AU2021103884A4

AU2021103884A4 - Epileptic Seizure Detection and Classification Using HOG feature based MSCA-ELM Model and Embedded Prototype Development

Info

Publication number: AU2021103884A4
Application number: AU2021103884A
Authority: AU
Inventors: Satyasis Mishra; Mihir Narayan Mohanty; Sreelekha Panda; Asif Basha Shaik
Original assignee: Mishra Satyasis Dr
Current assignee: Mishra Satyasis Dr
Priority date: 2021-07-06
Filing date: 2021-07-06
Publication date: 2022-04-07
Anticipated expiration: 2029-07-06

Abstract

"Epileptic Seizure Detection and classification Using HOG feature based MSCA-ELM Model and Embedded prototype development." Exemplary aspects of the present disclosure are directed towards the Epileptic Seizure Detection and classification Using HOG feature based MSCA-ELM Model and Embedded prototype development, consisting of a plurality of electrodes 101 capable of identifying the brain activity connected to the microcontroller 102. Microcontroller 102, capable of executing machine learning algorithms (MLA) and Deep learning algorithms (DLA) converts the EEG signals into relevant images and performs HOG feature based MSCA-ELM Model to predict different categories of infected and non-infected seizures due to head injury, brain infection, stroke and tumour. Further, images are undergone preprocessing by MWCA (Modified Water Cycle Algorithm) and Wavelet transform-based segmentation technique for detection of seizure from converted images that can inform the doctor about the details of seizure due to tumour or accident. Page 1 of 4 EEG Signal Synchrosqueezing Fejer-Korovkin Wavelet Transform for converting to image Image Dataset preparation and MWCA Preprocessing Image Wavelet segmentation and HOG feature extraction MSCA-ELM Classification Classification comparison Results Seizure Non-seizure Normal Fig-1 Research flow diagram

Description

Page 1 of 4

EEG Signal

Synchrosqueezing Fejer-Korovkin Wavelet Transform for converting to image

Image Dataset preparation and MWCA Preprocessing

Image Wavelet segmentation and HOG feature extraction

MSCA-ELM Classification

Classification comparison Results

Seizure Non-seizure Normal

Fig-1 Research flow diagram

I

TITLE Epileptic Seizure Detection and classification Using HOG feature based MSCA-ELM Model and Embedded prototype development

PREAMBLE TO THE DESCRIPTION The following specification particularly describes the invention and the manner in which it is to be performed.

DESCRIPTION TECHNICAL FIELD

[0001] The present disclosure generally relates to the biomedical domain. Precisely a Medical diagnosis and prediction system utilizing HOG feature based MSCA-ELM Model and implementing the same in a hardware model.

BACKGROUND

[0002] A neurological disorder that disables the brain electrical activity is termed as Epilepsy. If Epileptic seizures arise from a particular area of the brain, then the initial symptoms of the seizure often reflect the functions of that area. Head injury, brain infection, stroke, or tumour lead to brain seizures, but the cause is not known in most cases. Medical imaging is proved as an essential field for automated, reliable, fast and efficient medical diagnosis. It is a complex and challenging task for a doctor to detect and classify the seizure areas in the brain due to tumour or accidental injury. Factors like size, shape, and position of infection vary in different patient's brain.

[0003] Many efforts have been made for image detection and classification in the following prior art by considering machine learning and deep learning models. Motivated by the advancement of machine learning for classification of epilepsy, we are proposing a novel signal to image conversion techniques and a classifier for detection and classification of lung diseases.

[0004] In a prior art US8374696B2, Justin C. Sanchez and Paul R. Camey suggested a Closed-loop micro-control system for predicting and preventing epileptic seizures. The device includes a detection system that detects and collects electrophysiological information comprising

Z

action potentials from single neurons and ensembles of neurons in a neural structure such as an epileptogenic region of the brain in a subject. An analysis system included in the neuroprosthetic device evaluates the electrophysiological information and performs a real-time extraction of neuron firing features from which the system determines when stimulus intervention is required. The neuroprosthetic device further comprises a stimulation intervention system that provides stimulus output signals having a desired stimulation frequency and stimulation intensity directly to the neural structure in which abnormal neuronal activity is detected. The analysis system further analyzes collected electrophysiological information during or following stimulus intervention to assess the stimulation intervention's effects and provide outputs to maintain or modify the stimulation intervention.

[0005] Numerous prior arts have attempted to use automated image processing concepts and predict epileptic seizures but haven't integrated with modern technologies such as deep learning and machine learning in expecting the same and deploying it in a hardware.

