Raspberry Pi-based Data Archival System for EEG Signals from the SedLine^® Root Device

Abstract

BACKGROUND:

The retrospective analysis of EEG signals acquired from patients under general anesthesia is crucial in understanding the patient’s unconscious brain’s state. However, the creation of such database is often tedious, cumbersome and involves human labor. Hence, we developed a Raspberry Pi-based system for archiving Electroencephalogram (EEG) signals recorded from patients under anesthesia in operating rooms (ORs) with minimal human involvement.

METHODS:

Using this system, we archived patient EEG signals from over 500 unique surgeries at the Emory University Orthopaedics and Spine Hospital, Atlanta, USA for around 18 months. For this, we developed a software package that runs on a Raspberry Pi and archives patient EEG signals from a SedLine^® Root EEG Monitor (Masimo, Irvine, CA, USA) to a secure Health Insurance Portability and Accountability Act (HIPAA) compliant cloud storage. The OR number corresponding to each surgery was archived along with the EEG signal to facilitate retrospective EEG analysis. We retrospectively processed the archived EEG signals and performed signal quality checks. We also proposed a formula to compute the proportion of true EEG signal and calculated the corresponding statistics. Further, we curated and interleaved patient medical record information with the corresponding EEG signals.

RESULTS:

We retrospectively processed the EEG signals to demonstrate a statistically significant negative correlation between the relative alpha power (8–12Hz) of the EEG signal captured under anesthesia and the patient’s age.

CONCLUSIONS:

Our system is a standalone EEG archiver developed using low cost and readily available hardware. We demonstrated that one could create a large-scale EEG database with minimal human involvement. Moreover, we showed that the captured EEG signal is of good quality for retrospective analysis and combined the EEG signal with the patient medical records. This project’s software has been released under an open-source license to enable others to use and contribute.

INTRODUCTION

Over the past century, technology has played a key role in medicine, particularly in high-intensity environments such as operating rooms (ORs). Despite this, much of the equipment is poorly integrated for research, and many key events in the clinical environment remain undocumented with any level of spatiotemporal precision. One well-known signal that is often captured in isolation in the OR is the electroencephalogram (EEG), which captures the brain’s electrical activity to assess the patient’s brain’s state.

The retrospective analysis of EEG signals captured from unconscious patients undergoing surgery in conjunction with their patient medical records is highly beneficial in investigating potential correlations between the physiological (EEG) characteristics and medical history of the patients. For instance, Kreuzer et al.¹ explored the differences in EEG features in patients under sevoflurane anesthesia with respect to the patient age. The work by Chander et al.² investigated EEG variations during end maintenance and emergence and provided a systematized nomenclature for tracking brain states under general anesthesia from maintenance to emergence. In all these works, researchers strove to understand the brain’s state during anesthesia maintenance and emergence and its effect on post-operative outcomes, including delirium and pain. Therefore, creating an extensive database of EEG recordings from patients under anesthesia has wide-ranging applications.

The SedLine^® Root EEG Monitor (Masimo, Irvine, CA, USA) (Root) is a brain function monitor that helps anesthesiologists in the titration and maintenance of anesthetic drugs to patients in the OR.^3,4,5 It captures patient EEG from the patient’s forehead, specifically the Fp1, Fp2, F7, and F8 channels. The monitor then computes a proprietary metric known as the Patient State Index (PSi), a processed EEG parameter related to the anesthetic agents’ effect on the patient’s brain.

This study’s objective was to develop a system to archive raw EEG signals from the Root monitor together with quality metrics with minimal human involvement. Our EEG data archival system comprised a Universal Serial Bus (USB) switch connecting a Raspberry Pi to the Root monitor. The Raspberry Pi facilitated data-upload to a secure cloud-mediated database. We deployed this system at the Emory University Orthopaedics and Spine Hospital (Atlanta, GA, USA) to collect EEG signals from patients undergoing surgery. Further, we retrospectively processed the archived EEG signals. We calculated the EEG clipping ratio (eCR), interleaved patient medical records with the EEG recordings, and studied the variation of relative alpha power in the EEG signal with the patient age.

