Signal data processing method for diagnosis method of Alzheimer disease and frontotemporal dementia

Abstract

The invention belongs to the technical field of artificial intelligence, and discloses a signal data processing method of an Alzheimer's disease and frontotemporal dementia diagnosis method, which comprises the following steps: selecting an electroencephalogram channel by adopting PSO (Particle Swarm Optimization) and converting computer signal time sequence data into a time-frequency diagram by using wavelet transform, so as to realize accurate diagnosis of Alzheimer's disease patients and healthy subjects. And patients with frontotemporal dementia and healthy subjects as well as patients with Alzheimer's disease and patients with frontotemporal dementia can be accurately and efficiently classified. The PSO is used for selecting a channel combination with good electroencephalogram energy and reducing information interference caused by redundant electroencephalogram channels, a time-frequency graph obtained through wavelet transformation can provide time information and frequency information of electroencephalogram signals at the same time, the model can learn features more effectively, and the classification performance of the model is improved. According to the method disclosed by the invention, the accuracy of classification of the Alzheimer's disease patients and healthy subjects, the frontotemporal dementia patients and healthy subjects, and the Alzheimer's disease patients and the frontotemporal dementia patients is remarkably improved.

Description

Technical Field

[0001] The present invention belongs to the technical field of artificial intelligence, and particularly relates to a signal data processing method for diagnosing Alzheimer's disease and frontotemporal dementia. Background Art

[0002] Dementia is a term used to describe different brain diseases that affect memory, thinking, behavior, and mood. It mainly includes various types such as Alzheimer's disease, Lewy body dementia, and frontotemporal dementia. Among them, Alzheimer's disease is the most common type of dementia, and frontotemporal dementia is the second most common cause of early-onset dementia. With the aging of the world's population, dementia is becoming one of the main causes of death in the world. Although Alzheimer's disease is the most common dementia, its exact mechanism is not clear. Frontotemporal dementia has similar clinical symptoms to Alzheimer's disease (such as memory decline, behavior changes, and executive function disorders), and is often misdiagnosed as Alzheimer's disease, which will bring major safety hazards to the subsequent treatment of patients. Therefore, it is of great significance to effectively distinguish between Alzheimer's disease and frontotemporal dementia and provide corresponding treatment for patients. Traditional machine learning methods have good performance in electroencephalogram (EEG) signal classification tasks, but when faced with high-dimensional data, they often face challenges in feature extraction and limited model expression ability. With the continuous development of deep learning, more high-level abstract feature representations can be automatically learned from high-dimensional data, and complex data of this kind can be processed. Deep convolutional neural networks have become a powerful tool for processing EEG signal classification tasks.

[0003] Through the above analysis, the problems and defects existing in the prior art are as follows:

[0004] Traditional machine learning methods have good performance in EEG signal classification tasks, but when faced with high-dimensional data, they often face challenges in feature extraction and limited model expression ability. Summary of the Invention

[0005] In view of the problems existing in the prior art, the present invention provides a signal data processing method for diagnosing Alzheimer's disease and frontotemporal dementia.

[0006] The present invention is implemented as follows. A signal data processing method for diagnosing Alzheimer's disease and frontotemporal dementia includes:

[0007] S1. Define a classification framework for Alzheimer's disease and frontotemporal dementia based on EEG signal channel selection and deep convolutional neural network. Select 19 EEG signal channels through the particle swarm optimization algorithm, and use the selected EEG channel combination as the input data for wavelet transform, so as to facilitate the neural network model to extract EEG features related to the classification task;

[0008] S2. Use wavelet transform to convert the EEG data selected by channels into time-frequency maps corresponding to different frequency bands. By converting time series data into image data, both the time information and frequency information of EEG signals can be effectively utilized. Then, directly use the obtained time-frequency maps as the input of a deep convolutional neural network to achieve the classification task;

[0009] S3. Introduce a deep convolutional neural network to perform the classification task;

[0010] First, input the time-frequency map into the model. The model can capture more complex EEG features in time and frequency through the initial convolutional layer. Secondly, by increasing the number of network layers, the model can extract more abstract and high-level features, thereby improving the feature mapping relationship. The deep convolutional neural network model realizes the transfer and processing of features through learning the time-frequency map.

