Remote postoperative cognitive impairment monitoring method, storage medium and program product

Abstract

The invention relates to a remote postoperative cognitive impairment monitoring method, a storage medium and a program product. The remote postoperative cognitive impairment monitoring method comprises the following steps: acquiring three-dimensional space trajectory monitoring data of an arm of a patient, and acquiring voice data of the patient; abnormal value cleaning is carried out on the arm track data, and adaptive noise reduction filtering is carried out on the voice data; extracting motion amplitude, swing frequency and time sequence regularity features from the preprocessed arm track data, extracting voice features from the voice data after adaptive noise reduction filtering, and generating multi-modal joint features through weighted fusion or vector splicing of the arm track features and the voice features after dimension reduction processing; inputting into a pre-trained time sequence analysis model, and outputting a quantitative probability value of postoperative cognitive impairment of the patient in the monitoring process in real time by comparing feature combinations of normal and cognitive impairment modes. Compared with the prior art, remote, real-time and non-invasive postoperative cognitive impairment monitoring is realized, and the monitoring accuracy and efficiency are improved.

Description

Technical Field

[0001] The present invention relates to the field of medical monitoring technology, and in particular to a remote postoperative cognitive dysfunction monitoring method, storage medium, and program product. Background Art

[0002] Postoperative delirium (postoperative cognitive dysfunction) often occurs within one week after surgery, with a peak incidence 24-72 hours after surgery. More than half of patients with postoperative delirium experience high activity or mixed delirium, and episodes fluctuate. Currently, there is no equipment capable of remotely monitoring and quantifying delirium episodes. Existing methods for monitoring postoperative cognitive dysfunction rely primarily on subjective assessments by medical staff, which are subject to certain subjectivity and uncertainty and require significant manpower and material resources. Furthermore, existing methods lack remote, real-time, and non-invasive monitoring methods, making it impossible to promptly detect and intervene in patients' cognitive dysfunction.

[0003] Several patent applications have been filed regarding the problem of quantifying the probability of postoperative cognitive dysfunction in patients during real-time output monitoring. For example, CN118136232A discloses a method and system for early detection of Parkinson's disease based on multimodal deep learning. This patent collects audio and video data of patients performing speech tasks, extracts facial image sequences from video frames and audio mel-spectrograms, and uses the local feature extraction module and speech feature extraction module in the multimodal deep learning model to extract visual and speech features, respectively. These features are then fused using a cross-attention mechanism to output classification results for early detection of Parkinson's disease. However, this patent still suffers from the small dataset size and the need for improvement in the multimodal feature extraction and fusion methods. CN114724224A discloses a multimodal emotion recognition method for medical care robots. This patent collects video and audio information from patients, extracts emotion features from facial expressions, actions, speech, and text, and fuses these features using a mutual attention mechanism. Finally, a graph convolutional neural network is used to extract contextual emotion features, thereby outputting emotion labeling results. However, this patent still has problems with multimodal emotional information collection, multimodal feature extraction and fusion methods that need to be improved.

[0004] The existing technology has the following shortcomings and needs to be improved:

[0005] 1. There is a lack of remote, real-time, and non-invasive methods for monitoring postoperative cognitive dysfunction. Existing methods mainly rely on the subjective assessment of medical staff, which is subject to certain subjectivity and uncertainty and requires a lot of manpower and material resources.

[0006] 2. The existing multimodal data collection method is single and cannot fully reflect the patient's cognitive function status.

[0007] 3. The multimodal feature extraction and fusion methods in existing technologies need to be improved and cannot fully explore the potential information of multimodal data.

[0008] 4. The scale of multimodal datasets in existing technologies is small, making it difficult to train high-performance deep learning models.

[0009] 5. The existing technology lacks a cognitive dysfunction detection model based on time series analysis, which cannot effectively capture the dynamic changes in patients' cognitive functions. Summary of the Invention

[0010] The purpose of the present invention is to provide a remote postoperative cognitive dysfunction monitoring method, storage medium, and program product to overcome the defects of the above-mentioned prior art. By collecting the patient's three-dimensional arm trajectory data and voice data, the corresponding features are extracted, and these features are weighted fused or vector spliced ​​to generate multimodal joint features. The multimodal joint features are then input into a pre-trained time series analysis model, and the quantitative probability value of the patient's postoperative cognitive dysfunction is output in real time. This method overcomes the defects of the prior art, realizes remote, real-time, non-invasive postoperative cognitive dysfunction monitoring, and improves the accuracy and efficiency of monitoring.

