Intelligent state monitoring system and method for patient after radiotherapy

Abstract

The invention provides an intelligent state monitoring system and method for a post-radiotherapy patient, relates to the field of intelligent health monitoring, and utilizes a data analysis and coding technology based on machine learning to perform time sequence feature extraction and parameter fusion on each multi-dimensional physiological parameter of a post-radiotherapy patient object. Performing gating characteristic dynamic propagation on each multi-dimensional physiological parameter fusion information representation, and performing semantic coding on a symptom report text description of a patient at the same time; therefore, the intelligent judgment of the state of the patient object after radiotherapy is realized according to the interactive fusion representation between the multi-dimensional physiological parameter fusion information dynamic propagation characteristics and the symptom report text semantic coding characteristics. Therefore, the dynamic change of the state of the patient can be captured more comprehensively and meticulously, the internal relation and interaction among the indexes are fully considered, and the overall health state of the patient can be evaluated more accurately.

Description

Technical Field

[0001] The present application relates to the field of intelligent health monitoring, and more specifically, to a system and method for intelligent status monitoring of patients after radiotherapy. Background Art

[0002] Radiation therapy (RT) is an important method of cancer treatment. It uses high-energy rays or particles to kill cancer cells or shrink tumors. However, radiotherapy not only kills tumor cells, but may also damage normal tissues, leading to a series of side effects and complications. Therefore, it is very important to monitor the health status of patients after radiotherapy.

[0003] However, traditional post-radiotherapy patient monitoring mainly relies on periodic hospital examinations and doctor consultations. The data obtained in this way are discrete and discontinuous, making it difficult to fully capture subtle dynamic changes in the patient's condition, which may lead to delays in the discovery and treatment of early problems. In addition, various physiological indicators such as heart rate, blood pressure, and blood oxygen saturation are usually analyzed independently, ignoring the inherent connections and interactions between them. These indicators actually jointly affect the health of the human body, and the correlation between them is crucial for accurately assessing the patient's overall health. This segmented assessment method limits the accurate understanding of the patient's overall health status, makes it difficult to provide personalized treatment recommendations, and weakens the ability to predict potential health risks.

[0004] Therefore, an optimized monitoring scheme for patients' status after radiotherapy is desired. Summary of the invention

[0005] In order to solve the above technical problems, the present application is proposed. The embodiments of the present application provide a system and method for intelligent status monitoring of patients after radiotherapy.

[0006] According to one aspect of the present application, a system for monitoring the intelligent status of patients after radiotherapy is provided, which includes: a physiological parameter data acquisition module for collecting multi-dimensional physiological parameter time series data of a patient subject after radiotherapy from a sensor group, wherein the multi-dimensional physiological parameter time series data includes a time queue of heart rate values, a time queue of blood pressure values, a time queue of blood oxygen saturation, and a time queue of activity volume; a symptom report description acquisition module for acquiring a text description of a patient's symptom report; a multi-dimensional parameter fusion module for performing time series feature analysis and multi-dimensional parameter fusion on the time queue of heart rate values, the time queue of blood pressure values, the time queue of blood oxygen saturation, and the time queue of activity volume to obtain a sequence represented by multi-dimensional physiological parameter fusion information; and a multi-dimensional physiological parameter dynamic A propagation module is used to perform gated writing feature dynamic propagation on the sequence represented by the multi-dimensional physiological parameter fusion information to obtain the dynamic propagation feature of the multi-dimensional physiological parameter fusion information; a symptom report semantic encoding module is used to perform semantic encoding on the symptom report text description to obtain the semantic encoding feature of the symptom report text; a multi-dimensional physiological parameter time series dynamic aggregation module is used to perform attention feature interactive fusion on the dynamic propagation feature of the multi-dimensional physiological parameter fusion information and the semantic encoding feature of the symptom report text to obtain the multi-dimensional physiological parameter time series dynamic aggregation representation of the fused symptom semantic information; a state abnormality judgment module is used to judge whether the state of the patient object after radiotherapy is abnormal based on the multi-dimensional physiological parameter time series dynamic aggregation representation of the fused symptom semantic information.

