Interpreting Deep Neural Networks for Single-Lead ECG Arrhythmia Classification

Cardiac arrhythmia is a prevalent and significant cause of morbidity and mortality among cardiac ailments. Early diagnosis is crucial in providing intervention for patients suffering from cardiac arrhythmia. Traditionally, diagnosis is performed by examination of the Electrocardiogram (ECG) by a cardiologist. This method of diagnosis is hampered by the lack of accessibility to expert cardiologists. For quite some time, signal processing methods had been used to automate arrhythmia diagnosis. However, these traditional methods require expert knowledge and are unable to model a wide range of arrhythmia. Recently, Deep Learning methods have provided solutions to performing arrhythmia diagnosis at scale. However, the black-box nature of these models prohibit clinical interpretation of cardiac arrhythmia. There is a dire need to correlate the obtained model outputs to the corresponding segments of the ECG. To this end, two methods are proposed to provide interpretability to the models. The first method is a novel application of Gradient-weighted Class Activation Map (Grad-CAM) for visualizing the saliency of the CNN model. In the second approach, saliency is derived by learning the input deletion mask for the LSTM model. The visualizations are provided on a model whose competence is established by comparisons against baselines. The results of model saliency not only provide insight into the prediction capability of the model but also aligns with the medical literature for the classification of cardiac arrhythmia.Clinical relevance- Adapts interpretability modules for deep learning networks in ECG arrhythmia classfication, allowing for better clinical interpretation.

PPGnet: Deep Network for Device Independent Heart Rate Estimation from Photoplethysmogram

Photoplethysmogram (PPG) is increasingly used to provide monitoring of the cardiovascular system under ambulatory conditions. Wearable devices like smartwatches use PPG to allow long-term unobtrusive monitoring of heart rate in free-living conditions. PPG based heart rate measurement is unfortunately highly susceptible to motion artifacts, particularly when measured from the wrist. Traditional machine learning and deep learning approaches rely on tri-axial accelerometer data along with PPG to perform heart rate estimation. The conventional learning based approaches have not addressed the need for device-specific modeling due to differences in hardware design among PPG devices. In this paper, we propose a novel end-to-end deep learning model to perform heart rate estimation using 8-second length input PPG signal. We evaluate the proposed model on the IEEE SPC 2015 dataset, achieving a mean absolute error of 3.36±4.1BPM for HR estimation on 12 subjects without requiring patient-specific training. We also studied the feasibility of applying transfer learning along with sparse retraining from a comprehensive in-house PPG dataset for heart rate estimation across PPG devices with different hardware design.

Effectiveness of a continuous patient position monitoring system in improving hospital turn protocol compliance in an ICU: A multiphase multisite study in India

Purpose

Hospital-acquired pressure ulcers are a significant cause of morbidity and consume considerable financial resources. Turn protocols (repositioning patients at regular intervals) are utilized to reduce incidence of pressure ulcers. Adherence to turn protocols is particularly challenging for nursing teams, given the high number of interventions in intensive care unit, and lack of widely available tools to monitor patient position and generate alerts. We decided to develop and evaluate usefulness of a continuous patient position monitoring system to assist nurses in improving turn protocol compliance.

Methods

We conducted a prospective, non-randomized, multiphase, multicentre trial. In Phase I (control group), the function of the device was not revealed to nurses so as to observe their baseline adherence to turn protocol, while Phase II (intervention group) used continuous patient position monitoring system to generate alerts, when non-compliant with the turn protocol. All consecutive patients admitted to one of the two intensive care units during the study period were screened for enrolment. Patients at risk of acquiring pressure ulcers (Braden score < 18) were considered for the study (Phase I (N = 22), Phase II (N = 25)).

Results

We analysed over 1450 h of patient position data collected from 40 patients (Phase I (N = 20), Phase II (N = 20)). Turn protocol compliance was significantly higher in Phase II (80.15 ± 8.97%) compared to the Phase I (24.36 ± 12.67%); p < 0.001.

Conclusion

Using a continuous patient position monitoring system to provide alerts significantly improved compliance with hospital turn protocol. Nurses found the system to be useful in providing automated turn reminders and prioritising tasks.

Clinical Validation of a Wearable Respiratory Rate Device for Neonatal Monitoring

Respiratory rate monitoring is of paramount importance in neonatal care. Manual counting of expansions and contractions of the abdomen or diaphragm of the neonate is still the widely accepted measure of respiratory rate in most clinical settings. A practical, affordable, easy-to-use technology to continuously measure respiratory rate in neonates is essential to recognize the signs and symptoms of respiratory disorders. Clinical validation of a system for continuous and long term respiratory rate monitoring of neonates, in a wearable form factor with the capability of remote monitoring is presented in this paper. The respiratory rate monitor was validated in clinical settings on 10 premature babies with various disease conditions and respiratory rates varying from 25 to 90 breaths per minute. Results show a high degree of correlation between the respiratory rate measured by the device and reference measurements. An intelligent algorithm which can remove motion corruption from the accelerometer data and provide reliable results is essential for large-scale adoption of the technology for both clinical as well as home monitoring. The technical details of implementation, results and analysis of the clinical study and observations made during clinical study regarding the feasibility of integrating the device in neonatal care are covered in this paper.

