Funny & Jokes

A fully-asynchronous low-power implantable seizure detector for self-triggering treatment

A fully-asynchronous low-power implantable seizure detector for self-triggering treatment
of 10
All materials on our website are shared by users. If you have any questions about copyright issues, please report us to resolve them. We are always happy to assist you.
Related Documents
  IEEE TRANSACTIONS ON BIOMEDICAL CIRCUITS AND SYSTEMS, VOL. 7, NO. 5, OCTOBER 2013 563 A Fully-Asynchronous Low-Power ImplantableSeizure Detector for Self-Triggering Treatment Marjan Mirzaei, Muhammad Tariqus Salam  , Student Member, IEEE  , Dang K. Nguyen, andMohamad Sawan  , Fellow, IEEE   Abstract—  In this paper, we present a new asynchronous seizuredetector that is part of an implantable integrated device intendedto identify electrographic seizure onset and trigger a focal treat-ment to block the seizure progression. The proposed system hasa low-power front-end bioampli Þ er and a seizure detector withintelligent mechanism to reduce power dissipation. This systemeliminates the unnecessary clock gating during normal neuralactivity monitoring mode and reduces power dissipation in theseizure detector; as a result, this device is suitable for long-termimplantable applications. The proposed system includes analogand digital building blocks with programmable parameters forextracting electrographic seizure onset information from real-timeEEG recordings. Sensitivity of the detector is enhanced by opti-mizing the variable parameters based on speci Þ c electrographicseizure onset activities of each patient. The detection algorithmwas validated using Matlab tools and implemented in standard0.13 CMOS process with total die area of .The fabricated chip is validated of  ß ine using intracranial EEGrecordings from two patients with refractory epilepsy. Total powerconsumption of the chip is 9 and average detection delay is13.7 s after seizure onset, well before the onset of clinical man-ifestation. The proposed system achieves an accurate detectionperformance with 100% sensitivity and no false alarms during theanalyses of 15 seizures and 19 non-seizure datasets.  Index Terms—  Asynchronous, implantable device, seizure de-tector. I. I  NTRODUCTION E PILEPTIC patients suffer from a tendency to recurrentseizures as the result of abnormal neuronal discharges.First line of treatment for epilepsy is oral antiepileptic drugs;however, many of them have systemic side effects and re-maining of the patients are drug resistant [1]. Epilepsy surgeryis an alternative treatment option for the drug resistant (refrac-tory) patients. An accurate localization of the epileptogeniczone and precise analysis may increase success in the surgery. Manuscript received March 15, 2013; revised July 16, 2013 and September 19, 2013; accepted September 21, 2013. Date of publication October 16, 2013;dateofcurrentversionOctober24,2013.Thisworkwassupportedinpartbythe Natural Sciences and Engineering Research Council of Canada (NSERC) andin partby the Canada Research Chair in Smart Medical Devices.This paper wasrecommended by Associate Editor T.-P. Jung.M.Mirzaei,M.T.Salam,andM.SawanarewiththeDepartmentofElectricalEngineering,Polystim Neurotechnologies Laboratory, Polytechnique Montréal,Montréal, QC H3T 1J4, Canada (e-mail: K. Nguyen is with Neurology Service, Centre Hospitalier de L’Universitéde Montréal, Montréal, QC H2L 4M1, Canada (e-mail: versions of one or more of the  Þ gures in this paper are available onlineat Object Identi Þ er 10.1109/TBCAS.2013.2283502 However, patients who have multifocal epilepsy or have risksfrom surgery due to loss of brain functionalities are contin-uously disabling due to the lack of treatment option. Thus,researchers are looking for other alternative treatments for conventionally untreatable patients, such as implantable de-vices delivering focal treatment upon automated detection of electrographic seizures.Over the last decade, there has been a growing interest on im- plantable self-triggering microsystems for treatment of refrac-tory patients. Ef  Þ cacy of the self-triggering treatment dependson precise evaluation of EEG and accurate identi Þ cation of the prede Þ ned electrographic patterns. The automated evaluationand detection performance can be improved by using intracere- bral EEG (icEEG), which is advantageous over scalp EEG dueto less motion/muscle artifacts and lower sensitiveness for brainsignal recordings. However, several other noises may degradethe icEEG recordings and subsequent self-triggering treatmentmay not be effective. The front-end bioampli Þ er has a low dy-namic input range (microvolt) and the dc offset due to elec-trode-tissue interface and 60 Hz noise may saturate the bioam- pli Þ er  [2]. Moreover,  ß icker noise due