Bilateral acoustic sensing for predicting FEV1/FVC
A wearable system utilizing bilateral acoustic sensing for noninvasive monitoring and prediction of forced expiratory volume in one second (FEV1) and forced vital capacity (FVC) in patients with chronic obstructive pulmonary disease (COPD). The system collects and analyzes breathing sounds from both lungs, extracts relevant passive acoustic features, and employs machine learning algorithms to predict FEV1 or FVC values without requiring the subject to perform any forced expiratory maneuvers.
1 . A system for noninvasive monitoring and prediction of lung function in a patient with chronic obstructive pulmonary disease (COPD), the system comprising:
a wearable harness having bilateral sensors, a first and second of which are configured, when the harness is worn by the patient, to be proximate to a left lung and a right lung of the patient, respectively, each sensor including:
an acoustic sensor configured to produce an acoustic signal based on sounds in an environment of the acoustic sensor,
an inertial change sensor configured to produce an inertial change signal based on a change in inertia of the inertial change sensor,
a microprocessor coupled to the acoustic sensor and inertial change sensor, configured to process the acoustic signal into phonocardiogram data and integrate the inertial change signal into z-axis data, said data comprising sensor data, and
a wireless communication module coupled to the microprocessor, configured to transmit sensor data via a communications channel; and a remote server, configured to receive the sensor data from the bilateral sensors via the communications channel and configured with a machine learning (ML) algorithm to perform feature extraction from the sensor data by applying frequency analysis and normalization to said sensor data, including extraction of passive acoustic features Respiratory Rate (RR), Inhalation Duration (ID), Exhalation Duration (ED), Mean Airflow Velocity (AFV), Duration of Pauses (DP), Spectral centroid (SC), and Spectral bandwidth (SB), and prediction of a forced expiratory volume in one second (FEV1) and a forced vital capacity (FVC) of the patient based on the extracted features, wherein said ML algorithm has been trained with spirometry data.
2 . The system of claim 1 , wherein the remote server is further configured to compute a phonopulmogram (PPLG) waveform by combining the phonocardiogram data and the z-axis data from the bilateral sensors and to generate a representation of respiratory cycles and I: E ratios of the patient and wherein the bilateral sensors are adaptable for continuous monitoring in clinic and nonclinic settings.
3 . The system of claim 1 , wherein the acoustic sensor comprises an array of microphones.
4 . The system of claim 1 , wherein the inertial change sensor comprises a three axis accelerometer.
5 . The system of claim 1 , wherein the remote server is further configured to determine breathing pattern changes over a selectable period of time, wherein the selectable period of time may be selectable from periods of days, weeks, months, and years or bounded by a specified start date and by a specified end date.
6 . The system of claim 5 , wherein the remote server is further configured to identify exacerbated conditions related to the patient's COPD, including increased cough, shortness of breath, and changes in sputum production.
7 . The system of claim 1 , wherein the ML algorithm is trained using data gathered during a spirometry session of the patient and with spirometry data from multiple individuals with varying levels of COPD severity.
8 . The system of claim 7 , wherein the remote server is further configured to provide a reinforcement learning agent, trained to identify relevant features in the patient's sensor data for monitoring daily activities of the patient, and the identified features are used to develop a personalized model for the patient and wherein the reinforcement learning agent is rewarded for accurately predicting daily activities of the patient based on the data collected from the bilateral sensors using a Q-learning algorithm.
9 . The system of claim 8 , wherein the personalized model for the patient is continuously updated using learning algorithms based on data collected from the bilateral sensors.
10 . The system of claim 7 , wherein the remote server is further configured to determine and transmit a comprehensive assessment of the patient's functional status derived from the sensor data of the patient to a health care provider's dashboard.
11 . The system of claim 7 , further comprising a patient mobile device configured with a portal app to provide the patient with real-time feedback and visualizations of lung metrics, including real-time FEV 1 /FVC, I: E Ratio, RR, and HR outputs based on the sensors' data.