[0011] The purpose of the present invention can be achieved by the following technical solutions:

[0012] A first aspect of the present invention provides a method for remotely monitoring postoperative cognitive dysfunction, comprising the following steps:

[0013] S1: Acquire the patient's arm three-dimensional spatial trajectory monitoring data within a preset time period, including angular velocity, angular acceleration, and continuous displacement parameters, and simultaneously collect the patient's voice data, including the characteristics of voice content, tone, speed, and volume;

[0014] S2: Clean outliers on the arm trajectory data and perform adaptive noise reduction filtering on the voice data;

[0015] S3: Extract motion amplitude, swing frequency, and temporal regularity features from the preprocessed arm trajectory data and perform feature dimensionality reduction. Simultaneously, extract speech features from the adaptively denoised and filtered speech data. The speech features include acoustic features, semantic coherence features, and emotional fluctuation features. The arm trajectory features after dimensionality reduction and speech features are combined with the speech features through weighted fusion or vector concatenation to generate multimodal joint features.

[0016] S4: The multimodal joint features are input into a pre-trained time series analysis model, and by comparing the feature combinations of normal and cognitive impairment patterns, a quantitative probability value of the patient having postoperative cognitive dysfunction during the monitoring process is output in real time.

[0017] Furthermore, in S4, the time series analysis model is trained based on a labeled multimodal dataset.

[0018] Furthermore, in S4, the process of training the time series analysis model includes:

[0019] Collect and annotate multimodal data from patients with postoperative cognitive dysfunction and normal patients, including arm trajectory data and speech data;

[0020] Divide the collected data set into training set, validation set, and test set;

[0021] Choose an appropriate deep learning model architecture, such as Long Short-Term Memory (LSTM) or Gated Recurrent Unit (GRU);

[0022] Use the training set to train the model and adjust the model parameters through the back-propagation algorithm to minimize the loss function;

[0023] During the training process, the validation set is used to verify the model and adjust the hyperparameters to optimize the model performance;

[0024] Finally, the test set is used to evaluate the accuracy and generalization ability of the trained model.

[0025] Furthermore, in S2, the outlier cleaning includes identifying and correcting or eliminating outliers in the arm trajectory data by setting a threshold or a method based on statistical distribution.

[0026] Furthermore, in S2, the adaptive noise reduction filter adopts an adaptive filtering algorithm to dynamically adjust filter parameters according to the characteristics of the speech signal and background noise to remove the background noise.

[0027] Furthermore, in S3, the feature dimensionality reduction processing is: using the principal component analysis (PCA) algorithm, by calculating the covariance matrix and eigenvector of the features, the high-dimensional arm trajectory features are projected into a low-dimensional space while retaining the main variance information of the data.

[0028] Furthermore, in S3, the weighted fusion process is: assigning different weights according to the importance of the arm trajectory features and the speech features to the judgment of postoperative cognitive dysfunction, and the weight values ​​are determined through model training or expert experience.

[0029] Furthermore, in S3, the vector splicing process includes:

[0030] The arm trajectory feature vector after dimensionality reduction and the speech feature vector are connected end to end in a preset order to form a joint feature vector. During the splicing process, different feature dimensions are normalized or standardized according to actual needs to ensure the consistency of each feature after fusion.

[0031] A second aspect of the present invention provides a storage medium comprising computer-executable instructions, which, when executed by a computer processor, is used to perform the above-mentioned remote postoperative cognitive dysfunction monitoring method.

[0032] A third aspect of the present invention provides a computer program product, comprising a computer program, which, when executed by a processor, implements the above-mentioned remote postoperative cognitive dysfunction monitoring method.