[0007] Optionally, the multidimensional parameter fusion module includes: a physiological parameter data time series feature extraction unit, which is used to pass the time queue of the heart rate value, the time queue of the blood pressure value, the time queue of the blood oxygen saturation and the time queue of the activity amount through a physiological parameter time series feature extractor based on a 1D-FCN network to obtain a sequence of heart rate time series feature vectors, a sequence of blood pressure time series feature vectors, a sequence of blood oxygen saturation time series feature vectors and a sequence of activity amount time series feature vectors; a physiological parameter data time series feature cascade unit, which is used to cascade each group of corresponding heart rate time series feature vectors, blood pressure time series feature vectors, blood oxygen saturation time series feature vectors and activity amount time series feature vectors in the sequence of heart rate time series feature vectors, the sequence of blood pressure time series feature vectors, the sequence of blood oxygen saturation time series feature vectors and the sequence of activity amount time series feature vectors to obtain a sequence of multidimensional physiological parameter fusion information representation vectors as the sequence of multidimensional physiological parameter fusion information representation.

[0008] Optionally, the multidimensional physiological parameter dynamic propagation module includes: a multidimensional physiological parameter initial propagation unit, which is used to perform initial dynamic propagation on the sequence of the multidimensional physiological parameter fusion information representation vector to obtain an initial multidimensional physiological parameter fusion information feature dynamic propagation coding vector; a multidimensional physiological parameter private gating feature calculation unit, which is used to calculate the private gating features of each multidimensional physiological parameter fusion information representation vector in the sequence of the multidimensional physiological parameter fusion information representation vector based on the initial multidimensional physiological parameter fusion information feature dynamic propagation coding vector to obtain a sequence of gated-written multidimensional physiological parameter private feature coding vectors; a multidimensional physiological parameter time series dynamic aggregation representation generation unit, which is used to fuse the sequence of gated-written multidimensional physiological parameter private feature coding vectors and the initial multidimensional physiological parameter fusion information feature dynamic propagation coding vector to obtain a multidimensional physiological parameter fusion information dynamic propagation coding vector as a multidimensional physiological parameter fusion information dynamic propagation feature.

[0009] Optionally, the multidimensional physiological parameter initial propagation unit is used to: input the sequence of the multidimensional physiological parameter fusion information representation vector into a sequence reasoning module based on a forward LSTM model to obtain the initial multidimensional physiological parameter fusion information feature dynamic propagation coding vector.

[0010] Optionally, the multidimensional physiological parameter private gating feature calculation unit is used to: calculate the private features of each multidimensional physiological parameter fusion information representation vector in the sequence of multidimensional physiological parameter fusion information representation vectors relative to the initial multidimensional physiological parameter fusion information feature dynamic propagation coding vector to obtain a sequence of multidimensional physiological parameter private feature coding vectors; input each multidimensional physiological parameter private feature coding vector in the sequence of multidimensional physiological parameter private feature coding vectors into a gated writing unit based on a Softmax function to obtain the sequence of gated-written multidimensional physiological parameter private feature coding vectors.

[0011] Optionally, the symptom report semantic encoding module is used to: pass the symptom report text description through a semantic encoder including a Bert model to obtain a symptom report text semantic encoding feature vector as the symptom report text semantic encoding feature.

[0012] Optionally, the multidimensional physiological parameter time series dynamic aggregation module is used to: pass the multidimensional physiological parameter fusion information dynamic propagation coding vector and the symptom report text semantic coding feature vector through a feature interaction fusion device based on an X-linear attention network to obtain a multidimensional physiological parameter time series dynamic aggregation representation vector of the fused symptom semantic information as the multidimensional physiological parameter time series dynamic aggregation representation of the fused symptom semantic information.

[0013] Optionally, the abnormal state judgment module is used to: pass the multi-dimensional physiological parameter time-series dynamic aggregation representation vector that integrates the symptom semantic information through a classifier-based patient state monitor to obtain a monitoring result, and the monitoring result is used to indicate whether there is an abnormality in the state of the patient object after radiotherapy.

