Intelligent ward patient state detection method and system based on artificial intelligence

Abstract

The invention belongs to the technical field of smart wards, and discloses a smart ward patient state detection method and system based on artificial intelligence. The method comprises the following steps that: a data server constructs a patient state multi-modal detection model by using an artificial intelligence algorithm; the intelligent ward data acquisition device is used for acquiring real-time patient state multi-modal monitoring data and uploading the real-time patient state multi-modal monitoring data to the data server; the data server is used for detecting the real-time patient state multi-modal monitoring data by using a patient state multi-modal detection model to obtain a real-time patient state detection result; and the data server is used for generating a real-time alarm signal according to the real-time patient state detection result and returning the real-time alarm signal to the corresponding intelligent ward response device. According to the invention, the problems of low accuracy, low intelligent degree and insufficient real-time performance in the prior art are solved.

Description

Technical Field

[0001] This invention belongs to the field of smart ward technology, specifically relating to a method and system for detecting patient status in a smart ward based on artificial intelligence. Background Technology

[0002] A smart ward refers to a ward environment that utilizes modern information technology, the Internet of Things, and artificial intelligence to intelligently transform traditional wards, thereby improving the quality of medical services, enhancing the patient experience, and optimizing the workflow of medical staff. A smart ward can monitor patients' vital signs in real time, such as heart rate, blood pressure, and blood oxygen saturation, through wearable devices and bedside monitors, and transmit the data to a data server. The intelligent data server then uses artificial intelligence models to analyze and process the data.

[0003] The existing technology has the following drawbacks:

[0004] 1) Low accuracy: Existing technologies often rely on a single type of monitoring data, such as video surveillance or physiological parameter monitoring. This single data source method cannot fully capture the patient's overall condition and may ignore other important health indicators, resulting in low accuracy in the analysis and detection of the patient's condition.

[0005] 2) Low level of intelligence: The artificial intelligence models used in the existing technology are too simple, with low level of intelligence and insufficient training, resulting in low accuracy and high false alarm rate in detecting potential abnormal conditions of patients;

[0006] 3) Insufficient real-time performance: Existing technologies may have delays in processing and responding to changes in patient condition, which is a serious problem for emergencies that require immediate medical intervention. Summary of the Invention

[0007] To address the problems of low accuracy, low intelligence, and insufficient real-time performance in existing technologies, the present invention aims to provide a method and system for detecting patient status in smart wards based on artificial intelligence.

[0008] The technical solution adopted in this invention is as follows:

[0009] A method for detecting patient status in a smart ward based on artificial intelligence includes the following steps:

[0010] The data server uses artificial intelligence algorithms to construct a multimodal detection model of patient status based on several historical multimodal monitoring data of patient status.

[0011] The smart ward data acquisition device collects real-time multimodal monitoring data of patient status and uploads the real-time multimodal monitoring data of patient status to the data server;

[0012] The data server uses a multimodal patient status detection model to detect real-time multimodal patient status monitoring data and obtain real-time patient status detection results.

[0013] The data server generates real-time alarm signals based on real-time patient status detection results and returns these signals to the corresponding smart ward response devices.

[0014] Furthermore, the data server, based on several historical multimodal monitoring data of patient status, uses artificial intelligence algorithms to construct a multimodal detection model of patient status, including the following steps:

[0015] The data server collects historical multimodal monitoring data of several patients and preprocesses it to obtain several preprocessed historical multimodal monitoring data of patients.

[0016] Several preprocessed historical patient status multimodal monitoring data were divided into a model training set and a model test set in a 7:3 ratio;

[0017] An initial patient state multimodal detection model is constructed using artificial intelligence algorithms and input into the model training set. The initial patient state multimodal detection model is then optimized and trained to obtain an optimized patient state multimodal detection model.

[0018] Input the model test set to test the optimized patient state multimodal detection model and obtain the test accuracy. If the test accuracy is greater than the test accuracy threshold, the final patient state multimodal detection model is output.

