Leticia Silva
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Leticia Silva has a total of 35 co-authors. Here are the top co-authors by number of shared publications:
- Teodiano Freire Bastos-Filho
Federal University of Espírito Santo
11 shared publications - Denis Delisle-Rodriguez
Postgraduate Program in Neuroengineering, Edmond and Lily Safra International Institute of Neuroscience, Macaiba 59288-999, Brazil
7 shared publications - Carlos Valadão
Federal University of Espírito Santo
5 shared publications - Eliete Caldeira
Federal University of Espírito Santo
4 shared publications - Jéssica Lima
Postgraduate Program in Biotechnology, Universidade Federal do Espírito Santo, Vitória, Brazil
4 shared publications - Lucas Lampier
Federal University of Espírito Santo, Vitória, Brazil
4 shared publications
Latest Publications
- Open Access
- 29 July 2026
- Published by Springer Nature in Multimedia Tools and Applications
- Vol. 85 (8) , 656
Ensemble learning approach to support Manchester protocol triage
Emergency Departments (EDs) play an important role in providing critical healthcare services, especially in the context of acute conditions. The Manchester Triage System (MTS) was developed to assess the severity of patients arriving at EDs, ensuring that patients with more severe conditions receive immediate treatment. However, the presence of biases, resulting from factors such as a lack of adequate training and the influence of personal preferences of healthcare professionals, may compromise the waiting time for care. This study aims to improve the precision and efficiency of patient risk assessment within EDs, focusing on the MTS utilization. As a novel contribution, this work evaluates ensemble learning within the MTS framework and investigates the feasibility of risk classification using objective numerical features. Several Machine Learning (ML) models were explored using relevant patient information from a public MTS dataset, including Decision Tree, K-Nearest Neighbor, Random Forest, Multilayer Perceptron, and ensemble methods such as Soft and Hard Voting, and Stacking. Two experiments were conducted: the first combined categorical and numerical features; the second focused on numerical variables, except for the hospital identifier, included to control for inter-institutional variation. The results for the risk classification task were promising in terms of accuracy (ACC) and F1-Score, with up to 99.00% and 0.99, respectively, for Experiment #1, which achieved the highest predictive performance. For Experiment #2, an ACC of up to 74.68% and an F1-Score of 0.74 were achieved for the Stacking ensemble ML method, indicating clinically meaningful performance. These results suggest that the proposed approach may support risk classification in the MTS. However, further real-world evaluation is required due to the retrospective design and lack of prospective validation.
- 29 July 2026
- Published by Springer Nature in Multimedia Tools and Applications
- Vol. 85 (8) , 656
Open AccessEnsemble learning approach to support Manchester protocol triage
Emergency Departments (EDs) play an important role in providing critical healthcare services, especially in the context of acute conditions. The Manchester Triage System (MTS) was developed to assess the severity of patients arriving at EDs, ensuring that patients with more severe conditions receive immediate treatment. However, the presence of biases, resulting from factors such as a lack of adequate training and the influence of personal preferences of healthcare professionals, may compromise the waiting time for care. This study aims to improve the precision and efficiency of patient risk assessment within EDs, focusing on the MTS utilization. As a novel contribution, this work evaluates ensemble learning within the MTS framework and investigates the feasibility of risk classification using objective numerical features. Several Machine Learning (ML) models were explored using relevant patient information from a public MTS dataset, including Decision Tree, K-Nearest Neighbor, Random Forest, Multilayer Perceptron, and ensemble methods such as Soft and Hard Voting, and Stacking. Two experiments were conducted: the first combined categorical and numerical features; the second focused on numerical variables, except for the hospital identifier, included to control for inter-institutional variation. The results for the risk classification task were promising in terms of accuracy (ACC) and F1-Score, with up to 99.00% and 0.99, respectively, for Experiment #1, which achieved the highest predictive performance. For Experiment #2, an ACC of up to 74.68% and an F1-Score of 0.74 were achieved for the Stacking ensemble ML method, indicating clinically meaningful performance. These results suggest that the proposed approach may support risk classification in the MTS. However, further real-world evaluation is required due to the retrospective design and lack of prospective validation.
