System for detection and classification of individual capabilities

Abstract

The invention of intelligent system for detection and classification of individual capabilities through brain mapping and advanced EEG signal analysis using deep learning relates to a method capable of identifying and mapping an individual's cognitive strengths and weaknesses based on the impact of each brain region on the individual's performance In this invention, a combination of data and information from cognitive assessment databases, along with rules extracted from previous research and studies, and results obtained from individuals' brain signals in electroencephalography are aggregated to create an enhanced collective trained model for generating the individual's brain map and identifying the individual's cognitive strengths and weaknesses based on the obtained results. Rapid and accurate data processing, coupled with the use of modern deep learning and statistical techniques, has transformed this system into a powerful tool for better understanding brain function and its clinical and research applications.

Description

[0001] TITLE OF INVENTION

[0002] INTELLIGENT SYSTEM FOR DETECTION AND CLASSIFICATION OF INDIVIDUAL CAPABILITIES THROUGH BRAIN MAPPING AND ADVANCED EEG SIGNAL ANALYSIS USING DEEP LEARNING

[0003] TECHNICAL FIELD OF THE INVENTION

[0004] The present invention relates to the field of electroencephalography (EEG) signal processing, as well as to machine learning and cognitive data analysis, furthermore, this invention relates to brain mapping, intelligent brain analysis systems, and advanced filtering and noise reduction techniques in brain signals. Additionally, the invention includes GPU-based processing technologies and deep learning models for extracting and classifying brain features. Overall, this integrated system combines multidisciplinary technologies to detect and analyze individual capabilities through precise examination of brain signals and brain maps.

[0005] PRIOR ARTS

[0006] Electroencephalography (EEG), through the analysis of brain electrical activity, enables the identification of individual human characteristics. These characteristics include psychological states such as stress or relaxation, cognitive abilities such as concentration and memory, personality, and even biometric identity. Due to the uniqueness of EEG signals for each individual, they can be used to identify behavioral traits, decision-making processes, and even abnormal brain states (such as psychological disorders). Signal processing tools and machine learning techniques, such as neural networks and statisticalanalysis, convert this data into comprehensible patterns and provide wide-ranging applications in medicine, education, security, and neuro-marketing. The identification process involves preprocessing signals to remove noise, extracting temporal, frequency, and spatial features, and analyzing these features using machine learning models. These methods enable the design of personalized treatments, improvement of learning processes, biometric authentication, and the development of brain-computer interfaces. Despite significant advancements, challenges such as high noise levels, individual variability in signals, and the complexity of data interpretation remain. Nevertheless, EEG analysis, due to its high potential, is expanding and becoming an essential tool for understanding human characteristics.

[0007] Quantitative Electroencephalography (qEEG) is an advanced technique for the precise processing and analysis of EEG signals, converting brain electrical activities into numerical data and examining them using computational and statistical methods. This approach enables the identification of abnormal patterns, monitoring of brain changes, and analysis of connections between different brain regions. qEEG involves several stages: initially, the EEG signal is recorded and its noise is eliminated. Subsequently, frequency analysis is performed to examine brain activity across various frequency bands (delta, theta, alpha, beta, and gamma). The results are presented in the form of brain maps (Brain Mapping) that visually display brain activities. Finally, the extracted data are compared with normal databases to identify any deviations or anomalies.

[0008] qEEG offers numerous advantages, including providing more accurate analyses compared to traditional EEG, identifying patterns that are not observable in raw signals, and monitoring brain changes over time. This method is used in the diagnosis of psychiatric disorders (such as depression and ADHD), management of conditions like epilepsy and traumatic brain injury, and the design ofneurofeedback treatment protocols. Additionally, qEEG is utilized in scientific research and the monitoring of brain surgeries for analyzing brain function. However, it faces challenges such as dependence on the quality of initial data, high costs of equipment and software, and the need for specialized knowledge to interpret the results. Overall, due to its ability to provide precise and quantitative analyses, qEEG is considered a valuable tool in neuroscience and medicine.

[0009] Convolutional Neural Networks (CNNs) are among the prominent models in deep learning, widely utilized in image processing and pattern recognition. These models extract essential features from input data, such as edges or textures, using convolutional layers and leverage pooling layers to reduce dimensions and simplify computations. Additionally, non-linear activation layers, such as ReLU, enhance the model's ability to leam complex relationships. At the end of the network, fully connected layers are responsible for combining the extracted features and providing the final output, such as classification or object detection.

