1.1.3. Statistics

According to World Health Organization (WHO) [11], approximately 50 million people worldwide suffer from epilepsy. 80% of them live in low and middle-income countries. Among them 75% do not receive proper treatment. The estimated proportion of the general population with active epilepsy (continuing seizures or with the need for treatment) is between 4 and 10 per 1000 people. However, this number is much higher in low and middle-income countries, i.e., between 7 and 14 per 1000 people.

Each year, 2.4 million people are diagnosed with epilepsy [11]. The estimated proportion of new cases with epilepsy in high-income countries is between 30 and 50 per 100,000 people. This rate doubles in low and middle-income countries.

#### 1.1.4. Epilepsy Treatment

Several studies have explored the reason for the initiation of seizures [12,13]. The importance of finding the reasons and origin lies in the correct diagnosis and treatment of epilepsy. No medical treatment can be safely provided without having knowledge of the different sources that cause epilepsy. The development of modern medical devices and implantable systems participate in improving the understanding of the origins of epilepsy.

The treatment of epilepsy is composed of three levels. The first level is the treatment with medications. Several anti-epileptic drugs (AEDs) are available for epilepsy patients. There are approximately 25 different AEDs suitable for controlling seizures; different AEDs are suitable for different seizure types. The benefits of AEDs consist of the reduction of stopping seizures and risk of accidents. A first-line AED is an AED that is tried first in the therapy. The AEDs added to the first-line

AED are called second-line AEDs. The success rate of the third medication delivered after trying two different medications is approximately 5%.

Epilepsy is defined as a drug-resistant epilepsy when a patient has unsuccessfully tried two different anti-seizure medications. The second level of epilepsy treatment is lobectomy. Lobectomy is the removal of the part of the brain that is responsible for seizure initiation, if the region is uniquely detectable, and if it is not in a sensitive region of the brain. Before lobectomy, surgeons must find the precise location of the brain from which the seizures start. Following modern medical procedures, seizures can be identified and categorized using three Tesla magnetic resonance imaging (3T MRI), video-EEG, single-photon emission computed tomography (SPECT), magnetoencephalography (MEG) scan and positron emission tomography (PET) scan as prerequisites of further treatment steps. A resection surgery would be prescribed in cases of a focal seizure and if a single location can be identified [14]. In a resection surgery, a part of the skull is temporarily removed by craniotomy and the part of the brain which is engaged in the seizure initiation is removed. Laser thermal ablation is another way to stop the seizures, which is a minimally invasive method. This method is performed in the MRI in real time. In this surgery, during an MRI scan, the part of the brain that is engaged in the seizure initiation is heated up and destroyed, while local temperature is monitored using the MRI scanner. This method can only be employed in limited areas of the brain.

In order to precisely detect the part of the brain that is a target of resection surgery or laser thermal ablation, intracranial monitoring is necessary, which is a diagnostic surgery. Intracranial monitoring is carried out in two different ways, including recording using subdural grids and strips or stereo-EEG using SEEG electrodes.

Patients responding to prevalent cures like medications and surgeries are approximately 70%, while approximately 30% of the patients are untreated or poorly treated because of the following reasons:


Implantable electrical stimulators offer an alternative therapy to untreated or poorly treated patients whose seizures are not controlled using medication or surgery. These stimulator systems either operate in an open-loop or closed-loop mode. IEMDs operating in open-loop stimulation only consist of electrical stimulators. In contrast, closed-loop stimulators record neural signals from the brain and detect the seizure onsets. Upon seizure detection, closed-loop stimulators trigger electrical stimulation. Commercial stimulators are reviewed in Section 3.

Intracranial Pressure (ICP) is reported to influence drug-resistant epilepsy in some cases. In [15], a 23-year old patient was reported as the case study. In this study, an increase in ICP is shown to augmen<sup>t</sup> the efficiency of anti-seizure medications. The epileptic patient had undergone a shunt surgery in childhood to decrease the ICP. The patient had uncontrolled seizures in spite of three years of pharmacotherapy. However, the uncontrolled seizures suddenly stopped after shunt removal.

