Interactive music cognition training system for early Alzheimer's disease patients

Abstract

The invention discloses an interactive music cognition training system for early-stage senile dementia patients, and relates to the technical field of music cognition training, and the system comprises a multi-mode interactive training module which is configured to execute a music training task; collecting multi-modal interaction data; the intelligent evaluation module is configured to perform feature extraction on the multi-modal interaction data; performing trend analysis on the time sequence characteristic data by adopting a time sequence analysis model; a reinforcement learning algorithm is adopted to generate a dynamic optimal adjustment strategy according to the cognitive ability change trend; and the multi-modal interactive training module is further configured to dynamically adjust the music training task according to the dynamic optimal adjustment strategy. According to the method, the cognitive ability change trend based on the multi-modal interaction data is quantified by combining the time sequence analysis model and the reinforcement learning algorithm, the music training interactivity and reliability are improved, and a scientific basis is provided for long-term curative effect evaluation and treatment strategy optimization.

Description

Technical Field

[0001] This application relates to the technical field of music cognitive training, and particularly to an interactive music cognitive training system for early Alzheimer's disease patients. Background Art

[0002] With the increasing trend of population aging, Alzheimer's disease (especially Alzheimer's disease) has become a major public health problem that urgently needs to be solved globally. Existing research has widely confirmed the potential utility of music therapy in regulating emotions, stimulating memory, and improving cognitive function. Therefore, it is necessary to study music cognitive training technology.

[0003] In the prior art, Chinese Patent CN115240508A discloses a cognitive ability training method, terminal device, and storage medium based on music rhythm. The method includes: playing training audio data, and during the playing of the training audio data, setting beat points on the progress bar corresponding to the training audio data; obtaining the setting time of the beat points, and matching the setting time with the time information of the original beat points corresponding to the beat points carried in the training audio data to obtain the matching result of the beat points; determining the cognitive ability training result for the music rhythm of the training audio data according to the matching result of the beat points, and adjusting the training difficulty based on the cognitive ability training result, where different training difficulties correspond to different training audio data.

[0004] However, the above prior art lacks a quantitative analysis of the changing trend of cognitive ability based on multi-modal interaction data, and the interactivity and reliability of music training are poor. Summary of the Invention

[0005] This application provides an interactive music cognitive training system for early Alzheimer's disease patients to solve the problem that the existing music cognitive training technology lacks a quantitative analysis of the changing trend of cognitive ability based on multi-modal interaction data, and the interactivity and reliability of music training are poor.

[0006] On the one hand, this application provides an interactive music cognitive training system for early Alzheimer's disease patients, including: a multi-modal interaction training module and an intelligent evaluation module.

[0007] The multi-modal interaction training module is configured to: execute music training tasks; collect multi-modal interaction data during the music training process.

[0008] The intelligent evaluation module is configured to: extract features from the multi-modal interaction data to obtain time-series feature data; perform trend analysis on the time-series feature data using a time-series analysis model to obtain the changing trend of cognitive ability; generate a dynamic optimal adjustment strategy according to the changing trend of cognitive ability using a reinforcement learning algorithm.

[0009] The multi-modal interaction training module is further configured to: dynamically adjust the music training task according to the dynamic optimal adjustment strategy.

[0010] In a possible implementation, the multi-modal interaction training module includes: a music training execution unit and an interaction data acquisition unit.

[0011] The music training execution unit is used to execute the music training task.

[0012] The interaction data acquisition unit is used to collect multi-modal interaction data during the music training process.

[0013] The music training execution unit is further used to dynamically adjust the music training task according to the dynamic optimal adjustment strategy.

[0014] In a possible implementation, the music training task includes: a pitch recognition training task, a rhythm perception training task, an instrument recognition training task, a song memory training task, a music karaoke training task, a music imagination training task.

[0015] The multi-modal interaction data includes: interaction data, behavior data, physiological data, and emotional data.

[0016] In a possible implementation, the intelligent evaluation module includes: a feature extraction unit, a trend analysis unit, and a strategy adjustment unit.

[0017] The feature extraction unit is used to extract features from the multi-modal interaction data to obtain time-series feature data.

[0018] The trend analysis unit is used to perform trend analysis on the time-series feature data using a time series analysis model to obtain the changing trend of cognitive ability.

[0019] The strategy adjustment unit is used to generate a dynamic optimal adjustment strategy according to the changing trend of cognitive ability using a reinforcement learning algorithm.

