Classification of amyotrophic lateral sclerosis disease based on convolutional neural network and reinforcement sample learning algorithm

Artificial intelligence for screening and diagnosis of amyotrophic lateral sclerosis: a systematic review and meta-analysis

INTRODUCTION

Amyotrophic lateral sclerosis (ALS) is a rare and fatal neurological disease that leads to progressive motor function degeneration. Diagnosing ALS is challenging due to the absence of a specific detection test. The use of artificial intelligence (AI) can assist in the investigation and treatment of ALS.

METHODS

We searched seven databases for literature on the application of AI in the early diagnosis and screening of ALS in humans. The findings were summarized using random-effects summary receiver operating characteristic curve. The risk of bias (RoB) analysis was carried out using QUADAS-2 or QUADAS-C tools.

RESULTS

In the 34 analyzed studies, a meta-prevalence of 47% for ALS was noted. For ALS detection, the pooled sensitivity of AI models was 94.3% (95% CI - 63.2% to 99.4%) with a pooled specificity of 98.9% (95% CI - 92.4% to 99.9%). For ALS classification, the pooled sensitivity of AI models was 90.9% (95% CI - 86.5% to 93.9%) with a pooled specificity of 92.3% (95% CI - 84.8% to 96.3%). Based on type of input for classification, the pooled sensitivity of AI models for gait, electromyography, and magnetic resonance signals was 91.2%, 92.6%, and 82.2%, respectively. The pooled specificity for gait, electromyography, and magnetic resonance signals was 94.1%, 96.5%, and 77.3%, respectively.

CONCLUSIONS

Although AI can play a significant role in the screening and diagnosis of ALS due to its high sensitivities and specificities, concerns remain regarding quality of evidence reported in the literature.

Artificial intelligence for automatic classification of needle EMG signals: A scoping review

OBJECTIVE

This scoping review provides an overview of artificial intelligence (AI), including machine and deep learning techniques, in the interpretation of clinical needle electromyography (nEMG) signals.

METHODS

A comprehensive search of Medline, Embase and Web of Science was conducted to find peer-reviewed journal articles. All papers published after 2010 were included. The methodological quality of the included studies was assessed with CLAIM (checklist for artificial intelligence in medical imaging).

RESULTS

51 studies were identified that fulfilled the inclusion criteria. 61% used open-source EMGlab data set to develop models to classify nEMG signal in healthy, amyotrophic lateral sclerosis (ALS) and myopathy (25 subjects). Only two articles developed models to classify signals recorded at rest. Most articles reported high performance accuracies, but many were subject to bias and overtraining.

CONCLUSIONS

Current AI-models of nEMG signals are not sufficient for clinical implementation. Suggestions for future research include emphasizing the need for an optimal training and validation approach using large datasets of clinical nEMG data from a diverse patient population.

SIGNIFICANCE

The outcomes of this study and the suggestions made aim to contribute to developing AI-models that can effectively handle signal quality variability and are suitable for daily clinical practice in interpreting nEMG signals.

Artificial intelligence-based classification of motor unit action potentials in real-world needle EMG recordings

OBJECTIVE

To develop an artificial neural network (ANN) for classification of motor unit action potential (MUAP) duration in real-word, unselected and uncleaned needle electromyography (n-EMG) recordings.

METHODS

Two nested ANN models were trained, the first discerning muscle rest, contraction and artifacts in n-EMG recordings from 2674 individual muscles from 326 patients obtained as part of daily care. The second ANN model subsequently used segments labeled as contraction for prediction of prolonged, normal and shortened MUAPs. Model performance was assessed in one internal and two external validation datasets of 184, 30 and 50 muscles, respectively.

RESULTS

The first model discerned rest, contraction and artifacts with an accuracy of 96%. The second model predicted prolonged, normal and shortened MUAPs with an accuracy of 67%, 83% and 68% in the different validation sets.

CONCLUSIONS

We developed a two-step ANN that classifies rest, muscle contraction and artifacts from real-world n-EMG recordings with very high accuracy. MUAP duration classification had moderate accuracy.

SIGNIFICANCE

This is the first study to show that an ANN can classify MUAPs in real-world n-EMG recordings highlighting the potential for AI assisted MUAP classification as a clinical tool.

