Epilog

Status epilepticus is a dangerous medical condition that is incredibly difficult to detect outside of a medical setting.  According to Zack and Kobau (2017), approximately 470,000 children in the US were living with epilepsy as of 2015. These children could experience a seizure: a sudden spike in brain activity. Usually, a caretaker administers medication, and the seizure subsides, but for 5-7% of these children, their seizures don’t stop. It persists, or another follows quickly, leading to status epilepticus. An international review in 2023 found that status epilepticus is associated with a staggering 30% mortality rate. Status epilepticus occurs when a seizure exceeds five minutes or when multiple seizures strike within 5 minutes. Detecting it is crucial, yet profoundly challenging. Standard seizures are often noticeable due to convulsions, especially in the limbs. 

But, in many cases of status epilepticus, the child’s body may be too exhausted to exhibit such movements. They may appear to be recovering when, in reality, they're in a critical state. These cases of non-convulsive status epilepticus (NCSE) can present with the only symptoms being an altered state of awareness or minimal signs like slight eyelid movement. It is incredibly difficult for caregivers to diagnose whether NCSE is occurring, or if the child is suffering from post-seizure lethargy. Without EEG monitoring, NCSE can easily go unnoticed, yet the risk of brain injury escalates the longer it continues. This is an even larger problem in low- and middle-income countries. The World Health Organization estimates that around 50 million people worldwide have epilepsy, and nearly 80% of them live in low- and middle-income countries.

Currently, the only method to confirm status epilepticus is to describe the child's condition over the phone to a neurologist. The child may then be rushed to a hospital for EEG scans, which reveal brain activity irregularities. 

Devices on the market today either focus on acceleration of the limbs, such as through accelerometer wristwatches, or use full EEG helmets for monitoring. The first is unable to detect NCSE since convulsions are not present. The second is unable to be used for at-home settings since the application is slow and requires a medical professional to gel and position the multitude of electrodes. This means that for a non-clinical setting, the best diagnostic technology is a phone call between a panicking caregiver and a distant physician. 

A mortality rate of 30% for a condition that can easily be detected is inexcusable. This is the problem we have set out to solve. 

Sources:

Trinka, E., Rainer, L. J., Granbichler, C. A., Zimmermann, G., & Leitinger, M. (2023). Mortality, and life expectancy in Epilepsy and Status epilepticus—Current trends and future aspects. Frontiers in Epidemiology, 3, 1081757. https://doi.org/10.3389/fepid.2023.1081757

World Health Organization. (2019, February 8). Epilepsy. WHO. https://www.who.int/news-room/fact-sheets/detail/epilepsy

Zack, M. M., & Kobau, R. (2017). National and State Estimates of the Numbers of Adults and Children with Active Epilepsy—United States, 2015. MMWR. Morbidity and mortality weekly report, 66(31), 821–825. https://doi.org/10.15585/mmwr....

Epilog is a rapid-application EEG headband designed to help caretakers of epilepsy patients make data-driven decisions. The flexible headband offers clinical-level seizure monitoring with 8 hospital-grade electrodes for at-home use.

The device is designed for short-term use following a seizure event. Once the visual signs of a convulsive seizure have stopped, a caretaker applies Epilog to the patient's head. EEG data is instantly recorded and fed into a machine-learning model to extract spectral features relevant to diagnosis. Finally, the caregiver will be alerted if the child is in a state of status epilepticus. After approximately 5-15 minutes have passed, the headband is removed and placed back into its casing for future use. 

The rapid application is made possible through the use of a custom gel case. When not in use, Epilog rests wrapped in a ring inside a cylindrical case. The electrodes are positioned facing tubes radiating from the center, akin to spokes on a tire. Immediately before use, the gel is applied to a tube at the top, and the radiating tubes ensure that all electrodes on the headband are gelled simultaneously. This allows the device to be set up quickly by a caregiver and does not require expensive, low-fidelity dry electrodes.

Additionally, the electrodes on the headband are placed in adjustable brackets that can be locked into place. This means that the brackets can be adjusted as the child grows and for custom placement for different head shapes and sizes, but there is no risk of electrodes sliding out of place when locked and in use.