#### **2. Introduction to Epilepsy Control Using Implantable Microelectronic Systems**

Several electronic building blocks are essential to design and implement a low-power seizure detection system. These blocks are also some of the main building blocks of an epilepsy control system. Figure 2 shows the general overview of a low-power seizure detection (Figure 2a) and epilepsy control (Figure 2b) system. Both systems have common essential building blocks which are shown with gray boxes including the analog front-end (AFE), analog-to-digital converter (ADC), and seizure onset detector (SON). The Epilepsy control system has an additional essential building block which is the electrical neural stimulator aimed at suppressing or modulating the seizure electrical activity. In addition, data compression may be applied prior to feeding the data to the SON to lower the power consumption of the systems. Data compression can be done either in analog or digital domains.

In the following, the operation principles of some of these blocks are briefly reviewed, including the neural amplifiers, compressive sensing and feature extractors inside the SON block.

**Figure 2.** General overview of (**a**) a low-power seizure detection, and (**b**) a low-power seizure control systems.

#### *2.1. Electrical Stimulation*

The historical developments of electrical stimulators are considered to begin in the European post-Middle Ages, especially from the 16th century when a Dutch scientist, Jan Swammerdam, performed the first experiment on the muscle of a dissected frog [16]. He realized that the severed muscle of a frog is contracted by irritation. The idea of contracting a muscle by a stimuli had an important impact on neuroscience by demonstrating the fact that the observed behavior is based on the stimuli. More than one hundred years after Swammerdam's first experiment, on 6 November 1787, Luigi Galvani realized that a frog muscle can be contracted by placing an iron wire to the muscle and a copper wire to the nerve during a random experiment. He observed that an animal body performs convulsive movements when electricity is applied to it. The work of Galvani inspired Alessandro Volta to invent the voltaic pile in 1799. Using a voltaic pile, Luigi Rolando performed the first cortical stimulation experiment on an animal cortex in 1809. Indeed, Rolando is well-known for his pioneering research on brain localization of function. In 1825, J.-B. Sarlandiere published an extensive study on the benefits of electricity for pain relief by applying electricity to acupuncture needles. Gustav Fritsch and Eduard Hitzing published an article in 1870, showing that the stimulation of some specific part of the cortex leads to muscle contraction in dogs. Robert Bartholow, who was an American physician, was the first to report the findings of a study on electrical stimulation applied to the cerebral cortex of an awake human in 1874. All of these studies led to the design and introduction of different types of modern electrical stimulators and implantable stimulation devices.

Medical devices for electrical stimulation can be considered to belong to two classes, including implantable electrical stimulators and external electrical stimulators. External electrical stimulators are also categorized as transcranial and transcutaneous electrical stimulators. Some of the well-known external electrical stimulation methods include transcranial Alternating Current Stimulation (tACS), transcranial Direct Current Stimulation (tDCS), transcutaneous Trigeminal Nerve Stimulation (tTNS), transcutaneous Vagus Nerve Stimulation (tVNS) and transcutaneous Electric Nerve Stimulation (tENS). Each type of external electrical stimulator is developed, aiming at a specific application and target. For example, tTNS stimulators are currently commercialized for the relief and prevention of headaches. A commercialized tTNS is designed by Cefaly Technology. Implantable stimulators

including deep-brain stimulation (DBS), retinal implants, pacemakers, cochlear implants and functional electrical stimulation (FES) are employed to assist or restore the functionality of organs that are not properly functioning. Electrical stimulators operate on the principle of initiating an action potential (AP) upon the transfer of electrical charge into excitable tissue. Electrical stimulators operate in three distinct modes, namely voltage-mode stimulation, current-mode stimulation and charge-mode stimulation.

Safety issues of biological cells impose a strict condition on the electrical pulses, i.e., a biphasic stimulation is necessary to prevent tissue damage and any long-term effects such as pH shift (during the usage of IEMDs) and erosion of the electrodes. Biphasic stimulation consists of a cathodic phase followed by an anodic phase. During the cathodic phase, the cell membrane is depolarized. Then, the anodic phase neutralizes the charge which has been injected during the cathodic phase. For exerting a safe stimulation, the voltage across the electrode must be constraint within a specific window. In addition, to block any direct current passing through the tissue, a large off-chip capacitor, namely a blocking capacitor, is placed in series with the stimulation electrode. This capacitor blocks the flow of any dc current through the tissue in case of semiconductor failure which makes the stimulator fail-safe. However, in multichannel stimulators such as retinal or cochlear implants, a large silicon area cannot be allocated to the large blocking capacitors. Hence, various circuits are proposed in the literature to substitute blocking capacitors with active circuits and to reduce the overall size of the stimulation system.