[0020] In a possible implementation, before extracting features from the multi-modal interaction data, the feature extraction unit pre-processes the multi-modal interaction data by time synchronization alignment, smoothing, and standardization.

[0021] In a possible implementation, the feature extraction uses a principal component analysis algorithm.

[0022] In a possible implementation, before performing trend analysis on the time-series feature data using a time series analysis model, the trend analysis unit pre-processes the time-series feature data by first-order differencing.

[0023] In a possible implementation, the time series analysis model adopts an autoregressive integrated moving average model and combines with a vector autoregressive model.

[0024] In a possible implementation, the reinforcement learning algorithm adopts a Q-learning algorithm, takes different cognitive ability change trends as the states of the Q-learning algorithm, and takes different dynamic adjustment strategies as the actions of the Q-learning algorithm.

[0025] The trend analysis unit generates a dynamic optimal adjustment strategy by designing the reward function of the Q-learning algorithm and using the Q-learning algorithm to select the optimal action according to the current cognitive ability change trend.

[0026] The interactive music cognitive training system for early Alzheimer's patients in this application has the following advantages:

[0027] By combining the time series analysis model and the reinforcement learning algorithm, the change trend of cognitive ability based on multi-modal interaction data is quantified, the interactivity and reliability of music training are improved, and a scientific basis is provided for long-term efficacy evaluation and treatment strategy optimization.

[0028] By performing time synchronization alignment processing, smoothing processing, and standardization processing on the multi-modal interaction data, it is convenient for subsequent feature extraction.

[0029] The proposed feature extraction uses the principal component analysis algorithm, which improves the accuracy of feature extraction for multi-modal interaction data.

[0030] By performing first-order difference processing on the time series feature data, the data stationarity is improved.

[0031] The proposed time series analysis model adopts an autoregressive integrated moving average model and combines with a vector autoregressive model, which can capture the dynamic correlation between the features of multi-modal interaction data and improve the accuracy of the cognitive ability change trend.

[0032] The proposed use of the Q-learning algorithm to select the optimal action according to the current cognitive ability change trend and generate a dynamic optimal adjustment strategy ensures that the music training content is always adapted to the patient's cognitive state and improves the music training effect. BRIEF DESCRIPTION OF THE DRAWINGS

[0033] In order to more clearly illustrate the technical solutions in the embodiments of the present application or the prior art, the following will briefly introduce the drawings required for the description of the embodiments or the prior art. Obviously, the drawings in the following description are only some embodiments of the present application. For those of ordinary skill in the art, other drawings can be obtained based on these drawings without creative efforts.

[0034] Figure 1 This is a schematic diagram of the modules of the interactive music cognitive training system for early Alzheimer's patients provided by the embodiments of the present application. Detailed implementation manners

[0035] Next, the technical solutions in the embodiments of the present application will be clearly and completely described in conjunction with the accompanying drawings in the embodiments of the present application. Obviously, the described embodiments are only a part of the embodiments of the present application, rather than all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those of ordinary skill in the art without creative efforts shall fall within the protection scope of the present application.

[0036] As Figure 1 shown, the embodiments of the present application provide an interactive music cognitive training system for early Alzheimer's patients, including: a multi-modal interaction training module and an intelligent evaluation module.

[0037] The multi-modal interaction training module is configured to: execute music training tasks; collect multi-modal interaction data during the music training process.

[0038] The intelligent evaluation module is configured to: extract features from the multi-modal interaction data to obtain time-series feature data; perform trend analysis on the time-series feature data using a time-series analysis model to obtain the changing trend of cognitive ability; generate a dynamic optimal adjustment strategy according to the changing trend of cognitive ability using a reinforcement learning algorithm.

[0039] The multi-modal interaction training module is further configured to: dynamically adjust the music training tasks according to the dynamic optimal adjustment strategy.

[0040] Exemplarily, the multi-modal interaction training module includes: a music training execution unit and an interaction data collection unit.

[0041] The music training execution unit is used to execute music training tasks.

[0042] The interaction data collection unit is used to collect multi-modal interaction data during the music training process.

[0043] The music training execution unit is further used to dynamically adjust the music training tasks according to the dynamic optimal adjustment strategy.

[0044] Specifically, in this embodiment, both the music training execution unit and the interaction data acquisition unit are set on a tablet computer (in other possible embodiments, a laptop computer or other computers capable of implementing the functions of the music training execution unit and the interaction data acquisition unit can also be used). The tablet computer executes music training tasks through a touch interface and a player, and collects multi-modal interaction data during the music training process through a microphone, a camera, a touch screen sensor, and a background processor.