to fundamental physics propertyofinstrumentationandthermalnoiseduetointernalre-sistanceof instrumentation andwire resistance contributemajor amount.Following the ampli Þ cation and  Þ ltering stages, recordingsneed to be further evaluated to detect abnormal electrographic patterns.Frompatienttopatient,seizureonsetpatternsmayvaryintermsofonsetmorphology,dischargefrequency,focality,andspread pattern. The most common seizure onset pattern is char-acterized by a low-voltage high-frequency discharge [1]. In pa-tients with simple partial epilepsy, electrographic seizure ac-tivities begin from epileptogenic zone and spread to adjacentregions while in patients with generalized seizure, the electro-graphic seizure activities spread to a broad region of brain andusually lead to visible clinical manifestations. As a result, the behavioral changes such as shaking of the body, disability of motor functions, and loss of consciousness are evident.Sofar,severaltechniquesareinvestigatedforthetreatmentof refractory patients who suffer from focal epilepsy. Deep brainstimulation (DBS) for Parkinson and vagus nerve stimulation(VNS) for epilepsy disease are examples of open loop seizuretherapies. In these techniques the electrical stimulation is ap- plied to deep brain (DBS therapy) or to extracranial VNS (VNStherapy), respectively. The commercially available VNS pro-vide scheduled stimulation at predetermined time intervals toreduce seizure frequency. However, seizure freedom is rare andonly 30 to 40% of patients have attenuation in seizure [2]. Incontrast to these open-loop systems, the closed-loop devices 1932-4545 © 2013 IEEE  564 IEEE TRANSACTIONS ON BIOMEDICAL CIRCUITS AND SYSTEMS, VOL. 7, NO. 5, OCTOBER 2013 (detection and treatment) provide seizure alarms prior to clin-ical manifestations and further triggers focal treatment in order to abort seizures at their onset. Delivering focal treatment (e.g.,drug,electricalstimulation,cooling) canbeused byclosed-loopsystems at certain necessary times to provide effectively use of therapy and reduce the amount of medications. This may in-crease the ef  Þ ciency and safety of these systems compared toopen-loop devices. Responsive Neurostimulator (RNS) is a re-sponsive implantable device has been submitted for Food andDrug Administration (FDA) approval. This device is useful on patients with simple or complex focal seizure. As such, in thelast few years a growing interest for developing seizure de-tection methods can be seen. Some of them are desktop com- puter-based algorithm for of  ß ine detection [3]–[8] and the re-maining are integrated microsystems for real-time evaluationanddetection[2],[9]–[14].Generally,thecomputer-basedalgo- rithms use heavy mathematic computations to increase the sen-sitivity and speci Þ city of detectors [3]–[8]. However, tradeoff  between their accuracy in detection and complexity in analysismakes it complicated to implement them in an integrated circuitdue to the high power dissipation.Recently, several power ef  Þ cient detection algorithms have been proposed which were implemented in CMOS-basedmicrosystems and suitable for implantation [2], [9]–[19]. In [17], the patient-speci Þ c seizure onset activities are recordedand detected using 8 analog front-end channels and a ma-chine-learning seizure classi Þ cation. The seizure detector isimplemented in 0.18 1P6M CMOS process with total diearea of 25 . The functionality of the detector is veri Þ edwith ra pid eye blink patterns and children’s database thatshowed 84.4% accuracy in classi Þ cation test. However, power consumption of 66 is obtained only from analog front-endchannels. The seizure detector in [18] proposed patient speci Þ cseizure detection with multichannel feature extraction. Thissystem is trained with rapid-eye blinks characterized by 10 eye blinks within a 5 s window. The proposed system exhibitedmore than 95% detection accuracy and less than 1% falsealarm. However, the total power consumption of the seizuredetector is not mentioned. In [19], fully integrated seizuredetection with an adaptive neural stimulation is presented.This system is implemented in 0.18 CMOS occupying13.47 . It shows detection accuracy more than 92% within0.8 s and power dissipation of 2.798 mW. Power managementin these microsystems is an important issue for the implanta-tion. The power consumption is mainly dependent on signaltransitions in a device, such as charging and discharging of  parasitic capacitances in transistors and short-circuit currentsduring switching [20]. Thus, power consumption can be further reduced by avoiding the unnecessary signal transitions [20]. Allmodules of the synchronous designs [2], [10], [13]–[19] share a common clock signal distributed throughout the circuit, and theunnecessary signal transitions