[0010] Due to their flexibility and high efficiency, CNNs are applied to a broad range of tasks, including object detection in images, autonomous driving, medical image analysis, and robotic vision. Advanced architectures like AlexNet, VGGNet, ResNet, and DenseNet have improved the performance of these models in complex problems by enhancing network structures and employing innovative pooling methods. By appropriately combining different layers, CNNs can also analyze time- series or sequential data, making them one of the most powerful tools in deep learning.

[0011] CUDA, an architecture developed by NVIDIA, is a tool for leveraging the power of Graphics Processing Units (GPUs) to perform intensive computations and parallel processing. GPUs, with thousands of processing cores, have the ability to execute complex operations simultaneously and are significantly moreeffective than typical CPUs, which have a limited number of cores, for applications such as deep learning, image processing, and scientific simulations. CUDA programming involves defining kernels (functions executable on GPUs) and organizing computations as threads and blocks within multi-dimensional grids, enabling flexible management of complex data structures like images and matrices. Additionally, the CUDA memory model, which includes global, shared, constant, and local memories, plays a crucial role in optimizing computational speed and efficiency.

[0012] Utilizing CUDA and GPU acceleration offers several advantages, including increased speed of parallel computations, optimization in training deep learning models, and the development of custom algorithms. This platform is applied in diverse fields such as image and video processing, scientific simulations, signal processing, and big data analysis. Particularly in deep learning, frameworks like TensorFlow and PyTorch directly utilize GPUs to train neural networks more rapidly. Overall, CUDA provides advanced tools for parallel programming, enabling the execution of complex computations several times faster than traditional methods and is considered one of the key technologies in the development of modern computational systems.

[0013] Ensemble in Machine Learning refers to the combination of multiple models to enhance the accuracy and stability of results. This approach reduces the errors of a single model and creates better generalizability. In deep learning, Ensemble is employed by using several deep neural networks that are trained either in parallel or sequentially. The primary techniques in this area include Bagging, Boosting, and Stacking. In Bagging, models are trained in parallel on different subsets of the training data, and their outputs are combined. Boosting trains models sequentially, with each new model focusing on the errors of the previous one. Stacking takes the outputs of multiple models as inputs to a higher-level model to provide an optimal combination of predictions.The application of Ensemble in deep learning offers high accuracy due to its flexibility in combining diverse models. This approach can improve overall performance by introducing diversity in model architectures and hyperparameters or by varying the training data. However, higher computational costs and the need for proper management of model weights are its challenges. Ensemble is utilized in complex tasks such as Kaggle competitions, object recognition, and industrial forecasting, and is considered one of the key tools in deep learning due to its ability to combine the strengths of different models.

[0014] Analysis of Variance (ANOVA) is a powerful statistical method for comparing the means of multiple groups and examining significant differences between them. This method, by analyzing within-group variance (the dispersion of data within a group) and between-group variance (the distance between group means), determines whether the differences between means are due to actual differences or merely random fluctuations. In ANOVA, the null hypothesis (Ho) states that all groups have the same mean, and the alternative hypothesis (Hi) indicates that at least one group has a different mean. Types of ANOVA include one-way (for examining one independent factor), two-way (for two independent factors and their interaction), and repeated measures (for dependent and repeated data). This method requires adherence to assumptions such as data independence, normal distribution, and homogeneity of variances.

[0015] To perform ANOVA, the F-statistic is calculated, which is the ratio of between-group variance to within-group variance. If the p-value is less than the significance level (typically 0.05), the null hypothesis is rejected, indicating that a significant difference exists between the groups. If the null hypothesis is rejected, follow-up tests (such as Tukey or Bonferroni) are conducted to identify the differences between specific groups. ANOVA is applied in many fields, including social sciences, medicine, and industrial analyses, and is a powerfultool for examining the impact of various factors on outcomes. However, strict adherence to the assumptions is essential for the validity of the results.

[0016] Event-Related Potentials (PRCs) are electrical markers of the brain that are recorded as changes in brain electrical activity following a specific event or stimulus (such as sound, light, or a cognitive task). PRC components typically appear within specific time windows after the occurrence of the stimulus and include positive or negative peaks, which are named based on their timing (such as N 100 or P300). These components reflect the brain's cognitive, attentional, or sensory processing in response to the presented stimulus. Key features of PRCs include amplitude, latency, waveform shape, and sensitivity to the type of stimulus. These features provide information about the brain's performance in processing stimuli and the individual's cognitive state.