[0006] In a prior art titled Patient monitoring system for the real-time detection of epileptic seizures by Ohannes Bernardus et al, in US20090124870A1, they disclosed that a patient monitoring system for the real-time detection of epileptic seizures suffered a user of the monitoring system. The system comprises control means for receiving measuring signals and, conditional upon the measuring signals, generating an alarm signal, at least one hart rate sensor for measuring the heart rate of the user and, subject to the measurement, generating a heart rate signal, and at least one muscular tension sensor for measuring the change and the intensity of the contraction of at least one muscle of the user and, subject to the measurement, generating a muscle contraction signal. The control means are designed for, subject to both the heart signal and the muscle contraction signal, generating an alarm signal when the user suffers an epileptic seizure.

[0007] Various attempts at predicting epileptic seizures are described in the literature. An example of an apparatus and a method is described in WO 99/56821 Al, which describes an implantable electrode and sensor implanted in the body of a patient, preferably in the head, connected to an implantable processor and signal generator which is capable of emitting a stimulating signal via an implantable electrode. The processor is capable of recognizing various patterns and of applying a signal to the signal generator to generate a stimulating signal, so that the person is warned of a seizure.

[0008] In a prior art WO 2007/072425 A2 describes a cuff/band containing a plurality of sensors connected to a processor, which is capable of recognizing various patterns characteristic of an epileptic seizure and of applying an alarm signal to an alarm unit which generates an alarm.

[0009] In a prior art US 2008/0161713 Al describes a system for the prediction of epileptic seizures and comprising a plurality of electrodes connected to a communications unit, which is capable of analyzing the data measured, and which communicates with an external data unit which is capable of giving an alarm or instructions to the user. The communications unit comprises several processor units which extract parameters from the signal measured. The parameters are used by several classification units for classifying the state measured. The classification units are capable of giving a weighted answer which can indicate the probability of an imminent seizure within a predetermined time frame. The system determines the time interval of the coming measurements on the basis of the value of the weighted answer.

[00010] Another prior art attempt at predicting seizures in newborns based on ECG measurements of the heart rate are described in the literature (MALARVILI & MESBAH, "Newborn seizure detection based on heart rate variability"). In the attempt, the entire measurement is stored, following which the heart rate parameters are extracted by means of a QRS algorithm. Then, the heart rate parameters are analyzed both in the time domain and in the time frequency domain to select the suitable data by means of a selection process. The data are then classified in a plurality of classes, which define seizure and non-seizure.

[0011] In Simillar prior art by Tetzlaff in Automated Detection of a Preseizure State: non-linear EEG analysis in epilepsy by Cellular Nonlinear Networks and Volterra systems. Int. J. Ore. Theor. Appl, Vol. 34, 2006 discloses a method of detection of epileptic seizure precursors from EEG data based on deterministic modeling of neurons to capture the behavior of neural networks, seeking to find a time series with the same characteristics of the pre-seizure, which should be previously known.

[0012] In an prior art by Carney in Seizure prediction: Methods. Epilepsy & Behavior, vol. 22, 2011 presents a review of methods of predicting epileptic seizures from EEG, dividing them into univariate and multivariate. Among the univariate are the short-time Fourier transform, accumulated energy, autocorrelation and self-regression modeling, discrete wavelet transform, statistical moments, correlation dimension, correlation density, Kolmogorov entropy, dynamic similarity index, loss of recurrence and local flow and Lyapunov exponent. Among the multivariate are the simple synchronization measure, correlation structure, phase correlation, autoregressive measures of synchrony, index of maximum short term Lyapunov and phase synchronization.

[0013] Similarly in a prior art by Ramgopal in Seizure detection, seizure prediction, and closed-loop warning systems in epilepsy. Epilepsy & Behavior, vol. 37, 2014 also review methods for predicting epileptic seizures from EEG, emphasizing non-linear processing methods, which analyze the spontaneous formation of spatial, temporal and spatiotemporal patterns of brain waves. The methods of entropy of permutation, Kolmogorov entropy, correlation dimension, relative energy wavelet and approximate entropy are cited. [010] Moghim and Came in Predicting Epileptic Seizures in Advance, 2014 present the ASPPR method (Advance Seizure Prediction via Pre-ictal Relabeling), divided into three components: i) selection of 14 characteristics out of 204 for each patient according to the ReliefF criterion from EEG data; ii) preparing the data by separating a part of the data for training the prediction algorithm and iii) each predictive model is trained using a multi-class support vector machine

[0014] Another prior art document US2015282755 deals with a system and method for detecting the occurrence of seizures from EEG signals combined with electrocardiogram (ECG) signals. The method described does not detect the seizure with anticipation. It is proposed a probabilistic model that classifies an event as a seizure or non-seizure.