arise from clock gating. Becauseof these limitations in the conventional synchronous circuitdesign, an asynchronous technique (data-dependent analysis)is likely to become more popular.In this paper, a low-power asynchronous seizure detector (ASD) is proposed for self-triggering treatment microsystems.Power dissipation in the detector is managed by eliminating the Fig. 1. A diagram showing the application of proposed ASD in a closed-loopseizure detection and therapy. Seizure detector identi Þ es the progress increaseof low-voltage fast-activity at seizure onset and then triggers local therapy inorder to abort seizure prior to clinical manifestations. unnecessary clock skew and clock tree; as a result, transistorsdonotchangetheirtransientstateinpowersavingmode(icEEGmonitoring period) unless an abnormal event detected. Thissystem contains low noise front-end bioampli Þ er, digital signal processor and a detector. The ASD can identify electrographicseizure onset precisely and trigger an electrical stimulator for focal treatment prior to clinical manifestations. Fig. 1 showsthe high-level diagram of the proposed asynchronous seizuredetector in a closed loop detection and therapy system.Inthefollowingsections,thebackgroundworkandalgorithmof the proposedintegrated ASDare described in Section II.Sec-tion III highlights the design and implementation of the pro- posed circuits. The experimental tests and validation results areillustrated in Section IV and conclusion is summarized in Sec-tion V.II. P ROPOSED  S YSTEM  A. Background Works Our previous work proposed several low-power seizure de-tectors and stimulator/inhibitor for self-triggering treatment inrefractory epilepsy patients [11]. Time-frequency and time-am- plitudeanalysiswerethemaintoolsfor detecting electrographicseizure onset while minimizing the false recognition of unre-latedseizure activities.Moreover,compared to synchronousde-sign [14], asynchronous design [11] reduced the total power dissipation. These systems were assembled with discrete com- ponents on printed circuit board (PCB) illustrated in Fig. 2(a)and (b). Furthermore, the synchronous design was fabricatedin micro-chip [Fig. 2(c)] in order to reduce power dissipationand miniaturization [2]. These detectors monitor icEEG record-ings using electrodes [Fig. 2(e)] on real-time and upon seizureonset detection, trigger a current stimulator [Fig. 2(d)] [16] or direct drug delivery system [Fig. 2(f)] [11] in order to suppressthe seizure. Among these detectors, the asynchronous detector has demonstrated smart power saving technique (reduced 45%in power saving mode compared to synchronous prototyped ar-chitecture)[11].However,thePCB-basedasynchronousseizuredetector was neither implantable (area 1963.5 ) nor low- power (47200 ) for long-term implantable device therapies.Thus, a novel integrated circuit design is used to improve the  MIRZAEI  et al. : A FULLY-ASYNCHRONOUS LOW-POWER IMPLANTABLE SEIZURE DETECTOR 565 Fig. 2. Background works on seizure detection and stimulation/inhibition.(a) Synchronous seizure detection prototype made up of discrete components[14]. (b) Asynchronous seizure detection prototype made up of discretecomponents [11]. (c) Synchronous seizure detection micro-chip [2]. (d) Twochannels current stimulator [16]. (e) Hybrid surface electrodes compatible for stimulation and drug delivery [11]. (f) Drug delivery system [11].  power management, noise, size, and performance of the pro- posed asynchronous seizure detector.  B. Detection Algorithm The proposed power ef  Þ cient detection algorithm processesanalog and digital signals and detect an electrographic seizureonset. At  Þ rst, input analog signals are ampli Þ ed to the desiredamplitude range and  Þ ltered out all unnecessary frequency con-tentsandnoises[Fig.3(a)].Theoutputofprocessedsignals passes through four voltage window detectors (VWDs) to  Þ ndneural activities in lower and higher amplitudes, in both posi-tive and negative sides of the signal amplitudes. Outputs of theVWDs de Þ ne the strength of neural signal activities indifferent amplitude ranges (1).'' (1)where , 2, 3, 4. is the threshold voltage of VWDs. Several frequency analyzers extract and measuresignal frequency using a time frame , as shown in(2).(2)If is greater than speci Þ c seizure onset frequencyin both lower and higher amplitudes, the output is consid-ered fast activity (3).'' (3)Seizure detection logic analyzes and quanti Þ es speci Þ cfeatures characterized by a progressive increase in amplitudeand high frequency. Thus, a seizure onset is declared based onthe following conditions (4).'' (4)and are tuned to the speci Þ c patient to min-imize false alarms. The proposed system requires manual ad- justment. The digital-to-analog converters (DAC) and buffersare used to adjust the threshold voltages. DAC and buffers areoff chip. These parameters can optimize the decision boundaryand enhance sensitivity and speci Þ city of the system.III. I MPLEMENTATION OF THE  P ROPOSED  A LGORITHM This ASD consists of two main analog and digital building blocks. The analog one is composed of bioampli Þ er,  Þ lteringlevel, gain stage, and fully differential to single-ended con-verter. The digital part contains VWDs and high-frequencydetectors. Details are given below.  