User login
The new findings have the potential to revolutionize the management of patients with epilepsy, according to the researchers. “We have set a first benchmark for automatic detection of a variety of epileptic seizures using wearable sensors and deep-learning algorithms. In other words, we have shown for the first time that it’s possible to do this,” said study investigator Jianbin Tang, MA, data science project lead, IBM Research Australia, Victoria.
The findings were presented at the American Epilepsy Society’s annual meeting, held online this year because of the COVID-19 pandemic.
Accurate monitoring of seizures is important for assessing risk, injury prevention, and treatment response evaluation. Currently, video EEG is the gold standard for seizure detection, but it requires a hospital stay, is often costly, and can be stigmatizing, said Mr. Tang.
An advance in detecting seizure types
Recent advances in non-EEG wearable devices show promise in detecting generalized onset tonic-clonic and focal to bilateral tonic-clonic seizures, but it’s not clear if they have the ability to detect other seizure types. “We hope to fill this gap by expanding wearable seizure detection to additional seizure types,” said Mr. Tang.
Seizure tracking outside the hospital setting largely “relies on manually annotated family and patient reports, which often can be unreliable due to missed seizures and problems recalling seizures,” he said.
The study included 75 children (44% were female; mean age was 11.1 years) admitted to a long-term EEG monitoring unit at a single center for a 24-hour stay. Patients wore the detector on the ankle or wrist. The device continuously collected data on functions such as sweating, heart rate, movement, and temperature.
With part of the dataset, researchers trained deep-learning algorithms to automatically detect seizure segments. They then validated the performance of the detection algorithms on the remainder of the dataset.
The analysis was based on data from 722 epileptic seizures of all types including focal and generalized, motor and nonmotor. Seizures occurred throughout the day and during the night while patients were awake or asleep.
When a seizure is detected, the system triggers a real-time alert and will store the information about the detected seizure in a repository, said Mr. Tang.
The signals were initially stored in the wristband and then securely uploaded to the Cloud. From there, the signal files were downloaded by the investigators for analysis and interpretation. All data were entirely anonymized and de-identified. Researchers used Area Under Curve–Receiver Operating Characteristic (AUC-ROC) to assess performance.
“Our best performing detection models reach an AUC-ROC of 67.59%, which represents a decent performance level,” said Mr. Tang. “There certainly is room for performance improvement and we are already working on this,” he added.
The device performed “better than chance,” which is a “standard technical term” in the field of machine learning and is “the first hurdle any machine-learning model needs to take to be considered useful.” The investigators noted that such automatic seizure detection “is feasible across a broad spectrum of epileptic seizure types,” said Mr. Tang. “This is a first and has not been shown before.”
The study suggests that the noninvasive wearable device could be used at home, at school, and in other everyday settings outside the clinic. “This could one day provide patients, caregivers, and clinicians with reliable seizure reports,” said Mr. Tang.
He said he believes the device might be especially useful in detecting frequent or subtle seizures, which are easy to miss. Patients requiring medication evaluation and rescue medication and those at risk of status epilepticus may be good candidates.
The researchers don’t expect wearable technology to totally replace EEG but see it as “a useful complementary tool to track seizures continuously at times or in settings where EEG monitoring is not available,” said Mr. Tang.
‘Important milestone’
Commenting on the research, Benjamin H. Brinkmann, PhD, associate professor of neurology at the Mayo Clinic in Rochester, Minn., said the investigators “have done very good work applying state of the art machine learning techniques” to the “important problem” of accurately detecting seizures.
Dr. Brinkmann is part of the Epilepsy Foundation–sponsored “My Seizure Gauge” project that’s evaluating various wearable devices, including the Empatica E4 wristband and the Fitbit Charge 3, to determine what measurements are needed for reliable seizure forecasting.
“Previously, no one knew whether seizure prediction was possible with these devices, and the fact that this group was able to achieve ‘better-than-chance’ prediction accuracy is an important milestone.”
However, he emphasized that there is still a great deal of work to be done to determine, for example, if seizure prediction with these devices can be accurate enough to be clinically useful. “For example, if the system generates too many false-positive predictions, patients won’t use it.”
In addition, the findings need to be replicated and recordings extended to 6 months or more to determine whether they are helpful to patients long term and in the home environment, said Dr. Brinkmann.
The investigators and Dr. Brinkmann have disclosed having no relevant financial relationships.
A version of this article originally appeared on Medscape.com.