[0033] Compared with the prior art, the present invention has the following beneficial effects:

[0034] 1. It enables remote, real-time, and non-invasive monitoring of postoperative cognitive dysfunction, overcoming the shortcomings of existing technologies that rely on subjective assessments by medical staff, are subjective and uncertain, and consume a lot of manpower and material resources;

[0035] 2. By collecting the patient's three-dimensional arm trajectory data and voice data, it can fully reflect the patient's cognitive function status, making up for the shortcomings of single-modality data collection in existing technologies;

[0036] 3. Using advanced feature extraction and multimodal fusion methods, we fully tapped the potential information of multimodal data and improved the accuracy of cognitive dysfunction detection;

[0037] 4. Training a time series analysis model based on annotated multimodal datasets effectively captures the dynamic changes in patients' cognitive functions and improves the model's generalization capabilities;

[0038] 5. Real-time output of the quantitative probability value of the patient's postoperative cognitive dysfunction provides an objective basis for clinical diagnosis and intervention, and improves the efficiency and reliability of monitoring. BRIEF DESCRIPTION OF THE DRAWINGS

[0039] Figure 1 Schematic diagram of the process of remote postoperative cognitive dysfunction monitoring method of the present invention. DETAILED DESCRIPTION

[0040] The present invention will be described in detail below with reference to the accompanying drawings and specific embodiments. Component models, material names, connection structures, control methods, algorithms, and other features not explicitly described in this technical solution are considered common technical features disclosed in the prior art.

[0041] Example 1

[0042] In this embodiment, a remote postoperative cognitive dysfunction monitoring method is provided. Figure 1 , including the following steps:

[0043] Step 1: Collect the patient's arm three-dimensional space trajectory data and voice data.

[0044] Step 101: Use an inertial measurement unit-based sensor device (such as a wristband with a gyroscope) to collect the patient's arm three-dimensional spatial trajectory data, including angular velocity range of 0.1-10 rad / s and angular acceleration range of 0.5-20 rad / s. 2 , the continuous displacement parameter range is 0.01~1m;

[0045] Step 102: Use a microphone or voice acquisition device to synchronously collect patient voice data, including voice content, intonation frequency range of 80-500 Hz, speech speed range of 100-200 words / minute, and volume range of 40-90 dB;

[0046] Step 103: align and synchronize the timestamps of the collected arm trajectory data and voice data to ensure that the timestamps of the two data are consistent.

[0047] Step 2: Clean the arm trajectory data for outliers and perform adaptive noise reduction filtering on the voice data.

[0048] Step 201: Perform outlier detection on the arm trajectory data, setting the angular velocity threshold to 8 rad / s and the angular acceleration threshold to 15 rad / s. 2 ,The displacement threshold is 0.8m, and data points exceeding the threshold are identified as outliers and eliminated;

[0049] Step 202: Perform adaptive noise reduction filtering on the speech data, using a minimum mean square error adaptive filtering algorithm to dynamically adjust the filter coefficients according to the characteristics of the speech signal and background noise to remove the background noise.

[0050] Step 3: Extract the motion amplitude, swing frequency, and temporal regularity features from the preprocessed arm trajectory data and perform feature dimensionality reduction. At the same time, extract the acoustic features, semantic coherence features, and emotional fluctuation features from the speech data after adaptive noise reduction filtering. The arm trajectory features after dimensionality reduction and the speech features are weightedly fused to generate multimodal joint features.

[0051] Step 301: extract the motion amplitude range of 0.05-0.8m, the swing frequency range of 0.2-2Hz, and the temporal regularity eigenvalue range of 0.1-0.9 from the pre-processed arm trajectory data, and use the principal component analysis algorithm to perform feature dimensionality reduction processing, retaining more than 90% of the variance information;

[0052] Step 302: Extracting acoustic features including Mel-frequency cepstral coefficients and speech energy, semantic coherence features including word vectors and sentence vectors, and emotional fluctuation features including pitch changes and speech rate changes from the speech data after adaptive noise reduction filtering.

[0053] Step 303: assign a weight ratio of 3:7 based on the importance of the arm trajectory feature and the voice feature in judging postoperative cognitive dysfunction, and generate a multimodal joint feature vector through weighted fusion.

[0054] Step 4: Input the multimodal joint features into the pre-trained time series analysis model, and output the quantitative probability value of the patient having postoperative cognitive dysfunction in the monitoring process in real time by comparing the feature combination of normal and cognitive impairment modes.

[0055] Step 401: Collect a multimodal dataset of 1,000 patients with postoperative cognitive dysfunction and normal patients, including arm trajectory data and speech data, and have experts annotate the data.