#### *2.2. Physiological Signal Recording*

The current healthcare systems are expected to deal with two major issues including chronic diseases and global population aging [17–19]. Early detection, as well as timely treatment of diseases, require monitoring systems that allow physicians to closely monitor the physiological signals of their patients. Acquiring physiological signals requires robust, light-weight and low-power wearable or implantable electronic medical devices. There are several important physiological signals that provide vital information of the human body such as electromyogram signals (EMG), electroencephalogram signals (EEG), electrocardiogram signals (ECG), phonocardiogram signals (PCG), electroretinography signals (ERG) and photoplethysmograms (PPG). Each of these signals has its own electrical characteristics, i.e., they have specific bandwidth and maximum amplitude. The bandwidths of some of these signals are depicted in Figure 3.

**Figure 3.** Bandwidths of some vital physiological signals (modified from [20]).

Although epilepsy can be detected from different physiological signals, the most important signal to predict or to detect a seizure onset are neural signals recorded from the brain. The following Subsection describes the characteristics of different types of neural signals that can be recorded from different types of electrodes.

#### Neural Signal Recording

Historically, electrophysiology is based on the discovery of Italian scientist Galvani (1737–1798). Galvani realized that the tissues of frog muscles exhibit electrical potential. Studies on the living tissues continued until Hans Berger (1873–1941) discovered the human electroencephalogram (EEG). Berger could record the first human EEG signal using a Siemens double-coil galvanometer in 1924. When Berger's work was confirmed, EEG started to be used in clinical trials. Although vacuum tubes amplifiers were available since 1906, it took long time before they were used in the recording of neural activities. A group of researchers including Fredrick Gibbs (1903–1992), Hallowell Davis (1896–1992) and WG Lennox (1884–1960) with an EEG technician, namely Erna Gibbs (1906–1987), demonstrated EEG signals corresponding to a clinical absence attack [21] in 1935. They showed that the inter-ictal signals corresponding to an absence seizure attack have a specific signature of three spikes per second. Shortly after, in 1936, F. Gibbs demonstrated the importance of EEG in diagnosis and localization of epileptic seizures. Collaboration between F. Gibbs and Albert Grass, a MIT graduate, resulted in the development of EEG recording systems which smoothed the way for Grass Instrument Company. Grass Instrument Company was founded by A. Grass and his wife, Ellen, in 1945 and was acquired by Astro-Med Inc. in 1994, two years after the death of A. Grass. In 1946, the American EEG society (AEEGS) was founded and a year after, in 1947, the first annual meeting of the American EEG society was held in Atlantic City, NJ on 13–15 June [22]. In 1950s, William Grey Walter (1910–1977) developed the first EEG topography machine which can show a map of brain activity. During this decade, Wilder Penfield (1891–1976) and Herbert Jasper (1906–1999), neurosurgeons at the Montreal Neurological Institute, developed electro-corticography (ECoG) as a part of a surgical procedure for treating patients with severe epilepsy. The brain activity is recorded from the cerebral cortex using ECoG. It is shown that recording using ECoG has higher spacial resolution than EEG. Thus, this method is preferred for finding the regions of the cortex that generates epileptic seizures. Until now, recording brain waves has evolved thanks to the improvements in the technology of the electrodes. Nowadays, recording the activity of a single neuron is possible.

Decoding the functional operation of the brain requires recording of the electrical activity of neurons of the central nervous system [23]. Understanding and diagnosing neurological disorders such as epilepsy are based on the recording of the brain electrical activities. Furthermore, neural recording is a major module of brain–machine interfaces and neuroprosthetic technologies that aim at aiding paralyzed patients [24–26]. The demand for technologies that empower the neuroscientist and clinicians to observe the electrical activity of a large population of neurons in the brain has increased in the last decade. Extremely complex circuit solutions are needed to simultaneously monitor a large population of neurons in the brain that exceeds hundreds of cites in some applications [27], Simultaneous recording of the neural activities requires low-noise, low-power and area-efficient amplifiers. Good gain matching over the various channels and low crosstalk over the channels are other criteria.