METHODS

This study was approved by the Emory University Institutional Review Board (IRB00070900) – ‘Probing the overlap between sleep and anesthesia to enhance human cognition’. The requirement for written informed consent was waived by the IRB. This manuscript was prepared in accordance with the Strengthening the Reporting of Observational Studies in Epidemiology for the improved reporting of observational studies.⁶

EEG Archiving System

The Root monitor is a data capture and display unit. In conjunction with the SedLine^® EEG sensor, the Root monitor allows for capturing four-channel EEG signal from the patient’s forehead and displaying it. A signal processing engine that resides in the monitor processes this EEG signal. It generates a real-time signal spectrogram and computes the PSi - a processed EEG parameter related to the anesthetic agents’ effect on the patient’s brain’s state.

The Raspberry Pi 3 Model B is a $35 computer that is about the size of a deck of cards.⁷ It functioned as the central hub in our data collection pipeline. It was released in February 2016⁸ and was the most current model during the project’s software development stage. The Debian-based operating system called Raspbian-Jessie that is optimized⁹ for the Raspberry Pi was installed for the developers and users to interact with the hardware. The USB sharing switch is a plug and play device that provides the capability to operate multiple USB ports simultaneously. In our work, we used the Sabrent USB 2.0 Sharing Switch to facilitate the transfer of EEG records from Root monitor to Raspberry Pi. We used the USB sharing switch in conjunction with a 64GB flash-drive that functioned as the intermediary data storage unit. Among the onboard peripherals on the Raspberry Pi are four USB A 2.0 host ports for data transfer, 40 General Purpose Input-Output (GPIO) pins, and a micro-USB B power connector. We used the USB 2.0 port to connect the USB sharing switch to Raspberry Pi and the GPIO pins to connect a 3.5-inch display to the Raspberry Pi. We powered the Raspberry Pi via the micro-USB B power connector with a 5V 2.5A power adapter.

Figure 1 illustrates the data archival system. The SedLine^® EEG sensor (Fig. 1A (b)) connected to the patient’s head acquired EEG signal from the frontal lobe of the brain. This data was transmitted to the Root monitor (Fig. 1A (a)) for PSi computation, EEG storage and visualization. The Sabrent USB 2.0 sharing switch (Fig. 1A (e)) was connected to both the Root monitor and the Raspberry Pi (Fig. 1A (c)) forming a bridge between the two devices. A 64 GB flash-drive (Fig. 1A (d)) connected to the USB sharing switch acted as an intermediary data storage device. The data flow direction is illustrated in Figure 1B. Once the data was transferred to the Raspberry Pi, utilizing the graphical user interface (GUI) displayed on the screen attached to the Raspberry Pi, users could upload the data to a cloud storage at a click of a button. We utilized the Health Insurance Portability and Accountability Act (HIPAA) complaint Box cloud storage in our project to store the EEG records. We chose the Box service over other cloud storage services as we had a subsidized access to it through our university. The Box cloud storage can be replaced with Amazon Web Services’ Simple Storage Service (S3) or Google cloud or Microsoft OneDrive by modifying the backup script in our software.

Figure 1.

(A) The data archival system setup comprised of (a) SedLine^® Root Monitor (b) SedLine^® EEG sensor (c) Raspberry Pi with a touchscreen display (d) 64 GB flash-drive (e) Sabrent USB 2.0 sharing switch. A Python 3.7.3 based graphical user interface running on the Raspberry Pi facilitated user interaction with the data archival system. The exemplar subject in the figure is one of the authors. (B) The data flow diagram illustrating the movement of EEG signal from the SedLine^® EEG sensor to the Raspberry Pi for cloud storage. The letters a, b, c, d, and e correspond to the labels as marked on Fig. 1A. The EEG signal recorded using the sensor was first stored by the SedLine^® Root monitor. The USB sharing switch when clicked connected the SedLine^® monitor to the 64 GB flash-drive and triggered the data transfer from the monitor to the flash-drive. Another click on the sharing switch then connected the flash-drive to the Raspberry Pi. The GUI on the Raspberry Pi then allowed users to upload this data to a cloud storage. Image (c) Emory University, CC-BY-SA.