- Open Access
- 11 February 2026
- Published by Springer Nature in Medical & Biological Engineering & Computing
- Vol. 64 (4) , 1311-1322
A lower-limb motor imagery BCI using virtual reality and novel calibration strategy in post-stroke patients
This study proposes a novel two-step calibration strategy in a Motor Imagery (MI)-based Brain-Computer Interface (BCI) system for lower-limb post-stroke patients rehabilitation, using electroencephalography (EEG) signals, a virtual reality serious game, and a pedal end-effector. This research proposes a novel MI feature extraction algorithm combining EEG data from pedaling MI and actual movements using k-Nearest Neighbors (k-NN) and probability analysis. The extracted features are used to re-calibrate the BCI system, and improve its accuracy. Initially, participants performed 20 MI trials without feedback (open-loop, Calibration Mode #1) while receiving passive movement. During this phase, EEG data were processed using the Riemannian Geometry method to train a Linear Discriminant Analysis (LDA) classifier. In the second step (closed-loop, Calibration Mode #2), feedback through passive movement was provided, changing the pedal’s speed according to the MI classification consistency. A chronic post-stroke patient tested the proposed BCI system, first receiving transcranial Direct Current Stimulation (tDCS) before each session. The calibration strategy improved classification accuracy from 52.60% (Mode #1) to 86.74% (Mode #2). The classification remained effective throughout sessions, allowing for immediate feedback-driven training while collecting more reliable EEG MI data. The BCI developed here is able to provide post-stroke patient rehabilitation by engaging lower-limb central and peripheral mechanisms through simultaneous MI and passive pedaling during extended workout sessions.
- 11 February 2026
- Published by Springer Nature in Medical & Biological Engineering & Computing
- Vol. 64 (4) , 1311-1322
Open AccessA lower-limb motor imagery BCI using virtual reality and novel calibration strategy in post-stroke patients
This study proposes a novel two-step calibration strategy in a Motor Imagery (MI)-based Brain-Computer Interface (BCI) system for lower-limb post-stroke patients rehabilitation, using electroencephalography (EEG) signals, a virtual reality serious game, and a pedal end-effector. This research proposes a novel MI feature extraction algorithm combining EEG data from pedaling MI and actual movements using k-Nearest Neighbors (k-NN) and probability analysis. The extracted features are used to re-calibrate the BCI system, and improve its accuracy. Initially, participants performed 20 MI trials without feedback (open-loop, Calibration Mode #1) while receiving passive movement. During this phase, EEG data were processed using the Riemannian Geometry method to train a Linear Discriminant Analysis (LDA) classifier. In the second step (closed-loop, Calibration Mode #2), feedback through passive movement was provided, changing the pedal’s speed according to the MI classification consistency. A chronic post-stroke patient tested the proposed BCI system, first receiving transcranial Direct Current Stimulation (tDCS) before each session. The calibration strategy improved classification accuracy from 52.60% (Mode #1) to 86.74% (Mode #2). The classification remained effective throughout sessions, allowing for immediate feedback-driven training while collecting more reliable EEG MI data. The BCI developed here is able to provide post-stroke patient rehabilitation by engaging lower-limb central and peripheral mechanisms through simultaneous MI and passive pedaling during extended workout sessions.