The neural signals of interest for the recording of the brain activity using implantable devices have a frequency band of 1 Hz to 5 KHz [28]. As shown in Figure 4 [25,28–30], these signals consist of local field potentials (LFPs) and action potentials (APs) which are shown to include biomarkers that are useful for diagnosis and therapy of neurological disorders. LFP signals occupy a frequency band of 1 Hz to 100/200 Hz and a voltage range of 0.5–5 mV. AP signals have a frequency band of 100/200 Hz to 5 kHz with a voltage range of 50–500 μV. In addition, recording electrodes introduce background noise due to their resistance. This noise is thermal noise. Typically, noise integrated in the LFP bandwidth is smaller than 2 *μVrms* whereas noise integrated into the AP bandwidth is smaller than 5 *μVrms*. On account of the dynamic range of the LFPs and APs, an ADC with an effective number of bits (ENOB) larger than 12 to 14 bits is required for recording LFPs and an ADC with an ENOB higher than 8 to 10 bits is required for recording APs.

**Figure 4.** Signals characteristics in neural recording systems.

Signal conditioning is the first step to accommodate the signal before digitizing it to ensure accurate reading of neural activities. As shown in Figure 5, a low-noise amplifier (LNA) is the first stage in the front-end acquisition chain, in which neural signals of small amplitudes are amplified. In this stage, the amplifier should offer a high-pass filtering behavior in order to filter out the large electrode dc offset (EDO). Classically, the gain and bandwidth of the LNA is fixed. Following the LNA, a programmable-gain and bandwidth amplifier (PGA) is used to maximally cover the input range of the ADC that follows in the signal conditioning chain. Analog-to-digital converters (ADCs) are normally used to digitize acquired signals into data prior to its further processing or transmission from the implanted device to the outside of the body. One of the most important features of a neural amplifier is its input impedance. Since most of the electrodes present a 1 kHz impedance (*Z*1*kHz*) of less than 200 kΩ [31], the input impedance of the amplifier should be much larger than this value to accurately record the neural signals.

In a conventional sensing system, one amplifier is designed at each sensor site. To ensure matching and to reduce the level of the noise, the input stage of the amplifier consumes the largest portion of the power that is provided which allows it to satisfy matching constraints while reducing the electronic noise level. Different signal-to-noise ratio (SNR) are required to record different types of neural signals, and thus, the amplifiers are tuned for the specific targets of recording. Reducing the amplifier internal noise and its power are a trade-off that must be optimized in careful consideration of the target signals and application.

Recording electrodes are required to transduce neural signals that consist of transmembrane ion exchanges into electrical signals that can be processed by microe-elctronic and information systems. The type of electrode is adapted to the target of the neural recording in the brain. Different types of neural recording electrodes are depicted in Figure 6. A brief description of different types of electrodes is provided in the following:

• Electroencephalography (EEG) Electrode ([32–34]): EEG electrodes are placed on the surface of the scalp. The international 10–20 system is a well-known and internationally recognized distribution of each of the EEG electrodes on the scalp. EEG recording offers several applications including brain-machine interfaces (BMI), polysomnography (PSG) for a sleep study, seizure detection, as well as other medical applications aiming at brain research. EEG recording is not an invasive method. The amplitude and bandwidth of the neural signals recorded by EEG electrodes are significantly smaller than the signals recorded by implantable electrodes due to

the filtering behavior of cerebrospinal fluid (CSF), dura, skull and scalp. Furthermore, the fragile EEG signals are more exposed to different sources of artifacts including patient-related artifacts (e.g., movement, sweating, ECG, eye movements) and technical artifacts (50/60 Hz artifact, cable movements, electrode paste-related). The bandwidth of the EEG signals lies in the bandwidth of the LFP signals.