A New Hybrid Approach Based on Time Frequency Images and Deep Learning Methods for Diagnosis of Migraine Disease and Investigation of Stimulus Effect

Migraine is a neurological disorder that is associated with severe headaches and seriously affects the lives of patients. Diagnosing Migraine Disease (MD) can be laborious and time-consuming for specialists. For this reason, systems that can assist specialists in the early diagnosis of MD are important. Although migraine is one of the most common neurological diseases, there are very few studies on the diagnosis of MD, especially electroencephalogram (EEG)-and deep learning (DL)-based studies. For this reason, in this study, a new system has been proposed for the early diagnosis of EEG- and DL-based MD. In the proposed study, EEG signals obtained from the resting state (R), visual stimulus (V), and auditory stimulus (A) from 18 migraine patients and 21 healthy control (HC) groups were used. By applying continuous wavelet transform (CWT) and short-time Fourier transform (STFT) methods to these EEG signals, scalogram-spectrogram images were obtained in the time-frequency (T-F) plane. Then, these images were applied as inputs in three different convolutional neural networks (CNN) architectures (AlexNet, ResNet50, SqueezeNet) that proposed deep convolutional neural network (DCNN) models and classification was performed. The results of the classification process were evaluated, taking into account accuracy (acc.), sensitivity (sens.), specificity (spec.), and performance criteria, and the performances of the preferred methods and models in this study were compared. In this way, the situation, method, and model that showed the most successful performance for the early diagnosis of MD were determined. Although the classification results are close to each other, the resting state, CWT method, and AlexNet classifier showed the most successful performance (Acc: 99.74%, Sens: 99.9%, Spec: 99.52%). We think that the results obtained in this study are promising for the early diagnosis of MD and can be of help to experts.

Distinguishing normal, neuropathic and myopathic EMG with an automated machine learning approach

OBJECTIVE

Distinguishing normal, neuropathic and myopathic electromyography (EMG) traces can be challenging. We aimed to create an automated time series classification algorithm.

METHODS

EMGs of healthy controls (HC, n = 25), patients with amyotrophic lateral sclerosis (ALS, n = 20) and inclusion body myositis (IBM, n = 20), were retrospectively selected based on longitudinal clinical follow-up data (ALS and HC) or muscle biopsy (IBM). A machine learning pipeline was applied based on 5-second EMG fragments of each muscle. Diagnostic yield expressed as area under the curve (AUC) of a receiver-operator characteristics curve, accuracy, sensitivity, and specificity were determined per muscle (muscle-level) and per patient (patient-level).

RESULTS

Diagnostic yield of the classification ALS vs. HC was: AUC 0.834 ± 0.014 at muscle-level and 0.856 ± 0.009 at patient-level. For the classification HC vs. IBM, AUC was 0.744 ± 0.043 at muscle-level and 0.735 ± 0.029 at patient-level. For the classification ALS vs. IBM, AUC was 0.569 ± 0.024 at muscle-level and 0.689 ± 0.035 at patient-level.

CONCLUSIONS

An automated time series classification algorithm can distinguish EMGs from healthy individuals from those of patients with ALS with a high diagnostic yield. Using longer EMG fragments with different levels of muscle activation may improve performance.

SIGNIFICANCE

In the future, machine learning algorithms may help improve the diagnostic accuracy of EMG examinations.

Intersection of network medicine and machine learning towards investigating the key biomarkers and pathways underlying amyotrophic lateral sclerosis: a systematic review

BACKGROUND

Network medicine is an emerging area of research that focuses on delving into the molecular complexity of the disease, leading to the discovery of network biomarkers and therapeutic target discovery. Amyotrophic lateral sclerosis (ALS) is a complicated rare disease with unknown pathogenesis and no available treatment. In ALS, network properties appear to be potential biomarkers that can be beneficial in disease-related applications when explored independently or in tandem with machine learning (ML) techniques.

OBJECTIVE

This systematic literature review explores recent trends in network medicine and implementations of network-based ML algorithms in ALS. We aim to provide an overview of the identified primary studies and gather details on identifying the potential biomarkers and delineated pathways.

METHODS

The current study consists of searching for and investigating primary studies from PubMed and Dimensions.ai, published between 2018 and 2022 that reported network medicine perspectives and the coupling of ML techniques. Each abstract and full-text study was individually evaluated, and the relevant studies were finally included in the review for discussion once they met the inclusion and exclusion criteria.