[0019] Furthermore, the historical patient status multimodal monitoring data includes historical patient status image monitoring data, historical patient status audio monitoring data, and historical patient status vital sign monitoring data;

[0020] Real-time patient status multimodal monitoring data includes real-time patient status image monitoring data, real-time patient status audio monitoring data, and real-time patient status vital sign monitoring data.

[0021] Furthermore, the patient state multimodal detection model is constructed based on the MM-NN algorithm, and the patient state multimodal detection model includes an input layer, a multimodal feature extraction layer, a feature fusion layer, a state detection layer, and an output layer connected in sequence.

[0022] Furthermore, the multimodal feature extraction layer includes an image feature extraction module based on the CNN algorithm, an audio feature extraction module based on the RNN algorithm, and a sequence feature extraction module based on the LSTM algorithm. The image feature extraction module, the audio feature extraction module, and the sequence feature extraction module are connected in parallel between the input layer and the feature fusion layer.

[0023] The feature fusion layer is built based on the Attention mechanism and includes a weighted fusion module. The input of the weighted fusion module is connected to the image feature extraction module, the audio feature extraction module, and the sequence feature extraction module, respectively, and the output of the weighted fusion module is connected to the state detection layer.

[0024] The state detection layer is built based on the MLP algorithm and includes a label prediction module. The input of the label prediction module is connected to the output of the weighted fusion module, and the output of the label prediction module is connected to the output layer.

[0025] Furthermore, using artificial intelligence algorithms, an initial patient state multimodal detection model is constructed and input into the model training set. The initial patient state multimodal detection model is then optimized and trained to obtain an optimized patient state multimodal detection model, including the following steps:

[0026] The MM-NN algorithm is used to construct an initial patient state multimodal detection model. The initial patient state multimodal detection model includes an input layer, an initial image feature extraction module, an initial audio feature extraction module, an initial sequence feature extraction module, an initial weighted fusion module, an initial label prediction module, and an output layer.

[0027] By combining the loss functions of the initial image feature extraction module, the initial audio feature extraction module, the initial sequence feature extraction module, and the initial label prediction module, a comprehensive loss function is obtained.

[0028] The initial multimodal detection model for patient status is optimized and trained, and the real-time comprehensive loss value during the optimization training process is obtained using a comprehensive loss function.

[0029] If the real-time comprehensive loss value is less than the comprehensive loss value threshold, the optimized patient state multimodal detection model is output; otherwise, the optimization training continues.

[0030] Furthermore, the smart ward data acquisition device collects real-time multimodal monitoring data of patient status and uploads this data to the data server, including the following steps:

[0031] The smart ward data acquisition device collects real-time multimodal monitoring data on patient status;

[0032] The real-time multimodal monitoring data of patient status is compressed and packaged to obtain a real-time compressed package;

[0033] Upload the compressed file to the data server in real time.

[0034] Furthermore, the data server uses a patient status multimodal detection model to detect real-time patient status multimodal monitoring data and obtain real-time patient status detection results, including the following steps:

[0035] The data server receives real-time compressed packages uploaded by the smart ward data acquisition device, and decompresses the real-time compressed packages to obtain decompressed real-time multimodal monitoring data of patient status.

[0036] The real-time patient status multimodal monitoring data after decompression is analyzed to obtain real-time patient status image monitoring data, real-time patient status audio monitoring data, and real-time patient status vital sign monitoring data.

[0037] The input layer of the patient status multimodal detection model receives decompressed real-time patient status image monitoring data, decompressed real-time patient status audio monitoring data, and decompressed real-time patient status vital sign monitoring data.

[0038] The image feature extraction module of the patient status multimodal detection model extracts real-time image features from the decompressed real-time patient status image monitoring data, and inputs the real-time image features into the weighted fusion module.

[0039] The audio feature extraction module of the patient status multimodal detection model extracts the real-time audio features of the decompressed real-time patient status audio monitoring data and inputs the real-time audio features into the weighted fusion module.

[0040] The sequence feature extraction module of the patient status multimodal detection model is used to extract the real-time sequence features of the decompressed real-time patient status and vital signs monitoring data, and the real-time sequence features are input into the weighted fusion module.