- Open Access
- 2 January 2025
- Published by Brazilian Journals in Brazilian Journal of Biological Sciences
- Vol. 12 (26) , e132
Análise dos Efeitos da Ativação Cerebral por Eletroencefalografia em Pacientes Pós-AVC Durante Imagética Motora, Estimulação Elétrica Funcional e Luva Robótica
Objetivo: analisar os efeitos da ativação cerebral por eletroencefalografia (EEG) em pacientes pós-AVC durante estimulações com Imagética Motora (IM), Estimulação Elétrica Funcional (FES) e Luva Robótica (LR). Metodologia: Foram analisados e comparados registros de EEG de dois (02) sujeitos pós-AVC, com acometimento de movimento de mão durante intervenções de IM, FES, LR e associação de FES+LR. Resultados: A análise do ritmo Mu demonstrou que as terapias com LR, FES e FES+LR foram capazes de reduzir a PR de forma significativa esse ritmo no voluntário 1 (V1), quando comparadas a terapia de IM. No voluntário 2 (V2), observamos o mesmo padrão. Após a análise do ritmo Beta Low, foi demonstrado que a terapia de FES+LR foi superior e significativa quando comparada a IM nos dois indivíduos estudados. A terapia apenas com FES também se mostrou significativa em V2 em comparação a IM. Já durante a análise da frequência Beta High, observou-se que a terapia de FES+LR foi superior e significativa quando comparada a IM em V1. Conclusões: Foram obtidos resultados relevantes utilizando as estimulações FES+LR associadas. Diante disso, conclui-se que a associação dessas terapias pode auxiliar no processo de reabilitação neuro-motora pós-AVC, possibilitando a elaboração de novos protocolos de neuro-reabilitação que sejam acessíveis e seguros.
- 2 January 2025
- Published by Brazilian Journals in Brazilian Journal of Biological Sciences
- Vol. 12 (26) , e132
Open AccessAnálise dos Efeitos da Ativação Cerebral por Eletroencefalografia em Pacientes Pós-AVC Durante Imagética Motora, Estimulação Elétrica Funcional e Luva Robótica
Objetivo: analisar os efeitos da ativação cerebral por eletroencefalografia (EEG) em pacientes pós-AVC durante estimulações com Imagética Motora (IM), Estimulação Elétrica Funcional (FES) e Luva Robótica (LR). Metodologia: Foram analisados e comparados registros de EEG de dois (02) sujeitos pós-AVC, com acometimento de movimento de mão durante intervenções de IM, FES, LR e associação de FES+LR. Resultados: A análise do ritmo Mu demonstrou que as terapias com LR, FES e FES+LR foram capazes de reduzir a PR de forma significativa esse ritmo no voluntário 1 (V1), quando comparadas a terapia de IM. No voluntário 2 (V2), observamos o mesmo padrão. Após a análise do ritmo Beta Low, foi demonstrado que a terapia de FES+LR foi superior e significativa quando comparada a IM nos dois indivíduos estudados. A terapia apenas com FES também se mostrou significativa em V2 em comparação a IM. Já durante a análise da frequência Beta High, observou-se que a terapia de FES+LR foi superior e significativa quando comparada a IM em V1. Conclusões: Foram obtidos resultados relevantes utilizando as estimulações FES+LR associadas. Diante disso, conclui-se que a associação dessas terapias pode auxiliar no processo de reabilitação neuro-motora pós-AVC, possibilitando a elaboração de novos protocolos de neuro-reabilitação que sejam acessíveis e seguros.
- 26 June 2024
- Published by Institute of Electrical and Electronics Engineers (IEEE)
- p. 1-6
Towards a Multimodal Automatic Equipment for Predictive Health Monitoring of Industrial Workers: A Usability Perspective
Ensuring the health and safety of workers in indus-trial environments is of utmost importance. In Brazil, Regula-tory Normatives (RNs) mandate Periodic Medical Examinations (PMEs), with the frequency of these examinations primarily determined by the level of risk to which workers are exposed. However, normally these examinations do not provide a real-time assessment of workers' health status. The aim of this study is to introduce the Integrated Portable Medical Assistant (IPMA), which is a multimodal equipment capable of automatically collecting biomedical data, such as oxygen saturation level, body temperature, and heart rate of users. This equipment aids in monitoring the health of industry workers through Machine Learning (ML) algorithms, and can store biomedical data for ongoing research and tracking the health condition of workers. Preliminary findings evaluating 14 workers and 2 doctors suggest promising results using this equipment in industries. In terms of equipment usability, the workers reported the IPMA to be highly user-friendly through evaluation conducted through the System Usability Scale (SUS) and the Post-Study System Usability Questionnaire (PSSUQ), which also provided average scores above 85 and 1.9, respectively. Overall, the IPMA shows promising potential for improving safety and well-being in industrial workplaces, and it may also contribute to reducing worker absenteeism due to some illness.
