AD diagnosis method fusing electroencephalogram frequency spectrum standardization deviation and brain network

Abstract

The invention discloses an AD diagnosis method fusing electroencephalogram frequency spectrum standardization deviation and a brain network. The AD diagnosis method comprises the steps that S1, band-pass filtering and fragment segmentation are conducted on multi-channel EEG signals of five brain regions; s2, for each segment, calculating brain region-frequency band characteristics of five different brain regions; s3, performing standardized modeling by adopting a GAMLSS model, identifying individual deviation information of a disease group about a specific brain region rPSD, training a LightGBM classifier by utilizing the individual deviation information, and extracting a structured feature vector by utilizing an internal structure of the classifier; s4, dividing each segment into a plurality of non-overlapping frequency boxes according to frequency resolution, calculating an FCN graph in a matrix form of each frequency box, constructing a three-dimensional image according to the FCN graph, and extracting a visual feature vector by using a Swin transformer network; and S5, constructing a connection vector, and inputting the connection vector into a classifier for classification. The method provides a new technical approach for accurate diagnosis and differential diagnosis of neurodegenerative diseases.

Description

Technical Field

[0001] This invention relates to the field of modality recognition technology, and more particularly to a diagnostic method for AD (Alzheimer's disease). Background Technology

[0002] Alzheimer's disease (AD) and frontotemporal dementia (FTD) are highly prevalent neurodegenerative diseases, accounting for 60-80% and 10% of all dementia cases, respectively. The complex clinical and pathological presentations of these two diseases pose significant challenges to diagnosis and treatment.

[0003] Alzheimer's disease (AD) is the most common type of dementia, typically characterized by cognitive decline, memory loss, and behavioral changes. Frontotemporal dementia (FTD), while less common, tends to affect younger populations and is primarily characterized by personality changes and language impairments. Currently, there is no cure or treatment to halt the progression of AD or FTD, which has spurred increasing research interest in age-related brain changes associated with pathological aging.

[0004] Traditional diagnostic methods for Alzheimer's disease (AD) and free radical damage (FTD) rely on clinical assessment, neuroimaging studies (such as MRI, CT, PET, etc.), and cognitive tests. However, these methods are often costly and time-consuming. Electroencephalography (EEG), due to its high temporal resolution, non-invasiveness, and low cost, has become a promising tool for the early diagnosis and monitoring of abnormal brain activity.

[0005] In recent years, EEG has become a widely used modality for investigating and clinically supporting neurodegenerative diseases such as AD, FTD, mild cognitive impairment (MCI), and Parkinson's disease (PD). A growing body of research demonstrates that EEG signals can provide valuable biomarkers that can distinguish these diseases from healthy aging. Traditional EEG analysis methods, including spectral power analysis, coherence, phase synchronization, and event-related potentials (ERPs), have been commonly used to characterize brain rhythm changes associated with disease states.

[0006] However, by averaging population data, population analysis tends to mask individual differences, which may obscure significant variations in disease progression and individual-level treatment responses. From a precision medicine perspective, this "one-size-fits-all" approach fails to consider the unique genetic, environmental, and lifestyle factors that influence each patient's condition.

[0007] Normalized modeling has become a powerful method for achieving subject-specific inferences by characterizing population-level norms and quantifying individual deviations from these norms. While normalized modeling is established in other fields, such as the use of height or weight growth charts in pediatrics, it remains relatively novel in neuroimaging research. In existing research, normalized modeling has been used to predict brain age, thereby identifying potential problems in brain development and aging processes.

[0008] Recent MRI-based studies have employed standardized modeling to map age-related trajectories of brain phenotypes throughout the human lifespan. These efforts have been used not only to characterize standardized age-related changes in brain structure but also to describe structural and functional heterogeneity in mental illnesses such as schizophrenia and bipolar disorder, as well as neurodegenerative diseases such as Alzheimer's disease (AD).