RESULTS

We identified 109 eligible publications from primary studies representing this systematic review. The data coalesced into two themes: application of network science to identify disease modules and promising biomarkers in ALS, along with network-based ML approaches. Conclusion This systematic review gives an overview of the network medicine approaches and implementations of network-based ML algorithms in ALS to determine new disease genes, and identify critical pathways and therapeutic target discovery for personalized treatment.

Deep learning methods to predict amyotrophic lateral sclerosis disease progression

Amyotrophic lateral sclerosis (ALS) is a highly complex and heterogeneous neurodegenerative disease that affects motor neurons. Since life expectancy is relatively low, it is essential to promptly understand the course of the disease to better target the patient’s treatment. Predictive models for disease progression are thus of great interest. One of the most extensive and well-studied open-access data resources for ALS is the Pooled Resource Open-Access ALS Clinical Trials (PRO-ACT) repository. In 2015, the DREAM-Phil Bowen ALS Prediction Prize4Life Challenge was held on PRO-ACT data, where competitors were asked to develop machine learning algorithms to predict disease progression measured through the slope of the ALSFRS score between 3 and 12 months. However, although it has already been successfully applied in several studies on ALS patients, to the best of our knowledge deep learning approaches still remain unexplored on the ALSFRS slope prediction in PRO-ACT cohort. Here, we investigate how deep learning models perform in predicting ALS progression using the PRO-ACT data. We developed three models based on different architectures that showed comparable or better performance with respect to the state-of-the-art models, thus representing a valid alternative to predict ALS disease progression.

An Effective and Lightweight Deep Electrocardiography Arrhythmia Recognition Model Using Novel Special and Native Structural Regularization Techniques on Cardiac Signal

Recently, cardiac arrhythmia recognition from electrocardiography (ECG) with deep learning approaches is becoming popular in clinical diagnosis systems due to its good prognosis findings, where expert data preprocessing and feature engineering are not usually required. But a lightweight and effective deep model is highly demanded to face the challenges of deploying the model in real-life applications and diagnosis accurately. In this work, two effective and lightweight deep learning models named Deep-SR and Deep-NSR are proposed to recognize ECG beats, which are based on two-dimensional convolution neural networks (2D CNNs) while using different structural regularizations. First, 97720 ECG beats extracted from all records of a benchmark MIT-BIH arrhythmia dataset have been transformed into 2D RGB (red, green, and blue) images that act as the inputs to the proposed 2D CNN models. Then, the optimization of the proposed models is performed through the proper initialization of model layers, on-the-fly augmentation, regularization techniques, Adam optimizer, and weighted random sampler. Finally, the performance of the proposed models is evaluated by a stratified 5-fold cross-validation strategy along with callback features. The obtained overall accuracy of recognizing normal beat and three arrhythmias (V-ventricular ectopic, S-supraventricular ectopic, and F-fusion) based on the Association for the Advancement of Medical Instrumentation (AAMI) is 99.93%, and 99.96% for the proposed Deep-SR model and Deep-NSR model, which demonstrate that the effectiveness of the proposed models has surpassed the state-of-the-art models and also expresses the higher model generalization. The received results with model size suggest that the proposed CNN models especially Deep-NSR could be more useful in wearable devices such as medical vests, bracelets for long-term monitoring of cardiac conditions, and in telemedicine to accurate diagnose the arrhythmia from ECG automatically. As a result, medical costs of patients and work pressure on physicians in medicals and clinics would be reduced effectively.

Phenotypic categorisation of individual subjects with motor neuron disease based on radiological disease burden patterns: A machine-learning approach

Motor neuron disease is an umbrella term encompassing a multitude of clinically heterogeneous phenotypes. The early and accurate categorisation of patients is hugely important, as MND phenotypes are associated with markedly different prognoses, progression rates, care needs and benefit from divergent management strategies. The categorisation of patients shortly after symptom onset is challenging, and often lengthy clinical monitoring is needed to assign patients to the appropriate phenotypic subgroup. In this study, a multi-class machine-learning strategy was implemented to classify 300 patients based on their radiological profile into diagnostic labels along the UMN-LMN spectrum. A comprehensive panel of cortical thickness measures, subcortical grey matter variables, and white matter integrity metrics were evaluated in a multilayer perceptron (MLP) model. Additional exploratory analyses were also carried out using discriminant function analyses (DFA). Excellent classification accuracy was achieved for amyotrophic lateral sclerosis in the testing cohort (93.7%) using the MLP model, but poor diagnostic accuracy was detected for primary lateral sclerosis (43.8%) and poliomyelitis survivors (60%). Feature importance analyses highlighted the relevance of white matter diffusivity metrics and the evaluation of cerebellar indices, cingulate measures and thalamic radiation variables to discriminate MND phenotypes. Our data suggest that radiological data from single patients may be meaningfully interpreted if large training data sets are available and the provision of diagnostic probability outcomes may be clinically useful in patients with short symptom duration. The computational interpretation of multimodal radiology datasets herald viable diagnostic, prognostic and clinical trial applications.