[0015] Referring to another document US patent document US5857978 relates to a method and apparatus for anticipated and automatic prediction of epileptic seizures from EEG brain wave analysis. The method uses linear metrics (standard deviation, absolute mean deviation, asymmetry, kurtosis) and nonlinear metrics (time-cycle steps, Kolmogorov entropy, first minimum in the mutual information function and correlation dimension) calculated from data from 16 channels of EEG and using four versions of the dataset with different mathematical transformations.

[0016] KR1O1375673 document is an epileptic seizure alert method and device that applies epileptic brain wave data to an excitatory-inhibitory model based on neuronal chaos, calculating an optimal value for the connection coefficient, comparing it with the population data of brain waves of healthy people.

D

[0017] In another prior art document RU249S769 discloses an apparatus for detecting and preventing epileptic activity by means of a microelectrode system, containing a processor, a preamp, a filter, an information stimulation unit and a power source. The microelectrode system is surgically implanted at the site of epileptic activity and functions as a diagnostic recorder or as a neurostimulator. It has a transceiver and micro antenna for transferring and receiving the recorded data. There are no details on data processing, which suggests the use of a conventional technique for this.

[0018] The present invention provides a HOG (Histogram of Oriented Graph) feature based hybrid MSCA (Modified Sine Cosine Algorithm)-Extreme Learning Machine model to classify different categories of infected and non-infected seizures due to head injury, brain infection, stroke and tumor. Further, a novel Synchrosqueezing Fejer-Korovkin Wavelet Transform has been proposed for converting EEG signals to images. The images are undergone preprocessing by MWCA (Modified Water Cycle Algorithm) and Wavelet transform based segmentation technique for detection of seizure from converted images that can inform the doctor about the details of seizure due to tumor or accident.

[0019] The present invention addresses the shortcomings mentioned above of the prior art.

[0020] All publications herein are incorporated by reference to the same extent as if each publication or patent application were specifically and individually indicated to be incorporated by reference. Where a definition or use of a term in an incorporated reference is inconsistent or contrary to the definition of that term provided herein, the definition of that term provided herein applies, and the definition of that term in the reference does not apply.

SUMMARY

[0021] The following presents a simplified summary of the disclosure in order to provide a basic understanding of the reader. This summary is not an extensive overview of the disclosure, and it does not identify key/critical elements of the invention or delineate the scope of the invention. Its sole purpose is to present some concepts disclosed herein in a simplified form as a prelude to the more detailed description that is presented later.

[0022] Exemplary embodiments of the present disclosure are directed towards the machine learning-based Automatic brain seizure identification and classification at hospitals.

[0023] An exemplary object of the present disclosure is directed towards a system that deployment of new optimized machine learning and deep learning model and its application to predicting types of seizure present in the brain.

[0024] Another exemplary object of the present disclosure is directed towards the extraction features for identification of different types of seizures in the brain using Image Wavelet segmentation and HOG feature extraction.

[0025] Another exemplary object of the present disclosure is directed towards the implementation of the Modified Sine Cosine Algorithm (MSCA).MSCA is proposed to optimize the weights of the ELM model to enhance the performance of the conventional ELM model.

[0026] An exemplary aspect of the present subject matter is directed towards Image transformation from EEG signal by Synchrosqueezing Fejer-Korovkin Wavelet Transform.

[0027] An exemplary aspect of the present subject matter is directed towards using HOG (Histogram of Oriented Graph) Feature extraction and fed as input to MSCA-ELM Model.

[0028] An exemplary aspect of the present subject matter is directed towards the HOG (Histogram of Oriented Graph) feature-based hybrid MSCA (Modified Sine Cosine Algorithm) Extreme Learning Machine model to classify different categories of seizure.

[0029] Another exemplary aspect of the present disclosure is directed towards the use of MWCA (Modified Water Cycle Algorithm), and Wavelet transform-based segmentation technique on EEG images to detect a seizure.

[0030] Another exemplary aspect of the present disclosure is directed towards using Deep Neural networks and Machine learning models to predict seizures.

[0031] Another exemplary aspect of the present disclosure directed towards the development of Prototype Design Using Embedded Platform for detection of seizure.

BRIEF DESCRIPTION OF THE DRAWINGS

[0032] In the following, numerous specific details are set forth to provide a thorough description of various embodiments. Certain embodiments may be practised without these specific details or with some variations in detail. In some instances, certain features are described in less detail so as not to obscure other aspects. The level of detail associated with each of the elements or features should not be construed to qualify the novelty or importance of one feature over the others.