[0009] Deep learning has gained popularity due to its ability to automatically learn features from raw EEG data. While numerous deep learning methods have been applied to EEG-based disease diagnosis, few currently utilize multimodal models. In neuroscience, many studies have adopted multimodal frameworks. Despite significant successes in brain disease applications, multimodal approaches to EEG data remain underexplored due to limited sample sizes and insufficient fusion strategies. Summary of the Invention

[0010] To address the technical problems in existing technologies, such as the masking of individual differences by group analysis, insufficient multimodal fusion strategies, and the need to improve diagnostic accuracy, this invention aims to provide an AD diagnostic method that integrates EEG spectrum normalization bias and brain network, so as to achieve individualized neurophysiological assessment, effectively integrate heterogeneous data modalities, and significantly improve diagnostic accuracy.

[0011] To this end, the present invention provides an AD diagnosis method that integrates EEG spectral normalization bias and brain network, comprising the following steps: S1, bandpass filtering is performed on multi-channel EEG signals from five brain regions to segment continuous EEG signals into non-overlapping segments; S2, for each segment in step S1, the relative power spectrum (rPSD) of all channels in five typical frequency bands is calculated, and all channels within each brain region are averaged to obtain brain region-frequency band features; S3, normalization modeling is performed using the GAMLSS model, and individual bias information regarding rPSD of specific brain regions in the disease group is identified, and this individual bias information is used to train LightGB. The M classifier is used to extract structured feature vectors, specifically by extracting the indexes of each decision tree leading to a leaf node. These leaf node indices are then concatenated into feature vectors. Step S4 involves dividing each segment from step S1 into several non-overlapping frequency bins according to a set frequency resolution. The matrix form of the functional connectivity network (FCN) for each frequency bin is calculated, and the FCNs of all frequency bins are stacked along the frequency dimension to construct a three-dimensional image. A Swintransformer network is then used to extract visual feature vectors. Step S5 involves concatenating the structured feature vectors extracted from the LightGBM classifier with the visual feature vectors extracted from the Swin transformer network, and inputting this concatenation into the classifier for classification, resulting in a three-class discrimination between AD, FTD, and HC.

[0012] This invention proposes a multimodal integrated standardized diagnostic network (MIND-NET) framework. This network framework combines standardized modeling with multimodal feature extraction for the classification of neurodegenerative disease subtypes. It can not only achieve individualized neurophysiological assessment, but also significantly improve diagnostic accuracy, providing a new technical approach for the accurate and differentiated diagnosis of neurodegenerative diseases.

[0013] In addition to the objectives, features, and advantages described above, the present invention has other objectives, features, and advantages. The invention will now be described in further detail with reference to the figures. Attached Figure Description

[0014] The accompanying drawings, which form part of this application, are used to provide a further understanding of the invention. The illustrative embodiments of the invention and their descriptions are used to explain the invention and do not constitute an undue limitation of the invention. In the drawings:

[0015] Figure 1 This is a flowchart of the AD diagnosis method of the present invention that integrates EEG spectrum normalization deviation and brain network;

[0016] Figure 2 The distribution of extreme biases in AD and FTD across five brain regions is shown;

[0017] Figure 3 The differences in deviation values ​​among the AD group, FTD group, and HC group are shown;

[0018] Figure 4 The positive and negative deviation plots for the AD and FTD groups are shown;

[0019] Figure 5 The paper presents a comparison of the classification performance of raw features without normalization modeling and features with normalization modeling in four classic machine learning algorithms. Detailed Implementation

[0020] The present invention will now be described in detail with reference to the accompanying drawings and embodiments.

[0021] To address the technical problems in existing technologies, such as population analysis masking individual differences, insufficient multimodal fusion strategies, and the need to improve diagnostic accuracy, this invention proposes a multimodal integrated normalized diagnostic network (MIND-NET) framework. This is a novel framework that combines normalized modeling with multimodal feature extraction for the classification of neurodegenerative disease subtypes.

