Automatic detection of poor quality signals as a pre-processing scheme in the analysis of sEMG in swallowing. (January 2022)

• Record Type: Journal Article
• Authors: Cuadros-Acosta, J.; Orozco-Duque, A.
• Is Part Of: Biomedical signal processing and control. Volume 71(2022)Part A
• Publication Date: 2022-01
• Subjects: Signal quality -- Surface electromyography -- Swallowing -- Noise; BW bandwidth -- E envelope -- KNN k-nearest neighbors -- MLP multilayer perceptron -- NB naive Bayes -- P percentile -- PSD power spectral density -- RF random forest -- RMS root mean square -- ROC AUC area under the ROC curve -- SBS sequential backward selection -- SFS sequential forward selection -- SBFS sequential backward floating selection -- SFFS sequential forward floating selection -- SD standard deviation -- sEMG surface electromyography -- SQI signal quality indices -- SVM support vector machine; Signal processing -- Periodicals; Biomedical engineering -- Periodicals; Signal Processing, Computer-Assisted -- Periodicals; Image Processing, Computer-Assisted -- Periodicals; Biomedical Engineering -- Periodicals; 610.28
• DOI: 10.1016/j.bspc.2021.103122
• Languages: English
• ISSNs: 1746-8094

Abstract

Highlights: Supervised machine learning models can classify bad quality signals regarding noise. Time-domain and frequency-domain features can be estimators of sEMG signal quality. Signal quality validation may improve the performance in muscular segmentation. sEMG features can help to reject bad quality signals in swallowing. Abstract: Dysphagia, a swallowing disorder, has a high incidence worldwide. However, it is also one of the most under-diagnosed pathologies because its diagnosis includes invasive procedures or exposure to radiation. Currently, non-invasive approaches based on the analysis of surface electromyography (sEMG) signals are being developed. Since the sEMG technique is susceptible to noise, this study developed an automated pre-processing signal quality validation stage to assess contaminated signals. First, a dataset was generated from signals acquired using a bilateral protocol with surface electrodes placed on four muscle groups involved in swallowing. The dataset was labeled into two groups regarding signal quality classes ("good" and "poor"). Second, features in the time and frequency domains were evaluated as signal quality indices (SQI) using the area under the roc curve. Finally, we compared multiple supervised machine learning models trained in 5-fold cross validation. Hence, we included hyper-parameter optimization within a sequential feature selection. Our results demonstrate how the proposed three-stage scheme can automate the signal quality analysis of a swallowing dataset obtained from patients diagnosed with dysphagia by implementing a random forest classifier that uses three features achieving an accuracy of 98 ± 1.74%. In addition, to validate the reproducibility of the model, its prediction performance was measured with two different datasets of healthy patients, achieving an accuracy of 98% and 85%. This method could be applied as a pre-processing step to improve the study of sEMG signals (e.g., segmentation) obtained during swallowing tasks.