The new findings have the potential to revolutionize the management of patients with epilepsy, according to the researchers. “We have set a first benchmark for automatic detection of a variety of epileptic seizures using wearable sensors and deep-learning algorithms. In other words, we have shown for the first time that it’s possible to do this,” said study investigator Jianbin Tang, MA, data science project lead, IBM Research Australia, Victoria.
The findings were presented at the American Epilepsy Society’s annual meeting, held online this year because of the COVID-19 pandemic.
Accurate monitoring of seizures is important for assessing risk, injury prevention, and treatment response evaluation. Currently, video EEG is the gold standard for seizure detection, but it requires a hospital stay, is often costly, and can be stigmatizing, said Mr. Tang.
An advance in detecting seizure types
Recent advances in non-EEG wearable devices show promise in detecting generalized onset tonic-clonic and focal to bilateral tonic-clonic seizures, but it’s not clear if they have the ability to detect other seizure types. “We hope to fill this gap by expanding wearable seizure detection to additional seizure types,” said Mr. Tang.
Seizure tracking outside the hospital setting largely “relies on manually annotated family and patient reports, which often can be unreliable due to missed seizures and problems recalling seizures,” he said.
The study included 75 children (44% were female; mean age was 11.1 years) admitted to a long-term EEG monitoring unit at a single center for a 24-hour stay. Patients wore the detector on the ankle or wrist. The device continuously collected data on functions such as sweating, heart rate, movement, and temperature.
With part of the dataset, researchers trained deep-learning algorithms to automatically detect seizure segments. They then validated the performance of the detection algorithms on the remainder of the dataset.
The analysis was based on data from 722 epileptic seizures of all types including focal and generalized, motor and nonmotor. Seizures occurred throughout the day and during the night while patients were awake or asleep.
When a seizure is detected, the system triggers a real-time alert and will store the information about the detected seizure in a repository, said Mr. Tang.
The signals were initially stored in the wristband and then securely uploaded to the Cloud. From there, the signal files were downloaded by the investigators for analysis and interpretation. All data were entirely anonymized and de-identified. Researchers used Area Under Curve–Receiver Operating Characteristic (AUC-ROC) to assess performance.
“Our best performing detection models reach an AUC-ROC of 67.59%, which represents a decent performance level,” said Mr. Tang. “There certainly is room for performance improvement and we are already working on this,” he added.
The device performed “better than chance,” which is a “standard technical term” in the field of machine learning and is “the first hurdle any machine-learning model needs to take to be considered useful.” The investigators noted that such automatic seizure detection “is feasible across a broad spectrum of epileptic seizure types,” said Mr. Tang. “This is a first and has not been shown before.”
The study suggests that the noninvasive wearable device could be used at home, at school, and in other everyday settings outside the clinic. “This could one day provide patients, caregivers, and clinicians with reliable seizure reports,” said Mr. Tang.
He said he believes the device might be especially useful in detecting frequent or subtle seizures, which are easy to miss. Patients requiring medication evaluation and rescue medication and those at risk of status epilepticus may be good candidates.
The researchers don’t expect wearable technology to totally replace EEG but see it as “a useful complementary tool to track seizures continuously at times or in settings where EEG monitoring is not available,” said Mr. Tang.
‘Important milestone’
Commenting on the research, Benjamin H. Brinkmann, PhD, associate professor of neurology at the Mayo Clinic in Rochester, Minn., said the investigators “have done very good work applying state of the art machine learning techniques” to the “important problem” of accurately detecting seizures.
Dr. Brinkmann is part of the Epilepsy Foundation–sponsored “My Seizure Gauge” project that’s evaluating various wearable devices, including the Empatica E4 wristband and the Fitbit Charge 3, to determine what measurements are needed for reliable seizure forecasting.
“Previously, no one knew whether seizure prediction was possible with these devices, and the fact that this group was able to achieve ‘better-than-chance’ prediction accuracy is an important milestone.”
However, he emphasized that there is still a great deal of work to be done to determine, for example, if seizure prediction with these devices can be accurate enough to be clinically useful. “For example, if the system generates too many false-positive predictions, patients won’t use it.”
In addition, the findings need to be replicated and recordings extended to 6 months or more to determine whether they are helpful to patients long term and in the home environment, said Dr. Brinkmann.
The investigators and Dr. Brinkmann have disclosed having no relevant financial relationships.
A version of this article originally appeared on Medscape.com.