- 26 June 2024
- Published by Institute of Electrical and Electronics Engineers (IEEE)
- p. 1-6
Towards a Multimodal Automatic Equipment for Predictive Health Monitoring of Industrial Workers: A Usability Perspective
Ensuring the health and safety of workers in indus-trial environments is of utmost importance. In Brazil, Regula-tory Normatives (RNs) mandate Periodic Medical Examinations (PMEs), with the frequency of these examinations primarily determined by the level of risk to which workers are exposed. However, normally these examinations do not provide a real-time assessment of workers' health status. The aim of this study is to introduce the Integrated Portable Medical Assistant (IPMA), which is a multimodal equipment capable of automatically collecting biomedical data, such as oxygen saturation level, body temperature, and heart rate of users. This equipment aids in monitoring the health of industry workers through Machine Learning (ML) algorithms, and can store biomedical data for ongoing research and tracking the health condition of workers. Preliminary findings evaluating 14 workers and 2 doctors suggest promising results using this equipment in industries. In terms of equipment usability, the workers reported the IPMA to be highly user-friendly through evaluation conducted through the System Usability Scale (SUS) and the Post-Study System Usability Questionnaire (PSSUQ), which also provided average scores above 85 and 1.9, respectively. Overall, the IPMA shows promising potential for improving safety and well-being in industrial workplaces, and it may also contribute to reducing worker absenteeism due to some illness.
- 4 January 2024
- Published by Springer Nature
- p. 409-419
IPMA—An Automated System to Capture Biomedical Signals and Help the Diagnosis of Respiratory Diseases
- 4 January 2024
- Published by Springer Nature
- p. 409-419
IPMA—An Automated System to Capture Biomedical Signals and Help the Diagnosis of Respiratory Diseases
- Open Access
Unraveling Transformative Effects after tDCS and BCI Intervention in Chronic Post-Stroke Patient Rehabilitation—An Alternative Treatment Design Study
Stroke is a debilitating clinical condition resulting from a brain infarction or hemorrhage that poses significant challenges for motor function restoration. Previous studies have shown the potential of applying transcranial direct current stimulation (tDCS) to improve neuroplasticity in patients with neurological diseases or disorders. By modulating the cortical excitability, tDCS can enhance the effects of conventional therapies. While upper-limb recovery has been extensively studied, research on lower limbs is still limited, despite their important role in locomotion, independence, and good quality of life. As the life and social costs due to neuromuscular disability are significant, the relatively low cost, safety, and portability of tDCS devices, combined with low-cost robotic systems, can optimize therapy and reduce rehabilitation costs, increasing access to cutting-edge technologies for neuromuscular rehabilitation. This study explores a novel approach by utilizing the following processes in sequence: tDCS, a motor imagery (MI)-based brain-computer interface (BCI) with virtual reality (VR), and a motorized pedal end-effector. These are applied to enhance the brain plasticity and accelerate the motor recovery of post-stroke patients. The results are particularly relevant for post-stroke patients with severe lower-limb impairments, as the system proposed here provides motor training in a real-time closed-loop design, promoting cortical excitability around the foot area (Cz) while the patient directly commands with his/her brain signals the motorized pedal. This strategy has the potential to significantly improve rehabilitation outcomes. The study design follows an alternating treatment design (ATD), which involves a double-blind approach to measure improvements in both physical function and brain activity in post-stroke patients. The results indicate positive trends in the motor function, coordination, and speed of the affected limb, as well as sensory improvements. The analysis of event-related desynchronization (ERD) from EEG signals reveals significant modulations in Mu, low beta, and high beta rhythms. Although this study does not provide conclusive evidence for the superiority of adjuvant mental practice training over conventional therapy alone, it highlights the need for larger-scale investigations.