[0056] Step 402: Divide the collected data set into a training set, a validation set, and a test set in a ratio of 7:2:1;

[0057] Step 403: Select a bidirectional long short-term memory network as the time series analysis model architecture;

[0058] Step 404: Use the training set to train the model, adopt the Adam optimization algorithm, use cross entropy as the loss function, and adjust the model parameters through the back propagation algorithm;

[0059] Step 405: During the training process, the model performance is evaluated on the validation set every 10 epochs. When the loss function value on the validation set no longer decreases, the training is stopped and the model parameters are saved.

[0060] Step 406: Use the test set to evaluate the accuracy of the trained model. The accuracy of the model on the test set reaches 92%;

[0061] Step 407: Input the multimodal joint features into the trained time series analysis model, and output in real time the quantitative probability value of the patient having postoperative cognitive dysfunction during the monitoring process by comparing the feature combination of normal and cognitive impairment modes.

[0062] Example 2

[0063] In this embodiment, a remote postoperative cognitive dysfunction monitoring method includes the following steps:

[0064] Step 1: Collect the patient's arm three-dimensional space trajectory data and voice data.

[0065] Step 101: Use a wearable device based on an inertial measurement unit to collect three-dimensional spatial trajectory data of the patient's arm, including angular velocity range of 0.2 to 8 rad / s and angular acceleration range of 1 to 15 rad / s. 2 , the continuous displacement parameter range is 0.02~0.8m;

[0066] Step 102: Using a microphone on a smartphone or tablet computer, synchronously collect the patient's voice data, including voice content, intonation frequency range of 100-400 Hz, speech speed range of 120-180 words / minute, and volume range of 50-80 dB;

[0067] Step 103: The collected arm trajectory data and voice data are transmitted to the server via Bluetooth or WiFi, and timestamp alignment and synchronization are performed.

[0068] Step 2: Clean the arm trajectory data for outliers and perform adaptive noise reduction filtering on the voice data.

[0069] Step 201: Perform outlier detection on the arm trajectory data. Based on the Gaussian distribution assumption, data points that deviate from the mean by more than 3 standard deviations are identified as outliers and corrected.

[0070] Step 202: Adaptively perform noise reduction filtering on the speech data, using a wavelet transform domain adaptive filtering algorithm to dynamically adjust the filtering threshold according to the wavelet coefficient characteristics of the speech signal and background noise to remove the background noise.

[0071] Step 3: Extract the motion amplitude, swing frequency, and timing regularity features from the preprocessed arm trajectory data, and perform feature dimensionality reduction. At the same time, extract the acoustic features, semantic coherence features, and emotional fluctuation features from the speech data after adaptive noise reduction filtering. The arm trajectory features after dimensionality reduction and the speech features are vector-concatenated to generate multimodal joint features.

[0072] Step 301: extract the motion amplitude range of 0.1 to 0.6 m, the swing frequency range of 0.3 to 1.8 Hz, and the temporal regularity eigenvalue range of 0.2 to 0.8 from the pre-processed arm trajectory data, and use the principal component analysis algorithm to perform feature dimensionality reduction processing, retaining more than 95% of the variance information;

[0073] Step 302: extracting acoustic features including linear prediction coefficients and fundamental frequencies, semantic coherence features including topic models and semantic similarity, and emotional fluctuation features including energy changes and speech rate changes from the speech data after adaptive noise reduction filtering.

[0074] Step 303: Connect the arm trajectory feature vector and the speech feature vector after dimensionality reduction processing end to end in a preset order to form a joint feature vector, and perform normalization processing on different feature dimensions.

[0075] Step 4: Input the multimodal joint features into the pre-trained time series analysis model, and output the quantitative probability value of the patient having postoperative cognitive dysfunction in the monitoring process in real time by comparing the feature combination of normal and cognitive impairment modes.

[0076] Step 401: Collect a multimodal dataset of 1,500 patients with postoperative cognitive dysfunction and normal patients, including arm trajectory data and speech data, and have doctors and language experts annotate the data.