[0041] Based on the preset attention weight values, the weighted fusion module of the patient state multimodal detection model is used to perform weighted fusion of real-time image features, real-time audio features, and real-time sequence features to obtain real-time weighted fusion features.

[0042] The label prediction module of the patient status multimodal detection model is used to predict labels based on real-time weighted fusion features, and thus obtain real-time patient status prediction labels.

[0043] The output layer of the patient status multimodal detection model is used to output real-time patient status detection results based on the real-time patient status prediction labels.

[0044] Furthermore, the data server generates a real-time alarm signal based on the real-time patient status detection results and returns the real-time alarm signal to the corresponding smart ward response device, including the following steps:

[0045] The data server analyzes the real-time patient status detection results. If the real-time patient status detection results indicate an abnormal state, a real-time alarm signal is generated, and the process proceeds to the next step.

[0046] Based on the target data source number of the smart ward data acquisition device, obtain the target smart ward number of the real-time patient status detection result;

[0047] Based on the target smart ward number, the real-time alarm signal will be returned to the corresponding smart ward response device.

[0048] An AI-based smart ward patient status detection system is provided to implement a method for detecting patient status in a smart ward. The system includes a data server, a smart ward data acquisition device, and a smart ward response device. The data server is communicatively connected to both the smart ward data acquisition device and the smart ward response device. Both the smart ward data acquisition device and the smart ward response device are located in the smart ward.

[0049] The beneficial effects of this invention are as follows:

[0050] This invention provides an AI-based smart ward patient status detection method and system. By integrating multimodal data, including image data, audio data, and vital sign data, it more comprehensively monitors the patient's status, reduces reliance on a single data source, and improves the comprehensiveness and accuracy of detection. The constructed multimodal patient status detection model has a high degree of intelligence and strong generalization ability, capable of mining deep information from multimodal patient status monitoring data, discovering potential abnormal patient states, and fully leveraging the complementary role of multi-source data to enhance the ability to identify patient states, significantly reducing the false alarm rate and improving the reliability of smart ward patient status detection. It realizes real-time data acquisition, processing, and alarm, ensuring a rapid response when abnormal patient states occur, thus gaining valuable time for emergency medical intervention.

[0051] Other beneficial effects of the present invention will be further explained in the specific embodiments. Attached Figure Description

[0052] Figure 1 This is a flowchart of the AI-based smart ward patient status detection method in this invention.

[0053] Figure 2 This is a structural block diagram of the AI-based smart ward patient status detection system of the present invention. Detailed Implementation

[0054] The present invention will be further explained below with reference to the accompanying drawings and specific embodiments.

[0055] Example 1:

[0056] like Figure 1 As shown in the figure, this embodiment provides a patient status detection method for smart wards based on artificial intelligence, including the following steps:

[0057] S1: Data server, based on several historical patient status multimodal monitoring data, uses artificial intelligence algorithms to construct a patient status multimodal detection model, including the following steps:

[0058] S1-1: Data server, which collects historical multimodal monitoring data of several patients and performs preprocessing to obtain several preprocessed historical multimodal monitoring data of patients; data preprocessing includes data cleaning (removing invalid or erroneous data), normalization (unifying data scale), etc., to ensure the data quality of input model and reduce the impact of noise and outliers on model performance;

[0059] Historical patient status multimodal monitoring data includes historical patient status image monitoring data, historical patient status audio monitoring data, and historical patient status vital sign monitoring data;

[0060] S1-2: Several preprocessed historical patient status multimodal monitoring data are divided into a model training set and a model test set in a 7:3 ratio. By dividing the training set and the test set, the generalization ability of the model can be objectively evaluated, that is, the model's performance on unseen data, thus avoiding model overfitting, i.e., the model only remembers the training data and cannot generalize to new data.

[0061] S1-3: Using artificial intelligence algorithms, an initial multimodal detection model of patient status is constructed and input into the model training set. The initial multimodal detection model of patient status is optimized and trained to obtain an optimized multimodal detection model of patient status. Through training, the model can learn complex patterns and correlations in the data, thereby improving the ability to predict patient status.