[0014] According to another aspect of the present application, a method for intelligent status monitoring of patients after radiotherapy is provided, which comprises: collecting multi-dimensional physiological parameter time series data of the patient object after radiotherapy from a sensor group, the multi-dimensional physiological parameter time series data comprising a time queue of heart rate value, a time queue of blood pressure value, a time queue of blood oxygen saturation and a time queue of activity amount; obtaining a text description of the patient's symptom report; performing time series feature analysis and multi-dimensional parameter fusion on the time queue of heart rate value, the time queue of blood pressure value, the time queue of blood oxygen saturation and the time queue of activity amount to obtain a sequence represented by multi-dimensional physiological parameter fusion information; performing gated writing feature dynamic propagation on the sequence represented by multi-dimensional physiological parameter fusion information to obtain a multi-dimensional physiological parameter fusion information dynamic propagation feature; performing semantic encoding on the symptom report text description to obtain a symptom report text semantic encoding feature; performing attention feature interactive fusion on the multi-dimensional physiological parameter fusion information dynamic propagation feature and the symptom report text semantic encoding feature to obtain a multi-dimensional physiological parameter time series dynamic aggregation representation of fused symptom semantic information; judging whether the status of the patient object after radiotherapy is abnormal based on the multi-dimensional physiological parameter time series dynamic aggregation representation of the fused symptom semantic information.

[0015] Optionally, the time queue of the heart rate values, the time queue of the blood pressure values, the time queue of the blood oxygen saturation and the time queue of the activity volume are subjected to timing feature analysis and multi-dimensional parameter fusion to obtain a sequence of multi-dimensional physiological parameter fusion information representation, including: passing the time queue of the heart rate values, the time queue of the blood pressure values, the time queue of the blood oxygen saturation and the time queue of the activity volume through a physiological parameter timing feature extractor based on a 1D-FCN network to obtain a sequence of heart rate timing feature vectors, a sequence of blood pressure timing feature vectors, a sequence of blood oxygen saturation timing feature vectors and a sequence of activity volume timing feature vectors; cascading each group of corresponding heart rate timing feature vectors, blood pressure timing feature vectors, blood oxygen saturation timing feature vectors and activity volume timing feature vectors in the sequence of heart rate timing feature vectors, the sequence of blood pressure timing feature vectors, the sequence of blood oxygen saturation timing feature vectors and the sequence of activity volume timing feature vectors to obtain a sequence of multi-dimensional physiological parameter fusion information representation vectors as the sequence of multi-dimensional physiological parameter fusion information representation.

[0016] Compared with the prior art, the intelligent state monitoring system and method for patients after radiotherapy provided by the present application utilizes data analysis and coding technology based on machine learning to extract time series features and fuse parameters of various multidimensional physiological parameters of patients after radiotherapy, and then dynamically propagates gated features of the fusion information representation of each multidimensional physiological parameter, and semantically encodes the text description of the patient's symptom report, thereby realizing intelligent judgment of the state of the patient after radiotherapy based on the interactive fusion representation between the dynamic propagation features of the fusion information of the multidimensional physiological parameters and the semantic coding features of the symptom report text. In this way, the dynamic changes of the patient's state can be captured more comprehensively and meticulously, and the internal connection and interaction between the various indicators are fully considered, which helps to more accurately evaluate the overall health status of the patient. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments or the prior art descriptions. Obviously, the drawings described below are some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work. In the drawings: Figure 1 This is a system block diagram of an intelligent status monitoring system for patients after radiotherapy according to an embodiment of the present application.

[0018] Figure 2 It is a block diagram of a multi-dimensional parameter fusion module in the intelligent status monitoring system for patients after radiotherapy according to an embodiment of the present application.

[0019] Figure 3 It is a block diagram of a multi-dimensional physiological parameter dynamic propagation module in an intelligent status monitoring system for patients after radiotherapy according to an embodiment of the present application.

[0020] Figure 4 Flow chart of a method for intelligent status monitoring of patients after radiotherapy according to an embodiment of the present application. DETAILED DESCRIPTION

[0021] Below, the exemplary embodiments according to the present application will be described in detail with reference to the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present application, rather than all the embodiments of the present application, and it should be understood that the present application is not limited to the exemplary embodiments described here.

