Study: Wearable sensors and machine learning may well (one day) help detect a broad range of epileptic seizures

Wear­able Device Clears a First ‘Mile­stone’ in Seizure Detec­tion (Med­scape; requires subscription):

A wrist-worn device that uses machine learn­ing accu­rate­ly detects dif­fer­ent seizure types in find­ings that have the poten­tial to rev­o­lu­tion­ize the man­age­ment of patients with epilepsy.

“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,” study inves­ti­ga­tor Jian­bin Tang, MA, Data Sci­ence Project Lead, IBM Research Aus­tralia, Vic­to­ria, told Med­scape Med­ical News.

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 Tang.

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…

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 Tang.

The Study:

SEIZURE DETECTION USING WEARABLE SENSORS AND MACHINE LEARNING: SETTING A BENCHMARK (poster pre­sent­ed at the Annu­al Meet­ing of the Amer­i­can Epilep­sy Soci­ety). From the Abstract:

• Con­clu­sion: Auto­mat­ic epilep­tic seizure detec­tion using machine learn­ing and wear­ables data is fea­si­ble. Pre­lim­i­nary results show bet­ter-than-chance seizure detec­tion across a broad range of epilep­tic seizures. Future improve­ments may con­sid­er clin­i­cal chrono-epilep­to­log­i­cal vari­ables, such as seizure dura­tion, semi­ol­o­gy, and eti­ol­o­gy or syn­drome, and alter­na­tive data bal­anc­ing, pre- and post-pro­cess­ing, fusion and ensem­ble learn­ing meth­ods. Thus, while find­ings sug­gest fea­si­bil­i­ty, per­for­mance fol­low­ing future adjust­ments may improve further.

News in Context:

• Spe­cial Issue: Seizure detec­tion and mobile health devices in epilepsy
• Study: Brain scans map­ping lan­guage and mem­o­ry areas can help guide epilep­sy-relat­ed surgeries
• Seizure detec­tion at home: Do devices on the mar­ket match the needs of peo­ple liv­ing with epilep­sy and their care­givers?. From the Abstract: … Cur­rent­ly mar­ket­ed devices are focused pri­mar­i­ly on the record­ing of non–electroencephalography (EEG) sig­nals asso­ci­at­ed with tonic?clonic seizures, where­as the detec­tion of focal seizures with­out major motor fea­tures remains a clear evi­dence gap. More­over, there is pauci­ty of evi­dence com­ing from real?life set­tings. A joint effort of clin­i­cal and non­clin­i­cal experts, patients, and care­givers is required to ensure an opti­mal lev­el of accept­abil­i­ty and usabil­i­ty, which are key aspects for a suc­cess­ful con­tin­u­ous mon­i­tor­ing aimed at seizure detec­tion at home.