[0017] PRC components like N100, P200, or P300 are each associated with specific brain processes. For example, N100 represents early sensory processing, while P300, as a later component, is related to attentional processes and cognitive evaluation of the stimulus. The amplitude of the components can indicate the intensity of processing or the level of attention to the stimulus, whereas their latency often provides information about the speed of brain processing. These features make PRCs a powerful tool for studying sensory and cognitive functions and even for identifying neuropsychiatric disorders.

[0018] Ensemble Models in Deep Learning (Deep Learning Ensemble Models) are a powerful approach that enhances the accuracy and stability of predictions by combining multiple deep learning models. These models utilize several independent neural networks that are trained separately, merging their diverse outputs to deliver a more accurate final result. The primary objective of this approach is to reduce the errors of individual models and improve

[0019] generalization.The main methods for constructing ensemble models include Bagging, Boosting, and Stacking. In Bagging, multiple models are trained in parallel, and their outputs are typically combined through averaging or voting. Boosting operates sequentially, with each subsequent model focusing on the errors of the previous one. Stacking involves using the outputs of initial models as inputs to a higher-level model to create an optimal combination.

[0020] Ensemble models in deep learning offer high flexibility due to the diversity in model structures, such as variations in network architectures, hyperparameter tuning, or the use of different datasets. These models perform exceptionally well in complex tasks like object detection, language translation, and time-series forecasting, often outperforming single models. However, they present challenges such as higher computational costs and the need for more resources. In practical applications, ensemble models are frequently utilized in artificial intelligence competitions, complex data analysis, and optimizing model performance in industrial projects. They are recognized as an efficient tool for increasing accuracy in deep learning.

[0021] In this regard, numerous technologies and inventions have been registered, some of which are mentioned below:

[0022] an invention with publication No. W02015040532A2 which was filed in WIPO on 12 / 09 / 2014 titled "System and method for evaluating a cognitive load on a user corresponding to a stimulus" provides a method and system for assessing a user's cognitive load in response to a given stimulus. Initially, the user's EEG data is received and divided into smaller time segments. Subsequently, EEG features are extracted from these segments, which can be represented in the frequency or time domain. These features are then clustered into two or more groups using unsupervised learning techniques such as Fuzzy c-Means or K-Means clustering algorithms. Each cluster represents a level of cognitive load (e.g., high or low), enabling the evaluation of cognitive load.This process may include data filtering using CSP filters, calculating cognitive scores based on segments with high or low cognitive load, and labeling the clusters. The system comprises modules for receiving, segmenting, extracting, clustering, and calculating cognitive scores, and is designed to detect the difficulty levels of mental tasks or the degree of users' mental engagement. This invention has applications in analyzing users' cognitive performance in response to various mental tasks.

[0023] A US invention with patent No. US7580742B2 which was granted on 25 / 08 / 2009 titled "Using electroencephalograph signals for task classification and activity recognition" relates to a method for classifying brain states based on electroencephalography (EEG) signals without the need to remove noise and artifacts from the EEG signals. This method includes collecting labeled EEG data that contain noise, dividing this data into overlapping time windows, removing the temporal dimension from each window, extracting features from these windows, and building a classification model using these features. Initially, the features are calculated in their basic form, then combined into a larger set and optimized and reduced for use in machine learning techniques. The constructed model is used to classify brain states in unlabeled EEG data, which involves dividing the data into time windows, extracting the necessary features for the model, classifying brain states, and averaging the results across adjacent time windows. This method is capable of detecting both cognitive and non-cognitive noise and can identify brain states related to various user conditions, including levels of interaction, cognitive load, task engagement, satisfaction, or fatigue. Additionally, this method allows for the differentiation, continuity, and transition between different brain states and is specifically designed for evaluating user interfaces by healthy individuals.