[0022] Radiation therapy (RT) is an important means of treating cancer, but it may also damage normal tissues while killing tumor cells, leading to side effects and complications. Therefore, monitoring the health status of patients after radiotherapy is crucial. However, traditional monitoring methods mainly rely on periodic hospital examinations and doctor consultations. Data acquisition is discrete and discontinuous, making it difficult to capture the dynamic changes in the patient's condition in a timely manner, which may lead to delays in the discovery and treatment of early problems. In addition, physiological indicators such as heart rate, blood pressure, and blood oxygen saturation are usually analyzed independently, ignoring their inherent connections and interactions. This segmented assessment limits the accurate assessment of the patient's health status, which in turn affects the ability to make personalized treatment recommendations and predict potential health risks.

[0023] In response to the above technical problems, the technical concept of this application is to collect the time series data of multi-dimensional physiological parameters (heart rate value, blood pressure value, blood oxygen saturation and activity level) of the patient object after radiotherapy from the sensor group, and obtain the text description of the patient's symptom report, and use the data analysis and coding technology based on machine learning to extract the time series features and fuse the parameters of each multi-dimensional physiological parameter, and then perform gated feature dynamic propagation on the fusion information representation of each multi-dimensional physiological parameter, and at the same time, perform semantic encoding on the symptom report text description, so as to realize the intelligent judgment of the state of the patient object after radiotherapy according to the interactive fusion representation between the dynamic propagation characteristics of the fusion information of the multi-dimensional physiological parameter and the semantic coding characteristics of the symptom report text. This application overcomes the problems of data discreteness and discontinuity in traditional methods by collecting multi-dimensional physiological parameter data in real time, and can carefully capture the subtle dynamic changes of the patient's state and their mutual influence, which helps to find problems early and deal with them in time. In addition, this scheme not only analyzes the characteristics of each physiological indicator, but also reveals the intrinsic connection between them by integrating the semantic information of the symptom description, so as to achieve a more comprehensive and accurate diagnosis.

[0024] Figure 1 FIG. 1 is a system block diagram of an intelligent status monitoring system for patients after radiotherapy according to an embodiment of the present application. Figure 1As shown, in the intelligent state monitoring system 100 for patients after radiotherapy, it includes: a physiological parameter data acquisition module 110, which is used to collect multi-dimensional physiological parameter time series data of the patient object after radiotherapy from the sensor group, and the multi-dimensional physiological parameter time series data includes a time queue of heart rate value, a time queue of blood pressure value, a time queue of blood oxygen saturation and a time queue of activity; a symptom report description acquisition module 120, which is used to obtain a text description of the patient's symptom report; a multi-dimensional parameter fusion module 130, which is used to perform time series feature analysis and multi-dimensional parameter fusion on the time queue of the heart rate value, the time queue of the blood pressure value, the time queue of the blood oxygen saturation and the time queue of the activity to obtain a sequence represented by multi-dimensional physiological parameter fusion information; a multi-dimensional physiological parameter dynamic propagation module 14 0, used for performing gated writing of feature dynamic propagation of the sequence represented by the multi-dimensional physiological parameter fusion information to obtain the dynamic propagation feature of the multi-dimensional physiological parameter fusion information; symptom report semantic encoding module 150, used for performing semantic encoding of the symptom report text description to obtain the semantic encoding feature of the symptom report text; multi-dimensional physiological parameter time series dynamic aggregation module 160, used for performing attention feature interactive fusion of the dynamic propagation feature of the multi-dimensional physiological parameter fusion information and the semantic encoding feature of the symptom report text to obtain the multi-dimensional physiological parameter time series dynamic aggregation representation of the fusion symptom semantic information; state abnormality judgment module 170, used for judging whether the state of the patient object after radiotherapy is abnormal based on the multi-dimensional physiological parameter time series dynamic aggregation representation of the fusion symptom semantic information.