Software for EEG Data Archiving

The associated software that runs on the Raspberry Pi was written using a combination of Python 3.7.3 and shell scripting. A Python based GUI functioned as the interface for users to interact with the data archival system. The GUI contained the following four software buttons which could be clicked by the users via the touchscreen attached to the Raspberry Pi: (1) Backup; (2) Record; (3) Visualize; (4) Close. The Backup button when clicked, checked if there was EEG data available for upload and if available then commands were automatically run to upload data to a cloud storage. The Record and Visualize buttons were provided for recording and visualizing audio signals which were not utilized in this project. The Close button when clicked closed the GUI display. Further, we provided various MATLAB scripts to process the archived EEG signals. This included scripts to combine EEG recordings with the corresponding patient information and to replicate the results presented in this work. Our software is made available under a Berkeley Software Distribution (BSD) license.

Data Collection

All EEG and patient data presented in this paper were collected at the Emory University Orthopaedic and Spine Hospital (EUOSH) Atlanta, USA. We collected data from seven ORs at this facility where patients underwent a host of orthopedics and spine surgeries. At the beginning of surgery, general anesthesia was administered to the patients, and EEG was collected before the administration of anesthesia until the culmination of the surgery via the Root monitor. Our Raspberry Pi based EEG auto-archival system was designed to archive these EEG signals. The EEG archiving was performed three times between October 2017 and March 2019 – specifically in (1) October 2017; (2) March 2018; and (3) March 2019. The data archival process during each of the three visits was as follows. First, the researcher approached and switched ON a Root device in the OR which was used for monitoring PSi by anesthesiologists during surgeries. Once the monitor was live, the button on the USB sharing switch was pressed to connect the Root monitor to the 64 GB flash-drive. The following message appeared on the Root monitor’s display: “Copying EDF data has started”. The EEG data in the monitor was transferred automatically by the monitor to a folder named edf in the 64 GB flash-drive. Once all the data was transferred to the flash-drive, the following message appeared on the display of the Root monitor: “Transferring EDF data is done”. Next, the researcher pressed the button on the USB sharing switch to now connect the flash-drive to the Raspberry Pi. The GUI in the Raspberry Pi was opened and the Backup button was clicked. This transferred all EEG data in the edf folder of the flash-drive to the cloud storage via Wi-Fi. This procedure was repeated for all the Root monitors at EUOSH. Previously, in order to extract EEG recordings from the Root monitor one had to physically plug a flash-drive to the back of the monitor, extract the recordings, attach the flash-drive to a computer to upload this data to a local drive or a cloud drive. Our method improved upon this process by providing a Raspberry Pi based solution. We automated the process of transferring EEG records to a cloud storage via Wi-Fi. However, we still needed a person to click the button on the USB sharing switch to extract the data from the monitors onto the Raspberry Pi. Hence, our solution is semi-automated.

The Root monitor provided EEG signal data in two formats: (a) High-resolution encrypted data format, (b) Binary unencrypted European Data Format (EDF). We did not have access to the decryption key to decrypt the high-resolution encrypted data, and as a result, we used the EDF data in our analyses. While high-resolution encrypted data corresponded to the raw underlying EEG signal, EDF data, however, was the signal as visible on the Root monitor screen. As a result, based on the display resolution setting on the monitor screen, different recordings possibly had different physical maximums and physical minimums. The signal appeared clipped at locations where the raw data surpassed the physical maximums or the physical minimums. Further, the data from this device did not provide positive and negative deflections of the zero. Table 1 lists the different physical maximum and minimum values for different EEG channels for different display resolutions in our dataset. Von Dincklage et al.¹⁰ provide a comprehensive analysis on the effects of the display settings on the changes in EEG amplitude, sampling rate, and signal quality that can occur in Root monitor. The EEG signals had a sampling frequency equal to either 178Hz or 63Hz based on an internal setting in the Root monitor¹⁰. Hence, all EEG signals were resampled to 100Hz by upsampling the 63Hz signals and downsampling the 178Hz signals using a finite impulse response antialiasing lowpass filter.