Surface electromyography for testing motor dysfunction in amyotrophic lateral sclerosis

OBJECTIVES

To investigate the use of a set of dynamical features, extracted from surface electromyography, to study upper motor neuron (UMN) degeneration in amyotrophic lateral sclerosis (ALS).

METHODS

We acquired surface EMG signals from the upper limb muscles of 13 ALS patients and 20 control subjects and classified them according to a novel set of muscle activity features, describing the temporal and frequency dynamic behavior of the signals, as well as measures of its complexity. Using a battery of classification approaches, we searched for the most discriminating combination of those features, as well as a suitable strategy to identify ALS.

RESULTS

We observed significant differences between ALS patients and controls, in particular when considering features highlighting differences between forearm and hand recordings, for which classification accuracies of up to 94% were achieved. The most robust discriminations were achieved using features based on detrended fluctuation analysis and peak frequency, and classifiers such as decision trees, random forest and Adaboost.

CONCLUSION

The current work shows that it is possible to achieve good identification of UMN changes in ALS by taking into consideration the dynamical behavior of surface electromyographic (sEMG) data.

Adaptive Algorithms as Control Strategies of Smart Upper Limb Orthosis: A Protocol for a Systematic Scoping Review

Introduction: Adaptive algorithms for controlling orthosis emerged to overcome significant problems with automatic biosignal classification and personalized rehabilitation. Smart orthoses are evolving fast and need a better human-machine interaction performance since biosignals, feedback, and motor control dynamically change and must be adaptive. This manuscript outlines a scoping review protocol to systematically review the smart upper limb (UL) orthoses based on adaptive algorithms and feasibility tests. Materials and Methods: This protocol was developed based on the York framework. A field-specific structure was defined to achieve each phase. Eleven scientific databases (PubMed, Web of Science, SciELO, Koreamed, Jstage, AMED, CENTRAL, PEDro, IEEE, Scopus, and Arxiv) and five patent databases (Patentscope, Patentlens, Google Patents, Kripis, J-platpat) were searched. The developed framework will extract data (i.e., orthosis description, adaptive algorithms, tools used in the usability test, and benefits to the general population) from the selected studies using a rigorous approach. Data will be described quantitatively using frequency and trend analysis methods. Heterogeneity between the included studies will be assessed using the Chi-test and I-statistic. The risk of bias will be summarized using the latest Prediction Model Study Risk of Bias Assessment Tool. Discussion: This review will identify, map, and synthesize the advances about the description of adaptive algorithms for control strategies of smart UL orthosis using data extracted from patents and articles.

Augmenting Neuromuscular Disease Detection Using Optimally Parameterized Weighted Visibility Graph

In this contribution, we propose a novel neuromuscular disease detection framework employing weighted visibility graph (WVG) aided analysis of electromyography signals. WVG converts a time series into an undirected graph, while preserving the signal properties. However, conventional WVG is sensitive to noise and has high computational complexity which is problematic for lengthy and noisy time series analysis. To address this issue in this article, we investigate the performance of WVG by varying two important parameters, namely penetrable distance and scale factor, both of which have shown promising results by eliminating the problem of signal adulteration and decreasing the computational complexity, respectively. We also aim to unfold the combined effect of these two aforesaid parameters on the WVG performance to discriminate between myopathy, amyotrophic lateral sclerosis (ALS) and healthy EMG signals. Using graph theory based features we demonstrated that the discriminating capability between the three classes increased significantly with the increase in both penetrable distance and scale factor values. Three binary (healthy vs. myopathy, myopathy vs. ALS and healthy vs. ALS) and one multiclass problems (healthy vs. myopathy vs. ALS) have been addressed in this study and for each problem, we obtained optimum parameter values determined on the basis of F-value computed using one way analysis of variance (ANOVA) test. Using optimal parameter values, we obtained mean accuracy of 98.57%, 98.09% and 99.45%, respectively for three binary and 99.05% for the multi-class classification problem. Additionally, the computational time was reduced by 96% with optimally selected WVG parameters compared to traditional WVG.