[0022] The method of this invention comprises two main stages: First, based on HC (healthy control group) data, this invention constructs normalized trajectories of EEG spectral features using a generalized additive location, scale, and shape model (GAMLSS). Second, this invention applies individual-level bias derived from this normalized model to assess heterogeneity within the AD and FTD cohorts. The MIND-NET framework utilizes a Swin transformer to extract features from EEG-derived functional connectivity network (FCN) images and LightGBM (LGB) to extract features from structured bias data, thereby creating a comprehensive multimodal representation and enhancing the classification accuracy of neurodegenerative disease subtypes.

[0023] The overall workflow of this invention is as follows: Figure 1 As shown, the complete architecture of the MIND-NET model is illustrated. It begins with EEG preprocessing; (a) and (b) show normalized modeling across brain regions and structured data feature extraction using LGB; (c) is the image feature extraction stage based on the Swing transformer, which uses a traditional Swing transformer to generate feature vectors. Finally, the connected features are input into a classifier to achieve tri-class discrimination between AD, FTD, and HC.

[0024] Specifically, this AD diagnostic method includes the following steps S1 to S5.

[0025] S1. Obtain the EEG dataset

[0026] This invention utilizes the publicly available EEG dataset OpenNeuro ds004504. For more reasonable and standardized modeling, this invention uses data from subjects within the dataset's concentrated age range (61-78 years), including 25 AD patients, 17 FTD patients, and 28 HC individuals. Electrode placement followed the internationally standardized 10-20 system, with 19 channels located at Fp1, Fp2, F7, F3, Fz, F4, F8, T3, C3, Cz, C4, T4, T5, P3, Pz, P4, T6, O1, and O2. Electrodes A1 and A2 were designated as reference electrodes.

[0027] S2, Data Preprocessing

[0028] The EEG channel signal is bandpass filtered from 0.5 to 45 Hz to remove low-frequency drift and high-frequency electromyography artifacts, and uniformly resampled from 500 Hz to 200 Hz. The continuous EEG signal is segmented into non-overlapping 4-second epochs, and each segment is normalized to zero mean and unit variance within the channel to eliminate the influence of overall power differences between channels.

[0029] S3, Structured Data Feature Extraction

[0030] For each segment, this invention uses the Welch method to calculate the power spectral density (PSD) from 1 to 45 Hz, and calculates the relative power spectrum (rPSD) of all channels in five typical frequency bands: δ (1-4 Hz), θ (4-8 Hz), α (8-13 Hz), β (13-30 Hz), and γ (30-45 Hz). Subsequently, this invention arranges the EEG channels according to five different brain regions: frontal lobe (Fp1, Fp2, F7, F3, Fz, F4, F8), central lobe (C3, Cz, C4), temporal lobe (T3, T4, T5, T6), parietal lobe (P3, P4, Pz), and occipital lobe (O1, O2). This invention then averages all channels within each brain region to obtain the final brain region-frequency band characteristics.

[0031] rPSD eliminates the influence of inter-individual differences in overall EEG power and electrode impedance variations through normalization, enhancing comparability between different subjects and time periods, while reducing the impact of experimental environment and individual physiological fluctuations. This contributes to the stability of subsequent standardized modeling in this invention.

[0032] To construct reference trajectories of EEG signals in healthy individuals at different age stages and to identify bias characteristics of rPSD in specific brain regions within disease groups, this invention employs GAMLSS for standardized modeling. This method can comprehensively model nonlinearity, heteroscedasticity, skewness, and kurtosis variations in neurophysiological data, making it suitable for population modeling and individual anomaly detection of complex EEG biomarkers. Here, the primary covariate is age, and the response variable is the rPSD value of a specific frequency band in a specific brain region. This index reflects the intensity of neural activity in a brain region within a specific frequency range and has clear neurophysiological significance.

[0033] This invention uses the gamlss package in R to fit a GAMLSS model based on HC data. Through comparison of fitting candidate distribution families, this invention ultimately selects the Box-Cox-t distribution (BCT) as the response variable distribution for GAMLSS. The mathematical formula for GAMLSS is as follows:

[0034]

[0035] Where F represents the BCT distribution, which is parameterized by the location parameter μ, the scale parameter σ > 0, the power shape parameter ν, and the degree of freedom parameter τ > 0.