[0011] Further, the specific steps of S1 are as follows:

[0012] EEG channel selection helps to exclude the interference of redundant channels, reduce the computational complexity of data processing, and lower the computational cost. The energy of EEG signals directly reflects the activities of the brain and provides the overall information of EEG signals. Through the energy of EEG signals, the overall features of EEG signals can be captured;

[0013] The specific process is to use the particle swarm optimization algorithm to select the best combination of EEG channels from 19 EEG channels. The velocity and position

[0014]

[0015] Among them, is the individual optimal solution at time t, and gBest k (t) is the global optimal solution at time t. is the position of the i-th particle of the k-th control variable at iteration t. is the velocity of the i-th particle under the k-th control variable at iteration t. w(t) is the inertia weight, C1 and C2 are learning factors, and r1 and r2 are random numbers from 0 to 1. When randomly initializing the particles, the selected EEG channels are recorded as 1, and the unselected EEG channels are recorded as 0. The performance index of channel selection is the energy E of EEG signals i :

[0016]

[0017] Among them, i is each frequency band. is the squared value for each frequency band i. After EEG signal channel selection, the interference of redundant information on EEG data analysis is reduced.

[0018] Furthermore, S2 specifically includes the following steps:

[0019] The time-frequency diagram can provide both the time and frequency information of the electroencephalogram (EEG) signal, enabling the model to fully extract the EEG signal features and achieve the purpose of improving the model classification performance. For a signal f(t), wavelet transform is applied to it:

[0020]

[0021] where is the conjugate function of the wavelet function, a is the scale factor that can amplify or reduce the amplitude of the wavelet, and b is the translation factor used to correspond to different time periods in the signal. Through wavelet transformation, the obtained time-frequency diagrams of different frequency bands are used as the final input of the neural network model.

[0022] Another object of the present invention is to provide a diagnostic system for Alzheimer's disease and frontotemporal dementia, including:

[0023] A data input module for defining a classification framework for Alzheimer's disease and frontotemporal dementia based on EEG signal channel selection and deep convolutional neural network. Through the particle swarm optimization algorithm, 19 EEG signal channels are selected, and the selected EEG channel combinations are used as the input data for wavelet transformation, facilitating the neural network model to extract EEG features related to the classification task.

[0024] A data conversion module for using wavelet transformation to convert the EEG data selected by channels into time-frequency diagrams corresponding to different frequency bands. By converting time series data into image data, both the time information and frequency information of the EEG signal can be effectively utilized. Then, the obtained time-frequency diagrams are directly used as the input of the deep convolutional neural network to implement the classification task.

[0025] An introduction module for introducing a deep convolutional neural network to perform the classification task. First, the time-frequency diagram is input into the model, and the model can capture more complex EEG features in time and frequency through the initial convolutional layer. Secondly, by increasing the number of network layers, the model can extract more abstract and high-level features, thereby improving the feature mapping relationship. The deep convolutional neural network model realizes the transmission and processing of features through the learning of the time-frequency diagram.

[0026] Another object of the present invention is to provide a computer device, which includes a memory and a processor. When the computer program stored in the memory is executed by the processor, the processor executes the steps of the signal data processing method of the diagnostic method for Alzheimer's disease and frontotemporal dementia.

[0027] Another object of the present invention is to provide a computer-readable storage medium storing a computer program, which, when executed by a processor, causes the processor to execute the steps of the signal data processing method of the Alzheimer's disease and frontotemporal dementia diagnosis method.

[0028] Another object of the present invention is to provide an information data processing terminal for implementing the Alzheimer's disease and frontotemporal dementia diagnosis system.

[0029] Combined with the above technical solutions and the technical problems solved, the advantages and positive effects of the technical solutions to be protected by the present invention are as follows:

[0030] The present invention proposes an algorithm that combines wavelet transform and deep convolutional neural network classification to achieve the classification tasks of healthy subjects and Alzheimer's disease patients, healthy subjects and frontotemporal dementia patients, and Alzheimer's disease patients and frontotemporal dementia patients. This framework mainly uses the particle swarm optimization algorithm to select the electroencephalogram signal channels, reducing the influence of other redundant channels. Then, the wavelet transform is used to convert the electroencephalogram signal time series data into time-frequency diagrams of different frequency bands, so as to use the deep convolutional neural network for training to make classifications and complete the classification tasks of healthy subjects and Alzheimer's disease patients, healthy subjects and frontotemporal dementia patients, and Alzheimer's disease patients and frontotemporal dementia patients.