[0045] Exemplarily, the music training tasks include: pitch identification training tasks, rhythm perception training tasks, instrument recognition training tasks, song memory training tasks, music singing-along training tasks, and music imagination training tasks.

[0046] The multi-modal interaction data includes: interaction data, behavior data, physiological data, and emotional data.

[0047] Specifically, in this embodiment, the execution process of the pitch identification training task is as follows: The tablet computer randomly plays tones of low, medium, and high pitches. The patient selects the corresponding option displayed on the touch interface according to the sound heard. In addition, the tablet computer can also play tone change sounds that gradually change from low pitch to high pitch and from high pitch to low pitch. Similarly, the patient selects the corresponding option displayed on the touch interface according to the sound heard. Additionally, if the patient performs well in this training, real or synthesized music with multiple reciprocating changes between high and low pitches can be further added. The patient selects the correct number of high-low pitch switches displayed on the touch interface according to the sound heard. The tablet computer scores according to the answering accuracy rate and reaction time, etc.

[0048] In this embodiment, the execution process of the rhythm perception training task is as follows: The tablet computer plays real or synthesized music segments with different rhythms (such as lively, soothing, etc.). The patient judges the fast or slow rhythm type of the music through the options on the touch interface. The tablet computer records the patient's answering situation and adjusts the music style and task difficulty of the subsequent training in combination with their facial expression changes, etc.

[0049] In this embodiment, the execution process of the instrument recognition training task is as follows: The tablet computer sequentially plays the sounds of various instruments such as a piano, a violin, and a flute. The patient selects the correct instrument name from the alternative answers through the touch interface. The answering accuracy rate is calculated according to the patient's selection, and their reaction time, etc. is recorded. In subsequent training, the tablet computer can add instrument options with similar timbres (such as a guitar and a ukulele) to increase the task difficulty, and dynamically adjust the track selection according to the patient's training performance.

[0050] In this embodiment, the execution process of the song memory training task is as follows: The tablet computer plays classic song clips that match the patient's age group, and then displays several alternative song names for the patient to select the correct answer. If the patient selects the wrong answer, the tablet computer provides the correct answer and replays the corresponding clip to strengthen the patient's memory, while retaining the wrong answer for subsequent retesting. During the training process, the tablet computer records the patient's answering accuracy rate and reaction time for each question, etc., and dynamically adjusts the confusion difficulty of the alternative answers to adapt to the patient's state.

[0051] In this embodiment, the execution process of the music following-singing training task is as follows: The tablet computer plays the music that the patient likes, and the patient synchronously follows and sings or hums the melody. The microphone collects the patient's voice signal in real time, and the background processor calculates the difference between its pitch and the standard pitch, and visually displays the difference between the two on the screen, so that the patient can sing the corresponding pitch more accurately and achieve a better training effect. For patients who have difficulty with this training, music with simple tunes can also be synthesized using artificial intelligence for following-singing training. The tablet computer scores the patient's following-singing effect according to the pitch matching degree, etc., and provides improvement suggestions.

[0052] In this embodiment, the execution process of the music imagination training task is as follows: The tablet computer randomly plays singing and dancing music, symphonies or movie soundtracks with a story background, and at the same time displays multiple adjectives related to music and scenes (such as "romantic", "pleasant", "exciting", or "starry sky", "ranch" and "city", etc.) on the touch interface, and the patient selects the adjective that best matches the music situation. The tablet computer evaluates the patient's understanding of music emotions based on the selection results, and at the same time monitors the patient's concentration and emotional state through camera-based facial expression recognition technology, and records their cognitive state, etc.

[0053] Specifically, in this embodiment, the interaction data in the multi-modal interaction data includes the answering reaction time and answering accuracy rate, the behavior data includes the touch trajectory and operation frequency, the physiological data includes the pitch difference and rhythm synchronization deviation, and the emotion data includes the facial expression data and voice emotion data.

[0054] Exemplarily, the intelligent evaluation module includes: a feature extraction unit, a trend analysis unit, and a strategy adjustment unit.

[0055] The feature extraction unit is used to extract features from the multi-modal interaction data to obtain time-series feature data.

[0056] The trend analysis unit is used to perform trend analysis on the time-series feature data using a time series analysis model to obtain the changing trend of cognitive ability.

