Longitudinally tracking personal physiomes for precision management of childhood epilepsy

Abstract

Our current understanding of human physiology and activities is largely derived from sparse and discrete individual clinical measurements. To achieve precise, proactive, and effective health management of an individual, longitudinal, and dense tracking of personal physiomes and activities is required, which is only feasible by utilizing wearable biosensors. As a pilot study, we implemented a cloud computing infrastructure to integrate wearable sensors, mobile computing, digital signal processing, and machine learning to improve early detection of seizure onsets in children. We recruited 99 children diagnosed with epilepsy and longitudinally tracked them at single-second resolution using a wearable wristband, and prospectively acquired more than one billion data points. This unique dataset offered us an opportunity to quantify physiological dynamics (e.g., heart rate, stress response) across age groups and to identify physiological irregularities upon epilepsy onset. The high-dimensional personal physiome and activity profiles displayed a clustering pattern anchored by patient age groups. These signatory patterns included strong age and sex-specific effects on varying circadian rhythms and stress responses across major childhood developmental stages. For each patient, we further compared the physiological and activity profiles associated with seizure onsets with the personal baseline and developed a machine learning framework to accurately capture these onset moments. The performance of this framework was further replicated in another independent patient cohort. We next referenced our predictions with the electroencephalogram (EEG) signals on selected patients and demonstrated that our approach could detect subtle seizures not recognized by humans and could detect seizures prior to clinical onset. Our work demonstrated the feasibility of a real-time mobile infrastructure in a clinical setting, which has the potential to be valuable in caring for epileptic patients. Extension of such a system has the potential to be leveraged as a health management device or longitudinal phenotyping tool in clinical cohort studies.

Fig 1. Overview of the LOOP system.

The Microsoft Band wristband device is connected via Bluetooth to a nearby smartphone, and the sensor data is transmitted to the smartphone at the real time. The smartphone has a pre-installed dedicated app, which allows the caregiver to record it when the child is having a seizure or other unusual events. This smartphone is also connected to a remote central data server in the cloud via Wi-Fi or cellular network; the recorded data is stored on the remote server for storage and analysis. The smartphone app has an intuitive user interface, allowing the user to access and display historical data, either recent data stored locally on the cell phone or remotely on the cloud server.

Fig 2. Physiological dynamics associated with childhood development.

(A) Multiple Linear Regression (MLR) analysis conducted on 4 physiological signals (HR, GSR, ACC magnitude and SDNN variables in HRV analysis) by controlling for age, sex, and the time of the day. (B) HR signals associated with childhood development. (C) SDNN signals associated with childhood development. (D) GSR signals associated with childhood development. (E) HR signals associated with childhood development during the time of the day. (F) SDNN signals associated with childhood development during the time of the day.

Fig 3. Clustering and visualization of physiological features along the axes of gender and age groups.

Each data point represents a vector of a subjects’ HR, ACC, GSR and GYR data. (A) Clustering by Principal Component Analysis (PCA), (B) Clustering results after applying UMAP (Uniform Manifold Approximation and Projection).

Fig 4. Physiological dynamics associated with seizure events.

(A). Top panel: Heart rate measurements from a patient in a period of 600 seconds. The measurements during a short time seizure are marked in red, while the signals in the normal periods are marked in blue. Middle panel: The time series measurements were converted by STFT to frequency domain. Bottom panel: EEG signals in the same period are shown in 21 channels. (B) Zoomed in view of EEG signal in time interval 1 (TL1). (C) Zoomed in view of EEG signal in time interval 2 (TL2). (D) The measurement of HR, GSR, ACC during seizure significantly deviated from normal periods. (E). The measurement of SDNN, rMSSD, HF/LF during seizure significantly deviated from normal periods.

Fig 5. Performance of the machine learning model.

(A) ROC curve of 10-fold leave-one-seizure-out cross-validation on 66 patients, the threshold is marked in red. (B) ROC curve of real-time testing on recorded seizure events of 11 selected subject in the validation cohort, the threshold is marked in red. (C) Precision-Recall curves of the prediction accuracy. (D) Precision-Recall curves of the validation cohort.

Fig 6. Real-time monitoring during a seizure event.

(A) Real-time monitoring during a seizure event. The red horizontal bar in the grey box represents a true seizure event, the dark yellow horizontal bars represent “false alarms”, and the blue horizontal bars represent true negative events. (B). EEG signals during a seizure event. (C). Zoomed in view of EEG signals during the seizure event (TI1).

Fig 7. The App interface of LOOP system.