[0031] The present invention proposes an electroencephalogram signal processing method based on the combination of channel selection optimization and wavelet time-frequency analysis. Aiming at the problems of highly overlapping electroencephalogram manifestations, low feature signal-to-noise ratio, and channel dimension redundancy in the clinical diagnosis of Alzheimer's disease (AD) and frontotemporal dementia (FTD), an efficient classification framework is constructed. By introducing the particle swarm optimization algorithm (PSO) to globally screen the features of 19 standard electroencephalogram channels, the adverse effects of channel redundancy on the generalization ability of the model are effectively avoided, and at the same time, the computational efficiency and recognition robustness of model training are improved.

[0032] Aiming at the problem that traditional electroencephalogram signals are difficult to capture time-domain and frequency-domain features simultaneously, the present invention introduces wavelet transform (Wavelet Transform) in the feature conversion stage to convert the original multi-channel time series signal into a multi-resolution time-frequency diagram, realizing the visual coding of the dynamic changes in frequency bands such as δ, θ, α, and β, providing a high-dimensional representation containing time-frequency joint information for the deep neural network, and significantly enhancing the model's ability to recognize pathological electroencephalogram patterns.

[0033] In the feature extraction and classification stage, the present invention constructs a multi-level deep convolutional neural network (DeepCNN) structure, which combines convolution, pooling, and non-linear activation mechanisms to extract the spatial local patterns and cross-channel features of the imaged EEG signals layer by layer, and uses the deep network structure to improve the model's abstract expression ability for complex EEG patterns. The network automatically learns the non-linear mapping relationship between the time-frequency map and the disease label in an end-to-end manner, significantly improving the accuracy and stability compared with the traditional classifier based on handcrafted features.

[0034] This technical solution breaks through the problem of relying on manual experience to select channels and features in conventional EEG classification methods, realizes the full-process automatic modeling from channel optimization, time-frequency feature construction to deep feature learning, shows high sensitivity and specificity for the differential diagnosis of AD and FTD on actual clinical data, provides an intelligent solution with industrial feasibility for the early auxiliary diagnosis of cognitive disorder-related neurological diseases, and has significant medical application value and promotion potential. Brief Description of the Drawings

[0035] Figure 1 It is a flowchart of the signal data processing method for the diagnosis method of Alzheimer's disease and frontotemporal dementia provided by the embodiment of the present invention.

[0036] Figure 2 It is a block diagram of the structure of the diagnosis system for Alzheimer's disease and frontotemporal dementia provided by the embodiment of the present invention.

[0037] Figure 3 It is a flowchart of the signal data processing method for the diagnosis method of Alzheimer's disease and frontotemporal dementia based on EEG signal channel selection and deep convolutional neural network provided by the embodiment of the present invention.

[0038] Figure 4 It is a combined EEG channel diagram selected by the particle swarm optimization algorithm provided by the embodiment of the present invention.

[0039] Figure 5 It is a confusion matrix diagram for the classification task of Alzheimer's disease patients and healthy subjects provided by the embodiment of the present invention.

[0040] Figure 6 It is a confusion matrix diagram for the classification task of frontotemporal dementia patients and healthy subjects provided by the embodiment of the present invention.

[0041] Figure 7 It is a confusion matrix diagram for the classification task of Alzheimer's disease patients and frontotemporal dementia patients provided by the embodiment of the present invention. Detailed Embodiments

[0042] In order to make the objectives, technical solutions and advantages of the present invention more clear and understandable, the present invention will be further described in detail below in conjunction with embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not used to limit the present invention.

