AU2020103304A4

AU2020103304A4 - A cnn based model for disease classification using high temporal and spatial resolution images of simultaneous eeg-mri

Info

Publication number: AU2020103304A4
Application number: AU2020103304A
Authority: AU
Inventors: P. Anandan; Udayini Chandana; G. Naga Rama Devi; Sai Krishna Kodali; Ravaleedhar Murthy; S. Palanivel Rajan; Sivakumar Rajagopal; K. Sathishkumar; A. Suban; Saravanakumar V.
Original assignee: Individual
Current assignee: Individual
Priority date: 2020-11-07
Filing date: 2020-11-07
Publication date: 2021-01-14
Anticipated expiration: 2028-11-07

Abstract

A CNN BASED MODEL FOR DISEASE CLASSIFICATION USING HIGH TEMPORAL AND SPATIAL RESOLUTION IMAGES OF SIMULTANEOUS EEG-MRI ABSTRACT The major challenging task from the brain activity data is about classifying the visual information. Most of the research analysis on the basis of brain activity pattern is based on either the MRI or EEG. In order to map the brain activity, the EEG and MRI is considered as complementary neuroimaging models with regard to temporal and spatial resolution image The Simultaneous EEG-MRI will be effective to get the high temporal and spatial resolution image of the brain at the same time. The proposed method is based on simultaneous EEG-MRI data and the approach of machine learning in order to classify the patterns of the visual brain activity. The data of EEG and MRI is merged using the data fusion method. The classification is done by using the machine learning classifier. The simultaneous EEG-MRI method is effective in classifying the pattern of brain activity which is helpful to predict and decode the patterns of brain activity. 11 P a g e A CNN BASED MODEL FOR DISEASE CLASSIFICATION USING HIGH TEMPORAL AND SPATIAL RESOLUTION IMAGES OF SIMULTANEOUS EEG-MRI Drawings: ' EEG Data MRI Data Pre-processing (Rernoval of (Motion correction, Artifact) spatial smoothing) General Linear A~ Brain Activity Pattern RO _trcion Classifier Figure 1: Fusion of EEG and MRI 1 P a g e

Description

A CNN BASED MODEL FOR DISEASE CLASSIFICATION USING HIGH TEMPORAL AND SPATIAL RESOLUTION IMAGES OF SIMULTANEOUS EEG-MRI

Drawings: '

EEG Data MRI Data

Pre-processing (Rernoval of (Motion correction, Artifact) spatial smoothing)

General Linear A~

Brain Activity Pattern

RO _trcion Classifier

Figure 1: Fusion of EEG and MRI

1 Pa g e

Description

Field of Invention:

This field of invention addresses the Convolutional Neural Network (CNN) based model to classify the brain activity patterns using high temporal and spatial resolution images of simultaneous EEG-MRI. The novel approach to simultaneously access the EEG and MRI signals by integrating both temporal and spatial resolution of the both techniques in order to investigate the healthy as well as pathological brain function.

Background of invention:

For the past decades, the disease like neurodegenerative and psychiatric has increased and it requires finer and advanced tools which ranges from the electrophysiology to neuroimaging for the purpose of precise diagnostic accuracy. The Electrophysiology and Electroencephalogram (EEG) need a tool which is widely used to support the diagnosis of disease like neurological. The Electroencephalogram (EEG) will provide the good temporal resolution unlike the imaging techniques, and it has capability to record the electric brain activities in terms of milliseconds through the electrodes which is placed on the scalp. By using EEG signal processing techniques, the parameters like frequency spectrum, amplitude and the coherence can be achieved. The morphological view of the brain can be provided using the Computed Tomography (CT) and MRI with good spatial resolution and that allows the multiparametric evaluation of the properties for the brain tissue regarding to structural and functional information. Similar to EEG at the various temporal scales, the MRI will allow the non-invasive evaluation related to brain function activation while in rest state and execution of task. The complementary information will make use of the multimodal accession system which is developed to avoid drawbacks of single model and to improve the yield of patient. The drawback of EEG is low spatial resolution and it is weak in detecting epileptic activities from deep brain structures. The MRI is good in spatial resolution and it is sensitive to the signals of deep and superficial brain structures. In combining both of EEG

1| P a g e along with the MRI, it is possible to detect the blood oxygenated level dependent signal changes with the detection of spikes on the scalp EEG and even if the structure of deep brain is involved. The simultaneous analysis of EEG and MRI data in the human systems and clinical neurosciences is evolving rapidly. The importance of brain imaging is increasing now by using the simultaneous EEG-MRI and it is aiming to achieve the temporal as well as spatial resolution of the human brain function. The most important challenge with the simultaneous EEG-MRI is about decreasing the quality of the signal in the both models. Many of the neuroimaging has provided the relationship between the mind and brain, so that, there is possibility to decode from the brain activity about the thinking process in each individual. The non-invasive of the brain activity can give the proper extent of information in order to classify the various mental state and the visual pattern. The patterns of brain activity will tell about what the person can view, perceive or recall in the mind. When the person creates the visual image of the apple, then the required information for the purpose of constructing apple is available mentally as the person perceives in reality.