[0025] In an embodiment of the present application, the physiological parameter data acquisition module 110 is used to collect multi-dimensional physiological parameter time series data of the patient object after radiotherapy from the sensor group, and the multi-dimensional physiological parameter time series data includes a time queue of heart rate values, a time queue of blood pressure values, a time queue of blood oxygen saturation, and a time queue of activity. It should be understood that the time queue of heart rate values ​​contains the heart rate value information of patients after radiotherapy recorded at different time points, which can reflect the fluctuation information of the patient's heart rate over time, such as whether there is a heart rate that is too fast, too slow or irregular heartbeat rhythm changes; the time queue of blood pressure values ​​is composed of the patient's blood pressure values ​​measured at multiple time nodes in sequence, covering the relevant data of systolic and diastolic blood pressure, which can reflect whether the patient's blood pressure is in a high or low state for a long time, as well as the frequency and amplitude of blood pressure fluctuations; the time queue of blood oxygen saturation records the oxygen saturation in the patient's blood at each time point, expressed in the form of percentage, which can reflect whether the patient's blood's ability to carry oxygen is stable, whether there are abnormal manifestations such as a sudden drop in blood oxygen content or a long-term low level, and then can reflect the patient's respiratory system and blood circulation system The oxygen transport and supply status; the time queue of activity volume mainly reflects the patient's physical activity level at different time points. This data can be quantitative data such as the number of steps walked and the speed of body movement. It can help understand the regularity of the patient's daily activities and the degree of activity, and assist in judging their physical function status and physical recovery. In general, the time cohort of heart rate values, the time cohort of blood pressure values, the time cohort of blood oxygen saturation and the time cohort of activity reflect the physical condition of patients after radiotherapy from different angles. They are interrelated and mutually confirmed, and together provide a comprehensive and critical basis for accurately judging whether there is any abnormality in the patient's condition.

[0026] The following is a detailed description of an implementation method of "collecting multi-dimensional physiological parameter time series data of a patient subject after radiotherapy from a sensor group, wherein the multi-dimensional physiological parameter time series data includes a time queue of heart rate values, a time queue of blood pressure values, a time queue of blood oxygen saturation, and a time queue of activity amount": The first is the selection of sensors. For the collection of heart rate values, devices such as wearable bracelets or chest strap heart rate monitors can be selected. These sensors use their built-in sensing devices to measure heart rate by sensing the electrical signals or blood flow changes generated when the heart beats. When in use, it needs to be worn on the patient's body in a suitable position, such as the wrist or chest, and ensure that the sensor is in full contact with the skin to ensure that the heart rate data can be accurately obtained. In this way, during the patient's daily activities or rest, the sensor can also continuously record the heart rate values ​​at different time points, thereby forming a time queue of heart rate values. For the measurement of blood pressure values, electronic sphygmomanometers are usually used, and the common types are arm type and wrist type. When measuring blood pressure, the patient needs to remain quiet, wear the blood pressure sensor accurately on the arm or wrist, and start the measurement program strictly in accordance with the operating instructions of the device. The working principle of the blood pressure sensor is to obtain blood pressure values, including systolic and diastolic pressures, by detecting the pressure of blood on the blood vessel wall. After multiple measurements, the blood pressure data at different time points will be arranged in sequence to form a time queue of blood pressure values. The monitoring of blood oxygen saturation mainly relies on a clip-on blood oxygen saturation monitor, which is usually clipped on the patient's fingers or earlobes. The sensor uses infrared technology to calculate blood oxygen saturation by detecting the absorption of different wavelengths of light by oxygenated hemoglobin and reduced hemoglobin in the blood. During the continuous monitoring process, the blood oxygen saturation data obtained at each time point will form a time queue of blood oxygen saturation. The collection of activity volume is assisted by devices such as accelerometers built into wearable devices (such as smart watches, sports bracelets, etc.). These sensors can keenly sense the patient's body movement state and quantify the patient's activity volume through precise analysis of acceleration data. The activity volume data covers a wealth of information such as the number of steps walked, body movement speed, and exercise intensity, and its time queue clearly shows the patient's activity level at different times.