[0057] The strategy adjustment unit is used to generate a dynamic optimal adjustment strategy according to the changing trend of cognitive ability using a reinforcement learning algorithm.

[0058] Specifically, in this embodiment, the intelligent evaluation module is set on the cloud server, and the cloud server is communicatively connected to the tablet computer. In other possible embodiments, the intelligent evaluation module can also be set on the local server or directly on the tablet computer.

[0059] Exemplarily, before extracting features from the multimodal interaction data, the feature extraction unit performs time synchronization alignment processing, smoothing processing, and normalization processing on the multimodal interaction data in advance.

[0060] Specifically, in this embodiment, the time synchronization alignment processing improves the temporal correspondence of various multimodal interaction data; the smoothing processing can smooth short-term fluctuations and reduce errors; the normalization processing can eliminate the dimensional differences between various multimodal interaction data.

[0061] Exemplarily, the feature extraction uses the principal component analysis algorithm.

[0062] Specifically, by mapping the multimodal interaction data from a high-dimensional space to a low-dimensional space through the principal component analysis algorithm, key indicators of the multimodal interaction data (such as the standard deviation of reaction time, emotional entropy value, pitch stability, etc.) can be extracted to obtain time-series feature data.

[0063] Exemplarily, before performing trend analysis on the time-series feature data using a time series analysis model, the trend analysis unit performs first-order difference processing on the time-series feature data in advance.

[0064] Specifically, the first-order difference processing can improve the stationarity of the time-series feature data, lay a foundation for subsequent trend analysis, and reduce errors.

[0065] Exemplarily, the time series analysis model uses an autoregressive integrated moving average model combined with a vector autoregressive model.

[0066] Specifically, the modeling process of the autoregressive integrated moving average model (ARIMA) is as follows: For a single time-series feature data (such as the standard deviation of reaction time), first perform a stationarity test. If the data is non-stationary, eliminate the trend through first-order difference processing. Subsequently, determine the order (p, d, q) of the ARIMA model according to the autocorrelation function (ACF) and partial autocorrelation function (PACF), where p is the order of the autoregressive term, d is the number of difference times, and q is the order of the moving average term. Optimize the model parameters by minimizing the AIC criterion to complete univariate time-series prediction.

[0067] Specifically, the modeling process of the Vector Auto-Regression (VAR) model is as follows: Multimodal time-series feature data (such as pitch stability, emotional entropy value, etc.) is used as multivariate inputs. The long-term equilibrium relationship between variables is confirmed through cointegration tests, and a VAR model is constructed to capture the dynamic correlation between features. The optimal lag order is determined using lag order selection criteria (such as LR, AIC), and the model coefficients are estimated by the least squares method.

[0068] Specifically, the model combination and trend analysis process of the Auto-Regressive Integrated Moving Average (ARIMA) model and the Vector Auto-Regression (VAR) model are as follows: The univariate prediction results of ARIMA are used as one of the inputs of the VAR model. Combining the multivariate interaction effects, the changing trend of comprehensive cognitive ability is output. For example, the trend of pitch stability predicted by ARIMA is combined with the dynamic impact of the emotional entropy value analyzed by VAR on reaction time, and finally a multi-dimensional trend curve of the patient's cognitive ability is generated. The accuracy of the model is verified through residual analysis and rolling prediction to ensure the reliability of trend analysis.

[0069] Exemplarily, the reinforcement learning algorithm adopts the Q-learning algorithm, taking different changing trends of cognitive ability as the states of the Q-learning algorithm and different dynamic adjustment strategies as the actions of the Q-learning algorithm.

[0070] The trend analysis unit generates a dynamic optimal adjustment strategy by designing the reward function of the Q-learning algorithm and using the Q-learning algorithm to select the optimal action according to the current changing trend of cognitive ability.

[0071] Specifically, in this embodiment, the reinforcement learning algorithm adopting the Q-learning algorithm includes the following steps:

[0072] State space definition: The changing trend of cognitive ability output by the time series analysis module is discretized into multiple states, including: State S1: Significant improvement in cognitive ability (daily growth rate of correct rate ≥ 2%, decrease in emotional entropy value ≥ 10%); State S2: Slight fluctuation in cognitive ability (-1% < daily growth rate of correct rate < 1%, fluctuation in emotional entropy value < 5%); State S3: Slow decline in cognitive ability (daily growth rate of correct rate ≤ -2%, increase in emotional entropy value ≥ 15%); State S4: High-risk warning state (continuous decline in training correct rate for 3 times and abnormal physiological data).