[0024] A US invention with patent No. US8849727B2 which was granted on 30 / 09 / 2014 titled "Method and system for classifying brain signals in a BCIusing a subject- specific model" relates to a method and system for classifying brain signals in a Brain-Computer Interface (BCI) that involves the creation of subject- specific and individual-independent models for each user. Initially, an individual-independent model is constructed using labeled brain signals collected from previous users. Subsequently, a preliminary model specific to a new individual is created using the independent model and a portion of the new individual's unlabeled brain signals. These models are optimized through processing and feature extraction from the brain signals, utilizing techniques such as low-pass filtering, downsampling, and the removal of ocular artifacts. The process of updating the subject- specific model continues until the model achieves a stable confidence score. At this point, the optimized model is used for more accurate classification of the signals. This method employs structured stimuli arranged in rows and columns to collect data and analyzes the individual's brain signals in response to these stimuli to detect differences between elicited features (such as P300). The application of this method lies in the development of BCI systems that offer individual adaptability and high accuracy, enhancing the interaction between users and computer systems by tailoring the classification models to each user's unique brain signal patterns. An invention with publication No. WO2015111331A1 which was filed in WIPO on 16 / 12 / 2014 titled "Cognitive function evaluation apparatus, method, system, and program" related to a device, method, and system for evaluating cognitive performance through the analysis of electroencephalography (EEG) signals resulting from target and non-target stimulus events. In this method, stimulus events with varying levels of difficulty in distinguishing between target and non-target are presented, and the EEG signals associated with these events are measured. The processing device analyzes these signals, estimates the selected target event, and calculates a detection score or decoding accuracy. Cognitive performance is evaluated based on the values of this detection score,decoding accuracy, elapsed time, and decoding speed. This system includes a stimulus presentation device, an EEG device, and a processing unit capable of analyzing the relationship between discrimination difficulty and evaluation parameters, thereby providing cognitive evaluation results. Additionally, adjusting the difficulty of stimulus events through image processing is possible. This invention is applicable in analyzing cognitive performance and evaluating the mental abilities of users.

[0025] A US invention with publication No. US20230200741A1 which was filed on 27 / 12 / 2021 titled "Method for estimating physiological events from physiological signals, a non-transitory computer-readable medium, and, an apparatus" pertains to a method and device for accurately detecting peaks in physiological signals, enhancing the precision of signal analysis without the need to increase the sampling frequency. The proposed method utilizes machine learning algorithms to initially identify peaks and then precisely adjust their positions, minimizing errors compared to reference signals with higher sampling frequencies. Features of the signal, such as the amplitude and the time intervals between peaks and troughs, are extracted and employed to classify the desired peaks. This method is particularly suitable for analyzing Photo plethysmography (PPG) signals from smartwatches and calculating Inter-Beat Intervals (IBI). Additionally, it can be extended to other applications, such as estimating Cerebral Blood Flow Velocity (CBFV) using Transcranial Doppler (TCD) signals or detecting epileptic discharges with Electroencephalography (EEG) signals. The process includes pre-processing the signals to remove noise and artifacts, assessing the quality of signal segments, and utilizing advanced algorithms like one-dimensional Convolutional Neural Networks (ID CNN). The invention is also provided as a wearable device or executable software, designed for medical and health monitoring applications. This approach improves the accuracy of physiological event detection, facilitates real-timehealth monitoring, and supports various medical diagnostics and health management systems.

[0026] A Japanese invention with publication No. JP2004535231A which was filed on 07 / 06 / 2002 titled "Method and apparatus for brain fingerprint identification, measurement, evaluation, and analysis in brain function" pertains to a device and method for evaluating an individual's cognitive skills through the measurement and analysis of electroencephalography (EEG) brain signals. The device includes a user interface, memory for storing the processor's executable instructions, and a processing unit connected to these components. During the execution of the instructions, the device presents a task with specific interference to the individual, requiring a response to the task and at least one stimulus element. The user interface measures data that indicate the individual's reactions to the stimulus elements, including metrics related to sensory processing capabilities under emotional load. The device simultaneously records the individual's responses to both the task and the stimulus elements, receiving and analyzing this data to calculate performance metrics. These metrics include quantitative indices of the individual's cognitive abilities under emotional load. These indices can be used for precise evaluation of cognitive abilities under various emotional conditions, monitoring the progression of cognitive diseases, assessing the effectiveness of treatments, and developing educational programs. Additionally, the obtained scores may be displayed as games, wherein the game parameters are adaptively adjusted based on the scores. This system leverages machine learning methods and the precise analysis of behavioral and brain data to enable the assessment and enhancement of an individual's cognitive skills. It finds extensive applications in the fields of medicine, education, and personal development, providing a robust tool for evaluating and improving cognitive performance in diverse settings.A Chinese invention with patent No. CN109784242B which was granted on 25 / 10 / 2022 titled "Electroencephalogram signal denoising method based on one-dimensional residual convolution neural network" pertains to a method for denoising electroencephalography (EEG) signals using a one-dimensional convolutional neural network (ID CNN) with residual blocks. In this method, EEG samples are initially selected from the PhysioNet database and normalized. Subsequently, various types of noise, such as Gaussian white noise and muscle noise, are added to create noisy samples. These noisy data are then divided into training and testing sets, and overlapping time windows are generated using data augmentation techniques. The temporal dimension is removed, and features are extracted from these windows. A one-dimensional convolutional neural network with residual blocks is constructed and trained using the Adam optimization algorithm and mean squared error (MSE) minimization. Once trained, the network is utilized to reconstruct denoised EEG signals. The advantages of this method include a simple network structure, increased learning capacity, improved signal-to-noise ratio, reduced mean squared error, and the capability for real-time noise removal. This approach is highly effective for preprocessing and denoising EEG signals and is well-suited for integration into wearable devices such as smartwatches. Due to its high efficiency and optimal quality, this invention has extensive applications in the preprocessing and noise reduction of EEG signals, making it ideal for use in wearable health monitoring devices.