Table 1:

Different physical maximum (p-max) and physical minimum (p-min) values of EEG signals in our database. The numerical span of possible voltage measurements is provided for all channels. The total range is given by “p-max – p-min” which is the same for all four channels. However, the individual p-max and p-min values vary for the EEG channels due to the arrangement of Fp1, Fp2, F7 and F8 channels one below the other in the monitor display. The monitor display setting is manually set by the anesthesiologists and thus vary for different recordings.

Monitor Display Settings (μV/mm)	Total Range (μV)	Fp1		Fp2		F7		F8
		p-min (μV)	p-max (μV)	p-min (μV)	p-max (μV)	p-min (μV)	p-max (μV)	p-min (μV)	p-max (μV)
3	160.5	−139.7	20.8	−99.7	60.8	−59.7	100.8	−19.7	140.8
5	250.8	−218.3	32.5	−155.8	95.0	−93.3	157.5	−30.8	220.0
10	501.7	−436.7	65.0	−311.7	190.0	−186.7	315.0	−61.7	440.0

EEG Signal Quality Analysis

For each EEG recording in the database, we computed the lengths of flat-line signal, clipping signal, and true EEG signal. The flat-line signal was present in EEG signal recordings when EEG sensors were connected to the monitor but were not attached to the patient’s head. The EEG recordings contained a clipping signal when the signal displayed on the monitor was clipped due to display resolution constraints.

Let a given EEG recording be denoted as S_eeg. First, we removed the flat-line signal at the beginning and end of the recording in all four channels. Let this signal be denoted as S_eeg′. Next, we computed the length of S_eeg′ denoted by L_Total, the length of remaining flat-line signal in S_eeg′ denoted by L_Flat, and the length of clipping signal in S_eeg′ denoted by L_Clip. The length of true EEG signal was computed using the following expression:

$$L_{\mathit{\text{True}}}=L_{\mathit{\text{Total}}}-(L_{\mathit{\text{Flat}}}+L_{\mathit{\text{Clip}}})$$

The EEG clipping ratio (eCR) for an EEG recording was computed as the ratio of length of clipped EEG signal and the total length of EEG signal, i.e.

$$eCR=\frac{L_{\mathit{\text{Clip}}}}{L_{\mathit{\text{Total}}}}$$

The EEG flat-line ratio (eFR) for an EEG recording was computed as the ratio of length of flat-line EEG signal and the total length of EEG signal, i.e.

$$eFR=\frac{L_{\mathit{\text{Flat}}}}{L_{\mathit{\text{Total}}}}$$

Thus, the proportion of true EEG signal (eTR) given by the ratio of length of true EEG signal and the total length of EEG signal was computed as

$$eTR=\frac{L_{\mathit{\text{True}}}}{L_{\mathit{\text{Total}}}}=1-\frac{L_{\mathit{\text{Clip}}}}{L_{\mathit{\text{Total}}}}-\frac{L_{\mathit{\text{Flat}}}}{L_{\mathit{\text{Total}}}}=1-eCR-eFR$$

During the data collection period (October 2017 to March 2019) the shortest surgery duration was 32 minutes. Thus, we set L_Total = 30 minutes as a threshold to filter EEG records, and we discarded EEG records with a total length less than 30 minutes. Moreover, to make sure at least 50% of EEG record contained true EEG signal, we set eTR = 0.50 as a threshold to filter EEG records further, and we discarded EEG records with eTR ≤ 0.50.