[0036] in It is a suitable join function, X μ and Z μ These are the known fixed-effects and random-effects design matrices, with corresponding coefficient vectors β and β. μ and γ μ ,and Represents smooth basis functions used to capture covariates. The potential nonlinear effects. Here, the present invention writes a predictor for the location parameter μ; similar formulas apply to other distribution parameters that are modeled by nonlinear smoothing terms.

[0037] After fitting the GAMLSS model, this invention manually calculates the 5%, 50%, and 95th percentiles for each age point based on the four distribution parameters returned by the model, representing the reference range of EEG characteristics in healthy individuals. Subsequently, this invention uses linear interpolation to smoothly map these percentile estimates to a regular age sequence, thereby constructing a complete normalized trajectory and confidence interval.

[0038] To assess the extent to which AD and FTD patients deviate from a standardized reference trajectory, this invention compares their corresponding EEG characteristics with 5%–95% confidence intervals for age-matched healthy individuals. This invention calculates the percentage of each category falling within and outside the reference interval to quantify the probability of EEG abnormalities for each category.

[0039] Individual bias information is further used to construct classification models. Specifically, this invention uses each subject's bias value (i.e., bias data calculated from normalized values ​​derived from age-matched HC) as input features to train multiple machine learning classifiers to distinguish between AD and FTD. The best-performing classifier is then used to extract structured tabular data. This workflow enables normalized modeling not only to reveal subject-specific neural biases but also to provide structured input for downstream multimodal models, thereby enhancing the integration of individualized neural signatures into a broader computational framework for neurodegenerative disease analysis.

[0040] S4. Image Feature Extraction

[0041] After segmenting the raw EEG signal into non-overlapping 4-second windows, this invention uses 4-second segments and a frequency resolution of 0.25 Hz, taking into account the impact of segment length on spectral resolution in the Welch method. The frequency resolution Δf is inversely proportional to the analysis window duration T. At this resolution, the signal from 0.25 to 25 Hz is divided into 100 non-overlapping frequency bins, covering the critical range from delta waves to low gamma waves in the diagnosis of mental illness. Within each frequency band, this invention calculates the coherence function between all channel pairs to obtain the functional connectivity strength between channels. Given EEG data with a dimension of C×T (where C is the number of channels and T is the number of time points), each matrix corresponds to the brain functional connectivity network within the frequency bin.

[0042] The process calculates in-band coherence using the following formula:

[0043]

[0044] Where P xy (ω) represents the cross-spectral density between x(t) and y(t), while P xx (ω) and P yy (ω) represent their respective autospectral densities.

[0045] To encode this sequence into a standard image format suitable for input to the Swing transformer, this invention employs a matrix concatenation method, stacking the 100 frequency box concatenation maps along the frequency dimension to construct a three-dimensional image. The image is then resampled to 224×224×3 dimensions to meet the input requirements of the Swing transformer network.

[0046] This representation method not only preserves the brain network structure at different frequencies but also allows the network to capture cross-frequency functional connectivity patterns from the image space. Its advantage lies in preserving high-frequency resolution microstructural differences, rather than coarsely dividing frequency bands, thereby enhancing the model's ability to perceive subtle frequency domain features of neuropsychological states.

[0047] The input image is first processed through the initial L layer of the standard Swin transformer (patch segmentation, linear embedding, and shift window self-attention) to extract multi-scale visual features. Finally, a 1×768 feature vector is obtained in the feature extraction layer, which is fused in parallel with subsequent text features.

[0048] This invention uses rPSD bias based on normalized modeling as input to train an LGB classifier, with the goal of predicting the source label of each sample from the bias features. After training, this invention utilizes the internal structure of the model for feature extraction.