• Intracranial Electroencephalography (iEEG) [35]: Recording the neural signals inside the skull provides better signal quality in terms of signal-to-noise ratio and bandwidth. Intracranial EEG recording can be done using different types of electrodes including epidural electro-corticography (ECoG) electrodes, subdural ECoG electrodes, intracortical electrodes and depth electrodes.

**Figure 5.** Simplified block diagram of a practical neural recording system.

**Figure 6.** Placement of the different types of electrodes for neural recording.

Epidural ECoG electrodes [36,37]: this type of electrode is placed between the dura and skull. Hence, the dura need not be incised and opened for the placement of the electrodes on the cortex. This type of electrode is implemented in both micro-electrode and macro-electrode arrangements. Epidural electrodes are either strips of electrodes or a multicontact array as shown in Figure 7a. Epidural ECoG electrodes are suitable for both recording and stimulation.

**Figure 7.** Different types of electrodes for neural recording including (**a**) grid array for intradural or subdural recording, (**b**) laminar intracortical electrodes, (**c**) intracortical micro-electrode array (Reprinted with permission of Wiley Periodicals, Inc., © 2003 The American Laryngological, Rhinological and Otological Society, Inc.), [38]) and (**d**) SEEG electrode (Courtesy DIXI medical).

Subdural ECoG electrodes [39]: This type of electrode is placed between the dura and the surface of the cortex. In this case, the dura must be incised for the placement of the electrodes on the cortex. This type of electrode is implemented in both micro-electrode and macro-electrode arrangements. The design of the Subdural electrodes is identical to Epidural electrodes as shown in Figure 7a. Subdural electrodes record LFP and AP signals, also depending on the size of the electrode. The main advantage of Subdural electrodes is the large coverage of the brain which enables performing a wide range of cognitive studies. Subdural ECoG electrodes are suitable for both recording and stimulation.

Intracortical electrodes [40]: intracortical electrodes are mainly micro-electrodes that are designed to record the signals from different layers of the brain. This type of electrode can perform single-unit recording, where a unique integrated circuit embeds the amplifier(s) or multi-unit recording where several integrated circuits embed multiple amplifiers, depending on the size of the electrode. In general, there are two types of intracortical electrodes including Laminar electrodes (Figure 7b) and micro-electrode arrays. These are also commonly known as neural probes or shank-based electrodes, e.g., [41]. The Utah electrode (Figure 7c) is a micro-electrode array contains up to 96 electrodes which enable high-density multi-channels recording from a large population of neurons, providing valuable data by delivering high spatial resolution within a small area of the brain.

Depth electrodes [42]: depth electrodes are placed at a precise location in the brain using a stereotactic system. Hence, this type of electrodes is also called stereoelectroencephalography (SEEG) electrodes. SEEG electrodes are suitable for recording and stimulation. Figure 7d shows a SEEG electrode manufactured by DIXI Medical [43].

#### *2.3. Additional Blocks of Closed-Loop Epilepsy Control System*

Neural signals that originate from several electrodes that are distributed over the surface of the cortex are digitized and can then be processed. Filtering and compressing are two classical digital processing, with which encryption has recently been complemented. In terms of the core functionality of the epilepsy control implantable system, seizures should be detected from the multi-channel recorded data. Algorithms aiming at seizure onset detection and seizure prediction should be accurate in terms of sensitivity and specificity such as to conceive durable and long-lasting IEMDs for seizure prediction and closed-loop stimulation. In addition, algorithms that are used in closed-loop stimulation systems should have a tolerable latency. An ideal seizure detector used in a closed-loop system has a sensitivity of 100% and a specificity of 100% or a false alarm rate (FAR) of zero. The importance of sensitivity is higher than specificity. Indeed, with a perfect sensitivity of 100%, a closed-loop stimulation system can detect and suppress all seizures. Improving the specificity of the seizure detector enables the closed-loop stimulation system to save power by reducing the periods of unnecessary stimulation. Furthermore, the seizures should be detected in advance or with small latency such as to be suppressed by the stimulation. Failing to satisfy the latter constraint, stimulation may not be effective for suppressing the seizures [44].