Aggregating Patient Medical Information

For each surgery for which we had an accompanying EEG record, we extracted various corresponding patient information. First, we obtained the patient age and gender. Next, we accessed the time at which the patient entered the operating room (T_Enter) and exited the operating room (T_Exit). The duration of patient presence in surgery room (D_Presence) was calculated as D_Presence = T_Exit − T_Enter. Next, we obtained the start (T_A−Start) and stop (T_A−Stop) times of patient’s general anesthesia and computed the corresponding duration of Anesthesia (D_Anesthesia) as D_Anesthesia = T_A−Stop − T_A−Start. Next, we obtained the start (T_S−Start) and stop (T_S−Stop) times of the surgery and computed the duration of surgery (D_Surgery) as D_Surgery = T_S−Stop − T_S−Start Finally, we obtained the American Society of Anesthesiologists (ASA) physical status and the body weight of the patient. We aggregated the various durations listed above, the age, body weight and ASA physical status for all the patients, and we computed the following statistics for each of them: the minimum value, the 25^th percentile, the median value, the 75^th percentile and the maximum value. Moreover, we computed these statistics by segregating the patients into female and male groups. Further, we accessed and aggregated various comorbidities and surgery names associated with each patient in the cohort.

Variation of Relative Alpha Power with Age

For this experiment, we used the EEG signal from the Fp2 channel for all EEG records. We manually selected one 10-minute segment per surgery in the Fp2 channel by visually inspecting the signal and choosing a continuous region with no flat-line or clipping signal. This 10-minute segment was present within the start and stop time of the corresponding surgery. This process reduced the number of available unique surgeries from 533 to 502 as 31 EEG recordings did not have a continuous 10-minute-long EEG signal within the surgery start-stop times in the Fp2 channel. The alpha power in EEG is defined as the signal power in the frequency range of 8–12Hz. In order to compute the alpha power in each of the 502 10-minute segments, we constructed a Chebyshev Type II bandpass filter¹¹ in MATLAB, with the following parameters: Filter Order = 10; Stopband attenuation down from the peak passband value = 40dB; Lower stopband frequency = 7.5Hz; and Upper stopband frequency = 12.5Hz.

Next, we filtered all 502 10-minute EEG segments through this filter using a zero-phase forward-backward filter and computed the relative alpha power as follows:

$$P_{\propto }=\frac{\sum _{i=0}^{N-1}{S}_{\propto }{\left[i\right]}^2}{\sum _{i=0}^{N-1}{S}_o{\left[i\right]}^2}$$

where S_o is a 10-minute-long EEG signal along the Fp2 channel, S_∝ is the corresponding alpha bandpass filtered signal, N is the number of samples in both the original EEG signal and the alpha-filtered EEG signal, i is the index running from 0 to N − 1, and P_∝ is the relative alpha power which lies in the range [0,1]. To investigate the relationship between age and alpha power, we extracted the age in years for patients corresponding to each of the 502 surgeries and computed the least-squares linear-fit between age and alpha power. Further, we computed the R²-value and the p-value for the F-test on the least-squares linear-fit model (whether the model fits significantly better than a degenerate model consisting of only a constant term). We provide an illustration of the power spectra over time for four individuals of highly varying age (20, 40, 60, and 80 years) in Fig. 2. Note the drop in relative alpha power (P_∝) with age.

Figure 2.

EEG spectrograms of individual patients illustrating the diminishing relative Alpha power (P_∝) with patient biological age (a) 20 years:P_∝ = 0.31, (b) 40 years:P_∝ = 0.23, (c) 60 years:P_∝ = 0.15, (d) 80 years: P_∝ = 0.08. Each spectrogram corresponds to a 10-minute-long EEG signal captured in the frontal Fp2 EEG channel during the surgery while the patient is under general anesthesia. The dashed black lines depict the 8Hz and 12Hz lines and the area between these two lines is considered as the alpha band in this work. Image (c) Emory University, CC-BY-SA.