[0077] Step 402: Divide the collected data set into a training set, a validation set, and a test set in a ratio of 8:1:1;

[0078] Step 403: Select a gated recurrent unit as the timing analysis model architecture;

[0079] Step 404: Use the training set to train the model, adopt the RMSprop optimization algorithm, use cross entropy as the loss function, and adjust the model parameters through the back propagation algorithm;

[0080] Step 405: During the training process, the model performance is evaluated on the validation set every 20 epochs. When the loss function value on the validation set does not decrease for 5 consecutive epochs, the training is stopped and the model parameters are saved.

[0081] Step 406: Use the test set to evaluate the accuracy of the trained model. The accuracy of the model on the test set reaches 94%;

[0082] Step 407: Input the multimodal joint features into the trained time series analysis model, and output in real time the quantitative probability value of the patient having postoperative cognitive dysfunction during the monitoring process by comparing the feature combination of normal and cognitive impairment modes.

[0083] Example 3

[0084] In this embodiment, a remote postoperative cognitive dysfunction monitoring method includes the following steps:

[0085] Step 1: Collect the patient's arm three-dimensional space trajectory data and voice data.

[0086] Step 101: Use a sensor device based on an inertial measurement unit and computer vision to collect the three-dimensional trajectory data of the patient's arm, including the angular velocity range of 0.3 to 6 rad / s and the angular acceleration range of 2 to 12 rad / s. 2 , the continuous displacement parameter range is 0.05~0.5m;

[0087] Step 102: Use professional recording equipment to synchronously collect patient voice data, including voice content, intonation frequency range of 120-350 Hz, speech speed range of 130-170 words / minute, and volume range of 60-90 dB;

[0088] Step 103: Transmit the collected arm trajectory data and voice data to the server via a wired connection, and perform timestamp alignment and synchronization processing.

[0089] Step 2: Clean the arm trajectory data for outliers and perform adaptive noise reduction filtering on the voice data.

[0090] Step 201: Perform outlier detection on the arm trajectory data, set the upper threshold of angular velocity to 5 rad / s and the upper threshold of angular acceleration to 10 rad / s 2 ,The upper displacement threshold is 0.4m, and data points exceeding the threshold are identified as outliers and eliminated;

[0091] Step 202: Adaptively perform noise reduction filtering on the speech data, using a wavelet packet decomposition adaptive filtering algorithm, and dynamically adjust the filtering threshold according to the wavelet packet coefficient characteristics of the speech signal and background noise to remove the background noise.

[0092] Step 3: Extract the motion amplitude, swing frequency, and temporal regularity features from the preprocessed arm trajectory data and perform feature dimensionality reduction. At the same time, extract the acoustic features, semantic coherence features, and emotional fluctuation features from the speech data after adaptive noise reduction filtering. The arm trajectory features after dimensionality reduction and the speech features are weightedly fused to generate multimodal joint features.

[0093] Step 301: extract the motion amplitude range of 0.08-0.4m, the swing frequency range of 0.4-1.5Hz, and the temporal regularity eigenvalue range of 0.3-0.7 from the pre-processed arm trajectory data, and use the principal component analysis algorithm to perform feature dimensionality reduction processing, retaining more than 90% of the variance information;

[0094] Step 302: extracting acoustic features including mel-frequency cepstral coefficients and formants, semantic coherence features including word embedding vectors and sentence vectors, and emotional fluctuation features including pitch changes and energy changes from the speech data after adaptive noise reduction filtering;

[0095] Step 303: assign a weight ratio of 4:6 based on the importance of the arm trajectory feature and the voice feature in judging postoperative cognitive dysfunction, and generate a multimodal joint feature vector through weighted fusion.

[0096] Step 4: Input the multimodal joint features into the pre-trained time series analysis model, and output the quantitative probability value of the patient having postoperative cognitive dysfunction in the monitoring process in real time by comparing the feature combination of normal and cognitive impairment modes.

[0097] Step 401: Collect a multimodal dataset of 2,000 patients with postoperative cognitive dysfunction and normal patients, including arm trajectory data and speech data. The dataset is annotated by doctors, language experts, and behavior analysts.