[0049] For any given input sample, this invention does not retrieve its final classification probability, but instead records the indices of the leaf nodes reached in each decision tree. The LGB ensemble of this invention has 512 trees, and these leaf node indices are concatenated into a 1×512 feature vector. This vector is a high-dimensional nonlinear transformation of the original bias scores, capturing deep structural information in the data.

[0050] S5, Feature Fusion Prediction

[0051] To perform the final classification task, this invention fuses information from different modalities. The 1×512-dimensional feature vector extracted by LGB is concatenated with the 1×768-dimensional visual feature vector obtained from the visual model, producing a 1×1280-dimensional joint feature representation. This concatenated vector is then input into the downstream classifier to generate the final prediction.

[0052] Effect Comparison

[0053] After generating normalized trajectories, this invention maps data from AD and FTD subjects onto a model and calculates deviation scores based on the 5th to 95th percentile confidence intervals defined for healthy controls. In the frontal cortex, the deviation rate in the β band was 14.99% for AD and 17.59% for FTD, rising to 18.05% for AD and 25.42% for FTD in the γ band. In the temporal lobe, AD showed a deviation of 19.41% in the β band, while FTD peaked at 20.59% in the α band. The occipital lobe showed the most significant deviation, with an α band AD of 23.16% and an FTD of 34.46%. The parietal lobe showed an α band deviation of 15.85% for AD and 19.65% for FTD, while the central cortex remained relatively moderate across all bands. Overall, extreme deviations in AD and FTD were concentrated in the α, β, and γ bands of the temporal, occipital, and parietal lobes, and FTD also showed significant abnormalities in the β and γ bands of the frontal cortex, such as... Figure 2 .

[0054] The present invention then calculates the mean bias of rPSD in AD and FTD patients after normalization to HC-derived GAMLSS trajectories. Both the AD and FTD cohorts showed significantly elevated bias scores compared to the HC group, indicating significant heterogeneity in their EEG signatures. These group differences are as follows: Figure 3 As shown in the figure, the colored markers represent individual subjects from the AD, FTD, and HC groups, vividly highlighting the differences in rPSD bias between cohorts. The figure indicates that the bias values ​​in the AD and FTD groups are significantly higher than those in the HC group.

[0055] To identify shared bias patterns and explore heterogeneity between AD and FTD groups, this invention calculates the average power of positive and negative biases for each group and plots normalized (0-1) overlapping topographic maps.

[0056] Figure 4 Mid-section (a) and section (b) show the positive deviation plots for the AD and FTD groups, respectively. In both groups, the β-band activity in the occipital region showed the most significant difference, while additional substantial divergence was observed in the θ-band in the frontal and parietal regions.

[0057] Figure 4 The mid-section (c) and mid-section (d) show the negative deviation maps of AD and FTD, respectively, highlighting the δ and γ band activity in the frontal cortex, which provides the clearest distinction between the two groups.

[0058] These results indicate that the positive biases distinguishing AD from FTD are primarily characterized by the β-band activity of the occipital lobe electrodes and the θ-band activity of the frontal and parietal lobe electrodes. In contrast, the negative biases are most prominent in the δ and γ-band activities of the frontal lobe. This spatial-frequency signature provides an intuitive and informative comparison of shared bias patterns and group-specific heterogeneity.

[0059] This invention compares raw features without normalized modeling with features that have been modeled using normalized modeling, such as... Figure 5 The classification performance of four classic machine learning algorithms was evaluated: LGB (default parameters: 512 trees and 63 leaf nodes), Multilayer Perceptron (MLP), Random Forest, Support Vector Machine (SVM), and k-Nearest Neighbors (kNN). Four evaluation metrics were used: Accuracy (ACC), F1 score (F1), Sensitivity (SENS), and Specificity (SPEC).

[0060] It is evident that in traditional machine learning methods, normalized modeling-enhanced features produce better performance compared to raw features without normalized modeling. In particular, the LGB classifier stands out with an impressive accuracy of 92.05%. This is the main reason why this invention chooses LGB for structured data feature extraction.