A. Bioampli   Þ er Stage A fully differential bioampli Þ er with standard contin-uous-time common-mode feedback (CMFB) is used to amplifylow-voltage icEEGinputsignals.Althoughthefullydifferentialstructures are used with CMFB circuits and occupy more areacompared to similar single-ended designs, they exhibit lessnoise and show better performances. Fig. 3(b) shows schematicof the used bioampli Þ er that is able to reject the large andrandom dc offset voltages caused by electrode mismatches[21].  B. Filtering and Gain Stage The architecture of fully differential  Þ lter is shownin Fig. 4(a). Analyzing the frequency range of seizure signals[2], this low-pass  Þ lter is added at the output of front-endampli Þ er to attenuate the 60 Hz noise. Designing a low-pass Þ lter with very low cutoff frequency requires an operationaltransconductance ampli Þ er (OTA) with lower transconduc-tances or larger integrated capacitors. Concerning the chip area,low value transconductances are used. This leads to designtransistors with long lengths to minimize the current of OTAs.Since matching such geometries is dif  Þ cult from a layout per-spective, current splitting and source degeneration techniquesare used. Fig. 4(b) shows the schematic of a fully differentiallow-transconductance OTA where small-signal currents intransistors , and are divided by ratio in their sizes.The source degeneration technique increases the linearizationof   Þ lter. For this structure, the source-degeneration resistors arerealized by designing transistors and in triode region.of these transistors, and thus, their resistance is controlled by , , and .An additional ampli Þ cation stage is used at output of the Þ lter.This ampli Þ er usesthe same structure in Fig.3(b)with small capacitor ratio that provides the overallgain of 60 dB for analog building block.  566 IEEE TRANSACTIONS ON BIOMEDICAL CIRCUITS AND SYSTEMS, VOL. 7, NO. 5, OCTOBER 2013 Fig. 3. Proposed integrated seizure detector. (a) Block diagram of the asynchronous system, where is the recorded icEEG signal in the input of system, Vais the ampli Þ ed signal in the output of analog stage, are the outputs of voltage window detectors (VWDs), are the outputs of high-frequencydetectors, and is the  Þ nal output of integrated ASD. (b) General structure of the bioampli Þ er stage [21].Fig. 4. Schematic diagram of the low-pass  Þ lter. (a) low-pass  Þ lter.(b) Operational transconductance ampli Þ er used to implement the Þ lter. C. Differential Difference Ampli   Þ er  The fully differential inputs are converted to single endedoutput using a differential difference ampli Þ er (DDA). This block also works as a buffer between analog and digital stagesto minimize the effect of jitter and kickback noise. Fig. 5 illus-trates schematic diagram of the DDA that consists of two main parts, 1) a differential-input single-ended output transconduc-tance stage that converts two pairs of differential input voltagesinto a single subtracted output current and 2) a gain stage which provides the output voltage [22]. The compensating capacitor and resistor ( and ) are used to stabilize the circuit.  D. Seizure Detector  The detection stage contains four VWDs and four high-fre-quency detectors. The VWDs (Fig. 6) detect progressive in-crease in amplitude of icEEG recordings. The threshold volt-ages of VWDs are widely programmable whichare generated digitally using digital-to-analog converters and Fig. 5. Schematic diagram of the differential difference ampli Þ er. (a) Symbolof DDA where the  Þ rst port is used as an front-end op-amp and the second portconsists of a voltage adder. (b) The circuit design of DDA [22].Fig. 6. (a) Schematic diagram of the voltage window detector. (b) Circuitdesign of the comparator.  