[0043] As Figure 1 , Figure 3 shown, a signal data processing method for a diagnostic method of Alzheimer's disease and frontotemporal dementia provided by an embodiment of the present invention includes the following steps:

[0044] S1. Define a classification framework for Alzheimer's disease and frontotemporal dementia based on electroencephalogram (EEG) signal channel selection and deep convolutional neural network. Select 19 EEG signal channels through the particle swarm optimization algorithm, and use the selected EEG channel combination as the input data for wavelet transform, so as to facilitate the neural network model to extract EEG features related to the classification task;

[0045] S2. Use wavelet transform to convert the EEG data selected by the channel into time-frequency diagrams corresponding to different frequency bands. By converting time series data into image data, both the time information and frequency information of the EEG signal can be effectively utilized; then directly use the obtained time-frequency diagrams as the input of the deep convolutional neural network to achieve the classification task;

[0046] S3. Introduce a deep convolutional neural network to perform the classification task;

[0047] First, input the time-frequency diagram into the model. The model can capture more complex EEG features in time and frequency through the initial convolutional layer. Secondly, by increasing the number of network layers, the model can extract more abstract and high-level features, thereby improving the feature mapping relationship; this deep convolutional neural network model realizes the transfer and processing of features through the learning of the time-frequency diagram.

[0048] Specifically, S1 provided by the embodiment of the present invention includes the following steps:

[0049] EEG channel selection helps to exclude the interference of redundant channels, reduce the computational amount of data processing, and reduce the computational cost; the energy of the EEG signal directly reflects the activity of the brain and provides the overall information of the EEG signal. The overall characteristics of the EEG signal can be captured through the energy of the EEG signal;

[0050] The specific process is to use the particle swarm optimization algorithm to select the best EEG channel combination from 19 EEG channels. The velocity of the particle movement and position

[0051]

[0052] Among them, is the individual optimal solution at time t, gBest k (t) is the global optimal solution at time t, is the position of the i-th particle of the k-th control variable at iteration t, is the velocity of the ith particle under the kth control variable at iteration t, w(t) is the inertia weight, C1 and C2 are learning factors, r1 and r2 are random numbers between 0 and 1; when randomly initializing particles, the selected EEG channel is recorded as 1, and the unselected EEG channel is recorded as 0. The performance indicator of channel selection is the EEG signal energy E i :

[0053]

[0054] Among them, i is each band, is the square value of each band i; after the EEG signal channel selection, the interference of redundant information on EEG data analysis is reduced.

[0055] S2 provided in the embodiment of the present invention specifically includes the following steps:

[0056] The time-frequency diagram can provide the time and frequency information of the EEG signal at the same time, so that the model can fully extract the characteristics of the EEG signal and achieve the purpose of improving the classification performance of the model; for a signal f(t), use wavelet transform on it:

[0057]

[0058] in, It is the conjugate function of the wavelet function, a is the scale factor, which can amplify or reduce the amplitude of the wavelet, and b is the translation factor, which is used to correspond to different time periods in the signal. Through wavelet changes, the time-frequency diagrams of different frequency bands are used as the final input of the neural network model.

[0059] like Figure 2 As shown, an Alzheimer's disease and frontotemporal dementia diagnosis system provided by an embodiment of the present invention includes:

[0060] The data input module is used to define a classification framework for Alzheimer's disease and frontotemporal dementia based on EEG signal channel selection and deep convolutional neural network. The 19 EEG signal channels are selected by particle swarm algorithm, and the selected EEG channel combination is used as the input data of wavelet transformation, so that the neural network model can extract EEG features related to the classification task.

[0061] A data conversion module is used to convert the EEG data selected by channels into time-frequency diagrams corresponding to different frequency bands. By converting time-series data into image data, both the time information and frequency information of EEG signals can be effectively utilized. Then, the obtained time-frequency diagrams are directly used as the input of a deep convolutional neural network to achieve classification tasks.

[0062] An introduction module is used to introduce a deep convolutional neural network to perform classification tasks. First, the time-frequency diagrams are input into the model, and the model can capture more complex EEG features in terms of time and frequency through preliminary convolutional layers. Secondly, by increasing the number of network layers, the model can extract more abstract and high-level features, thereby improving the feature mapping relationship. The deep convolutional neural network model realizes the transfer and processing of features through the learning of time-frequency diagrams.

[0063] Another object of the present invention is to provide a computer device, which includes a memory and a processor. The memory stores a computer program. When the computer program is executed by the processor, the processor executes the steps of the signal data processing method of the Alzheimer's disease and frontotemporal dementia diagnosis method.

[0064] Another object of the present invention is to provide a computer-readable storage medium storing a computer program. When the computer program is executed by a processor, the processor executes the steps of the signal data processing method of the Alzheimer's disease and frontotemporal dementia diagnosis method.