[0098] Step 402: Divide the collected data set into a training set, a validation set, and a test set in a ratio of 6:2:2;

[0099] Step 403: Select a bidirectional gated recurrent unit based on an attention mechanism as the temporal analysis model architecture;

[0100] Step 404: Use the training set to train the model, adopt the Adam optimization algorithm, use cross entropy as the loss function, and adjust the model parameters through the back propagation algorithm;

[0101] Step 405: During the training process, the model performance is evaluated on the validation set every 15 epochs. When the loss function value on the validation set does not decrease for three consecutive epochs, the training is stopped and the model parameters are saved.

[0102] Step 406: Use the test set to evaluate the accuracy of the trained model. The accuracy of the model on the test set reaches 96%;

[0103] Step 407: Input the multimodal joint features into the trained time series analysis model, and output in real time the quantitative probability value of the patient having postoperative cognitive dysfunction during the monitoring process by comparing the feature combination of normal and cognitive impairment modes.

[0104] Example 4

[0105] In this embodiment, a storage medium containing computer-executable instructions is provided, which, when executed by a computer processor, is used to execute a remote postoperative cognitive dysfunction monitoring method. The storage medium may be a flash memory, a hard disk, or an optical disk, and the computer processor may be a processor in a device such as a personal computer, a server, or a mobile terminal. When the processor executes the instructions in the storage medium, monitoring will be performed according to the following steps: first, a wristband with a gyroscope and a voice analysis device are used to collect the patient's arm three-dimensional space trajectory data and voice data respectively; then, the arm trajectory data is cleaned of outliers, and the voice data is adaptively denoised and filtered; then, arm trajectory features and voice features are extracted from the preprocessed data, and feature dimensionality reduction processing is performed; finally, the reduced dimensionality features are generated by weighted fusion or vector splicing to generate multimodal joint features, and are input into a pre-trained time series analysis model to output the quantitative probability value of the patient having postoperative cognitive dysfunction in real time.

[0106] During the specific implementation process, the data collection stage needs to ensure the accuracy and completeness of the arm trajectory data and voice data in order to provide a reliable basis for subsequent analysis. In the data preprocessing stage, outlier cleaning and adaptive noise reduction filtering are key steps that can effectively improve data quality. In the feature extraction and fusion stage, the dimensionality reduction of features through algorithms such as principal component analysis, and the generation of multimodal joint features using methods such as weighted fusion or vector splicing are the key to achieving multimodal data fusion. In the model training and probability evaluation stage, a suitable deep learning model architecture is selected, and training and optimization are carried out through a large number of annotated multimodal data sets to ensure the accuracy and generalization ability of the model. Ultimately, this embodiment can provide medical staff with timely and quantitative postoperative cognitive dysfunction monitoring results, which has significant application advantages and broad development prospects.

[0107] Example 5

[0108] In this embodiment, a computer program product is provided, including a computer program that, when executed by a processor, implements the above-mentioned remote postoperative cognitive dysfunction monitoring method. The computer program product can be installed on various computer devices or mobile terminals, and can realize remote monitoring of cognitive dysfunction in postoperative patients by calling the gyroscope data and voice analysis device of the bracelet. During execution, the program first obtains the three-dimensional spatial trajectory data and voice data of the patient's arm, and then performs data preprocessing, including outlier cleaning and adaptive noise reduction filtering, to improve the accuracy and reliability of the data. Next, the program extracts key features from the processed data, such as the amplitude, frequency and regularity of the arm movement, as well as the acoustic, semantic coherence and emotional fluctuation features of the voice, and generates multimodal joint features through feature dimensionality reduction and fusion. Finally, these joint features are input into the pre-trained time series analysis model, and the quantitative probability value of the patient's postoperative cognitive dysfunction is output in real time, providing medical staff with timely and objective evaluation basis.

[0109] This computer program product offers significant advantages and flexibility in practical applications. It can adapt to diverse medical environments and equipment configurations, enabling deployment and use on any computer, whether a server system in a large hospital or a regular computer in a primary care unit, as long as it possesses basic computing and data transmission capabilities. Furthermore, the program's modular design allows medical staff to flexibly select data acquisition equipment and analysis parameters based on specific needs and patient conditions, improving the targetedness and effectiveness of monitoring. In this way, this computer program product not only provides advanced technical means for remote monitoring of postoperative cognitive dysfunction but also promotes the rational allocation of medical resources and improves the quality of medical services.