Open AccessUnraveling Transformative Effects after tDCS and BCI Intervention in Chronic Post-Stroke Patient Rehabilitation—An Alternative Treatment Design Study
Stroke is a debilitating clinical condition resulting from a brain infarction or hemorrhage that poses significant challenges for motor function restoration. Previous studies have shown the potential of applying transcranial direct current stimulation (tDCS) to improve neuroplasticity in patients with neurological diseases or disorders. By modulating the cortical excitability, tDCS can enhance the effects of conventional therapies. While upper-limb recovery has been extensively studied, research on lower limbs is still limited, despite their important role in locomotion, independence, and good quality of life. As the life and social costs due to neuromuscular disability are significant, the relatively low cost, safety, and portability of tDCS devices, combined with low-cost robotic systems, can optimize therapy and reduce rehabilitation costs, increasing access to cutting-edge technologies for neuromuscular rehabilitation. This study explores a novel approach by utilizing the following processes in sequence: tDCS, a motor imagery (MI)-based brain-computer interface (BCI) with virtual reality (VR), and a motorized pedal end-effector. These are applied to enhance the brain plasticity and accelerate the motor recovery of post-stroke patients. The results are particularly relevant for post-stroke patients with severe lower-limb impairments, as the system proposed here provides motor training in a real-time closed-loop design, promoting cortical excitability around the foot area (Cz) while the patient directly commands with his/her brain signals the motorized pedal. This strategy has the potential to significantly improve rehabilitation outcomes. The study design follows an alternating treatment design (ATD), which involves a double-blind approach to measure improvements in both physical function and brain activity in post-stroke patients. The results indicate positive trends in the motor function, coordination, and speed of the affected limb, as well as sensory improvements. The analysis of event-related desynchronization (ERD) from EEG signals reveals significant modulations in Mu, low beta, and high beta rhythms. Although this study does not provide conclusive evidence for the superiority of adjuvant mental practice training over conventional therapy alone, it highlights the need for larger-scale investigations.
- 1 September 2023
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Instrumentation and Measurement
- Vol. 72 (00189456) , 1-8
A Deep Learning Approach for Estimating SpO2 Using a Smartphone Camera
The COVID-19 pandemic has highlighted the significance of both telemedicine and measuring physiological signals as peripheral Oxygen Saturation (SpO 2 ) at home. Aiming to increase the confiability of SpO 2 readings performed by smartphones, this study presents a novel Deep Neural Network (DNN) model that predicts SpO 2 levels using fingertip videos captured by smartphones. The model was trained and evaluated using the leave-one-out method, and compared against the SpO 2 values measured by a set of oximeters. The dataset used in this work is a differential of the study, unlike most of similar studies which use a breath-holding protocol that presents only a few samples of low SpO 2 values, our tests were evaluated using a publicly available dataset in which volunteers’ SpO 2 levels were measured in a range from almost 100% to approximately 70% by manipulating the percentage of oxygen inhaled by the volunteers, and the SpO 2 levels were kept on lower values for a considerable period of time. The proposed DNN model was compared against multiple variations of the Ratio-of-Rations (RoR) method and the model proposed by the dataset creators. The results show that the DNN outputs have a strong relationship with the true SpO 2 measurements, as indicated by a Pearson coefficient (PC) of 0.95 and a Concordance Correlation Coefficient (CCC) of 0.95, and also a small error, measured by a Root Mean Squared Error (RMSE) of 2.92% and a Mean Absolute Error (MAE) of 1.97%. Compared to the RoR method, which achieved a PC of 0.29, CCC of 0.21, RMSE of 8.83% and MAE of 7.03%, the proposed DNN model exhibits superior performance. Moreover, the proposed DNN model outperforms the model proposed by the dataset authors, which achieved a MAE of 5.00%. These findings demonstrate that the proposed DNN model can accurately predict SpO 2 levels using smartphone-recorded fingertip videos with lower error and higher precision than a widely used classical method.