[0062] The patient state multimodal detection model is constructed based on the Multimodal Neural Network (MM-NN) algorithm, and the patient state multimodal detection model includes an input layer, a multimodal feature extraction layer, a feature fusion layer, a state detection layer, and an output layer connected in sequence.

[0063] The multimodal feature extraction layer includes an image feature extraction module based on the Convolutional Neural Network (CNN) algorithm, an audio feature extraction module based on the Recurrent Neural Network (RNN) algorithm, and a sequence feature extraction module based on the Long Short-Term Memory Network (LSTM) algorithm. The image feature extraction module, audio feature extraction module, and sequence feature extraction module are connected in parallel between the input layer and the feature fusion layer.

[0064] The image feature extraction module uses a convolutional neural network (CNN) to extract key features from the input image data, such as edges, textures, and shapes. Through multi-layer convolution and pooling operations, CNN can automatically learn hierarchical representations in images, improving the model's ability to understand image data. This enables the model to identify useful state information from complex patient state image monitoring data, including the patient's posture, body shape, and movements.

[0065] The audio feature extraction module uses a recurrent neural network (RNN) to process time-series audio data, capturing temporal features and patterns in the audio signal. RNNs can handle variable-length sequence data and are suitable for the dynamic characteristics of sound signals, enabling the model to identify and analyze the sound features of patients in different states, such as breathing sounds and heartbeat sounds, providing additional information for state detection. By learning the temporal relationship of audio signals, the model can better understand the correlation between sound and patient state.

[0066] The sequence feature extraction module uses a Long Short-Term Memory (LSTM) network to process and extract time series data, namely long-term dependencies in patient condition and vital sign monitoring data. Through its special gating mechanism, LSTM can effectively avoid the gradient vanishing problem that occurs in traditional neural networks in long sequence learning, thus improving the model's ability to process time series data. It can extract key features related to changes in patient condition from continuous physiological monitoring data. By memorizing long-term dependencies, the model can better predict changes in the patient's health status.

[0067] The feature fusion layer is built based on the Attention mechanism and includes a weighted fusion module. The input of the weighted fusion module is connected to the image feature extraction module, the audio feature extraction module, and the sequence feature extraction module, respectively, and the output of the weighted fusion module is connected to the state detection layer.

[0068] The feature fusion layer integrates the outputs from different feature extraction modules to form a unified multimodal feature representation. By fusing features from different modalities, the model can obtain more comprehensive information, thereby improving the accuracy and robustness of detection. The fusion of multimodal features enables the model to understand the patient's state from multiple perspectives, improving the comprehensiveness and accuracy of state detection. By integrating information from different sources, the model's prediction of the patient's state is more reliable, reducing the bias that may be caused by a single modality.

[0069] The state detection layer is built based on the multilayer perceptron (MLP) algorithm, and the state detection layer is equipped with a label prediction module. The input of the label prediction module is connected to the output of the weighted fusion module, and the output of the label prediction module is connected to the output layer.

[0070] The state detection layer, based on the fused multimodal features, uses MLP to predict the patient's specific state, such as normal, abnormal or emergency state, achieving accurate classification or continuous prediction of patient state, providing timely decision support for medical staff. Through highly accurate state detection, the timeliness and effectiveness of medical intervention can be effectively improved.

[0071] An initial multimodal detection model for patient status is constructed using artificial intelligence algorithms and input into the model training set. The initial multimodal detection model for patient status is then optimized and trained to obtain an optimized multimodal detection model for patient status. The process includes the following steps:

[0072] S1-3-1: Using the MM-NN algorithm, an initial multimodal detection model for patient status is constructed. The initial multimodal detection model for patient status includes an input layer, an initial image feature extraction module, an initial audio feature extraction module, an initial sequence feature extraction module, an initial weighted fusion module, an initial label prediction module, and an output layer. A model framework capable of handling multi-source heterogeneous data is initialized, laying the foundation for subsequent training and optimization. The settings of each module ensure that the model can extract and integrate information from data of different modalities, improving the comprehensive understanding of patient status.