[0073] Action set design: According to the state definition, the action set of the dynamic adjustment strategy is designed, including: Action A1: Increase task difficulty (such as increasing the similarity of instrument timbres or shortening the reaction time limit); Action A2: Maintain the current task parameters; Action A3: Decrease difficulty (switch to a simple rhythm or extend the reaction time); Action A4: Cross-task switching (for example, switch from tone identification to music imagination training); Action A5: Trigger a guiding prompt (such as "Fatigue detected, it is recommended to pause for 1 minute").

[0074] Quantification of Reward Letters: The reward value is calculated in real time based on multimodal interaction data, and the formula is:

[0075] R = w1 * Δ 正确率 + w2 * (-Δ 反应时间 ) + w3 * (-Δ 情绪熵 ) + R penalty .

[0076] Among them, Δ represents the change amount compared with the previous training, and R penalty is the penalty term. Reference values of weight coefficients: w1 = 0.5, w2 = 0.3, w3 = 0.2.

[0077] Q-table Update and Policy Generation: Initialization: The initial Q value of all state-action pairs is set to 0; Exploration and Exploitation: The ε-greedy policy (ε = 0.3) is adopted, and a new action is randomly explored with a 30% probability; Q-value iteration formula:

[0078] Q(s,a) ← Q(s,a) + α[R + γ * max a′ Q(s′,a′) - Q(s,a)].

[0079] Among them, Q(s,a) represents the current Q value (i.e., the expected cumulative reward) of performing action a in state s; R represents the immediate reward obtained after performing action a; s' is the next state, that is, the new cognitive ability state that the patient enters after performing action a; a' is the next action, that is, all candidate actions that the system may select in state s'; the discount factor γ (0 ≤ γ ≤ 1) is used to balance the importance of the current reward and future rewards (usually set to 0.9); the learning rate α (0 ≤ α ≤ 1) controls the amplitude of Q-value update. Example of parameter settings: learning rate α = 0.1, discount factor γ = 0.9.

[0080] Convergence Judgment: Check the change of the Q-table after every 15 trainings. If the fluctuation of the maximum Q value < 1%, it is determined that the policy is stable.

[0081] In this embodiment, after each music cognitive training, the autoregressive integrated moving average model, vector autoregressive model, and Q-learning algorithm are updated and trained again, which helps to improve the effect of the next music cognitive training.

[0082] In a possible embodiment, the intelligent evaluation module generates a detailed patient report based on multimodal interaction data, cognitive ability change trends, and dynamic optimal adjustment strategies, combined with internationally recognized cognitive assessment scales, providing scientific efficacy evaluation and diagnostic support for the medical team to help optimize treatment strategies.

[0083] In a possible embodiment, the interactive music cognitive training system of the present application further includes an adjustment guidance unit, which is used to display the adjustment logic through a touch interface or a player when dynamically adjusting a music training task according to a dynamic optimal adjustment strategy. For example, in a music following training task, when the difficulty is reduced, a prompt is given: "It is detected that your concentration has decreased, and the rhythm has been switched to a soothing one"; when the difficulty is increased, a prompt is given: "The rhythm will be a bit faster next. Are you ready?" The prompts through the touch interface or the player can avoid abrupt changes from affecting the experience.

[0084] In a possible embodiment, the interactive music cognitive training system of the present application further includes a trend warning unit. For example, when the trend analysis unit performs trend analysis on time series feature data and the obtained change trend of cognitive ability shows that the decline slope of the correct rate exceeds 5% in the next 30 days, an alarm is immediately triggered. According to the alarm, the training content can be adjusted: increasing the frequency of song memory tasks; or generating a report prompt: suggesting that the medical team strengthen memory intervention measures, etc.

[0085] In a possible embodiment, in an instrument recognition training task, if the patient's correct rate exceeds 90% for three consecutive times, the system automatically increases the similarity of interference items (such as replacing "violin" with "viola"), and the trend analysis unit synchronously monitors the change trend of the adjusted cognitive ability (such as the stability of the correct rate, etc.).

[0086] The embodiment of the present application combines a time series analysis model and a reinforcement learning algorithm to quantify the change trend of cognitive ability based on multi-modal interaction data, improve the interactivity and reliability of music training, and provide a scientific basis for long-term efficacy evaluation and treatment strategy optimization.