- 1 September 2023
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Instrumentation and Measurement
- Vol. 72 (00189456) , 1-8
A Deep Learning Approach for Estimating SpO2 Using a Smartphone Camera
The COVID-19 pandemic has highlighted the significance of both telemedicine and measuring physiological signals as peripheral Oxygen Saturation (SpO 2 ) at home. Aiming to increase the confiability of SpO 2 readings performed by smartphones, this study presents a novel Deep Neural Network (DNN) model that predicts SpO 2 levels using fingertip videos captured by smartphones. The model was trained and evaluated using the leave-one-out method, and compared against the SpO 2 values measured by a set of oximeters. The dataset used in this work is a differential of the study, unlike most of similar studies which use a breath-holding protocol that presents only a few samples of low SpO 2 values, our tests were evaluated using a publicly available dataset in which volunteers’ SpO 2 levels were measured in a range from almost 100% to approximately 70% by manipulating the percentage of oxygen inhaled by the volunteers, and the SpO 2 levels were kept on lower values for a considerable period of time. The proposed DNN model was compared against multiple variations of the Ratio-of-Rations (RoR) method and the model proposed by the dataset creators. The results show that the DNN outputs have a strong relationship with the true SpO 2 measurements, as indicated by a Pearson coefficient (PC) of 0.95 and a Concordance Correlation Coefficient (CCC) of 0.95, and also a small error, measured by a Root Mean Squared Error (RMSE) of 2.92% and a Mean Absolute Error (MAE) of 1.97%. Compared to the RoR method, which achieved a PC of 0.29, CCC of 0.21, RMSE of 8.83% and MAE of 7.03%, the proposed DNN model exhibits superior performance. Moreover, the proposed DNN model outperforms the model proposed by the dataset authors, which achieved a MAE of 5.00%. These findings demonstrate that the proposed DNN model can accurately predict SpO 2 levels using smartphone-recorded fingertip videos with lower error and higher precision than a widely used classical method.
- 1 February 2023
- Published by IOP Publishing in Journal of Neural Engineering
- Vol. 20 (1) , 016047
EEG changes during passive movements improve the motor imagery feature extraction in BCIs-based sensory feedback calibration
Objective. This work proposes a method for two calibration schemes based on sensory feedback to extract reliable motor imagery (MI) features, and provide classification outputs more correlated to the user’s intention. Method. After filtering the raw electroencephalogram (EEG), a two-step method for spatial feature extraction by using the Riemannian covariance matrices (RCM) method and common spatial patterns is proposed here. It uses EEG data from trials providing feedback, in an intermediate step composed of both kth nearest neighbors and probability analyses, to find periods of time in which the user probably performed well the MI task without feedback. These periods are then used to extract features with better separability, and train a classifier for MI recognition. For evaluation, an in-house dataset with eight healthy volunteers and two post-stroke patients that performed lower-limb MI, and consequently received passive movements as feedback was used. Other popular public EEG datasets (such as BCI Competition IV dataset IIb, among others) from healthy subjects that executed upper-and lower-limbs MI tasks under continuous visual sensory feedback were further used. Results. The proposed system based on the Riemannian geometry method in two-steps (RCM–RCM) outperformed significantly baseline methods, reaching average accuracy up to 82.29%. These findings show that EEG data on periods providing passive movement can be used to contribute greatly during MI feature extraction. Significance. Unconscious brain responses elicited over the sensorimotor areas may be avoided or greatly reduced by applying our approach in MI-based brain–computer interfaces (BCIs). Therefore, BCI’s outputs more correlated to the user’s intention can be obtained.
- 1 February 2023
- Published by IOP Publishing in Journal of Neural Engineering
- Vol. 20 (1) , 016047
EEG changes during passive movements improve the motor imagery feature extraction in BCIs-based sensory feedback calibration
Objective. This work proposes a method for two calibration schemes based on sensory feedback to extract reliable motor imagery (MI) features, and provide classification outputs more correlated to the user’s intention. Method. After filtering the raw electroencephalogram (EEG), a two-step method for spatial feature extraction by using the Riemannian covariance matrices (RCM) method and common spatial patterns is proposed here. It uses EEG data from trials providing feedback, in an intermediate step composed of both kth nearest neighbors and probability analyses, to find periods of time in which the user probably performed well the MI task without feedback. These periods are then used to extract features with better separability, and train a classifier for MI recognition. For evaluation, an in-house dataset with eight healthy volunteers and two post-stroke patients that performed lower-limb MI, and consequently received passive movements as feedback was used. Other popular public EEG datasets (such as BCI Competition IV dataset IIb, among others) from healthy subjects that executed upper-and lower-limbs MI tasks under continuous visual sensory feedback were further used. Results. The proposed system based on the Riemannian geometry method in two-steps (RCM–RCM) outperformed significantly baseline methods, reaching average accuracy up to 82.29%. These findings show that EEG data on periods providing passive movement can be used to contribute greatly during MI feature extraction. Significance. Unconscious brain responses elicited over the sensorimotor areas may be avoided or greatly reduced by applying our approach in MI-based brain–computer interfaces (BCIs). Therefore, BCI’s outputs more correlated to the user’s intention can be obtained.