RESULTS

EEG Signal Quality Analysis

Figure 3A shows the average amount of flat-line, clipping, and true EEG signal in all the archived EEG records for which we had patient information. Further, it shows the total signal length of each EEG record. 25 EEG records had a total signal length less than 30 minutes. Figure 3B shows the eTR values for each EEG record and 60 EEG records had an eTR value less than the threshold value of 0.50. The filtering of EEG records based on total signal length and eTR resulted in a total of 533 EEG records with corresponding patient information. The statistics for the amount of flat-line, clipping, and true EEG signals in the filtered 533 EEG records are provided in the Supplementary Table 1. The median value for the amount of flat-line signal per record was in the range of 24 to 26 seconds for the four EEG channels. The 25^th percentile, median value, 75^th percentile, and the maximum value for the amount of clipping signal was higher for the F8 channel when compared to other EEG channels. The median eTR value was 1.00 for the Fp2 channel and 0.99 for the Fp1, F7, and F8 EEG channels.

Figure 3.

We used the total signal length and the proportion of true EEG signal (eTR) to filter EEG records for inclusion in the collation of various patient and surgery information that is presented in Supplementary Table 1. To begin with, we had 618 EEG records. (A) For each of the 618 EEG records, we computed the amount of flat-line signal, clipping signal, true EEG signal along each of the four EEG channels. Further, we computed the average amounts of flat-line signal, clipping signal, and true EEG signal across the four channels. Furthermore, we computed the total length of the EEG signal. For each record, we plotted the average amount of flat-line signal (x), the average amount of clipping signal (o), the average amount of true EEG signal (+) across four EEG channels, and the total signal length (.) in seconds. The y-axis was in the logarithmic scale to better visualize the values corresponding to different signal types. The plot was generated after sorting the total signal length of the records in ascending order. We set a threshold on the total signal length equal to 30 minutes and denoted it with a horizontal dashed line. (B) For all the archived EEG records, the proportion of true EEG signal (eTR) was computed and plotted for all four channels after sorting them in ascending order. This sorting of eTR for each of the four channels resulted in the same ascending order of EEG records. We set a threshold on eTR equal to 0.50 and depicted it in the figure in the form of the horizontal dashed line. Image (c) Emory University, CC-BY-SA.

Aggregating Patient Medical Information

Table 2 illustrates the statistics of various patient information metrics. The patients’ age was in the range [20, 91] years, and the median patient age was equal to 67 years. With respect to the median values corresponding to different durations, we had, Duration of Surgery < Duration of EEG Recording < Duration of Presence in Surgery Room < Duration of Anesthesia. The patient body weight was in the range [40, 155] kilograms. The median patient body weight was equal to 85 kilograms. Further, “Spinal stenosis, lumbar region without neurogenic claudication” was the most common comorbidity among the patients and “Lumbar Posterior Laminectomy/Hemi-Laminectomy/Discectomy/Decompression” was the most common surgical procedure performed on the patients in the curated dataset.

Table 2:

The collated patient and surgery information for 533 patients from October 2017 to March 2019 for which we had corresponding EEG records. In this table, we provide the computed statistics corresponding to (a) Patient age; (b) Duration of patient’s presence in surgery room; (c) Duration of anesthesia; (d) Duration of surgery; (e) American Society of Anesthesiologists (ASA) physical status; and (f) Patient body weight. For each of these categories, we provide each statistic separately for female patients, male patients, and all patients. The various statistics that were computed and presented here include the minimum value, the 25^th percentile, the median value, the 75^th percentile, and the maximum value.