[0027] A Chinese invention with publication No. CN114861702 A which was filed on 27 / 03 / 2022 titled "Noise identification method for EEG data by using deep neural network" pertains to an advanced method and system for identifying and removing noise from electroencephalography (EEG) signals using deep neural networks. In this method, EEG signals resulting from the simultaneous activities of multiple joints, such as the shoulder, elbow, wrist, and fingers, are firstcollected through an EEG cap and then divided into two categories: single -joint training and comprehensive training. The collected signals undergo preprocessing steps, including downsampling, initial filtering, and Independent Component Analysis (ICA), to obtain the activation of independent components and their topographical maps. The topographical maps are segmented, and the activations of the independent components are divided into multiple tests. These data are then fed into a deep neural network that includes one-dimensional and two-dimensional convolutional networks as well as a Deep Denoising Network (DDN). The signals are processed using various convolutional layers and activation functions. The outputs of these networks are connected to a deep neural network, and using a Softmax classifier, the noises present in the EEG signals are identified and removed. This method offers high noise removal accuracy and a simple neural network structure, providing high learning capacity and effective noise elimination. It is suitable for preprocessing and denoising EEG signals. Additionally, by introducing an error generator and a detector within the neural network, the accuracy and quality of noise removal are enhanced, and processing time is reduced, making it suitable for medical applications and health monitoring.

[0028] A Chinese invention with patent No. CN116982993B which was granted on 02 / 04 / 2024 titled "Electroencephalogram signal classification method and system based on high-dimensional random matrix theory" relates to a method and system for classifying electroencephalography (EEG) signals based on high-dimensional random matrix theory. In this method, the principle of covariance matrix similarity judgment in high-dimensional random matrix theory is utilized to compare the similarity and differences between the current patient's observed EEG signal, the reference hypothetical state signal, and the patient's normal state EEG signal. Based on this comparison, a composite index is designed and compared with a specific threshold value. If the compositeindices meet the threshold conditions, the patient is determined to be in the hypothetical EEG state. This invention considers the comprehensive features of the EEG signal sample data, preserving more of the primary information of the EEG signal and the correlation information between channels. It leverages the advantage of high-dimensional random matrix theory, which is more suitable for analyzing small datasets with high dimensions. Consequently, the invention can identify the patient's condition more quickly and accurately, even when the size of the observed EEG signal samples is relatively small. The presented system includes modules for signal processing, similarity index calculation, composite index design, and brain state judgment, enabling precise and faster analysis of the patient's brain conditions. It has extensive applications in the fields of medicine and neuroscience, providing a robust tool for the accurate and efficient classification of EEG signals.

[0029] A US invention with publication No. US20190159716A1 which was filed on 03 / 08 / 2017 titled "Cognitive platform including computerized evocative elements" relates to a device and method for creating a representation of an individual's cognitive skill level. The device includes a user interface, memory for storing the processor's executable instructions, and a processing unit connected to the user interface and memory. When executing the instructions stored by the processing unit, the device first presents a sample task with specific interference on the user interface that requires the individual's response to the task in the presence of interference and also a response to at least one stimulus element. The user interface is configured to measure data that indicate the individual's responses to the stimulus elements, including at least one metric of the individual's sensory processing capabilities under emotional load. The device simultaneously records the individual's responses to both the task and the stimulus elements, receives and analyzes this data to calculate at least one performance metric that includes quantitative indices of the individual'scognitive abilities under emotional load. These performance metrics can be used for a more precise evaluation of the individual's cognitive abilities under various emotional conditions and are displayed to the individual as output, for example, in the form of a game, where the game parameters are adaptively adjusted based on the obtained scores. This system utilizes machine learning methods and precise analysis of response data to enable the assessment and improvement of an individual's cognitive skills and has extensive applications in the fields of medicine, education, and personal development.