The data is collected by Ag/AgCl electrodes and is processed using hardware circuitry to amplify, denoise, and filter the signal. The signal from the 8 channels is then sent over Bluetooth to a smartphone, where the Epilog app (developed for both iOS and Android) accesses a custom machine learning model to detect whether status epilepticus is occurring. The data is first passed through a Discrete Wavelet Transform one second at a time to convert the data to the wavelet domain and obtain coefficients for various frequency bands. These coefficients are then passed into a Balanced Random Forest model, trained and tuned on the publicly available CHB-MIT dataset, to obtain the probability of a seizure occurring at every given second. The 8 channels are analyzed independently before aggregating and thresholding the resultant probability. If a seizure is detected, the application alerts the user and Epilog’s indicator light turns red. This detection algorithm can be fine-tuned for each individual user if needed, which is especially useful in patients with irregular EEG baselines. 

Epilog is battery-powered by rechargeable 9V batteries and is a portable, lightweight solution, fitting in a 9” x 9” x 6” travel case, ensuring it can be used in a variety of environments. 

Our demo recording is unavailable due to restrictions from the University of Pennsylvania. Instead, below is a link to our slide deck: 

https://docs.google.com/presentation/d/1ojUISffBjWhbcwQ9q5qpeAVfn2_e-tsg/edit?usp=sharing&ouid=106316594561361713993&rtpof=true&sd=true

The solution is designed to serve children with epilepsy, especially those without reliable, 24/7 access to a neurology clinic. 

First, the pediatric space for medical devices has been historically underdeveloped. Because of a lack of revenue and a significantly smaller market size, children often have to use medical devices designed for adults. These devices, such as a full EEG headcap, are often uncomfortable and provide inaccurate results. We recently attended the Pediatric Device Innovation Consortium (PDIC) in Minneapolis and found that many doctors were upset about the lack of EEG systems at the pediatric level. Epilog is designed with these pediatric patients in mind. The headband is made out of a flexible fabric and allows for adjustable electrode placements that can be moved and then locked into place. This ensures the device can be used for a variety of head shapes and sizes and even as the child grows. There are also a limited number of electrodes used to increase comfort and ease of use by an untrained caregiver since a full helmet is much more difficult and invasive than a headband. 

Additionally, Epilog is designed for at-home use. This benefits everyone, but especially those who live far away from a hospital or cannot have reliable access to a pediatric neurologist. We have talked to a pediatric neurologist in a rural setting, who echoed that this is a major problem that hurts patient outcomes. When caretakers are unsure if their child has entered a state of status epilepticus or not, their best option is to rush to the hospital. However, many of these cases are false positives, as the symptoms of recovery often match the symptoms of status epilepticus. Having a device for at-home use reduces the need for these individuals to travel to a hospital, which is a key pain point. 

It is important to note, though, that Epilog errs on the side of caution and prefers false positives (inversely related to specificity) to false negatives (inversely related to high sensitivity). Our algorithm metrics are a sensitivity of 0.94 and a specificity of 0.88, on par with products on the market today. This reflects a balance between ensuring that unnecessary hospital trips are reduced and that a child who needs care is not left without care.

Our development team reflects a diverse background that reflects the communities we are trying to serve. We have familiarity with medical device research and have worked in labs that directly target hospitals like the Children’s Hospital of Philadelphia (CHOP). Through this research, we have interacted directly with patients. Several members of our team are also heavily involved in the community, such as by volunteering at the HMS School for Children with Cerebral Palsy

For this project, we are also working closely with representatives who are close to our target communities. Our clinical mentor is Dr. Dennis Dlugos, a pediatric neurologist at CHOP. Since our planned pilot program will be at CHOP and at Philadelphia-area hospitals, working with these leaders is especially helpful. We are also working with Dr. Julie Karand from the University of Delaware, who started the Devices for the Developing World conference. She helps us ensure that we keep low-resource settings in mind by remaining battery-powered and therefore outlet-independent, reusable, and durable. 

Our designs and implementation are heavily influenced by what Dr. Dlugos hears and sees when talking to his patients and their caregivers on a daily basis. The original idea for Epilog as a rapid-application, at-home device stemmed from these patient interactions. We hope to become more integrated into these parent groups and involve them in human factors testing to ensure that the target population is able to use and understand the device correctly. Through Solve, we hope to expand our close network beyond Philadelphia to make a nationwide impact with the same principles of patient- and caregiver-guided design.