RespNet: A deep learning model for extraction of respiration from photoplethysmogram

Respiratory ailments afflict a wide range of people and manifests itself through conditions like asthma and sleep apnea. Continuous monitoring of chronic respiratory ailments is seldom used outside the intensive care ward due to the large size and cost of the monitoring system. While Electrocardiogram (ECG) based respiration extraction is a validated approach, its adoption is limited by access to a suitable continuous ECG monitor. Recently, due to the widespread adoption of wearable smartwatches with in-built Photoplethysmogram (PPG) sensor, it is being considered as a viable candidate for continuous and unobtrusive respiration monitoring. Research in this domain, however, has been predominantly focussed on estimating respiration rate from PPG. In this work, a novel end-to-end deep learning network called RespNet is proposed to perform the task of extracting the respiration signal from a given input PPG as opposed to extracting respiration rate. The proposed network was trained and tested on two different datasets utilizing different modalities of reference respiration signal recordings. Also, the similarity and performance of the proposed network against two conventional signal processing approaches for extracting respiration signal were studied. The proposed method was tested on two independent datasets with a Mean Squared Error of 0.262 and 0.145. The cross-correlation coefficient of the respective datasets were found to be 0.933 and 0.931. The reported errors and similarity was found to be better than conventional approaches. The proposed approach would aid clinicians to provide comprehensive evaluation of sleep-related respiratory conditions and chronic respiratory ailments while being comfortable and inexpensive for the patient.

Differential effects of physical and psychological stressors on electrodermal activity

Stress being labelled by WHO as "the health epidemic of 21^st century" need to be treated as a clarion call for devising strategies that aim at its early detection, for the reason that stress is the cause as well as the catalyst for several chronic human health disorders. The work reported here in is a progression towards the development of a stress detection system based on the electrodermal activity (EDA) in humans, which can further be incorporated into a wearable vital signs monitor. The utility of EDA as a potential physiological measure for classifying physical and psychological stressors is analyzed in this paper. A group of 12 subjects (8 males and 4 females, age: 25.4 ± 3.1 years, mean ± SD) volunteered to participate in a laboratory stress task that included a psychological stressor close to real life work stress scenario and a physical stressor. The capability of stressors to elicit persistent stress response was validated by assessing variations in salivary cortisol levels. EDA was monitored throughout the experiment sessions as a measure of sympathetic activation in subjects. Six classification models were investigated concerning their usability to distinguish physical and psychological stressors based on EDA. A maximum accuracy of 95.1% was achieved using linear discriminat analysis (LDA) based classifier which imply that EDA is indeed a potential discriminate measure to classify physical and psychological stress responses. Furthermore, the best feature combination for maximum classification accuracy was also determined.

A wrist worn SpO2 monitor with custom finger probe for motion artifact removal

Continuous monitoring of blood oxygen saturation (SpO2) level and heart rate is critical in surgery, ICUs and patients suffering from Chronic Obstructive Pulmonary Diseases. Pulse oximeters which compute SpO2 using transmittance photoplethysmography (PPG), is widely accepted for continuous monitoring. Presence of motion artifacts in PPG signals is a major obstacle in the extraction of reliable cardiovascular parameters, in real time and continuous monitoring applications. In this paper, a wrist worn device with a custom finger probe with an integrated accelerometer to remove motion artifacts is presented. An algorithm which can run on low power systems with processing constraints is implemented on the device. The device does continuous acquisition of PPG and accelerometer waveforms and computes SpO2 using the proposed light weight algorithm. The measurement results are continuously synced with an Android tablet, which acts as a gateway and is pushed on to the cloud for further analysis. The accuracy in SpO2 measured by the device was validated using Fluke ProSim 8 SpO2 simulator and the efficiency in accurately computing SpO2 in the presence of motion was validated over 40 healthy volunteers in a controlled setting.

A novel system to tackle hospital acquired pressure ulcers

Hospital acquired pressure ulcers (HAPUs) is a major problem that affects around one in twenty patients who are admitted in hospital with sudden illness. These ulcers often occur when patients have limited mobility and cannot change positions in bed on their own. Traditionally, the occurrence of HAPUs has been minimized by turning the patient every 2 hours to alternating lateral and supine positions, and by using pressure redistributing mattresses. In many healthcare facilities, such a patient repositioning schedule is not always maintained owing to low caregiver compliance to turning protocols. Difficulty in monitoring patient position continuously, lack of turn reminders/alerts and suboptimal caregiver staffing ratio increases the occurrence of HAPUs. A novel method to address the need for improved pressure ulcer prevention is presented. The proposed method consists of a wearable device which continuously monitors the patient's position and communicates wirelessly with a tablet which enables alerts to be sent to the caregiver when a patient turn is due in accordance with the protocol adopted by the hospital. The patient's position is continuously monitored and the turning procedure carried out is logged and updated on the hospital's cloud system, thereby enabling centralized monitoring. Under a controlled setting, system was able to continuously monitor patient's position and can accurately detect standard patient positions.

Wearable ECG platform for continuous cardiac monitoring

An ultra-low power ECG platform for continuous and minimally intrusive monitoring for systems with minimal processing capabilities, is presented in this paper. The platform is capable of detecting abnormalities in the ECG signal by extracting and analyzing features related to various cardiac trends. The platform is built to continuously operate on any of the 12 leads and the presented work includes a single lead implementation that works on lead I or II. A single lead, wearable ECG patch that can detect rhythm based arrhythmias and continuously monitor beat-to-beat heart rate and respiratory rate has been developed. In addition, the device stores raw ECG waveform locally and is designed to run for 10 days on a single charge. The ECG patch works in conjunction with a front end device or tablet and updates the results on the tablet interface. Upon detection of an abnormality or an arrhythmia the device switches to an ECG visualization mode enabling manual analysis on the acquired signal. The front end device also functions as a gateway for remote monitoring. The functionality and processing capabilities of the platform along with the validation tests carried out in a controlled setting are presented.