Objects of the invention:

• The objective is about simultaneous access of the EEG and MRI signals by integrating both

temporal and spatial resolution of the both techniques in order to investigate the healthy as well as pathological brain function. • The simultaneous EEG-MRI and the machine learning approach is proposed in order to classify the patterns of visual brain activities. • The Data of EEG and MRI is fused together using the data fusion method. • The machine learning classifier is used for purpose of classification. • The simultaneous EEG-MRI method is productive in order to classify the pattern of brain activity which is helpful to predict and decode the patterns of brain activity.

Summary of the invention:

The sophisticated tools are needed for the purpose of diagnosis and monitoring the disease like neurodegenerative and psychiatric. The clinical estimation will take the advantage of main parameters which is extracted by the electroencephalogram (EEG) and magnetic resonance imaging (MRI) comparing to others, in order to support the clinical management for the reason of neurological diseases. The two tools are complementary and it can be highlighted by the possibility

21Page of combining two technologies in the hybrid way and allowing simultaneous accession of the two signals with the help of new approach of EEG-MRI technology. The EEG is most widely used technique for the purpose of brain electrical activity. The innovative technology has led to the development of high-density EEG system with improvement of high number of electrodes for the brain connectivity studies. The EEG will be used to categorize the disease like metabolic, sleep disorders, epileptic syndromes, traumatic brain injuries, tumor lesions, neurodegenerative disease and the characteristics of comatose patients and the brain death. The MRI is also one of the non invasive techniques which allows to measure the functioning of brain. The mechanism that focus the signal of MRI is known as the blood oxygen level dependent (BOLD) effect. The simultaneous EEG-MRI accession is used to estimate the correlation between the electrical brain activities and hemodynamic mutation. The MRI along with the high spatial resolution will not provide any sufficient temporal sampling because of the slow blood oxygen level dependent response in terms of seconds unlike the electroencephalography. Instead of that, it will offer the high temporal resolution in terms of milliseconds, but along with weak localization of the signal sources. The combination of these two tools in the hybrid simultaneous accession will allow to overcome the limitations present in the both techniques and increases the analyses which can be performed. The simultaneous accession also makes sures of the identical registration with regard of subject in mental state task execution and the inference recording of the environment. But this will not happen in recording the two methods in separate way, in case if recording is taken in different environment along with the cognitive unstable patients. The accession of simultaneous EEG and MRI will use the specialized EEG hardware which is safe and compatible with the MRI environment and it will be comfortable to the participant. The data which is obtained from the simultaneous accession EEG-MRI will be strongly influenced by the artifacts. The applications of EEG-MRI are represented by the feedback of neuro, that will allow the modulation of brain activities. The information will come back to the patient which belongs to EEG or MRI scan. For EEG neurofeedback, the authors had compared the brain activation during the motion imagination and the movement execution and suggests the roles for rehabilitation of the patients which is affected by the paralysis. The Data analysis is the basic step for EEG-MRI research and for simultaneous multimodal accessions. The two macro areas are present i.e. symmetric and integrated analysis. The symmetric approach will involve in simultaneous analysis of the data which is extracted from the two methods, while the integrated analysis will make use of the data

31Page which is collected by one of the two methods in order to understand and validate the collected data from the other one. The integrated method will analyze the two methods of applications. The first method will use the brain electrical activity in order to predict the hemodynamic variations. Another method will use the activation map which is extracted by the MRI in order to correct and analyze the EEG source. The EEG signal which is recorded from the scalp will have high temporal resolution and the weak spatial resolution because of the limited number of electrodes. The MRI will have high spatial resolution and the low temporal resolution. The integration of EEG and MRI will get high temporal and spatial resolution in same time. The technical challenges in simultaneous EEG and MRI will be in terms of data accession and analysis.