[0073] S1-3-2: Combining the loss functions of the initial image feature extraction module, the initial audio feature extraction module, the initial sequence feature extraction module, and the initial label prediction module, a comprehensive loss function is obtained. The loss function is an indicator that measures the difference between the model's predicted value and the true value. Each module (image, audio, sequence, label prediction) has its own loss function. Combining the loss functions of these modules into a comprehensive loss function allows for an overall evaluation of the model's performance. The comprehensive loss function provides a unified optimization objective, enabling the model to balance the importance of different modalities during training. By comprehensively considering the performance of all modules, the model can be guided to learn more effectively, improving overall performance.

[0074] S1-3-3: Optimize and train the initial patient state multimodal detection model, and use the comprehensive loss function to obtain the real-time comprehensive loss value during the optimization training process;

[0075] S1-3-4: If the real-time comprehensive loss value is less than the comprehensive loss value threshold, output the optimized patient state multimodal detection model; otherwise, continue optimization training.

[0076] S1-4: Input the model test set to test the optimized patient state multimodal detection model and obtain the test accuracy. If the test accuracy is greater than the test accuracy threshold, the final patient state multimodal detection model is output. Through testing and evaluation, the model's performance on independent data can be verified to ensure that the model has good generalization ability.

[0077] S2: Smart ward data acquisition device, which collects real-time multimodal monitoring data of patient status and uploads the real-time multimodal monitoring data of patient status to the data server, including the following steps:

[0078] S2-1: Smart ward data acquisition device, which collects real-time multimodal monitoring data of patient status;

[0079] Real-time patient status multimodal monitoring data includes real-time patient status image monitoring data, real-time patient status audio monitoring data, and real-time patient status vital sign monitoring data;

[0080] The smart ward data acquisition device includes cameras, microphones, wearable devices, and bedside monitors.

[0081] Cameras are used to collect image monitoring data of patient status, microphones are used to collect audio monitoring data of patient status, and wearable devices and bedside monitors are used to collect vital signs monitoring data of patient status; real-time monitoring data provides medical staff with current monitoring data of the patient, providing data support for subsequent status detection;

[0082] S2-2: Compress and package the real-time multimodal monitoring data of the patient status to obtain a real-time compressed package; after data acquisition, in order to reduce the bandwidth requirements of data transmission and speed up the upload speed, it is usually necessary to compress the data, which reduces the amount of data transmitted over the network, saves bandwidth, and improves the efficiency of data transmission.

[0083] S2-3: Upload the real-time compressed file to the data server; rapid uploading of the real-time compressed file enables the data server to process and analyze the data in a timely manner, providing patients with timely medical intervention;

[0084] S3: Data server, using a patient status multimodal detection model to detect real-time patient status multimodal monitoring data and obtain real-time patient status detection results, including the following steps:

[0085] S3-1: Data server, which receives real-time compressed packages uploaded by the smart ward data acquisition device, and decompresses the real-time compressed packages to obtain decompressed real-time multimodal monitoring data of patient status; ensuring the efficiency of data transmission, while maintaining the original integrity and accuracy of the decompressed data;

[0086] S3-2: The decompressed real-time patient status multimodal monitoring data is parsed to obtain decompressed real-time patient status image monitoring data, decompressed real-time patient status audio monitoring data, and decompressed real-time patient status vital sign monitoring data; the decompressed data is separated into different modalities to facilitate subsequent feature extraction, thereby improving the professionalism and efficiency of data processing.

[0087] S3-3: The input layer of the patient status multimodal detection model receives decompressed real-time patient status image monitoring data, decompressed real-time patient status audio monitoring data, and decompressed real-time patient status vital sign monitoring data. The input layer provides a unified entry point for data from different modalities, facilitating subsequent processing and ensuring that the model can handle multiple types of data input.