Delivering power to implanted devices is one of the major challenges in the design of IEMDs. Powering should be carried out over a wireless link, through the living tissues consisting of the scalp or skin layers. If the powering is not efficient, excessive heat due to power loss may damage the tissues around the IEMDs or may cause strange and unwanted sensations to the patients. Powering an IEMD requires applying several technologies including implantable batteries, energy harvesting or wireless power transfer. Two types of batteries are suitable for IEMDs consisting of primary batteries that are non-rechargeable, and secondary batteries that are rechargeable. Medical-grade battery equipped IEMDs are designed to include such technology. Several companies in the world provide medical batteries such as Eaglepicher [45]. Energy harvesting, which is also known as energy scavenging or ambient powering, is a process in which energy is captured from the environment and stored for further usage in small and ultra low-power systems. Energy sources may originate from external sources including solar power, thermal energy, or kinetic energy. Wireless power transfer is also a method that is suitable to power an implantable electronic system. Wireless power transfer (WPT) is also used to recharge the secondary medical-grades batteries in IEMDs.

Implantable electronic medical devices also require a wireless data transceiver. A data transceiver or a data receiver are implemented depending on the type of IEMD that is used for epilepsy control. An IEMD operating in an open-loop stimulation does not necessarily require the presence of a wireless data transmitter; however, a data receiver is needed for setting the stimulation parameters. On the other hand, an IEMD used for intracranial recording or closed-loop stimulation requires a wireless

data transceiver to set the internal parameters as well as to send the recorded data to an external base station.

#### **3. Commercial Systems and Products for Epilepsy Control**

Commercial devices for seizure alerting and epilepsy control are reviewed in this section. These devices are considered to be partitioned into two categories, namely invasive medical devices, or implantable electronic medical devices, and non-invasive medical devices.

#### *3.1. FDA Approved Implantable Electronic Medical Devices*

Commercial IEMDs that are proposed in epilepsy control therapy apply electrical stimulation. These devices may operate as open-loop or closed-loop devices. In the case of open-loop stimulation, electrical stimulation is applied to the brain or group(s) of nerves without detecting any feedback from the body. On the other hand, closed-loop devices record physiological signal(s) and process them to adapt the stimuli. A closed-loop system would only stimulate the brain or a group(s) of nerves upon detection of seizure onset.

Although the quality of epilepsy detection and control of IEMDs is significantly better than it is in non-invasive medical devices, employing IEMDs presents several disadvantages, including the necessary surgery, perioperative risks, as well as side-effects such as hoarseness, throat pain, coughing, dyspnea, paresthesia, to quote the most prominent [46]. The potential of causing these symptoms enters into the decision of implanting and the selection of the device type. Governmental agencies approve devices and systems that can be implanted; these agencies are local to a country or group of countries, and include the U.S. Food and Drug Administration (FDA, USA), the Chinese National Medical Products Administration (NMPA, China; formerly the China Food and Drug Administration, CFDA), the European Economic Area CE Marking (CE marking, Europe).

#### 3.1.1. Vagus Nerve Stimulation Therapy

Vagus nerve stimulation (VNS) therapy was approved by the FDA in July 1997, for the treatment of epilepsy in patients who suffer from drug-resistant epilepsy. The request was presented by Cyberonics Inc. [47] which was subsequently renamed LivaNova. The VNS therapy consists of sending mild electrical pulses to the brain through the vagus nerve (generally the left vagus nerve) using an electrical pulse generator as shown in Figure 8. The stimulation parameters are set within a frequency range of 1 Hz to 30 Hz, a current range of 0 mA (no stimuli) to 3.5 mA and a pulse-width range of 130 μs to 1000 μs [48]. The vagus nerve is one of the longest nerves in the body that originates from the brain stem located on both sides of the brain. The vagus nerve is a part of the parasympathetic nervous system which is responsible for recovery and digestion, in particular. The benefits of VNS therapy may include fewer seizures, shorter seizures, better seizure recovery time, decreased seizure severity, less medication, improved mood and memory and generally improved quality of life of epileptic patients. The major adverse effects of VNS include voice alteration, coughing and shortness of breath as well as headache and neck pain. A study [49] of 60 patients with VNS therapy shows that the most common adverse effects that affect more than 20% of the patients were voice alteration (55%) and headaches (22%). In 1997, the premarket approval application (PMA) from FDA for VNS therapy (PMA number of P970003) indicated that this device is approved for use as a therapy in adults and adolescents over 12 years old. In the new version of this PMA with PMA number P970003/S207, the FDA added the patients within 4–11 years old to the VNS therapy approval. Hence, this made VNS therapy the only candidate of IEMDs that are suitable for patients less than 18 years old.