[0065] Another object of the present invention is to provide an information data processing terminal, which is used to implement the Alzheimer's disease and frontotemporal dementia diagnosis system.

[0066] Such as Figure 4 It is a combined EEG channel diagram selected by the particle swarm optimization algorithm.

[0067] Such as Figure 5 It is a confusion matrix diagram for the classification task of Alzheimer's disease patients and healthy subjects.

[0068] Such as Figure 6 It is a confusion matrix diagram for the classification task of frontotemporal dementia patients and healthy subjects.

[0069] Such as Figure 7 It is a confusion matrix diagram for the classification task of Alzheimer's disease patients and frontotemporal dementia patients.

[0070] The specific application field of the present invention focuses on the auxiliary diagnosis system for neurodegenerative diseases, and is particularly suitable for the early screening, subtype identification and disease progression assessment of Alzheimer's disease (AD) and frontotemporal dementia (FTD). This technical solution can be deployed in the electroencephalogram monitoring system in the hospital's neurology department, the intelligent diagnosis platform in the cognitive impairment specialty center, and the background data processing module of the wearable electroencephalogram acquisition device combined with telemedicine, providing doctors with a quantitative diagnosis basis based on objective electroencephalogram signals, improving the diagnosis efficiency and reducing the subjective error of manual intervention.

[0071] In the verification of the actual data set, the signal data processing method designed by the present invention shows excellent discrimination ability in multiple binary classification and multi-classification tasks. Through Figure 5 , Figure 6 , Figure 7 The confusion matrix given in, it can be seen that this method has achieved high classification accuracy, sensitivity and specificity in distinguishing AD patients from healthy controls, FTD patients from healthy controls, and between AD and FTD patients, reflecting the good adaptability and discrimination ability of this model to different electroencephalogram feature patterns. Especially when facing the discrimination task between AD and FTD with overlapping clinical manifestations, it has outstanding clinical practical value.

[0072] As Figure 4 shown, the channel combination strategy selected based on the particle swarm optimization algorithm (PSO) effectively improves the feature utilization efficiency of the model, avoids the noise interference brought by redundant channels, and at the same time reduces the model calculation complexity and training time, laying a foundation for real-time electroencephalogram analysis and embedded deployment. This channel selection scheme combined with the wavelet time-frequency transform processing flow can significantly enhance the expression clarity of pathological electroencephalogram signals in the time-frequency diagram, which is beneficial for the deep model to capture electrophysiological features with diagnostic value.

[0073] The present invention also supports the system-level implementation based on computer-readable storage media and information processing terminals. The computer program can be embedded in the diagnostic software platform, and the whole process of data preprocessing, feature construction, model inference and result output is completed by the processor. Cooperating with the data acquisition device to form a complete electroencephalogram intelligent auxiliary diagnosis system. This system has strong scalability and deployment flexibility, and can be widely applied to clinical scenarios such as early identification of neurocognitive disorders, monitoring of drug intervention effects and recommendation of personalized treatment paths, reflecting significant interdisciplinary integration advantages and industrial transformation potential.

[0074] It should be noted that the embodiments of the present invention can be implemented by hardware, software, or a combination of software and hardware. The hardware part can be implemented using dedicated logic; the software part can be stored in a memory and executed by a suitable instruction execution system, such as a microprocessor or dedicated design hardware. Those of ordinary skill in the art can understand that the above devices and methods can be implemented using computer-executable instructions and / or included in processor control code, such as provided on a carrier medium such as a disk, CD, or DVD-ROM, a programmable memory such as read-only memory (firmware), or a data carrier such as an optical or electronic signal carrier. The devices and modules of the present invention can be implemented by hardware circuits of programmable hardware devices such as very large scale integrated circuits or gate arrays, semiconductors such as logic chips, transistors, etc., or field programmable gate arrays, programmable logic devices, etc., can also be implemented by software executed by various types of processors, or can be implemented by a combination of the above hardware circuits and software, such as firmware.

[0075] As described above, the above are only specific embodiments of the present invention, but the protection scope of the present invention is not limited thereto. Any person skilled in the art within the technical scope disclosed by the present invention, any modifications, equivalent replacements, and improvements made within the spirit and principle of the present invention shall all be covered by the protection scope of the present invention.