buffers [16]. These threshold voltages divide the ampli Þ ed and Þ ltered icEEG signals into delimited intervals named voltagewindows. Our previous studies showed that instead of usingsigni Þ cant number of VWDs, two VWDs in positive side of the icEEG and another two in the negative side are suf  Þ cientto detect seizure onset [11]. The VWDs are adjusted to the pa-tient speci Þ c seizure onset to prevent false seizure detections.Moreover, the implemented tuning parameters separate normaland hyperexcited activities; as a result, some modules of detec-tion stage are turned off during normal icEEG recordings. This processing method allows saving power dissipation in detection  MIRZAEI  et al. : A FULLY-ASYNCHRONOUS LOW-POWER IMPLANTABLE SEIZURE DETECTOR 567 Fig. 7. The proposed asynchronous seizure detector. (a) Layout of theintegrated chip. (b) Photograph of the fabricated chip. stage. Furthermore, the VWDs can detect small magnitude vari-ation in the icEEG recordings that enhances the sensitivity andspeci Þ city of the system for detecting the seizure activities withvariable patterns, magnitude, and spectral contents.The high frequency detectors contain the asynchronouscounters (using four D  ß ip- ß ops) and various logic gates. Thecounters detect high-frequency events by counting the number of neural activities in a variable time frame. The time frameis tuned externally based on the patient speci Þ c seizure onsetfrequency for accurate seizure detection with minimum falsealarms.IV. E XPERIMENTAL AND  V ALIDATION  R  ESULTS  A. Circuit Validation Results The asynchronous seizure detector has been validated  Þ rstusing discrete components in 50 mm diameter PCB [11], thenimplemented in area micro-chip using CMOS0.13 process. The circuit design and integrated circuitlayout were validated using Spectre simulator (Cadence tools)and seizure detection performance was evaluated [12]. Later,two samples of the fabricated micro-chip were used for mea-surements and the experimental tests promised the consistencyof test results.Fig. 7 illustrates both photographs of the layout and the fab-ricated micro-chip. Experimental result of the analog block’sfrequency response is demonstrated in Fig. 8(a). The randomdc offset of electrodes and  ß icker noise of circuits were re-duced in the front-end stage to prevent saturation of bioampli- Þ er.Thermal noisefrom analogcircuitswasattenuatedbyusinga low-pass  Þ lter. In Fig. 8(b), the performance of asyn-chronous seizure detector is shown for a set of 15 mV voltageinterval, 20 Hz signal frequency, and 1 s time frame. Table Isummarizes the measured speci Þ cations of the proposed ASD.  B. Patients Description This study was conducted at Notre-Dame Hospital, CentreHospitalier de l’Université de Montréal (CHUM) and exper-imental protocol were approved by ethics committee. Two patients with refractory focal epilepsy were quali Þ ed for these studies based on electrographic seizure onset. These Fig. 8. Measured results. (a) Gain and frequency response of the analog building block. (b) Validation of the asynchronous seizure detector, whereis the ampli Þ ed signal in the output of the analog stage, is the output of the  Þ rst voltage window detector, is the time frame, and is the outputof the  Þ rst channel changing state after counting 12 pulses.TABLE IM EASURED  F EATURES OF THE  F ABRICATED  A SYNCHRONOUS S EIZURE  D ETECTOR   patients had previously undergone a complementary nonin-vasive study of brain (e.g., video-scalp EEG, brain magneticresonance study (MRI), ictal single photon emission com- puted tomography (SPECT), positron emission tomography(PET), EEG-functional MRI (EEG-fMRI), and magnetoen-cephalographic (MEG) study) which were failed to localizethe epileptogenic zone effectively. Hence, the invasive studieswere recommended for better delineation of epileptogeniczone. In these studies, signals were recorded by implementingintracranial electrodes through a craniotomy or burr holes under general anesthesia. Each patient was monitored about 3 weeksand an average of 7 seizures per patient was recorded. IcEEGrecordings of these patients acquired using commerciallyavailable equipments (PRO-36 ampli Þ ers, Stellate HarmonieSystem) with a gain of 1000 V/V, a 0.1–70 Hz bandwidth,200 Hz sampling frequency, and 8 bits of digitization resolution[9]. Following the recordings, the seizures and epileptogeniczone were marked by epileptologist (DKN). The recordedicEEG of these patients were used for validation of our asyn-chronous seizure detector.The  Þ rst patient was a 36 year-old male with left frontotem- poralregionalonsetseizuresincetheageof30years.TheicEEGrecording of this patient was characterized by low and high fre-quency preictal spiking, and brief electrical seizure activities.Fig. 9(a) and (b) show the MRI and 3D images of the implantedelectrodes for this patient, respectively. The 3D image of elec-trodes is reconstructed using grid view software. The circum-scribed area in these  Þ gures shows the position of two contacts
Related Search
We Need Your Support
Thank you for visiting our website and your interest in our free products and services. We are nonprofit website to share and download documents. To the running of this website, we need your help to support us.

Thanks to everyone for your continued support.

No, Thanks