[0030] A US invention with patent No. US6434419B1 which was granted on 13 / 08 / 2002 titled "Neurocognitive ability EEG measurement method and system" pertains to a device and method for generating an index of an individual's cognitive skills. The device includes a user interface, memory for storing executable instructions by a processor, and a processing unit connected to both the user interface and memory. During the execution of the instructions by the processing unit, the device first displays a sample task with interference on the user interface, which requires the individual's initial response to the task in the presence of interference and a secondary response to at least one stimulus element. The user interface measures data that indicate the individual's reactions to the stimulus elements, including at least one metric of the individual's sensory processing capabilities under emotional load. The device simultaneously measures the individual's reactions to both the task and the stimulus elements and receives this data. The processing unit analyzes the data and calculates at least one performance metric, which includes quantitative indices of the individual's cognitive abilities under emotional load. These performance metrics can be utilized for a more precise evaluation of the individual's cognitive abilities under various emotional conditions and are displayed to the individual as output, for example, in the form of a game where the game parameters are adaptively adjusted based on the obtained scores.Additionally, the system can be used for monitoring cognitive conditions, assessing the impact of medications or educational and therapeutic programs, and recommending adjustments to treatments or educational programs. By employing machine learning methods and precise analysis of behavioral and brain data, this device facilitates the assessment and enhancement of an individual's cognitive skills and has extensive applications in the fields of medicine, education, and personal development.

[0031] A Singaporean invention with publication No. SG172262A1 which was filed on 14 / 09 / 2009 titled "Device and method for generating a representation of a subject's attention level" pertains to a device and method for creating a representation of an individual's attention level using electroencephalography (EEG) brain signals. The device includes units for measuring the individual's brain signals, extracting temporal features from these signals, and classifying these features using a classifier to generate score xl. Additionally, spectral-spatial features are extracted from the brain signals, and features containing discriminative information between focused and unfocused states are selected. These features are then classified using another classifier to produce score x2. Scores xl and x2 are combined to obtain a single score, which is displayed to the individual in the form of a game. This game adaptively adjusts its control parameters based on the received score. The process of extracting temporal features involves calculating statistics such as the standard deviation of brain waves in each electrode channel and combining these statistics into a common feature vector. Meanwhile, extracting spectral- spatial features includes using filter banks and the Common Spatial Pattern (CSP) algorithm to obtain spectral-spatial features. The selection of spectral-spatial features is based on the mutual dependence of these features on focused and unfocused states. Combining scores xl and x2 is performed through normalization, weighting based on the classifiers' accuracy, and weighted summation of the scores to obtain a singlescore. The classifiers can include various models such as Linear Discriminant Analysis, neural networks, Support Vector Machines, and fuzzy inference systems. This system determines the necessary parameters for classification and feature selection through machine learning methods using training data from individuals performing different tasks, thereby enabling accurate assessment and display of the individual's attention level.

[0032] DESCRIPTION OF THE INVENTION

[0033] The present invention relates to a system and method for examination, detection and classification of individual’s capabilities using brain mapping and brainwave analysis systems to achieve optimal results from cognitive information analysis and brainwave outputs. In this invention, a combination of data and information from cognitive assessment databases, along with rules extracted from previous research and studies, and results obtained from individual’s brain signals in electroencephalography (EEG), are aggregated to create an enhanced collective trained model. This model is used to generate the individual’s brain map and identify the individual’s strengths and weaknesses based on the obtained results.

[0034] In this regard, initially, by placing the EEG device electrodes on the individual's head, it becomes possible to record and examine the subject's brainwave changes in real-time. By collecting EEG signals within the hardware structure of the aforementioned system, a hardware filtration stage is first performed on the signal. In this unit, using a programmable processor, the preliminary filtration method for each brain channel is defined and executed according to the specific conditions of the respective channel and its placement. Subsequently, the processed information is regulated in terms of general noise using filters such as band-pass, high-pass and low-pass limiters, notch filters,and other usable preliminary filters through CPU core processing and depending on the conditions of the individual's vital elements and the information provided to them, further advanced filtration is performed.