[0033] FIG.1 is a diagram depicting 100, Process adopted in Epileptic Seizure Detection and classification Using HOG feature based MSCA-ELM Model and its Embedded prototype

[0034] FIG. 2 is a MSCA-ELM architecture, according to an exemplary embodiment of the present disclosure.

[0035] FIG. 3 represents the Implementation phase of detection and classification, according to an exemplary embodiment of the present disclosure.

[0036] FIG. 4 is a representation of Steps involved in prototype design, according to an exemplary embodiment of the present disclosure.

DETAILED DESCRIPTION OF EXAMPLE EMBODIMENTS

[0037] It is to be understood that the present disclosure is not limited in its application to the details of construction and the arrangement of components outlined in the following description or illustrated in the drawings. The present disclosure is capable of other embodiments and of being practised or of being carried out in various ways. Also, it is to be understood that the phraseology and terminology used herein is for the purpose of description and should not be regarded as limiting.

[0038] The use of "including," "comprising," or "having" and variations thereof herein is meant to encompass the items listed thereafter and equivalents thereof as well as additional items. The terms "a" and "an" herein do not denote a limitation of quantity, but rather denote the presence of at least one of the referenced item. Further, the use of terms "first," "second," and "third," and the like, herein do not denote any order, quantity, or importance, but rather are used to distinguish one element from another.

[0039] Referring to FIG. 1 is a 100 Process adopted in Epileptic Seizure Detection and classification Using HOG feature based MSCA-ELM Model and its Embedded prototype. It is a fact from the prior art that different classification models proposed by the researchers, but to date, the HOG (Histogram of Oriented Gradient) features are not yet utilized related to the classification of an epileptic seizure. In general EEG device detects the macroscopic motion of the superficial stratum of the brain beneath. These electrical surges are plotted on a graph sheet, and doctors make a diagnosis based on their understanding and experience. To enable soft computing tools to analyse these signals, the graph sheets/signals are transformed into relevant images that the soft computing tools can process.

[0040] Further to it, the Synchrosqueezing Fejer-Korovkin Wavelet Transform is a time frequency analysis (TFA) system for signal conversion system which converts the EEG signals to an image. The images are undergone preprocessing by MWCA (Modified Water Cycle Algorithm) and Wavelet transform-based segmentation technique for detecting seizure from converted images that can inform the doctor about the details of seizure due to tumor or accident.

[0041] Further, The "sine cosine algorithm (SCA)" is an optimization algorithm that preserves a population of "m search agents" and each agent is represented by "n-dimensions decision variable vector". According to "sine cosine algorithm" the position equation is updated as

<I ,X[ +ax sin(a 2)x a3phe' -X1, a4 <0.5 (1) a4 > 0.5 |Xin +al x co4a 2 )x a3 pgbes' - X

Where a,a2 ,a3 are the random variables and is given by

a =a I- ) (2)

Where "Xi is the i-th search agent" in the population, is the current iteration, K is the maximum number of iterations. For fast convergence of the parameter is modified as

a - (3) " 1+log af

The represents the current position and represents the update position. The position equations are mapped to the weights of the ELM model for optimization to improve the performance of the ELM model.

[0042] In accordance with a non-limiting exemplary embodiment of the present subject matter, FIG. 2 depicts the MSCA-ELM architecture. Extreme learning machine uses sophisticated mathematical modelling to process data in complex ways. ELM has been applied for the classification of different datasets like brain tumors, breast cancer etc. The MSCA is proposed to optimize the weights of the ELM model to enhance the performance of the conventional ELM model.

[0043] Further, given a "set of N training dataset " to N, each vector is the desired output. The output function of ELM with L hidden neurons is represented by

Y - =0 fkhk (Wk;X) (4)

Where h(w;x)=[1,h(w;x),.......hL (L;x)]isthe hidden units mapped to input x= x,....xN] and is the weight vector of all hidden neurons to an output neuron to be analytically analyzed. hk () which

is the activation function of the hidden layer. Equation (4) can be written as

H# = y (5)

Where H is the Nx (L +1)hidden layer feature-mapping matrix, whose elements are as follows:

Sh,(w x1) - hL (WN; X1 H= (6) 1 hlf(W; xN)- hL(WN N

r (-i )1 2

And hL (WN; xN)[W X1 +w X ....WNXNJ 2"

Where o is the parameter for "controlling the smoothness of the activation function"

p6 =Ht d, H t =(HT H )HT (7)

Where " His the Moore-Penrose generalized inverse" of matrix H

And d= ,=

_d,] Further the weights are mentioned as W=[wo + +......WINXN ]and the weigthts are mapped and updated using

1u

"+ a x 2 )x sin(a p heS' - ", a < 0.5 i (8) + a, x cosa 2 )x a3 pgheS' - ", a4 ;> 0.5

Traditionally, in order to train ELM, to minimizing the cost function, the minimum square error is given by

MSE= di -pih(w-xi) (9) J=1 i=1

When H is unknown "gradient-based learning" algorithms are generally used to search the minimum of||H8 - d||.