[0088] S3-4: The image feature extraction module of the patient status multimodal detection model extracts the real-time image features of the decompressed real-time patient status image monitoring data and inputs the real-time image features into the weighted fusion module; it extracts visual features such as texture and shape of the image, which helps the model understand the image content, reduces the data dimensionality, and retains the most important information;

[0089] S3-5: The audio feature extraction module of the patient status multimodal detection model extracts the real-time audio features of the decompressed real-time patient status audio monitoring data and inputs the real-time audio features into the weighted fusion module; it captures the time dynamic changes in the audio data, such as pitch and rhythm, and increases the model's comprehensive understanding of the patient status.

[0090] S3-6: The sequence feature extraction module of the patient status multimodal detection model extracts the real-time sequence features of the decompressed real-time patient status vital sign monitoring data and inputs the real-time sequence features into the weighted fusion module; by analyzing the changing trend of vital sign data, the patient's health status can be monitored, providing continuous physiological data support for the model's prediction;

[0091] S3-7: Based on the preset attention weight values, the weighted fusion module of the patient state multimodal detection model is used to perform weighted fusion of real-time image features, real-time audio features, and real-time sequence features to obtain real-time weighted fusion features; combining features from different sources provides a more comprehensive data representation, and through weighting, the influence of key features on the prediction results is emphasized;

[0092] S3-8: The label prediction module of the patient status multimodal detection model performs label prediction based on real-time weighted fusion features to obtain real-time patient status prediction labels.

[0093] S3-9: The output layer of the patient status multimodal detection model outputs the real-time patient status detection results based on the real-time patient status prediction labels.

[0094] S4: The data server generates real-time alarm signals based on real-time patient status detection results and returns the real-time alarm signals to the corresponding smart ward response device, including the following steps:

[0095] S4-1: Data server, which analyzes the real-time patient status detection results. If the real-time patient status detection results show an abnormal state, it generates a real-time alarm signal and proceeds to the next step; this ensures continuous monitoring of the patient's status, and any abnormal situation can be detected immediately. By generating an alarm signal, it can quickly notify medical staff to take necessary medical measures.

[0096] S4-2: Based on the target data source number of the smart ward data acquisition device, obtain the target smart ward number of the real-time patient status detection result; the data server identifies the ward where the patient with the abnormal status is located based on the target data source number attached when the smart ward data acquisition device uploads data. This number is usually unique and corresponds to a specific ward and patient, ensuring that the alarm signal can be accurately sent to the corresponding ward, avoiding false alarms or missed alarms. Through the association of the number, effective communication between the data server and the ward response device is realized.

[0097] S4-3: Based on the target smart ward number, the real-time alarm signal is returned to the corresponding smart ward response device; medical staff can receive the alarm immediately through the response device without waiting for manual information transmission. The response device can automatically execute certain predetermined response procedures, such as issuing an audible and visual alarm to notify the doctor.

[0098] Example 2:

[0099] like Figure 2 As shown, this embodiment provides an artificial intelligence-based smart ward patient status detection system to implement a smart ward patient status detection method. The system includes a data server, a smart ward data acquisition device, and a smart ward response device. The data server is communicatively connected to the smart ward data acquisition device and the smart ward response device, respectively. Both the smart ward data acquisition device and the smart ward response device are installed in the smart ward.

[0100] The data server is used to construct a multimodal patient status detection model based on several historical multimodal patient status monitoring data and artificial intelligence algorithms; the multimodal patient status detection model is used to detect real-time multimodal patient status monitoring data to obtain real-time patient status detection results; the data server generates real-time alarm signals based on the real-time patient status detection results and returns the real-time alarm signals to the corresponding smart ward response devices;

[0101] The smart ward data acquisition device is used to collect real-time multimodal monitoring data of patient status and upload the real-time multimodal monitoring data of patient status to the data server;

[0102] The smart ward response device is used to respond to real-time alarm signals.