[0035] The filtered signal from the first and second stages undergoes further advanced filtration, where noise resulting from eye blinking is removed based on a feature extraction engine that analyzes EEG data using time-frequency components and signal shape, assisted by inferences from blinking noise detection and additionally, a CNN deep learning model trained on at least 2,000 samples from individuals who were blinking at the moment and those who were not blinking using a CUDA (GPU) processor. Thus, the noise resulting from eye blinking is effectively eliminated from the aforementioned system (EEG signals received from the electroencephalograph). At this stage, the clarity and independence of the received signals compared to the raw signals extracted by the EEG device are significantly more accurate and notably independent of environmental conditions. However, the independence of the received signals may still be subject to considerable interference due to existing noise distortions, and the noise resulting from other stimuli such as muscle movements or environmental stimulations, especially the changes in signal amplitude and intensity due to the interference of brain signals with muscle signal variations, is minimized.

[0036] In this section, noise resulting from muscle movements is identified using a special protocol, and the inference of the distorted signal's shape, amplitude, and frequency due to muscle noise can be performed by the GPU and CUDA core processing unit using an Ensemble deep learning model based on at least 1,300 samples. The feature extraction engine can derive features based on the ANOVA model and feature selection based on the statistical importance of EAGENT and the features of PRCs components using the baseline signal Lt and appropriate rotation Ht and additionally, it can present a time-scale decomposition representation model. The regulated information can serve as afoundation for preparing more precise data as initial data for the analyzer system. The signal information is divided into two categories of quantitative EEG data based on numerical data and a machine learning-based feature generation engine and this information is processed in the CUDA processor unit in terms of power spectral density, amplitude, phase, channels Coherence, and the phase locking value. On the other hand, in the machine learning and initial signal analysis section, using 2,500 samples from the target population, the required features are separated and bolded by machine learning.

[0037] Additionally, using a machine learning-based graph clustering model, an initial analysis of signal categorization and feature detection is prepared. At this stage, the data resulting from machine learning-based feature generation synchronizing with the same data, however, with numerical and statistical analyses are merged in the next section and based on the amplitude and phasecorrelation of the channels, hidden features and graph-based signal features are categorized and prepared. The PSD (Power Spectral Density) feature value, derived from the numerical and statistical results of the CUDA processing section on the statistical information, is transferred to a statistical process integration engine. Here, cognitive assessment database information is merged with the PSD feature values and simultaneously compared with inferential information from scientific articles regarding cognitive risks, as well as existing documents based on their importance and in case of extraction rules from articles, patents, and the relationship between numerical data and cognitive risk indices exist, they are added to the statistical process integration engine in JSON data format.

[0038] In this section, comparative information between the numerical data extraction engine from the quantitative EEG process and the machine learning-based signal feature generation, which have been merged, is added to the statistical process integration values and by using a cognitive index / risk classificationmethodology that includes at least 3,000 statistical samples, it is analyzed with a deep learning ensemble model that has been subject-matter trained and compared with the quantitative EEG data in a trained ensemble model. In this section, simple EEG signals, after undergoing multiple filtration stages and then statistical analyses using data derived from prior knowledge and machine learning results (new knowledge), are transformed into a spider web graph of the strengths and weaknesses of various brain regions and the nature of the individual's different characteristics. Utilizing knowledge based on the impact of each brain region on individual performance, a precise analysis of the activity or inactivity of a specific brain region, or the intensity and weakness of its activity, can be provided.

[0039] This advanced system for processing and analyzing EEG (electroencephalography) signals leverages a combination of deep learning techniques, signal processing, and statistical analyses. Its primary objective is to accurately identify brain features and patterns, accelerate data processing using Graphics Processing Units (GPU), and perform precise statistical decomposition to extract useful information. The system preprocesses raw EEG data collected from various brain channels, eliminating additional noise. Subsequently, features such as signal energy, dominant frequency, and temporal scales are extracted using tools like Fourier Transform and Intrinsic Time-Scale Decomposition (ITD). These features are then fed into deep learning models, including Convolutional Neural Networks (CNN) and Ensemble models, to identify complex patterns.

[0040] Data processing is performed on the GPU, which, by leveraging the CUDA, significantly increases the processing speed of complex algorithms. The Convolutional Neural Network (CNN) is designed to identify spatiotemporal features in EEG signals. This network utilizes convolutional layers, pooling layers, and fully connected layers to extract and combine features. The output ofthe CNN can include classification of brain states such as sleep, wakefulness, or seizures. The Ensemble model enhances prediction accuracy and provides greater confidence in data analysis by combining multiple models. Additionally, the statistical method ANOVA is employed to examine significant differences among various EEG data groups, aiding in more precise analysis.