Information	Gender	Minimum	25th Percentile	Median	75th Percentile	Maximum
Age (years)	Female	20	58	67	73	86
	Male	22	56	66	73	91
	All	20	57	67	73	91
Duration of Presence in Surgery Room (minutes)	Female	46	156	218	321	722
	Male	64	171	222	305	767
	All	46	162	221	310	767
Duration of EEG Recording (minutes)	Female	42	138	200	286	709
	Male	62	152	201	273	722
	All	42	145	201	279	722
Duration of Anesthesia (minutes)	Female	59	164	227	331	736
	Male	72	181	230	312	796
	All	59	170	228	322	796
Duration of Surgery (minutes)	Female	32	108	160	258	637
	Male	34	119	166	244	689
	All	32	113	161	248	689
ASA Physical Status (Categorical: 1 to 6)	Female	1	2	3	3	4
	Male	1	2	3	3	4
	All	1	2	3	3	4
Body Weight (kilograms)	Female	40	65	78	91	141
	Male	45	81	91	104	155
	All	40	71	85	93	155

Variation of Relative Alpha Power with Age

Figure 4 shows the scatter plot illustrating the variation of relative alpha power (P_∝) in the Fp2 EEG channel with the patient’s age. The solid red line depicts the least-squares linear fit. A negative slope in the linear fit indicated that on average, the P_∝ decreased with age. The slope of the linear fit was equal to −0.000806 year⁻¹. The letters a, b, c, and d refer to the individuals illustrated in Fig. 2 with ages of 20, 40, 60, and 80 years, respectively. The p-value for the F-test on the linear fit model was equal to 0.0061, indicating a statistically significant linear fit at the p < 0.05 level. However, the coefficient of determination (R²) for this linear fit was equal to 0.015, indicating the linear fit’s low explanatory power.

Figure 4.

Variation of relative alpha power in 10-minute-long EEG signal with increasing patient biological age. The EEG signal was captured in the frontal Fp2 channel in the middle of the surgery while the patient was under general anesthesia. Each blue dot corresponds to EEG signal from a different surgery. The solid red line is the corresponding least-squares linear fit. The black circles correspond back to Fig. 2(a,b,c,d). Image (c) Emory University, CC-BY-SA.

DISCUSSION

The data archival system described in this work captures data from medical devices (the SedLine^® Root device in particular) via a USB interface using readily available and inexpensive hardware to automate data collection. Many clinical systems are either designed to retain data in a proprietary ‘walled-garden’ to reduce competition or are not designed for the high throughput needed to transmit the data. Our Raspberry Pi-based data archival system allows direct import via USB and uploads data asynchronously to overcome these issues.

Anesthesiologists have used the EEG signal in the titration of anesthetics for over two decades. We refer the readers to Purdon et al.¹² to obtain a good understanding of the neurophysiology of the EEG signatures with the usage of different anesthetics. Moreover, they provide an excellent introduction to the biophysics of the EEG. The work by Rampil¹³ provides a good primer on EEG signal processing in anesthesia. Multiple research groups have created EEG signal databases during anesthesia. Kaiser et al.¹⁴ recorded two-channel bihemispheric frontal EEG in 1072 patients undergoing cardiac surgery from a Narcotrend^® DoA-monitor (MonitorTechnik, Bad Bramstedt, Germany). Hesse et al.^15,16 enrolled and collected EEG from 626 unique subjects to study the association of EEG trajectories during anesthesia emergence with Post-Anesthesia Care Unit (PACU) delirium. The EEG recordings were collected from both a SedLine^® monitor (Masimo, Irvine, CA, USA) and a Bispectral Index^™ (BIS^™) XP monitor (Covidien-Medtronic, Dublin, Ireland). Our work simplifies such EEG data capture efforts in the OR by minimizing human involvement in the data collection process.

