Participants
In the pre-test phase, a 90% cough recognition accuracy was obtained by selecting the normal population as test subjects and enrolling 10 males and 5 females as subjects. Based on the pre-test cough recognition accuracy of 90%, it was assumed that the cardiac patch achieved 90% recognition in cough recognition in pneumoconiosis patients with α of 0.05 and efficacy of 80% (β of 0.2, p of 0.9, p0 of 0.72). The sample size required for the test was calculated, and at a 10% rejection rate, a total sample size of 25 was required.
25 pneumoconiosis patients were recruited from the Shanghai Pulmonary Hospital affiliated with Tongji University, and 25 healthy young adults were recruited from Hangzhou Dianzi University. This study was conducted in accordance with the guidelines of the Helsinki Declaration of the World Medical Association (2000) and was approved and supervised by the Ethics Committee of Shanghai Pulmonary Hospital (approval No. L23-226). After receiving a detailed description of the experiment, all participants signed informed consent forms. The patients’ demographic information is listed in Table 1. Among them, pneumoconiosis were staged based on X-ray chest radiographs of satisfactory technical quality as well as standard films for the diagnosis of pneumoconiosis. 19 of the patients were in stage I, 1 of the patients was in stage II and 5 of the patients were in stage III.
Table 1. The demographic information of the participants
Nature
|
Healthy adults
|
Patients
|
Number
|
25
|
25
|
Sex
|
Male: 20
|
Male: 20
|
Female: 5
|
Female: 5
|
Age
|
24(±2)
|
65(±11)
|
Devices
The experimental data were collected using flexible chest patch that integrates a tri-axial accelerometer and a single-lead ECG provided by Vivalink, Inc. of California, USA. During the experiment, the ECG patch was worn on the patient's upper and middle left chest using disposable medical backing. The ECG patch has an embedded battery which can support its continuous function for as long as three weeks, which make it suitable for remote monitoring of vital signs. The data collected by the patch were transferred to a matching smart phone through blue tooth and automatically uploaded to the clout platform for easy downloading and analysis after the experiment. The acceleration signal was sampled at 125HZ.
In the healthy adults group, the subject was required to take mocking coughs with different intensity of low, medium and high. A noise meter (SmartSensor, China) was used for the measurement of the intensity and classification. According to the noise meter, the decibel ranges of low intensity cough, medium intensity cough and high intensity cough in healthy people are 30-50dB, 50-70dB and 70-90dB respectively. In the group of pneumoconiosis patients, a Vital Signs Monitor (NT1D, China) together with the chest patch were used to monitor respiratory rate and heart rate for 10 minutes to determine whether the patient's condition was suitable for the experiment. Since audio signal is considered to be the gold standard for cough recognition, the ASR software in the SAMSUNG mobile phone was used for recording the audio information in the experiment. During the experimental phase when the patient was engaged in spontaneous coughing, the key mapper software in another SAMSUNG mobile phone was used to record the start time and the end time of each coughing session of the patient. Figure 1 shows the schematic of the experimental setup with all the devices used in the experiment, as well as the location where the ECG patch was pasted on the subject.
Experimental procedure
The experiment was divided into two stages. Healthy adults participated in the first phase of the experiment, and the behaviors collected were: breathing, deep breathing, rapid breath change, low-intensity coughing, medium-intensity coughing, high-intensity coughing, shutting up and coughing, whispering, normal talking, loud talking, yawning and sighing. The experiment was performed in the office environment. Each volunteer was required to complete three separate rounds of collection in the resting state (sitting in a chair) and in the moving state.
In the second phase, the subjects were patients with pneumoconiosis, and the behaviors collected were: breathing, deep breathing, shortness of breath, talking, yawning, sighing and coughing. The experiment was done in the ward of Shanghai Pulmonary Hospital, and each patient was required to complete three rounds of stationary state (sitting on a chair) and in the mobile state, and each round required the patient to complete 12 behaviors.
An acquisition time of 2 minutes was adopted for each behavior. When the subject was given a command for a specific behavior, he (she) need to complete in 5 seconds, and then rested for 5s before giving the next command. Figure 2 shows the entire experimental flow.
When the subject coughs, the tri-axial accelerometer can measure displacement along the X, Y, and Z axes, resulting in a distinct waveform for each direction. Figure 3 presents typical cough waveforms for healthy adults and patients with pneumoconiosis. Specifically, Figure 3(a) and (b) show the waveforms of low-intensity and high-intensity coughs in a healthy adult, respectively, revealing the distinct fluctuation ranges of intensity. Figure 3(c) and (d) present the waveforms of voluntary and spontaneous coughs in a patient, respectively, revealing a significant difference in fluctuation intensity.
Data Processing
Figure 4 illustrates the structural framework of the data processing. K-fold cross-validation was employed for all recognition tasks. The average of the k evaluation metrics served as the final evaluation criterion to mitigate the possibility of final recognition results being attributable to chance rather than being relatively stable. The metric values were normalized to expedite computation. Additionally, normalization confines the preprocessed data within a defined range, thereby mitigating the adverse effects arising from outlier data.
Data filtration was applied to ensure that the recorded information matched the actual signals, a verification made by comparing with concurrent audio signals during the experiment. 50 distinct features were extracted from the ACC signals to construct a dataset. Subsequently, feature selection was performed using three algorithms: Kruskal-Wallis, XGBoost, and Random Forest. Features were ranked based on metrics derived from each algorithm 28 out of the original 50 features were selected, and were ranked by their composite scores, as depicted in Figure 5. Effective features were then introduced into various classifiers for classification tasks.
Classification results and relevant metrics, including Accuracy, Precision, Recall, F1-scores, and AUC (there definition are given in Appendix) were obtained for the different classifiers. A decision matrix was constructed from the metrics, and two MCDM methods, namely, Technique for Order Preference by Similarity to Ideal Solution (TOPSIS) and VIseKriterijumska Optimizacija I Kompromisno Resenje (VIKOR) were employed to score and rank the classifiers. Lastly, the classification task was executed using the features selected through feature selection, and the top-ranked classifier was identified by the MCDM method. Figure 6 presents the performance evaluation criteria for the classifier and the MCDM method employed in the proposed approach. The correctness with its three sub-criteria (Accuracy, Precision, F1 score), verifies the Accuracy of the classification model's outcomes. The robustness metric (Recall) measures the classifier's capability to accurately identify positive samples. Subsequently, a high detection rate and accuracy were achieve by applying the MCDM method with appropriate weighting of each performance evaluation metric. To determine these weights, we referred to the methodology in [13] and determined the weights to be Accuracy: 0.12, Precision: 0.32, Recall: 0.32, F1 score: 0.12, AUC: 0.12.