involve the development and application of high-fidelity virtual reality simulators may play an important role in future surgical educational curricula. With the implementation of artificial intelligence (AI) virtual reality simulators have the potential to advance the understanding, training and assessment of psychomotor performance of surgical expertise. In this study, we apply four hybrid deep neural network algorithms called Fuzzy Clustering Neural Networks (FCNN-1, FCNN2, FCNN-3, and FCNN-4). The proposed study consists of four hybrid deep neural networks methods called Fuzzy Clustering Neural Network (FCNN) were employed to differentiate surgical expertise. Raw data was obtained from virtual reality tumor resection studies utilizing the NeuroVR simulator platform. The performance of neurosurgeons, senior residents, junior residents and medical students was assessed using a series of metrics. For two proposed algorithms (FCNN-3 and FCNN-4), we achieved a further improvement in classification accuracy Bekir Karlik, Recai Yilmaz, Alexander Winkler-Schwartz, Nykan Mirchi, Vincent Bissonnette, Nicole Ledwos & Rolando Del Maestro International Journal of Artificial Intelligence and Expert Systems (IJAE), Volume (10) : Issue (3) : 2021 48 ISSN: 2180-124X, https://www.cscjournals.org/journals/IJAE/description.php of 100% correctly classifying the level of expertise of all participants. The other 2 models were also better accuracies than previous studies. The results demonstrate that the four hybrid FCNN algorithms utilized have increased the accuracy of classification compared to previous studies utilizing the same dataset. 1. ІNTRОDUСTІОN Virtual reality technology is being increasingly implemented in many areas of medicine and surgery. The advantage of utilizing this technology is to provide safe simulated environments for healthcare trainees and professionals to develop their skillset and knowledge. In addition, surgical education can be enhanced by the use of virtual reality surgical simulators by allowing an objective assessment of performance and improved evaluation and training of skills. These studies have focused on defining novel metrics, classification of expertise and describing objective proficiency benchmarks (Fried et al., 2019; Gelinas-Phaneuf, and Del Maestro, 2013; Alotaibi et al., 2015). A series of machine learning algorithms have been employed to assess surgical skills performance on a variety of virtual reality simulators. Studies (Moorthy et al., 2003; Megali et al., 2006; Liang et al., 2011; Rhienmora et al., 2011) have used Hidden Markov Models (HMM) to differentiate expert surgical performances. Fuzzy classification methods have also been employed to assess bimanual skills and evaluate performance (Huang et al., 2005; Hajshirmohammadi and Payandeh, 2007). Jog et al. (2011) and Kerwin et al. (2012) have utilized decision tree algorithms to assess surgical simulator results. Jog et al also used Support Vector Machines (SVM) to compare their results with a decision tree algorithm. Loukas and Georgiou (2011) have presented 2 machine learning methods (SVM and HMM) for laparoscopic skills assessment of surgical trainees. Sewell et al. (2008) have proposed 3 different methods (HMM, Naïve Bayes, Logistic Regression) to assess performance of a surgical simulator. Richstone et al. (2010) have presented an Artificial Neural Network (ANN) method to assess surgical skill. Ershad et al. (2018) have used SVM to assess of robotic surgical performance. Wang and Fey (2019) have proposed a Convolutional Neural Network (CNN) for objective skill evaluation in robot-assisted surgery. Winkler-Schwartz et al. (2019) have assessed a series of machine learning algorisms used in virtual reality surgical studies to develop a series of best practice guidelines for these investigations. A number of studies completed by our group have explored the use of machine learning and artificial neural networks to develop novel metrics and to classify surgical expertise utilizing both cranial and spine simulators. The study by Winkler-Schwartz et al. (2019) on virtual reality brain tumor resection used different machine learning algorithms (KNearest Neighbor, Naive Bayes, Discriminant Analysis and SVM) obtaining a 90% accuracy in classifying participants into 4 groups of expertise utilizing a NeuroVR simulator platform. Siyar et al. (2018) utilized 4 machine learning algorithms to classify “skilled” and “novice” participants employing a different brain tumor resection model obtaining an average of 90% accuracy. Bissonnette et al. (2019) utilized SVM and achieved a 97.6% ability to classify level of expertise utilizing a virtual reality spinal task on the NeuroVR simulator. Ledwos et al. (2020) validated an anterior cervical discectomy and fusion (ACDF) scenario on a novel spine simulator developed by OSSimTech and this scenario has been utilized in a series of new studies. Mirchi et al. (2020) have employed ANN for assessment of the virtual reality anterior cervical discectomy and fusion performance obtaining a test accuracy of 83.3% classifying 3 groups of participants based on level of expertise. The artificial neural network allowed Bekir Karlik, Recai Yilmaz, Alexander Winkler-Schwartz, Nykan Mirchi, Vincent Bissonnette, Nicole Ledwos & Rolando Del Maestro International Journal of Artificial Intelligence and Expert Systems (IJAE), Volume (10) : Issue (3) : 2021 49 ISSN: 2180-124X, https://www.cscjournals.org/journals/IJAE/description.php for the differentiation of surgical expertise and provided insight into the relative importance of performance metrics. Mirchi et al. (2020) have proposed a virtual operative assistant as an explainable AI tool for simulation-based neurosurgical training employing SVM utilizing data from the study of Winkler-Schwartz et al. (2019) and they achieved a 92% accuracy in classifying “skilled” and “novice” groups. In this study, we apply four different types of hybrid deep neural network algorithms called Fuzzy Clustering Neural Networks (FCNN-1, FCNN-2, FCNN-3, and FCNN-4) to assess if these networks can further improve classification of neurosurgical expertise using similar data as employed by Winkler-Schwartz et al. (2019). 2. FUZZY CLUSTERİNG NEURAL NETWORK As seen in Fig. 1, Fuzzy Clustering Neural Network (FCNN) as hybrid learning method integrates both unsupervised Fuzzy C-means clustering and supervised artificial neural network. FCNN was defined and used by Karlik et al. (2002, 2003) and Ceylan et al. (2009, 2011) have improved this model and these methods have been named type-2 FCNN. Clustering is the process of grouping into clusters in a dataset according to an appropriate similarity range for the degree of similarity so that the data points in a cluster are strong and weak in different clusters. Similarities of the data in cluster analysis are found by determining the distances from each other. For this, different distance methods are used. The most commonly used distance methods are Euclid, Manhattan and Minkowski Distance. The most popular clustering algorithms are k-means and Fuzzy CMeans. Fuzzy clustering points out different grades of membership from different clusters. As seen in Fig. 1, the fuzzy self-organizing layer is responsible for the clustering of the input data. The neural network is responsible for the classifier of obtained new training set according to target values. Here target values represent four levels of neurosurgical expertise. Karlik (2016) has proposed the positive effects of Fuzzy C-Mean Clustering on different supervised machine learning algorithms. FIGURE 1: The block diagram of hybrid Fuzzy Clustering Neural Networks (FCNN). The FCM algorithm calculates a suitable set of clusters in a limited number of cycles as an effective unsupervised algorithm (Karlik et al., 2009). It is important to designate different classes of membership from different clusters for FCM. Let X= {x1, x2,...,xn}, xi ∈ R become the utilizing dataset. The FCM algorithm partitions the utilizing dataset into fuzzy groups (c) and computes cluster centers as V={v1,v2,.....,vc} ⊂ R so that it can be minimized by Bekir Karlik, Recai Yilmaz, Alexander Winkler-Schwartz, Nykan Mirchi, Vincent Bissonnette, Nicole Ledwos & Rolando Del Maestro International Journal of Artificial Intelligence and Expert Systems (IJAE), Volume (10) : Issue (3) : 2021 50 ISSN: 2180-124X, https://www.cscjournals.org/journals/IJAE/description.php J= (1) where, uij ∈ [0, 1] explains the degree of membership of the data point (xj) relating to the ith clustering, dij=║vi-xj║ is the Euclidean distance (Manhattan distance can be also used) between ith cluster center vi and jth data point xj, and ∈ [1, ∞) is an exponent of weighting which affects the fuzziness of the clusters. The equation (1) is minimized by