[0087] By performing time synchronization alignment processing, smoothing processing, and standardization processing on multi-modal interaction data, it is convenient for subsequent feature extraction.

[0088] The proposed feature extraction uses a principal component analysis algorithm, which improves the accuracy of feature extraction for multi-modal interaction data.

[0089] By performing first-order difference processing on time series feature data, the data stationarity is improved.

[0090] The proposed time series analysis model uses an autoregressive integrated moving average model and combines a vector autoregressive model to capture the dynamic correlation between the features of multi-modal interaction data, improving the accuracy of the change trend of cognitive ability.

[0091] The proposed use of the Q-learning algorithm selects the optimal action according to the current change trend of cognitive ability, generates a dynamic optimal adjustment strategy, ensures that the music training content is always adapted to the patient's cognitive state, and improves the music training effect.

[0092] Although the preferred embodiments of the present application have been described, additional changes and modifications can be made to these embodiments by those skilled in the art once they learn of the basic creative concept. Therefore, the appended claims are intended to be construed to include the preferred embodiments as well as all changes and modifications that fall within the scope of the present application.

[0093] Obviously, those skilled in the art can make various changes and modifications to the present application without departing from the spirit and scope of the present application. Thus, if these modifications and variations of the present application fall within the scope of the claims of the present application and their equivalent technologies, the present application also intends to include these modifications and variations.

Claims

1. An interactive music cognitive training system for early-stage Alzheimer's patients, characterized in that, Including: A multimodal interaction training module and an intelligent evaluation module; The multimodal interaction training module is configured to: execute a music training task; Collect multimodal interaction data during the music training process; The intelligent evaluation module is configured to: extract features from the multimodal interaction data to obtain time-series feature data; Use a time series analysis model to perform trend analysis on the time-series feature data to obtain the changing trend of cognitive ability; Use a reinforcement learning algorithm to generate a dynamic optimal adjustment strategy according to the changing trend of cognitive ability; The multimodal interaction training module is further configured to: dynamically adjust the music training task according to the dynamic optimal adjustment strategy.

2. The interactive music cognitive training system for early-stage Alzheimer's patients according to claim 1, characterized in that The multimodal interaction training module includes: a music training execution unit and an interaction data collection unit; The music training execution unit is used to execute the music training task; The interaction data collection unit is used to collect multimodal interaction data during the music training process; The music training execution unit is also used to dynamically adjust the music training task according to the dynamic optimal adjustment strategy.

3. The interactive music cognitive training system for early-stage Alzheimer's patients according to claim 2, wherein The music training tasks include: pitch identification training task, rhythm perception training task, instrument recognition training task, song memory training task, music singing-along training task, music imagination training task; The multimodal interaction data includes: interaction data, behavior data, physiological data, and emotion data.

4. The interactive music cognitive training system for early Alzheimer's patients according to claim 1, characterized in that, The intelligent evaluation module includes: a feature extraction unit, a trend analysis unit, and a strategy adjustment unit; The feature extraction unit is used to extract features from the multimodal interaction data to obtain time-series feature data; The trend analysis unit is used to perform trend analysis on the time-series feature data using a time series analysis model to obtain the changing trend of cognitive ability; The strategy adjustment unit is used to generate a dynamic optimal adjustment strategy using a reinforcement learning algorithm according to the changing trend of cognitive ability.

5. The interactive music cognitive training system for early Alzheimer's patients according to claim 4, characterized in that, Before extracting features from the multimodal interaction data, the feature extraction unit pre-processes the multimodal interaction data by time synchronization alignment, smoothing, and standardization.

6. The interactive music cognitive training system for early-stage Alzheimer's patients according to claim 4, characterized in that, The feature extraction uses the principal component analysis algorithm.

7. The interactive music cognitive training system for early-stage Alzheimer's patients according to claim 4, characterized in that, The time series analysis model uses the autoregressive integrated moving average model combined with the vector autoregressive model.

8. The interactive music cognitive training system for early-stage Alzheimer's patients according to claim 4, wherein The reinforcement learning algorithm uses the Q-learning algorithm, taking different changing trends of cognitive ability as the states of the Q-learning algorithm and different dynamic adjustment strategies as the actions of the Q-learning algorithm; The trend analysis unit generates a dynamic optimal adjustment strategy by designing the reward function of the Q-learning algorithm and using the Q-learning algorithm to select the optimal action according to the current changing trend of cognitive ability.

Images