The work by Von Dincklage et al.¹⁰ documents the data quality issues associated with the extracted EEG signals from the Root device. We note similar experiences with data quality and performed an extensive analysis to understand how much good quality EEG signal can be captured. On average, each of our EEG recordings contained between two and three hours of clean EEG signal captured during the surgery. Every recording does not contain EEG which includes the point in time at which the patient emerges from anesthesia. However, there is sufficient good data to perform a series of analyses such as the variation in relative alpha power with patient age, as we have shown in this work. A well-known characteristic of EEG under anesthesia is the frontal alpha (8–12Hz) oscillations.^17,18 Further, the signal energy in this frontal alpha frequency range is known to decline with aging, even when anesthetic doses are adjusted for patient age.^17,19,20 Shao et al.¹⁷ and Hight et al.²¹ have extensively studied the variations in the patterns of the frontal EEG signal strength in the alpha frequency range^12,17 and the extended alpha frequency range (7–17Hz)²¹ during the maintenance and the emergence of anesthesia. In the current work, similar to the work by Shao et al.,¹⁷ we examined the variation of frontal EEG alpha power with patient age. We observed a decreasing trend in the frontal EEG relative alpha power with increasing age. We also noted a weak but significant relationship, indicating that there is a large variation across the population.

The combining of patient information from electronic medical records with the archived EEG signals is potentially extremely valuable. It allows us to examine and analyze EEG patterns in sub-groups of patients where the sub-groups can be constructed based on the patient age, the anesthetic, the comorbidities, the surgical procedure experienced by the patients, and other relevant variables. We created a database of 533 unique surgeries with associated EEG recordings and patient medical records.

Electronic medical records are not designed to capture high-resolution data and labels with good temporal accuracy.²² The work described in this article provides a solid platform to extend the types of data that can be collected in OR and other clinical monitoring environments. Specifically, audio data, physical movement of patients and caregivers, and location of the caregivers during surgery are some of the non-traditional data modalities that can be collected using a Raspberry Pi-based system. A Raspberry Pi can act as a central hub; collect signals through different sensors (including EEG through the Root device), and archive them asynchronously to the cloud. We leave these enhancements for future work. Further, the usage of this database to study the association of metrics derived from the EEG signal with the patient’s condition, comorbidities, surgery and other metrics in the interleaved medical records is part of our future work.

In conclusion, this article presents an overview of a Raspberry Pi-based data archival system for EEG signals from a SedLine^® Root device. A method for direct USB transfer of EEG data from clinical monitors into a cloud-mediated database was implemented to facilitate multi-center trials in which manufacturers have not provided a method to pool or share data. We included metrics for EEG quality assessment to provide a generally acceptable system that conforms to privacy and security standards. We used this system to archive EEG recordings corresponding to over 500 unique surgeries. The patient medical records corresponding to each of these surgeries were extracted and interleaved with the EEG recordings. We analyzed the variation of relative alpha power in the frontal EEG with the patient age to demonstrate that the collected EEG data was clinically meaningful and useful for research. The data archival system itself was developed at a relatively low cost, and it provides a high degree of flexibility in the design. The bill-of-materials and open-source code have been made available²³ to enable replication of the work described here.

Key Points Summary.

Question:

Is it possible to automate the process of archiving patient EEG signal from a Sedline^® Root monitor and interleave patient medical information with the EEG recordings to do retrospective EEG analysis?

Findings:

The EEG archiving process can be semi-automated by utilizing a Raspberry Pi-based data archival system and the patient medical information can be interleaved with the EEG recordings using the OR number of the captured EEG records.

Meaning:

With this system, we can create large database (over 500 patients) of EEG recordings with associated patient medical records with minimal human labor for retrospective analysis.

GLOSSARY OF TERMS

OR

Operating Room

EEG

Electroencephalogram

Root

SedLine^® Root EEG Monitor

PSi

Patient State Index

USB

Universal Serial Bus

eCR

EEG Clipping Ratio

GPIO

General Purpose Input-Output

GUI

Graphical User Interface

HIPAA

Health Insurance Portability and Accountability Act

BSD

Berkeley Software Distribution

EUOSH

Emory University Orthopaedic and Spine Hospital

EDF

European Data Format

eFR

EEG Flat-line Ratio

eTR

Proportion of True EEG Signal

ASA

American Society of Anesthesiologists

PACU

Post-Anesthesia Care Unit

BIS

Bispectral Index