Assessment of Surgical Expertise in Virtual Reality Simulation by Hybrid Deep Neural Network Algorithms

http://ieeexplore.ieee.org/iel5/5687566/5697726/05698195.pdf

Tensor robust principal component analysis (robust PCA) has been applied to the lightning images. Robust PCA aims to classify the images into low-rank and sparse components. The low rank and sparse components correspond to static background separation and the dynamic (lightning) part of the images correspondingly. After classification, Singular Value Decomposition (SVD) unfold technique has been applied to sparse tensor, which transforms the tensor to spatial-temporal spaces in the form of vector matrixes. Spectral evolution shows the evolution of the polarization of the UV-Vis spectra. The contour maps of 2D energy density plots reveal the zero-point energy fluctuations of the bosons, fermions, and virtual matters from early to late stages. Detection of such fluctuations in the early stages can help to remote sensing of the lightning. 3D vector(ether) field representation of the sparse tensor of the brightest lightning reveals the butterfly effect of the black holes with the signatures of flapping. There is a correlation between the approximate entropy and the chaos trends of the 3D vector field analyzed by the Lyapunov exponent. In the brightest frame, the fractal dimension measures 0.8004, which is close to the Fibonacci Hamiltonian.

Bekir Karlik

Papers

Hierarchical neural network based compression of ECG signals

Diagnosing diabetes illness using pervasive computing and artificial neural networks

Assessment of Surgical Expertise in Virtual Reality Simulation by Hybrid Deep Neural Network Algorithms

Diagnosing diabetes illness using pervasive computing and artificial neural networks

Tensor robust principal component analysis of lightning images: butterfly effect of blackholes