[0044] In an embodiment, as depicted in Fig 3, the implementation strategy of the invention is disclosed. Wherein the data signals collected from the EEG electrodes are processed and converted into relevant images in the microcontroller. The microcontroller is the latest configured micro-chip capable of executing relevant machine learning (MLA) and Al algorithms. The microcontroller collects the signals and processes using relevant MLA and then these images are fed to modified and novel MLAs to predict the type of seizure. Once predicted am alarm intimation is sent to doctor.

[0045] In an embodiment, as depicted in Fig 4, the EEG signals are converted to images by employing Synchrosqueezing Fejer-Korovkin Wavelet Transform and the HOG features are extracted. The program is written with tensor flow python programming. After execution of the program it has been saved to MSCA-ELM.h5 file which can run any general python environment. This has been implemented to make user friendly. The completing programming has been enrooted through the ARM Cortex processor to detect and classify the epileptic seizures due to head injury, brain infection, stroke and tumor. There are nearly 5000 EEG seizure signals are collected and utilized for detection and classification of seizures due to head injury, brain infection, stroke and tumor.

Claims

We Claim

1. The Epileptic Seizure Detection and classification Using HOG feature based MSCA

ELM Model and Embedded prototype development consisting of;

a plurality of electrodes 101 capable of identifying the brain activity connected to the

microcontroller 102. Wherein Microcontroller 102, capable of executing machine

learning algorithms (MLA) and Deep learning algorithms (DLA) converts the EEG

signals into relevant images; and

A novel model is developed to predict different categories of infected and non

infected seizures due to head injury, brain infection, stroke and tumour.

2. As claimed in claim 1, Microcontroller 102, converts the EEG signals into relevant images by executing Synchrosqueezing Fejer-Korovkin Wavelet Transform and performs HOG feature based MSCA-ELM Model to predict different categories of infected and non-infected seizures due to head injury, brain infection, stroke and tumour.

3. As claimed in claim 1, Images are further undergone preprocessing by MWCA (Modified Water Cycle Algorithm) and Wavelet transform-based segmentation technique for detection of seizure from converted images that can inform the doctor about the details of seizure due to tumour or accident.

__ _***__ _

Page 1 of 4 06 Jul 2021

EEG Signal

Synchrosqueezing Fejer-Korovkin Wavelet Transform for converting to image 2021103884

Image Dataset preparation and MWCA Preprocessing

Image Wavelet segmentation and HOG feature extraction

MSCA-ELM Classification

Classification comparison Results

Seizure Non-seizure Normal

Fig-1 Research flow diagram

Page 2 of 4 06 Jul 2021 2021103884

Fig.2 MSCA-ELM architecture

Page 3 of 4 06 Jul 2021

Synchrosquee Conversion of MWCA Detection of seizure zing Fejer- EEG Epileptic preprocessing by using Wavelet Korovkin seizure signals HOG Feature Transform and Wavelet to image extraction resizing Transform conversion

EEG 2021103884

Signal

Resized images are Epileptic Seizure fed as input to and non- seizure MSCA-ELM Model classification for classification

Prototype

Comparison Epileptic Seizure and non -seizure classification

Synchrosquee Conversion of MWCA Deep Neural zing Fejer- EEG Epileptic preprocessing Network and Korovkin seizure signals by HOG Machine learning Wavelet to image Feature models Transform conversion extraction

EEG Signal

Fig.3 Implementation phase of detection and classification

Page 4 of 4 06 Jul 2021

Image transformation from EEG signal by Synchrosqueezing Fejer- Korovkin Wavelet Transform

Convert to rgb2gray and 2021103884

image conversion Automatic Types of epileptic detection and seizures and non- Classified seizures due to head output injury, brain HOG Feature extraction and fed infection, stroke and as input to MSCA-ELM Model tumor

Tensor flow -Python ARM Cortex- programming processor

Conversion to machine language and save to MSCA- ELM.h5 file

Fig.

4 Steps of prototype design