[0110] The above description of the embodiments is intended to facilitate understanding and use of the invention by those skilled in the art. It will be apparent that those skilled in the art can readily make various modifications to these embodiments and apply the general principles described herein to other embodiments without requiring inventive effort. Therefore, the present invention is not limited to the above-described embodiments. Improvements and modifications made by those skilled in the art based on the disclosure of the present invention, without departing from the scope of the present invention, should be within the scope of protection of the present invention.

Claims

1. A remote postoperative cognitive dysfunction monitoring method, characterized in that: The following steps are involved: S1: Acquire the patient's arm three-dimensional spatial trajectory monitoring data within a preset time period, including angular velocity, angular acceleration, and continuous displacement parameters, and simultaneously collect the patient's voice data, including the characteristics of voice content, tone, speed, and volume; S2: Clean outliers on the arm trajectory data and perform adaptive noise reduction filtering on the voice data; S3: Extract motion amplitude, swing frequency, and temporal regularity features from the preprocessed arm trajectory data and perform feature dimensionality reduction. Simultaneously, extract speech features from the adaptively denoised and filtered speech data. The speech features include acoustic features, semantic coherence features, and emotional fluctuation features. The arm trajectory features after dimensionality reduction and speech features are combined with the speech features through weighted fusion or vector concatenation to generate multimodal joint features. S4: The multimodal joint features are input into a pre-trained time series analysis model, and by comparing the feature combinations of normal and cognitive impairment modes, a quantitative probability value of the patient having postoperative cognitive dysfunction during the monitoring process is output in real time.

2. A remote postoperative cognitive dysfunction monitoring method according to claim 1, characterized in that: In S4, the time series analysis model is trained based on the labeled multimodal dataset.

3. A remote postoperative cognitive dysfunction monitoring method according to claim 2, characterized in that: In S4, the process of training the time series analysis model includes: Collect and annotate multimodal data from patients with postoperative cognitive dysfunction and normal patients, including arm trajectory data and speech data; Divide the collected data set into training set, validation set, and test set; Choose an appropriate deep learning model architecture, such as a long short-term memory network or a gated recurrent unit; Use the training set to train the model and adjust the model parameters through the back-propagation algorithm to minimize the loss function; During the training process, the validation set is used to verify the model and adjust the hyperparameters to optimize the model performance; Finally, the test set is used to evaluate the accuracy and generalization ability of the trained model.

4. The remote postoperative cognitive dysfunction monitoring method according to claim 1, characterized in that: In S2, the outlier cleaning includes identifying and correcting or eliminating outliers in the arm trajectory data by setting a threshold or a method based on statistical distribution.

5. The remote postoperative cognitive dysfunction monitoring method according to claim 2, characterized in that: In S2, the adaptive noise reduction filter adopts an adaptive filtering algorithm to dynamically adjust the filter parameters according to the characteristics of the speech signal and the background noise to remove the background noise.

6. A remote postoperative cognitive dysfunction monitoring method according to claim 2, characterized in that: In S3, the feature dimensionality reduction process is: using the principal component analysis algorithm, by calculating the covariance matrix and eigenvector of the features, the high-dimensional arm trajectory features are projected into a low-dimensional space, while retaining the main variance information of the data.

7. The remote postoperative cognitive dysfunction monitoring method according to claim 2, characterized in that: In S3, the weighted fusion process is: assigning different weights according to the importance of the arm trajectory features and the speech features to the judgment of postoperative cognitive dysfunction, and the weight values ​​are determined through model training or expert experience.

8. The remote postoperative cognitive dysfunction monitoring method according to claim 2, characterized in that: In S3, the vector splicing process includes: The arm trajectory feature vector after dimensionality reduction and the speech feature vector are connected end to end in a preset order to form a joint feature vector. During the splicing process, different feature dimensions are normalized or standardized according to actual needs to ensure the consistency of each feature after fusion.

9. A storage medium containing computer-executable instructions, characterized in that: When the storage medium of the computer executable instructions is executed by a computer processor, it is used to perform the remote postoperative cognitive dysfunction monitoring method according to any one of claims 1 to 8.

10. A computer program product comprising a computer program, characterized in that When the computer program is executed by a processor, the remote postoperative cognitive dysfunction monitoring method according to any one of claims 1 to 8 is implemented.

Images