We have built and tested our fully integrated prototype. The first part is the gel storage case that has been constructed out of acrylic and tubing and tested to ensure that a medically untrained user can properly apply it in under 15 seconds. Secondly, the elastic headband, complete with adjustable, lockable brackets, EEG electrodes, and postprocessing circuitry has been built and tested. The EEG signal quality was validated by visualizing the processed signal via Python and having the user blink, flutter their eyes, and close their eyes to see the Berger effect (a rise in alpha band magnitude when the eyes are closed). These effects were clear and easy to detect by a neurologist, our clinical mentor, who mentioned that the EEG collected by our headband and clinical-grade EEG were virtually indistinguishable. The signal-to-noise ratio of our device was also calculated to be around 25 dB among 5 test subjects and 20 trials each, which is on par with products like Emotiv that are currently on the market. 

Thirdly, our machine learning algorithm was able to effectively classify seizure events with a specificity of 0.883 and sensitivity of 0.946 (ROC=0.95) on a testing dataset from CHB-MIT. Additionally, the average latency after a convulsive seizure event for 20 seconds of 100% correct prediction (status or not) was 26.77 seconds. We have also implemented fine-tuning on this algorithm by allowing patients to record their baseline (regular EEG, non-seizure) and use that to calibrate the weights of the model for each user. This improved performance on ROC (0.03), specificity (0.01), and sensitivity (0.05) for a specific patient who had irregular baselines in the CHB-MIT dataset. This algorithm was made available on a smartphone application that can be easily used and connected to Epilog via Bluetooth. The latency of the whole system was consistently under 1 second.

The entire prototype has been built. The headband can be gelled, put on a patient, and analyzed in under a minute. We have demonstrated this device to professors at the University of Pennsylvania in a non-seizure setting and simulated what would occur in a seizure setting by feeding in data from a seizure event from the CHB-MIT testing dataset. 

We have also filed a provisional utility patent focusing on the rapid application of our headband through the use of a gel case and flexible headband. 

The Epilog team's expertise lies predominantly in bioengineering, with a focus on using data and AI to address healthcare challenges. However, when it comes to navigating the complex landscape of business operations, including sales and strategy, we recognize our limited experience. While we have taken courses like engineering entrepreneurship, we do seek guidance from seasoned professionals who have been through the process before. In this field, we believe that experience is arguably the most important factor in navigating the landscape, and Solve’s affiliates have just that. By joining Solve's community, we aim to connect with experienced entrepreneurs, industry experts, and mentors who can offer guidance, share insights, and provide support in areas where our team may lack expertise. This is especially important as we branch out beyond the Philadelphia area, which is where many of our current mentors and advisors are based. 

Through Solve, we anticipate gaining access to resources and connections that can help us address key business challenges, refine our strategies, and navigate regulatory requirements effectively. Whether it's seeking advice on crafting a robust business model, understanding the nuances of FDA approval processes, or devising effective product distribution strategies, we believe that Solve's network can provide invaluable support in accelerating our journey toward accessibility.

While the guidance and mentorship offered by Solve's affiliates are crucial for our success, we also recognize the significance of financial support, particularly as we embark on the FDA approval process. Securing funding is essential for sustaining operations, conducting clinical trials, and ultimately bringing our innovative medical device to market. In this regard, Solve presents a compelling opportunity to access funding and investment opportunities tailored to the needs of early-stage medical device startups like ours.

Epilog is uniquely designed to be rapidly applicable, making it ideal for at-home uses and short-term settings. The use of a gel case allows all 8 of Epilog’s electrodes to be rapidly gelled at the same time. The headband, consisting of flexible elastic, adjustable brackets, and a limited montage of electrodes allows the headband to be attached in under 30 seconds. Additionally, the rapid detection algorithm has a test runtime of under 1 second, ensuring that the decision can be performed with ease at every second. The unique combination of all of these tools, especially the gel case, allows Epilog to be used at home and for short-term wear. This improves comfort and patient quality of life, as existing solutions are meant to be for continuous wear and monitoring. Based on patient interviews, these devices are so uncomfortable that patients often choose not to wear them.

We hope to drive change and have rapid-application wearables become more commonplace during medical emergencies like status epilepticus. This allows detection systems to be used in an outpatient setting and even in their own home since they can be quickly applied and removed by a caregiver. As data becomes more prevalent, it is important to be able to use data that is collected outside of a medical setting to improve diagnostics, and at-home wearables can do just that. 