[0061] This invention uses LGB to perform classification between AD, FTD, and HC individually for each brain region. Five-fold cross-validation is used to evaluate the model's performance. As shown in Table 1, the experimental results demonstrate that the temporal and occipital lobe regions achieved classification accuracies of 82.00% and 82.58%, respectively, significantly outperforming the other three regions. This finding shows strong consistency with previously identified high heterogeneity features of AD and FTD, thus confirming the significant advantages and effectiveness of the normalized modeling method proposed in this invention for deep learning model training.

[0062]

[0063] Table 2 shows the impact of different feature extraction schemes on the overall performance of MIND-NET. Model 1 uses the Swin-T (tiny) Swin transformer architecture to extract image features, while Model 2 utilizes LGB to process structured data. Specifically, LGB (T:100) uses 100 trees to generate a 1×100 feature vector, and LGB (T:512) uses 512 trees to generate a 1×512 feature vector. The " / " symbol indicates that structured data features were not used in the model. This invention also includes AUC as an additional performance metric to evaluate classification ability. Experimental results show that the combination of Swin-T and LGB (T:512) achieves the best performance, reaching 97.12% accuracy, 0.998 AUC, and 97.01% F1 score under five-fold cross-validation. Using the Swin-T alone yielded 93.42% accuracy, 0.993 AUC, and 93.03% F1 score, while using the LGB (T:512) alone achieved 91.79% accuracy, 0.973 AUC, and 86.97% F1 score.

[0064]

[0065] To demonstrate the capabilities of the proposed MIND-NET (Swin-T and LGB (T:512)), Table 3 evaluates not only the three-class classification performance in Case 1 (AD vs FTD vs HC), but also three binary classification tasks: Case 2 (AD vs FTD), Case 3 (AD vs HC), and Case 4 (FTD vs HC). Evaluation metrics include ACC, F1, SEN, SPEC, and AUC. To further validate the effectiveness of the proposed model, it is compared with benchmarks of several state-of-the-art three-class classification methods on the same dataset.

[0066]

[0067] Experimental results show that the MIND-NET framework of this invention achieves excellent performance in all cases. In the three-class classification task (Case 1), the accuracy reaches 97.12%±0.31%, the AUC is 0.998±0.002, and the F1 score is 97.01%±0.39%. In the binary classification task, Case 2 (AD vs FTD) achieves an accuracy of 97.79%±0.25%, Case 3 (AD vs HC) achieves an accuracy of 98.11%±0.34%, and Case 4 (FTD vs HC) achieves an accuracy of 98.25%±0.24%, all indicators demonstrating excellent performance.

[0068] Comparisons with other state-of-the-art methods show that the MIND-NET framework of this invention significantly outperforms existing technologies, as shown in Table 4. Acharya et al.'s EEGConvNeXt method achieved 95.70% accuracy in three-class classification, while the method of this invention achieved 97.12%, an improvement of 1.42%. Chen et al.'s custom CNN combined with a visual transformer method only achieved 79.12% accuracy, while the method of this invention showed a significant improvement. Hachamnia et al.'s LightGBM method combined with wavelet transform achieved 93.21% accuracy in three-class classification, still lower than the 97.12% of this invention. These comparative results clearly demonstrate the technical advantages of the MIND-NET framework proposed in this invention.

[0069]

[0070] The above description is merely an embodiment of the present invention and is not intended to limit the invention. Various modifications and variations can be made to the present invention by those skilled in the art. Any modifications, substitutions, or improvements made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A method for diagnosing AD by integrating EEG spectral normalization bias and brain network, characterized in that, Includes the following steps: S1. Bandpass filtering is performed on the multi-channel EEG signals from five brain regions to segment the continuous EEG signals into non-overlapping segments. S2. For each segment in step S1, calculate the relative power spectrum rPSD of all channels in five typical frequency bands, and average all channels in each brain region to obtain brain region-frequency band characteristics. S3. The GAMLSS model is used for normalized modeling, and individual bias information about rPSD in specific brain regions of the disease group is identified. This individual bias information is used to train the LightGBM classifier, and the internal structure of the classifier is used to extract structured feature vectors. S4. Divide each segment in step S1 into several non-overlapping frequency boxes according to the set frequency resolution, calculate the matrix form of the functional connectivity network (FCN) of each frequency box, stack the FCNs of all frequency boxes along the frequency dimension to construct a three-dimensional image, and use the Swing transformer network to extract visual feature vectors. S5. The structured feature vector extracted from the LightGBM classifier and the visual feature vector extracted from the Swin transformer network are concatenated and input into the classifier for classification to obtain the three-class discrimination between AD, FTD and HC.