There are three types of vagus nerve stimulators for epilepsy control proposed by LivaNova including the standard model, the AspireSR and SenTiva. The standard model VNS is the earliest device designed for epilepsy therapy which only offers basic programming features. The most common stimulation pattern using this model consists of 30 s of stimulation every 5 min [50]. In the standard model, a magne<sup>t</sup> is provided to apply additional stimulation during a seizure. The new model of VNS

is a closed-loop VNS based on heart rate signal acquisition and processing. This model, namely the AspireSR (Figure 8a), received the CE mark in Europe in February 2014. A study [51] of 66 seizures from 16 patients using a closed-loop VNS shows that the cardiac-based seizure detection presents more than 80% sensitivity. In addition, the severity of the seizures was significantly reduced after 3–5 days of closed-loop stimulation. In general, this study confirmed that using a cardiac-based closed-loop VNS presents an acceptable sensitivity and specificity for triggering the stimulation. The AspireSR can also apply pre-programmed stimulation throughout the day and night. The third and the newest VNS product is SenTiva shown in Figure 8b. This device offers similar characteristics as the AspireSR with some additional features. For example, SenTiva can detect if a patient is lying after a seizure, revealing a potential loss of consciousness; in addition, it can be programmed to apply a different amount of stimulation at different times of the day. The VNS is FDA-approved for MRI under specific conditions. The estimated battery life for SenTiva is 4.9–10 years [52].

**Figure 8.** Vagus Nerve Stimulation (VNS) devices by LivaNova [47] including (**a**) AspireSR and (**b**) SenTiva, (Reprinted with permission; copyrighted material of LivaNova).

#### 3.1.2. Responsive Neurostimulation

The responsive neurostimulation (RNS) therapy is a method for epilepsy control that received FDA approval in 2013. The RNS system, which is placed on the skull of the patient as shown in Figure 9, monitors the neural activity of the brain using two leads. Each lead has four electrode contacts that are used for stimulation and recording. Four amplifiers are used to the aim of recording neural signals. Each lead contains four electrodes that can be assigned to one or two out of four amplifiers [53]. The RNS system can store up to 30 min of ECoG activity [53]. The stimulation parameters are set within a frequency range of 1 Hz to 333 Hz, a current range of 1 mA to 12 mA and a pulse-width range of 40 μs to 1000 μs [54]. The current density limit of the stimulation equals 25 μC/cm<sup>2</sup> in each phase, while the current density is usually much less than the limit [55]. The stimulation can be applied between any two stimulation electrodes or between the electrode and the neurostimulator case. Typical stimulation parameters include a current of 1.5–3 mA, a pulse width of 160 μs, a burst duration of 100–200 ms, and a frequency of 100–200 Hz [54].

The RNS system continuously monitors the brain signals and employs several methods to extract the appropriate feature required for seizure detection. Three algorithms are employed by the RNS system to detect seizures, namely the area, line-length and half-wave algorithms. Physicians can change the algorithms' parameters to obtain an appropriate sensitivity, specificity and latency.

**Figure 9.** FDA approved responsive neurostimulation (RNS) system for closed-loop epilepsy detection and stimulation (courtesy of NeuroPace, Inc.) [53].

Before implanting the RNS system, the patients should undergo several tests. In order to receive the treatment from the RNS system, the patients should be older than 18 years, and suffer from disabling partial-onset seizure from no more than two foci, and they should be refractory to more than two antiepileptic medications that are properly chosen [53].

In [56], 230 patients with implanted RNS system were studied over time to measure the average decrease in seizures. This study shows that the average decrease in seizures was 44% after the first year, 53% after the second year and up to 66% after 3 to 6 years from implanting the RNS. This trend was observed among the patients who were followed over 7 years and the seizures were decreased by 72%.