This aspect of our device has not been produced before, and we have filed a provisional utility patent (18/628,803). We are currently in talks with legal clinics at the University of Pennsylvania to file a utility patent application.

Epilog will provide caregivers of those suffering from a seizure with granular data to improve the decision-making process of whether medical intervention is necessary. Based on conversations with neurologists, the status quo of this process is to call a child’s pediatric neurologist and describe the situation. Epilog aims to change this diagnostic process between a stressed caregiver and a distant medical professional by injecting data into the equation. Epilog is not designed to provide a firm answer on whether status is occurring, but simply provide more information to ensure that caregivers can make the best decisions using wearable technologies and AI. With this information, caregivers can be more confident in their decisions to bring their child to the hospital or help them through post-seizure recovery at home. This will substantially improve patient outcomes and also reduce the burden on pediatric neurologists as they will no longer have to guide a caregiver over the phone. This allows them to see more patients and reduce their stress levels, which is a major problem contributing to burnout in the neurology field. 

We also hope to improve caregiver peace of mind and reduce the stress of a seizure event. When caregivers, who are typically medically untrained, are in high-stress situations, the chance of a mistake becomes much higher, 

Overall, we hope to make an impact on the patient, their caregiver, and their medical provider through the use of data collection and analysis with AI.

The main impact goals are to improve the patient outcomes for status epilepticus. The primary factor that we hope to reduce is to reduce the mortality rate of patients that use Epilog by a significant margin. This is a difficult statistic to measure, though, and before we roll it out to customers, it is impossible to calculate. Therefore, we have several intermediate impact goals in mind. The first is improving caregiver peace of mind. This can be measured qualitatively through surveys of caregivers regarding Epilog as a concept and then after providing them with a functional device. It can also be measured in the reduction of communication between caregivers and pediatric neurologists post-seizure to diagnose status epilepticus, since these medical professionals are often quite busy and many of these calls/visits are false positives. Surveys can also be used to measure a patient’s level of comfort and confidence with and without Epilog. 

During the development phase, our goal is to collect a quality EEG signal and have correct detection systems. For these, we have measured tools like application time (< 1 minute) the signal-to-noise ratio (temporally > 25 dB and spectrally > 1.0 with the Berger effect), specificity (> 0.85), and sensitivity (> 0.9). We have met these thresholds and are moving forward with the patient-specific impact goals to improve the length and quality of life.

The core technologies involved in our solution are as follows: 

Wearable devices: The core piece of Epilog that is responsible for data collection is a headband made of a flexible elastic material and has Ag/AgCl electrodes on it to collect clinical-grade EEG data. 

Custom circuitry: Our postprocessing circuit uses components such as operational amplifiers, bandpass filters, differential amplifiers, and passive components to filter and gain the signal selectively.

Machine Learning/Artificial Intelligence: The collected EEG data is fed into our custom machine learning algorithm that is trained on millions of seconds of data from seizure patients (https://physionet.org/content/chbmit/1.0.0/). The specific algorithm used is a Balanced Random Forest, a variant of a traditional random forest that is built specifically for class-imbalanced data like a medical dataset. Fine-tuning is also implemented to ensure the model is personalized to users who have irregular EEG baselines or unique seizure profiles.

Mobile Apps: The interface that allows users to see granular data and its classification every second. It also comes with tools to search through prior history, the ability to fine-tune the detection algorithm with a new baseline measurement and to reset the model to the out-of-the-box model without any finetuning. It is being designed for both iOS and Android through React Native.

IoT: The headband transmits data via Bluetooth to the paired smartphone, which then accesses the algorithm using the Internet and a custom API endpoint with authentication. The integration of sensors, Bluetooth, and the Internet embraces the idea of Industry 4.0 and the Internet of Things. 

5 full-time engineers are working on this solution.

We have worked on this project for approximately 1 year (starting in June 2023).

Our team of 5 individuals is incredibly diverse, representing broad backgrounds. Ethnically, team members are from Brazil, India, China, and the United States. Additionally, there are a total of 6 different languages spoken between us, ensuring that communication is possible with a variety of target audiences. There are also 4 women on the team to advocate for and support women in STEM. In the future, as our team expands, we hope to keep DEI practices in mind by ensuring voices from all sides of the conversation are heard and integrated. 