Detailed Description of the Invention:

Electroencephalography and the Magnetic Resonance Imaging are complementary neuroimaging models. Each of the model will have different some limitation with regard of temporal and spatial resolution. The simultaneous EEG-MRI will be productive with the regard of high spatial and temporal resolution at same time. The imaging and machine learning method will classify the patterns of visual brain activities which is not possible to be with imaging models like EEG or MRI. The integration of EEG and MRI is one of the promising solutions for higher resolution and spatial domain. The benefit in simultaneous EEG-MRI data accession is that, it will consist of common information related to the brain activities which is captured by using EEG and MRI which is further can be merged together using the data fusion method. As the overall system is consistent, the fusion can be done for both of the data i.e. the images of MRI and EEG signals. The fusion of EEG-MRI is divided into two classes i.e. 1. data driven approach and 2. Model driven approach. The model driven approach is difficult to implement and also tough to predicate the precise biophysical model of the human brain. The problem in integration of EEG-MRI data analysis is that, the absence of the model in blood oxygenated level dependent response. The Figure 1 describes the merging of EEG and MRI data using the linear model. The EEG signal is accessed inside the scanner and made it impure with artifacts. The gradient artifacts (GA) consist of higher amplitude and it is better suitable method comparing with others. The method to remove the artifact is template matching method. The waveform subtraction method is also to remove the gradient artifacts. The average GA templates is formed and then it will be subtracted from each of the occurrence of GA in the signals of EEG. The Ballistocardiogram is difficult to remove from the

41Page physiological activity as the main source which is occurring in human body and in the brain. In order to remove Ballistocardiogram, the optimal basis set on the basis of principal component analysis is used for removal. The principal component analysis is used for identifying the components of EEG and later removes the regression effect as well as fixed component. The data accession in EEG is contaminated along with the artifact of eye blink, movement of eyes, BCA and GA. The data is to be filtered with help of band pass filter and its frequency up to 30Hz In order to remove the DC components and higher frequency artifacts. Later, the gradient artifacts and Ballistocardiogram artifacts is removed with the help of average artifact subtraction. For further use, the artifacts like blinking of eyes, movement of eyes and the muscular movements which is removed from the EEG data. The MRI accession of data will consist of sequence of the brain functional volume. Every volume consists of the slices and dimension for each image. The MRI images can be re-ordered to the middle image in the series of time in order to make some corrections for the movement of head. The next process is about the brain activation map which infers to the brain activity recorded by the EEG and MRI. The maps of EEG will be generated from the clean data and the brain activity is possible to see in the optical region and the temporal region. The classification result has to be done for EEG-MRI data with the help of EEG as well as MRI data. The two conditions have to be classified. The extraction of region of interest and the performance of feature reduction of the EEG and MRI data before it is passed to the classifiers. The region of interest is to be extracted from all the region of brain for MRI and EEG-MRI data. The Region of interest will be extracted for EEG-MRI and MRI whereas, the extraction is carried out from the data of EEG after the removal of artifacts. The reduction of Feature dimension is important for any of the analysis of data either the EEG or MRI. With help of this, the learning efficiency can be improved and also improves the prediction performance. The Machine learning approaches are needed to employ the various types of brain and physiological signals. The object categories can be classified with related to the patterns of brain activity and the classifiers are selected as support vector machine, convolutional neural network and decision tree. The SVM is used widely for the purpose of classifying MRI images, and EEG data. This classifier is used to find the function of input data which enables the classification as well as regression. The SVM is the concept of finding out the hyper plane which is used to classify the data in order to separate the class. The convolutional neural networks use the multilayer perceptron on basis of neural network and it is a model of feed forward neural network. It uses the supervised learning approach

51Page by using the back propagation to train the network. The Decision Tree is the classifier and it performs the partition of the feature space to produce the boundary. It is like a tree with presence of nodes. For each node, the possibilities are two i.e. leaf or decision node. It starts from root node and ends at the leaf node and it gives the output class.

61Page

Claims

A CNN BASED MODEL FOR DISEASE CLASSIFICATION USING HIGH TEMPORAL AND SPATIAL RESOLUTION IMAGES OF SIMULTANEOUS EEG-MRI CLAIM We claim,

1. The new approach is present for the pattern of visual brain activity and its classification on basis of simultaneous EEG-MRI.

2. The machine learning approach will classify the patterns of visual brain activities which is not possible to be with imaging models like EEG or MRI.

3. The method claims, The Fusion of EEG with MRI is done to improve the brain resolution comparing to the individual EEG and MRI data.

4. The method of claims, it is possible to improve high spatial and temporal resolution in same time.

5. The method of claimI, the data is merged by using data fusion method.

6. The method claim2, Machine learning method classifies the patterns of visual brain activities which is not possible to be with imaging models like EEG or MRI.

1 Pag e

A CNN BASED MODEL FOR DISEASE CLASSIFICATION USING HIGH TEMPORAL 07 Nov 2020

AND SPATIAL RESOLUTION IMAGES OF SIMULTANEOUS EEG-MRI

Drawings: ` 2020103304

Figure 1: Fusion of EEG and MRI

1|Page