Machine-based patient-specific seizure classification system
View Patent ↗This disclosure is directed to a machine-based patient-specific seizure classification system. In general, an example system may comprise a non-linear SVM seizure classification system-on-chip (SoC) with multichannel EEG data acquisition and storage for epileptic patients is presented. The SoC may integrate a hardware-efficient log-linear Gaussian Basis Function engine, floating point piecewise linear natural log, and low-noise, high dynamic range readout circuits. In at least one example implementation, the SoC may consume 1.83 μJ/classification while classifying 8 channel results with an average detection rate, average false alarm rate and latency of 95.1%, 0.94% and <2 s, respectively.
1. A device, comprising:
an analog front-end (AFE) to generate at least one analog signal representative of electroencephalography (EEG) information;
a digital back end (DBE) including a feature extraction (FE) engine and a non-linear support vector machine (NL-SVM) classification engine comprising circuitry configured to perform an antilog function on a summation of a first input and a log conversion of a second input to determine onset of a seizure based on at least one digital signal corresponding to the at least one analog signal; and
an analog-to-digital converter (A/D) stage to convert the at least one analog signal provided by the AFE to the at least one digital signal for use by the DBE.
2. The device of claim 1 , wherein the device is composed entirely on an integrated circuit (IC) in a system-on-chip (SoC) configuration.
3. The device of claim 2 , wherein the SoC further comprises electrodes to sense the EEG information, wherein the SoC is configured to be worn by a patient with the electrodes coupled directly to the patient's skin.
4. The device of claim 1 , wherein the analog front end comprises a plurality of channels, each of the plurality of channels comprising at least a chopper-stabilized capacitive-coupled instrumentation amplifier (CS-CCIA) and a variable gain amplifier.
5. The device of claim 4 , wherein the A/D stage comprises a multiplexer to receive analog signals from the plurality of channels and an A/D to convert the received analog signals into digital signals.
6. The device of claim 5 , wherein the A/D is a 10-bit successive approximation A/D (10-b SAR-A/D).
7. The device of claim 1 , wherein the FE engine comprises a plurality of channels, each of the plurality of channels including a time division multiplex band-pass filter (TDM-BPF) to extract feature vectors (FVs) from the EEG information for use in determining the onset of a seizure.
8. The device of claim 1 , wherein the device further comprises a static random access memory to store the EEG information.
9. The device of claim 1 , wherein the circuitry configured to perform the log conversion comprises circuitry to linearize the log function into multiple linear segments to approximate the log function.
10. A method, comprising:
receiving at least one analog signal representative of encephalography (EEG) information;
converting the at least one analog signal to at least one digital signal;
extracting feature vectors (FVs) from the at least one digital signal; and
determining onset of a seizure based on the extracted FVs using a linear support vector machine (NL-SVM) classification engine configured to perform an antilog function on a summation of a first input and a log conversion of a second input; and
if the onset of a seizure is determined, causing electrical stimulation to commence to prevent occurrence of the seizure.
11. The method of claim 10 , wherein the analog signal comprises a plurality of electroencephalography (EEG) information signals received via an analog front-end (AFE) comprising a plurality of channels, each of the plurality of channels including at least a chopper-stabilized capacitive-coupled instrumentation amplifier (CS-CCIA) and a variable gain amplifier.
12. The method of claim 11 , wherein the analog signals are converted into digital signals via an analog-to-digital converter (A/D) stage comprising a multiplexer to receive the plurality of analog signals from the plurality of channels and an A/D to convert the received analog signals into digital signals.
13. The method of claim 12 , wherein the A/D is a 10-bit successive approximation A/D (10-b SAR-A/D).
14. The method of claim 10 , wherein the FVs are generated by a feature extraction (FE) engine comprising a plurality of channels, each of the plurality of channels comprising a time division multiplex band-pass filter (TDM-BPF) to extract the FVs.
15. The method of claim 14 , wherein determining the onset of a seizure based on the FVs comprises mapping the FVs into a seven-dimensional (7-D) feature space and classifying the FVs using said non-linear support vector machine (NL-SVM) classification engine.
16. The method of claim 10 , wherein the circuitry configured to perform the log conversion comprises circuitry to linearize the log function into multiple linear segments to approximate the log function.
17. One or more non-transitory computer-readable memories having stored thereon, individually or in combination, instructions that when executed by one or more processors result in the following operations comprising:
receiving at least one digital signal corresponding to at least one analog signal representative of a patient's electroencephalography (EEG) information;
extracting feature vectors (FVs) from the digital signal;
determining the onset of a seizure based on the extracted FVs using a non-linear support vector machine (NL-SVM) classification configured to perform an antilog function on a summation of a first input and a log conversion of a second input; and
if the onset of a seizure is determined, causing electrical stimulation to commence to prevent occurrence of the seizure.
18. The one or more non-transitory computer-readable memories of claim 17 , wherein determining on the onset of a seizure based on the FVs comprises mapping the FVs into a seven-dimensional (7-D) feature space and classifying the FVs using said non-linear support vector machine (NL-SVM) classification engine.
19. The one or more non-transitory computer-readable memories of claim 17 , wherein performing the log conversion comprises linearizing the log function into multiple linear segments to approximate the log function.