2. The AD diagnosis method integrating EEG spectrum normalization deviation and brain network as described in claim 1, characterized in that, The multichannel EEG signals from five brain regions include 19 channels: frontal lobe (Fp1, Fp2, F7, F3, Fz, F4, F8), central lobe (C3, Cz, C4), temporal lobe (T3, T4, T5, T6), parietal lobe (P3, P4, Pz), and occipital lobe (O1, O2).

3. The AD diagnostic method integrating EEG spectrum normalization deviation and brain network as described in claim 1, characterized in that, In step S1, the EEG signal is bandpass filtered from 0.5 to 45 Hz and then uniformly resampled from 500 Hz to 200 Hz. Each segment after segmentation is standardized with zero mean and unit variance within the channel.

4. The AD diagnosis method integrating EEG spectrum normalization deviation and brain network as described in claim 1, characterized in that, In step S2, five typical frequency bands are used: δ (1-4Hz), θ (4-8Hz), α (8-13Hz), β (13-30Hz), and γ (30-45Hz).

5. The AD diagnosis method integrating EEG spectrum normalization deviation and brain network as described in claim 1, characterized in that, In the GAMLSS model, the covariate is age, and the response variable is the rPSD value of a specific brain region and frequency band. The Box-Cox-t distribution is chosen as the distribution of the response variable in GAMLSS.

6. The AD diagnostic method integrating EEG spectrum normalization deviation and brain network as described in claim 5, characterized in that, Step S3 also includes: S31. Calculate the 5%, 50%, and 95th percentiles for each age point based on the distribution parameters returned by the GAMLSS model to represent the reference range of EEG characteristics in healthy individuals. Then, use linear interpolation to smoothly map these percentile estimates to regular age sequences. S32. Compare the EEG characteristics of patients with AD and FTD to be evaluated with the 5%-95% confidence interval of healthy individuals of the same age, calculate the percentage of each category falling within and outside the reference interval, to quantify the probability of EEG abnormality in each category, and thus assess the degree to which AD and FTD patients deviate from the standardized reference trajectory.

7. The AD diagnosis method integrating EEG spectrum normalization deviation and brain network as described in claim 1, characterized in that, Individual bias information regarding rPSD in specific brain regions of disease groups is identified as bias data calculated from normalized values ​​derived from age-matched HC.

8. The AD diagnosis method integrating EEG spectrum normalization deviation and brain network as described in claim 1, characterized in that, In step S4, the frequency resolution is set to 0.25 Hz. At this resolution, the signal from 0.25 to 25 Hz is divided into 100 non-overlapping frequency boxes, covering the key range from delta waves to low gamma waves in the diagnosis of mental illness.

9. The AD diagnosis method integrating EEG spectrum normalization deviation and brain network as described in claim 1, characterized in that, In step S4, the 3D image is first resampled to 224×224×3 dimensions, and then input into the Swin transformer model, outputting a 1×768 visual feature vector.

10. The AD diagnostic method integrating EEG spectrum normalization deviation and brain network as described in claim 1, characterized in that, The 1×512-dimensional feature vector extracted from the LightGBM classifier is concatenated with the 1×768-dimensional visual feature vector obtained from the Swin transformer model to produce a 1×1280-dimensional concatenation vector, which is used as input to the downstream classifier to produce a predicted classification.

Images