12 . A system for noninvasive monitoring and prediction of lung function in a patient with chronic obstructive pulmonary disease (COPD), the system comprising:
a wearable harness having first and second sensors configured, when the harness is worn by the patient, to be proximate to a left lung and a right lung of the patient, respectively, each sensor including:
an acoustic sensor configured to produce an acoustic signal based on sounds in an environment of the acoustic sensor,
an inertial change sensor configured to produce an inertial change signal based on a change in inertia of the inertial change sensor,
a microprocessor coupled to the acoustic sensor and inertial change sensor, configured to process the acoustic signal into phonocardiogram data and integrate the inertial change signal into z-axis data, said data comprising sensor data, and
a wireless communication module coupled to the microprocessor, configured to transmit sensor data via a communications channel; and
a remote server, configured to receive the sensor data from each of the first and second sensors via the communications channel and configured with a machine learning (ML) algorithm to perform feature extraction from the sensor data by applying frequency analysis and normalization to said sensor data, including extraction of passive acoustic features Respiratory Rate (RR), Inhalation Duration (ID), Exhalation Duration (ED), Mean Airflow Velocity (AFV), Duration of Pauses (DP), Spectral centroid (SC), and Spectral bandwidth (SB), and prediction of a forced expiratory volume in one second (FEV1) and a forced vital capacity (FVC) of the patient based on the extracted features, wherein said ML algorithm has been trained with spirometry data;
further wherein the remote server is further configured to compute a phonopulmogram (PPLG) waveform by combining the phonocardiogram data and the z-axis data from the first and second sensors and to generate a representation of respiratory cycles and I: E ratios of the patient and wherein the first and second sensors are adaptable for continuous monitoring in clinic and nonclinic settings; and
further wherein the prediction of FEV1 and FVC are based on the PPLG waveform, I: E ratios, and acoustic features extracted from the phonocardiogram data, according to the equation:
FEV
1
/
FVC
=
b
0
+
b
1
·
RR
+
b
2
·
ID
+
b
3
·
ED
+
b
4
·
AFV
+
b
5
·
DP
+
b
6
·
SC
+
b
7
·
SB
.
13 . A method for monitoring and prediction of lung function in a patient with chronic obstructive pulmonary disease (COPD) comprising the steps of:
disposing a wearable harness having first and second bilateral sensors configured, when the harness is worn by the patient, to be proximate to a left lung and a right lung of the patient, respectively, each sensor including:
an acoustic sensor configured to produce an acoustic signal based on sounds in an environment of the acoustic sensor,
an inertial change sensor configured to produce an inertial change signal based on a change in inertia of the inertial change sensor,
a microprocessor coupled to the acoustic sensor and inertial change sensor, configured to process the acoustic signal into phonocardiogram data and integrate the inertial change signal into z-axis data, said data comprising sensor data, and
a wireless communication module coupled to the microprocessor, configured to transmit sensor data via a communications channel;
transmitting the sensor data, by each of the wireless communications modules of the bilateral sensors;
receiving, by a remote server, the sensor data from each of the bilateral sensors via the communications channel; and
processing, by the remote server, the sensor data with a machine learning (ML) algorithm to perform feature extraction from the sensor data by applying frequency analysis and normalization to said sensor data, including extraction of passive acoustic features Respiratory Rate (RR), Inhalation Duration (ID), Exhalation Duration (ED), Mean Airflow Velocity (AFV), Duration of Pauses (DP), Spectral centroid (SC), and Spectral bandwidth (SB), and to predict a value representing a forced expiratory volume in one second (FEV1) and forced vital capacity (FVC) of the patient, wherein said ML algorithm has been trained with spirometry data.
14 . The method of claim 13 further comprising a step of transmitting the patient's predicted FEV 1 and FVC to a healthcare provider's device.
15 . The method of claim 13 where said step of processing further comprises analyzing collected data to extract relevant passive acoustic features and predict forced expiratory volume in one second (FEV1) and forced vital capacity (FVC).
16 . The method of claim 13 further comprising a step of training the ML algorithm with spirometry data from multiple individuals with varying levels of COPD severity and with data gathered during a spirometry session of the patient.
17 . The method of claim 13 further comprising encrypting the sensor data prior to the transmitting step to create encrypted sensor data and decrypting the encrypted sensor data after the receiving step.
18 . The method of claim 13 further comprising steps of determining breathing patterns changes or indicated exacerbations in the patient and transmitting the results of the determination step to a healthcare provider's device.
19 . The method of claim 13 further comprising steps of determining a functional status of the patient and transmitting the result of the determination step to a healthcare provider's device.
20 . The method of claim 13 , further comprising a step of providing the patient a portal app for execution on a patient mobile device, said portal app configured to provide the patient with real-time feedback and visualizations of lung metrics based on the sensors' data.