Femoral neck fracture detection in X-ray images using deep learning and genetic algorithm approaches

Objectives

This study aims to detect frontal pelvic radiograph femoral neck fracture using deep learning techniques.

Patients and methods

This retrospective study was conducted between January 2013 and January 2018. A total of 234 frontal pelvic X-ray images collected from 65 patients (32 males, 33 females; mean age 74.9 years; range, 33 to 89 years) were augmented to 2106 images to achieve a satisfactory dataset. A total of 1,341 images were fractured femoral necks while 765 were non-fractured ones. The proposed convolutional neural network (CNN) architecture contained five blocks, each containing a convolutional layer, batch normalization layer, rectified linear unit, and maximum pooling layer. After the last block, a dropout layer existed with a probability of 0.5. The last three layers of the architecture were a fully connected layer of two classes, a softmax layer and a classification layer that computes cross entropy loss. The training process was terminated after 50 epochs and an Adam Optimizer was used. Learning rate was dropped by a factor of 0.5 on every five epochs. To reduce overfitting, regularization term was added to the weights of the loss function. The training process was repeated for pixel sizes 50x50, 100x100, 200x200, and 400x400. The genetic algorithm (GA) approach was employed to optimize the hyperparameters of the CNN architecture and to minimize the error after testing the model created by the CNN architecture in the training phase.

Results

Performance in terms of sensitivity, specificity, accuracy, F1 score, and Cohen’s kappa coefficient were evaluated using five- fold cross validation tests. Best performance was obtained when cropped images were rescaled to 50x50 pixels. The kappa metric showed more reliable classifier performance when 50x50 pixels image size was used to feed the CNN. The classifier performance was more reliable according to other image sizes. Sensitivity and specificity rates were computed to be 83% and 73%, respectively. With the inclusion of the GA, this rate increased by 1.6%. The detection rate of fractured bones was found to be 83%. A kappa coefficient of 55% was obtained, indicating an acceptable agreement.

Conclusion

This experimental study utilized deep learning techniques in the detection of bone fractures in radiography. Although the dataset was unbalanced, the results can be considered promising. It was observed that use of smaller image size decreases computational cost and provides better results according to evaluation metrics.

Comparison of Bagging and Boosting Ensemble Machine Learning Methods for Automated EMG Signal Classification

The neuromuscular disorders are diagnosed using electromyographic (EMG) signals. Machine learning algorithms are employed as a decision support system to diagnose neuromuscular disorders. This paper compares bagging and boosting ensemble learning methods to classify EMG signals automatically. Even though ensemble classifiers' efficacy in relation to real-life issues has been presented in numerous studies, there are almost no studies which focus on the feasibility of bagging and boosting ensemble classifiers to diagnose the neuromuscular disorders. Therefore, the purpose of this paper is to assess the feasibility of bagging and boosting ensemble classifiers to diagnose neuromuscular disorders through the use of EMG signals. It should be understood that there are three steps to this method, where the step number one is to calculate the wavelet packed coefficients (WPC) for every type of EMG signal. After this, it is necessary to calculate statistical values of WPC so that the distribution of wavelet coefficients could be demonstrated. In the last step, an ensemble classifier used the extracted features as an input of the classifier to diagnose the neuromuscular disorders. Experimental results showed the ensemble classifiers achieved better performance for diagnosis of neuromuscular disorders. Results are promising and showed that the AdaBoost with random forest ensemble method achieved an accuracy of 99.08%, F-measure 0.99, AUC 1, and kappa statistic 0.99.

Deep learning for waveform identification of resting needle electromyography signals

OBJECTIVE

Given the recent advent in machine learning and artificial intelligence on medical data analysis, we hypothesized that the deep learning algorithm can classify resting needle electromyography (n-EMG) discharges.

METHODS

Six clinically observed resting n-EMG signals were used as a dataset. The data were converted to Mel-spectrogram. Data augmentation was then applied to the training data. Deep learning algorithms were applied to assess the accuracies of correct classification, with or without the use of pre-trained weights for deep-learning networks.