Claims

1. A signal data processing method for a diagnostic method of Alzheimer's disease and frontotemporal dementia, characterized in that Including the following steps: S1. Define a classification framework for Alzheimer's disease and frontotemporal dementia based on electroencephalogram (EEG) signal channel selection and deep convolutional neural network. Select 19 EEG signal channels through the particle swarm optimization algorithm, and use the selected EEG channel combination as the input data for wavelet transform, which is convenient for the neural network model to extract EEG features related to the classification task; S2. Use wavelet transform to convert the EEG data selected by channels into time-frequency diagrams corresponding to different frequency bands. By converting time series data into image data, both the time information and frequency information of the EEG signal can be effectively utilized; then directly use the obtained time-frequency diagrams as the input of the deep convolutional neural network to achieve the classification task; S3. Introduce a deep convolutional neural network to perform the classification task; First, input the time-frequency diagram into the model, and the model can capture more complex EEG features in time and frequency through the initial convolutional layer. Second, by increasing the number of network layers.

2. The signal data processing method of the Alzheimer's disease and frontotemporal dementia diagnosis method according to claim 1, characterized in that, The specific steps of S1 are as follows: Use the particle swarm optimization algorithm to select the best combination of EEG channels from 19 EEG channels. The velocity of the particle movement and position Among them, is the individual optimal solution at time t, is the global optimal solution at time t, is the position of the i-th particle of the k-th control variable at iteration t, is the velocity of the i-th particle under the k-th control variable at iteration t, w(t) is the inertia weight, C1 and C2 are learning factors, r1 and r2 are random numbers from 0 to 1; when randomly initializing particles, the selected EEG channels are marked as 1, and the unselected EEG channels are marked as 0. The performance index of channel selection is the EEG signal energy E i : Among them, i is each band, for each band i is the square value.

3. The signal data processing method of the Alzheimer's disease and frontotemporal dementia diagnosis method according to claim 1, characterized in that The specific steps of S2 are as follows: For a signal f(t), perform wavelet transform on it: wherein, is the conjugate function of the wavelet function, a is the scaling factor that can amplify or reduce the amplitude of the wavelet, and b is the translation factor used to correspond to different time periods in the signal.

4. A diagnostic system for Alzheimer's disease and frontotemporal dementia using a signal data processing method for implementing the diagnostic method for Alzheimer's disease and frontotemporal dementia according to any one of claims 1-3, characterized in that, The Alzheimer's disease and frontotemporal dementia diagnosis system includes: A data input module, which is used to define a classification framework for Alzheimer's disease and frontotemporal dementia based on EEG signal channel selection and deep convolutional neural network. Select 19 EEG signal channels through the particle swarm optimization algorithm, and use the selected EEG channel combination as the input data for wavelet transform, which is convenient for the neural network model to extract EEG features related to the classification task; A data conversion module, which is used to use wavelet transform to convert the EEG data selected by channels into time-frequency diagrams corresponding to different frequency bands. By converting time series data into image data, both the time information and frequency information of the EEG signal can be effectively utilized; then directly use the obtained time-frequency diagrams as the input of the deep convolutional neural network to achieve the classification task; An introduction module, which is used to introduce a deep convolutional neural network to perform the classification task; First, input the time-frequency diagram into the model, and the model can capture more complex EEG features in time and frequency through the initial convolutional layer. Second, by increasing the number of network layers, the model can extract more abstract and high-level features, thereby improving the feature mapping relationship; the deep convolutional neural network model realizes the transmission and processing of features through the learning of the time-frequency diagram.

5. A computer device, characterized in that, The computer device includes a memory and a processor. When the computer program stored in the memory is executed by the processor, the processor executes the steps of the signal data processing method of the Alzheimer's disease and frontotemporal dementia diagnosis method as described in any one of claims 1-3.

6. A computer-readable storage medium stores a computer program. When the computer program is executed by a processor, the processor executes the steps of the signal data processing method of the Alzheimer's disease and frontotemporal dementia diagnosis method as described in any one of claims 1-3.

7. An information data processing terminal, characterized in that, The information data processing terminal is used to implement the Alzheimer's disease and frontotemporal dementia diagnosis system as described in claim 4.

Images