- Open Access
- 5 October 2022
- Published by Frontiers Media SA in Frontiers in Signal Processing
- Vol. 2
COVID-19 respiratory sound analysis and classification using audio textures
Since the COVID-19 outbreak, a major scientific effort has been made by researchers and companies worldwide to develop a digital diagnostic tool to screen this disease through some biomedical signals, such as cough, and speech. Joint time–frequency feature extraction techniques and machine learning (ML)-based models have been widely explored in respiratory diseases such as influenza, pertussis, and COVID-19 to find biomarkers from human respiratory system-generated acoustic sounds. In recent years, a variety of techniques for discriminating textures and computationally efficient local texture descriptors have been introduced, such as local binary patterns and local ternary patterns, among others. In this work, we propose an audio texture analysis of sounds emitted by subjects in suspicion of COVID-19 infection using time–frequency spectrograms. This approach of the feature extraction method has not been widely used for biomedical sounds, particularly for COVID-19 or respiratory diseases. We hypothesize that this textural sound analysis based on local binary patterns and local ternary patterns enables us to obtain a better classification model by discriminating both people with COVID-19 and healthy subjects. Cough, speech, and breath sounds from the INTERSPEECH 2021 ComParE and Cambridge KDD databases have been processed and analyzed to evaluate our proposed feature extraction method with ML techniques in order to distinguish between positive or negative for COVID-19 sounds. The results have been evaluated in terms of an unweighted average recall (UAR). The results show that the proposed method has performed well for cough, speech, and breath sound classification, with a UAR up to 100.00%, 60.67%, and 95.00%, respectively, to infer COVID-19 infection, which serves as an effective tool to perform a preliminary screening of COVID-19.
- 5 October 2022
- Published by Frontiers Media SA in Frontiers in Signal Processing
- Vol. 2
Open AccessCOVID-19 respiratory sound analysis and classification using audio textures
Since the COVID-19 outbreak, a major scientific effort has been made by researchers and companies worldwide to develop a digital diagnostic tool to screen this disease through some biomedical signals, such as cough, and speech. Joint time–frequency feature extraction techniques and machine learning (ML)-based models have been widely explored in respiratory diseases such as influenza, pertussis, and COVID-19 to find biomarkers from human respiratory system-generated acoustic sounds. In recent years, a variety of techniques for discriminating textures and computationally efficient local texture descriptors have been introduced, such as local binary patterns and local ternary patterns, among others. In this work, we propose an audio texture analysis of sounds emitted by subjects in suspicion of COVID-19 infection using time–frequency spectrograms. This approach of the feature extraction method has not been widely used for biomedical sounds, particularly for COVID-19 or respiratory diseases. We hypothesize that this textural sound analysis based on local binary patterns and local ternary patterns enables us to obtain a better classification model by discriminating both people with COVID-19 and healthy subjects. Cough, speech, and breath sounds from the INTERSPEECH 2021 ComParE and Cambridge KDD databases have been processed and analyzed to evaluate our proposed feature extraction method with ML techniques in order to distinguish between positive or negative for COVID-19 sounds. The results have been evaluated in terms of an unweighted average recall (UAR). The results show that the proposed method has performed well for cough, speech, and breath sound classification, with a UAR up to 100.00%, 60.67%, and 95.00%, respectively, to infer COVID-19 infection, which serves as an effective tool to perform a preliminary screening of COVID-19.