RESULTS

While the original data yielded the accuracy up to 0.86 on the test dataset, data-augmentation up to 200,000 training images showed significant increase in the accuracy to 1.0. The use of pre-trained weights (fine tuning) showed greater accuracy than "training from scratch".

CONCLUSIONS

Resting n-EMG signals were successfully classified by deep-learning algorithm, especially with the use of data augmentation and transfer learning techniques.

SIGNIFICANCE

Computer-aided signal identification of clinical n-EMG testing might be possible by deep-learning algorithms.

Imaging Biomarkers for the Diagnosis and Prognosis of Neurodegenerative Diseases. The Example of Amyotrophic Lateral Sclerosis

The term amyotrophic lateral sclerosis (ALS) comprises a heterogeneous group of fatal neurodegenerative disorders of largely unknown etiology characterized by the upper motor neurons (UMN) and/or lower motor neurons (LMN) degeneration. The development of brain imaging biomarkers is essential to advance in the diagnosis, stratification and monitoring of ALS, both in the clinical practice and clinical trials. In this review, the characteristics of an optimal imaging biomarker and common pitfalls in biomarkers evaluation will be discussed. Moreover, the development and application of the most promising brain magnetic resonance (MR) imaging biomarkers will be reviewed. Finally, the integration of both qualitative and quantitative multimodal brain MR biomarkers in a structured report will be proposed as a support tool for ALS diagnosis and stratification.

Features based on variational mode decomposition for identification of neuromuscular disorder using EMG signals

Neuromuscular disorder is a muscular and nervous disorder resulting in muscular weakness and progressively damages nervous control, such as amyotrophic lateral sclerosis (ALS) and myopathy (MYO). Its diagnosis can be possible by classification of ALS, MYO, and normal electromyogram (EMG) signals. In this paper, an effective method based on variational mode decomposition (VMD) is proposed for identification of neuromuscular disorder of EMG signals. VMD is an adaptive signal decomposition which decomposes EMG signals nonrecursively into band-limited functions or modes. These modes are used for extraction of spectral features, particularly spectral flatness, spectral spread, spectral decrease and statistical features like kurtosis, mean absolute deviation, and interquartile range. The extracted features are fed to the extreme learning machine classifier in order to classify neuromuscular disorder of EMG signals. The performance of obtained results shows that the method used provides a better classification for neuromuscular disorder of EMG signals as compared to existing methods.

Deep Learning on 1-D Biosignals: a Taxonomy-based Survey

OBJECTIVES

 Deep learning models such as convolutional neural networks (CNNs) have been applied successfully to medical imaging, but biomedical signal analysis has yet to fully benefit from this novel approach. Our survey aims at (i) reviewing deep learning techniques for biosignal analysis in computer- aided diagnosis; and (ii) deriving a taxonomy for organizing the growing number of applications in the field.

METHODS

 A comprehensive literature research was performed using PubMed, Scopus, and ACM. Deep learning models were classified with respect to the (i) origin, (ii) dimension, and (iii) type of the biosignal as input to the deep learning model; (iv) the goal of the application; (v) the size and (vi) type of ground truth data; (vii) the type and (viii) schedule of learning the network; and (ix) the topology of the model.

RESULTS

 Between January 2010 and December 2017, a total 71 papers were published on the topic. The majority (n = 36) of papers are on electrocariography (ECG) signals. Most applications (n = 25) aim at detection of patterns, while only a few (n = 6) at predection of events. Out of 36 ECG-based works, many (n = 17) relate to multi-lead ECG. Other biosignals that have been identified in the survey are electromyography, phonocardiography, photoplethysmography, electrooculography, continuous glucose monitoring, acoustic respiratory signal, blood pressure, and electrodermal activity signal, while ballistocardiography or seismocardiography have yet to be analyzed using deep learning techniques. In supervised and unsupervised applications, CNNs and restricted Boltzmann machines are the most and least frequently used, (n = 34) and (n = 15), respectively.

CONCLUSION

 Our key-code classification of relevant papers was used to cluster the approaches that have been published to date and demonstrated a large variability of research with respect to data, application, and network topology. Future research is expected to focus on the standardization of deep learning architectures and on the optimization of the network parameters to increase performance and robustness. Furthermore, application-driven approaches and updated training data from mobile recordings are needed.