The new findings have the potential to revolutionize the management of patients with epilepsy, according to the researchers. “We have set a first benchmark for automatic detection of a variety of epileptic seizures using wearable sensors and deep-learning algorithms. In other words, we have shown for the first time that it’s possible to do this,” said study investigator Jianbin Tang, MA, data science project lead, IBM Research Australia, Victoria.
The findings were presented at the American Epilepsy Society’s annual meeting, held online this year because of the COVID-19 pandemic.
Accurate monitoring of seizures is important for assessing risk, injury prevention, and treatment response evaluation. Currently, video EEG is the gold standard for seizure detection, but it requires a hospital stay, is often costly, and can be stigmatizing, said Mr. Tang.
An advance in detecting seizure types
Recent advances in non-EEG wearable devices show promise in detecting generalized onset tonic-clonic and focal to bilateral tonic-clonic seizures, but it’s not clear if they have the ability to detect other seizure types. “We hope to fill this gap by expanding wearable seizure detection to additional seizure types,” said Mr. Tang.
Seizure tracking outside the hospital setting largely “relies on manually annotated family and patient reports, which often can be unreliable due to missed seizures and problems recalling seizures,” he said.
The study included 75 children (44% were female; mean age was 11.1 years) admitted to a long-term EEG monitoring unit at a single center for a 24-hour stay. Patients wore the detector on the ankle or wrist. The device continuously collected data on functions such as sweating, heart rate, movement, and temperature.
With part of the dataset, researchers trained deep-learning algorithms to automatically detect seizure segments. They then validated the performance of the detection algorithms on the remainder of the dataset.
The analysis was based on data from 722 epileptic seizures of all types including focal and generalized, motor and nonmotor. Seizures occurred throughout the day and during the night while patients were awake or asleep.
When a seizure is detected, the system triggers a real-time alert and will store the information about the detected seizure in a repository, said Mr. Tang.
The signals were initially stored in the wristband and then securely uploaded to the Cloud. From there, the signal files were downloaded by the investigators for analysis and interpretation. All data were entirely anonymized and de-identified. Researchers used Area Under Curve–Receiver Operating Characteristic (AUC-ROC) to assess performance.
“Our best performing detection models reach an AUC-ROC of 67.59%, which represents a decent performance level,” said Mr. Tang. “There certainly is room for performance improvement and we are already working on this,” he added.
The device performed “better than chance,” which is a “standard technical term” in the field of machine learning and is “the first hurdle any machine-learning model needs to take to be considered useful.” The investigators noted that such automatic seizure detection “is feasible across a broad spectrum of epileptic seizure types,” said Mr. Tang. “This is a first and has not been shown before.”
The study suggests that the noninvasive wearable device could be used at home, at school, and in other everyday settings outside the clinic. “This could one day provide patients, caregivers, and clinicians with reliable seizure reports,” said Mr. Tang.
He said he believes the device might be especially useful in detecting frequent or subtle seizures, which are easy to miss. Patients requiring medication evaluation and rescue medication and those at risk of status epilepticus may be good candidates.
The researchers don’t expect wearable technology to totally replace EEG but see it as “a useful complementary tool to track seizures continuously at times or in settings where EEG monitoring is not available,” said Mr. Tang.
‘Important milestone’
Commenting on the research, Benjamin H. Brinkmann, PhD, associate professor of neurology at the Mayo Clinic in Rochester, Minn., said the investigators “have done very good work applying state of the art machine learning techniques” to the “important problem” of accurately detecting seizures.
Dr. Brinkmann is part of the Epilepsy Foundation–sponsored “My Seizure Gauge” project that’s evaluating various wearable devices, including the Empatica E4 wristband and the Fitbit Charge 3, to determine what measurements are needed for reliable seizure forecasting.
“Previously, no one knew whether seizure prediction was possible with these devices, and the fact that this group was able to achieve ‘better-than-chance’ prediction accuracy is an important milestone.”
However, he emphasized that there is still a great deal of work to be done to determine, for example, if seizure prediction with these devices can be accurate enough to be clinically useful. “For example, if the system generates too many false-positive predictions, patients won’t use it.”
In addition, the findings need to be replicated and recordings extended to 6 months or more to determine whether they are helpful to patients long term and in the home environment, said Dr. Brinkmann.
The investigators and Dr. Brinkmann have disclosed having no relevant financial relationships.
A version of this article originally appeared on Medscape.com.
FROM AES 2020