Through the support of the University of Pennsylvania, equity is maintained. Our working space, the George H. Stephenson Foundation Educational Laboratory & Bio-MakerSpace, is open to everyone and provides resources equitably to ensure that all of us have all the tools necessary to thrive. Additionally, our team practices inclusion and equity by ensuring that more than one person is in charge of a certain task to ensure that no one is singled out. This also has the added benefit of having a partner to springboard ideas off of and has helped us stay on track through the development process. 

Our team environment is incredibly transparent and we practice good communication strategies, such as by having an open mind at all times and hosting weekly in-person meetings to plan for the week ahead and air out any grievances, if any. 

Our group also regularly engages with community organizations to stay involved in the Philadelphia community, both academically/medically and otherwise. This helps us keep in touch with users outside of the so-called “Penn bubble” that students find themselves in. As we expand, we hope to include more community leaders in the areas we target, to ensure that we maintain DEI principles and to enhance the participation of traditionally underrepresented groups. 

Through our training at Penn (ethics, anthropology, history, and other courses in social science)  as well as resources like the Office of Diversity, Equity, and Inclusion (ODEI) at Penn, we hope to continue having a diverse, equitable, and inclusive environment moving forward. 

Our business model is to focus on pediatric neurology clinics. Epilog can accurately monitor EEG activity in real-time, providing a precise assessment of a child’s condition. This precision helps avoid unnecessary hospital communication and hospital visits, ensuring only critical cases lead to hospital admission. This drastically reduces hospital costs, especially for neurologists. In 2021, Ney, Gururangan, and Parvizi calculated that the use of EEG monitoring to detect non-convulsive seizures in hospitals saves $3,971 to $17,290 per use per patient. The same principle applies to at-home EEG monitoring. It is economically advisable for hospitals to keep Epilog on hand and hand it out to patients who may need it. Our clinical advisor at the Children’s Hospital of Philadelphia has mentioned that he would recommend at-home EEG monitoring to all his patients, to ensure both data-driven decision-making and fewer hospital visits. 

This is a similar business plan to Ceribell, an EEG-monitoring system designed for hospital use. Epilog is meant to be a rapid application variant of this technology, meant to be applied and removed after a seizure event at home. However, the impact of reducing hospital costs still applies and we are confident that hospitals will be incentivized to purchase devices like Epilog to give to their patients. 

Citation: Ney, J. P., Gururangan, K., & Parvizi, J. (2021). Modeling the economic value of Ceribell Rapid Response EEG in the inpatient hospital setting. Journal of medical economics, 24(1), 318–327. https://doi.org/10.1080/136969...

Currently, we have received approximately $1500 in funding from the University of Pennsylvania Bioengineering Department. Our funding approach moving forward is to use resources from the University of Pennsylvania, such as the Detkin IP and Technology Legal Clinic, and non-dilutive competition funding over the next couple of months as we move forward in securing IP protection and human factors testing. After that, we plan to raise investment capital from angel and seed investors to sustain operations as we focus on the manufacturability of our device and draft a 510(k) approval application (using codes OMC and OLT and with PCCP, Predetermined Change Control Plans for Machine Learning-Enabled Medical Devices). Once approval has been granted, we plan to roll out a pilot program at the Children’s Hospital of Philadelphia (CHOP), as we have secured clinical interest from pediatric neurologists there. This is where the first revenue streams begin, and we are targeting neurology clinics as a point of sale. We do not calculate that we will be profitable at this point yet, and will rely on investor funding and likely open a Series A round. Following a pilot program at CHOP, we will incorporate the resulting feedback and start rolling out to local Philadelphia-area clinics. Since the team is based in Philadelphia, this will make distribution and supply chain logistics easier. We plan to slowly increase the distribution to other metropolitan areas on the East Coast and eventually throughout the East Coast. This is when we project profitability to first occur. Eventually, we hope to expand across the United States. We also plan to concurrently work on international approval pipelines to extend internationally and start to help the millions worldwide who suffer from pediatric seizures. 

Additionally, we also foresee potential business partnerships or acquisition opportunities from Ceribell and other EEG monitoring devices. Since our rapid-application, at-home technologies complement their hospital use, we do not wish to compete with these companies but rather work together to ensure that EEG monitoring becomes commonplace in all settings.