[0103] This invention provides an AI-based smart ward patient status detection method and system. By integrating multimodal data, including image data, audio data, and vital sign data, it more comprehensively monitors the patient's status, reduces reliance on a single data source, and improves the comprehensiveness and accuracy of detection. The constructed multimodal patient status detection model has a high degree of intelligence and strong generalization ability, capable of mining deep information from multimodal patient status monitoring data, discovering potential abnormal patient states, and fully leveraging the complementary role of multi-source data to enhance the ability to identify patient states, significantly reducing the false alarm rate and improving the reliability of smart ward patient status detection. It realizes real-time data acquisition, processing, and alarm, ensuring a rapid response when abnormal patient states occur, thus gaining valuable time for emergency medical intervention.

[0104] This invention is not limited to the optional embodiments described above, and anyone can derive other various forms of products based on the inspiration of this invention. The specific embodiments described above should not be construed as limiting the scope of protection of this invention; the scope of protection of this invention should be determined by the claims, and the specification can be used to interpret the claims.

Claims

1. A method for detecting patient status in a smart ward based on artificial intelligence, characterized in that: Includes the following steps: The data server uses artificial intelligence algorithms to construct a multimodal detection model of patient status based on several historical multimodal monitoring data of patient status. The smart ward data acquisition device collects real-time multimodal monitoring data of patient status and uploads the real-time multimodal monitoring data of patient status to the data server; The data server uses a multimodal patient status detection model to detect real-time multimodal patient status monitoring data and obtain real-time patient status detection results. The data server generates real-time alarm signals based on real-time patient status detection results and returns these signals to the corresponding smart ward response devices.

2. The method for detecting patient status in a smart ward based on artificial intelligence according to claim 1, characterized in that: The data server, based on historical multimodal monitoring data of patient status, uses artificial intelligence algorithms to construct a multimodal detection model of patient status, including the following steps: The data server collects historical multimodal monitoring data of several patients and preprocesses it to obtain several preprocessed historical multimodal monitoring data of patients. Several preprocessed historical patient status multimodal monitoring data were divided into a model training set and a model test set in a 7:3 ratio; An initial patient state multimodal detection model is constructed using artificial intelligence algorithms and input into the model training set. The initial patient state multimodal detection model is then optimized and trained to obtain an optimized patient state multimodal detection model. Input the model test set to test the optimized patient state multimodal detection model and obtain the test accuracy. If the test accuracy is greater than the test accuracy threshold, the final patient state multimodal detection model is output.

3. The method for detecting patient status in a smart ward based on artificial intelligence according to claim 2, characterized in that: The historical patient status multimodal monitoring data includes historical patient status image monitoring data, historical patient status audio monitoring data, and historical patient status vital sign monitoring data; The real-time patient status multimodal monitoring data includes real-time patient status image monitoring data, real-time patient status audio monitoring data, and real-time patient status vital sign monitoring data.

4. The method for detecting patient status in a smart ward based on artificial intelligence according to claim 3, characterized in that: The patient state multimodal detection model is constructed based on the MM-NN algorithm, and the patient state multimodal detection model includes an input layer, a multimodal feature extraction layer, a feature fusion layer, a state detection layer, and an output layer connected in sequence.

5. The method for detecting patient status in a smart ward based on artificial intelligence according to claim 4, characterized in that: The multimodal feature extraction layer includes an image feature extraction module based on the CNN algorithm, an audio feature extraction module based on the RNN algorithm, and a sequence feature extraction module based on the LSTM algorithm. The image feature extraction module, audio feature extraction module, and sequence feature extraction module are connected in parallel between the input layer and the feature fusion layer. The feature fusion layer is constructed based on the Attention mechanism and includes a weighted fusion module. The input of the weighted fusion module is connected to the image feature extraction module, the audio feature extraction module, and the sequence feature extraction module, respectively, and the output of the weighted fusion module is connected to the state detection layer. The state detection layer is constructed based on the MLP algorithm and includes a label prediction module. The input of the label prediction module is connected to the output of the weighted fusion module, and the output of the label prediction module is connected to the output layer.