[0041] This system, due to its flexible design and use of advanced technologies, has extensive applications in medicine, cognitive science research, and engineering. Its ability to identify brain disorders, perform real-time data analysis, and provide brain mapping (Brain Mapping) are among the prominent features of this invention. The rapid and accurate data processing, along with the use of modern deep learning and statistical techniques, makes this system a powerful tool for better understanding brain function and its clinical and research applications.

[0042] BRIEF DESCRIPTION OF FIGURES

[0043] Figure 1 is a flowchart illustrating the overall operation of the invention.

[0044] Figure 2 shows the spider web graph of the brain as the final report.

Claims

WHAT IS CLAIMED IS:

1. The invention of intelligent system for detection and classification of individual capabilities through brain mapping and advanced EEG signal analysis using deep learning, comprising at least one device for measuring electroencephalography (EEG) signals from an individual and at least one unit for initial signal filtration using a programmable processor, and at least one unit for advanced signal filtration to remove noise caused by eye blinking, muscle movements, and other interferences using trained deep learning CNN models and at least one unit for extracting features from the filtered signals, which comprises extracting time-frequency features, signal energy, amplitude, signal phase, and channel coherence, and at least one unit for analyzing and categorizing the extracted data using machine learning algorithms such as graph-based clustering and Ensemble models and at least one unit for data integration, which includes merging EEG numerical data with features extracted from machine learning and performing precise statistical analyses to create collective deep learning models, and at least one unit for generating the individual's brain map that identifies and maps the individual's cognitive strengths and weaknesses based on the analyzed data and at least one Graphics Processing Unit (GPU) and at least one deep learning algorithm.

2. The system according to claim 1, wherein the method of preliminary filtration for each brain channel is defined and executed based on the specific conditions of the respective channel and its placement.

3. The system according to claim 1, wherein the processed information is regulated in terms of general noise using filters such as band-pass filters, high-pass and low-pass limiters, notch filters, and other applicable preliminary filters through CPU core processing.

4. The system according to claim 1, wherein the advanced signal filtration is performed using eye blinking noise detection based on a feature extraction engine that analyzes EEG data using time-frequency components and signal shape, assisted by inferences from eye blinking noise detection and a trained deep learning CNN model utilizing a CUDA (GPU) processor.

5. The system according to claim 1, wherein the noise resulting from muscle movements is identified using a special protocol, and the inference of the distorted signal's shape, amplitude, and frequency due to muscle noise is performed by the GPU and CUDA core processing unit using an Ensemble deep learning model based on at least 1,300 samples.

6. The system according to claim 1, wherein the feature extraction engine operates based on the ANOVA model and feature selection is performed based on the statistical importance of EAGENT and the features of PRC components.

7. The system according to claim 1, wherein the signal information is divided into two categories of quantitative electroencephalography (EEG) data based on numerical data and a machine learning-based feature generation engine.

8. The system according to claim 1, wherein the information obtained in the CUDA processor unit pertains to power spectral density, signal amplitude, signal phase, channel coherence, and the phase locking value.

9. The system according to claim 1, wherein an initial analysis of signal categorization and feature detection is prepared using a machine learningbased graph clustering system.

10. The system according to claim 1, wherein the data resulting from machine learning-based feature generation synchronizing with the same data, however, with numerical and statistical analyses are merged in the next section.

11. The system according to claim 1, wherein the information based on the amplitude and phase coherence of the channels is categorized and prepared for hidden features as well as graph-based signal features.

12. The system according to claim 1, wherein the PSD feature value, which is derived from the numerical and statistical results of the CUDA processing section on the statistical information, is transferred to a statistical process integration engine.

13. The system according to claim 1, wherein cognitive assessment database information is merged with PSD feature values and inferential information from previous data as well as cognitive risk indices in JSON data format are added to the statistical process integration engine.

14. The system according to claim 1, wherein simple EEG signals, after undergoing multiple filtration stages and subsequent statistical analyses using data derived from prior knowledge and machine learning results (new knowledge), are transformed into a spider web graph illustrating the strengths and weaknesses of various brain regions and the nature of the individual's different characteristics.

15. The system according to claim 1, wherein the machine learning unit, utilizing various deep learning models and statistical analyses, has the capability to identify complex brain patterns.

Images