6. The method for detecting patient status in a smart ward based on artificial intelligence according to claim 5, characterized in that: An initial multimodal detection model for patient status is constructed using artificial intelligence algorithms and input into the model training set. The initial multimodal detection model for patient status is then optimized and trained to obtain an optimized multimodal detection model for patient status. The process includes the following steps: An initial patient state multimodal detection model is constructed using the MM-NN algorithm. The initial patient state multimodal detection model includes an input layer, an initial image feature extraction module, an initial audio feature extraction module, an initial sequence feature extraction module, an initial weighted fusion module, an initial label prediction module, and an output layer. By combining the loss functions of the initial image feature extraction module, the initial audio feature extraction module, the initial sequence feature extraction module, and the initial label prediction module, a comprehensive loss function is obtained. The initial multimodal detection model for patient status is optimized and trained, and the real-time comprehensive loss value during the optimization training process is obtained using a comprehensive loss function. If the real-time comprehensive loss value is less than the comprehensive loss value threshold, the optimized patient state multimodal detection model is output; otherwise, the optimization training continues.

7. The method for detecting patient status in a smart ward based on artificial intelligence according to claim 6, characterized in that: The smart ward data acquisition device collects real-time multimodal monitoring data of patient status and uploads this data to a data server, including the following steps: The smart ward data acquisition device collects real-time multimodal monitoring data on patient status; The real-time multimodal monitoring data of patient status is compressed and packaged to obtain a real-time compressed package; Upload the compressed file to the data server in real time.

8. The method for detecting patient status in a smart ward based on artificial intelligence according to claim 7, characterized in that: The data server uses a patient status multimodal detection model to detect real-time patient status multimodal monitoring data and obtain real-time patient status detection results, including the following steps: The data server receives real-time compressed packages uploaded by the smart ward data acquisition device, and decompresses the real-time compressed packages to obtain decompressed real-time multimodal monitoring data of patient status. The real-time patient status multimodal monitoring data after decompression is analyzed to obtain real-time patient status image monitoring data, real-time patient status audio monitoring data, and real-time patient status vital sign monitoring data. The input layer of the patient status multimodal detection model receives decompressed real-time patient status image monitoring data, decompressed real-time patient status audio monitoring data, and decompressed real-time patient status vital sign monitoring data. The image feature extraction module of the patient status multimodal detection model extracts real-time image features from the decompressed real-time patient status image monitoring data, and inputs the real-time image features into the weighted fusion module. The audio feature extraction module of the patient status multimodal detection model extracts the real-time audio features of the decompressed real-time patient status audio monitoring data and inputs the real-time audio features into the weighted fusion module. The sequence feature extraction module of the patient status multimodal detection model is used to extract the real-time sequence features of the decompressed real-time patient status and vital signs monitoring data, and the real-time sequence features are input into the weighted fusion module. Based on the preset attention weight values, the weighted fusion module of the patient state multimodal detection model is used to perform weighted fusion of real-time image features, real-time audio features, and real-time sequence features to obtain real-time weighted fusion features. The label prediction module of the patient status multimodal detection model is used to predict labels based on real-time weighted fusion features, and thus obtain real-time patient status prediction labels. The output layer of the patient status multimodal detection model is used to output real-time patient status detection results based on the real-time patient status prediction labels.

9. The method for detecting patient status in a smart ward based on artificial intelligence according to claim 8, characterized in that: The data server generates real-time alarm signals based on real-time patient status monitoring results and returns these signals to the corresponding smart ward response devices, including the following steps: The data server analyzes the real-time patient status detection results. If the real-time patient status detection results indicate an abnormal state, a real-time alarm signal is generated, and the process proceeds to the next step. Based on the target data source number of the smart ward data acquisition device, obtain the target smart ward number of the real-time patient status detection result; Based on the target smart ward number, the real-time alarm signal will be returned to the corresponding smart ward response device.

10. A smart ward patient status detection system based on artificial intelligence, used to implement the smart ward patient status detection method as described in any one of claims 1-9, characterized in that: The system includes a data server, a smart ward data acquisition device, and a smart ward response device. The data server is communicatively connected to both the smart ward data acquisition device and the smart ward response device. Both the smart ward data acquisition device and the smart ward response device are located in the smart ward.

Images