[0027] After completing the selection and arrangement of sensors, data collection and transmission are crucial. In order to ensure data validity while avoiding data redundancy and increased equipment energy consumption caused by excessive collection, it is necessary to reasonably set the collection frequency of various physiological parameters according to monitoring needs and equipment performance. Generally speaking, heart rate and blood oxygen saturation can be set to collect data once a minute, so that their dynamic changes can be captured more carefully. The collection frequency of blood pressure can be adjusted according to the patient's condition and doctor's advice, such as collecting once every half an hour or an hour, because blood pressure changes relatively slowly, but it still needs to be monitored regularly to grasp its long-term trend. Activity data can be collected continuously in real time due to its strong dynamics, such as updating acceleration data once a second, and then calculating the activity within a certain time interval through a special algorithm, so as to more accurately reflect the patient's activity status.

[0028] The data collected by the sensor is usually transmitted to the paired receiving device (specialized data acquisition terminal) with the help of wireless transmission technology (such as Bluetooth, Wi-Fi, etc.). During the transmission process, it is key to ensure the stability and accuracy of data transmission. To this end, security measures such as encryption are needed to protect the patient's privacy data and prevent it from being leaked during the transmission process. The receiving device is pre-installed with a corresponding data receiving and storage application, which can receive the time series data from each sensor in an orderly manner and store it properly. During the storage process, the data will be sorted in chronological order to provide a convenient data basis for subsequent data analysis and processing.

[0029] Finally, data preprocessing and sorting are indispensable. The collected raw data may contain abnormal values ​​and erroneous data. For example, the heart rate sensor may have a short maximum or minimum value (obviously beyond the normal heart rate range) due to poor contact and other reasons. These data will interfere with subsequent analysis, so they need to be identified and corrected or deleted by algorithms. For blood pressure data, if the patient moves during the measurement, resulting in inaccurate measurements, corresponding processing is also required. At the same time, for missing data, interpolation and other methods can be used to supplement to ensure the continuity and integrity of the data. After data cleaning, the data format needs to be unified and sorted, and data from different sensors and different formats need to be integrated. Ensure that the heart rate value, blood pressure value, blood oxygen saturation value and activity value are strictly and accurately corresponding in time sequence to form a standardized multi-dimensional physiological parameter time series data format.

[0030] In this way, the multi-dimensional physiological parameter time series data after a series of rigorous operations can provide a high-quality data source for subsequent analysis and processing, thereby more accurately evaluating the health status of patients after radiotherapy and providing strong support for patients' rehabilitation and treatment.

[0031] In the embodiment of the present application, the symptom report description acquisition module 120 is used to obtain the patient's symptom report text description. It should be understood that the patient's symptom report text description contains various physical condition related information that the patient himself subjectively feels. For example, the patient may describe the pain in a certain part of the body after radiotherapy, such as whether the chest pain is tingling, dull pain or dull pain, the degree of pain (mild, moderate or severe), the frequency of pain attacks (occasional attacks, frequent attacks, etc.); whether there is fatigue and the severity of fatigue, whether you feel extremely tired after a little activity, or only have slight fatigue after a long period of activity; and whether there are symptoms of the digestive system such as nausea, vomiting, loss of appetite, and whether the breathing is smooth, whether there are shortness of breath, coughing and other respiratory system related manifestations. In addition, it may also include information about sleep quality, such as the difficulty of falling asleep, whether it is easy to wake up at night, and the total length of sleep time. Considering that the data obtained by the sensor are mainly objective and quantifiable physiological indicators, although they can reflect some functional states of the body, they cannot fully present the patient's own subjective experience. Patients have the most direct feelings about their own physical discomfort, and the text description of the symptom report can convey these subjective feelings, such as the degree of fatigue felt by the patient, pain in specific parts, etc. This information is feedback on the patient's physical condition from the patient's perspective, and can provide another dimension of judgment basis different from objective physiological data. In other words, obtaining the text description of the patient's symptom report is to understand the patient's condition more comprehensively and accurately, and it can be closely combined with the data measured by the sensor to jointly judge the patient's condition after radiotherapy, thereby improving the accuracy of judging abnormal patient conditions after radiotherapy.

[0032] In the embodiment of the present application, the multidimensional parameter fusion module 130 is used to perform time series feature analysis and multidimensional parameter fusion on the time queue of the heart rate value, the time queue of the blood pressure value, the time queue of the blood oxygen saturation, and the time queue of the activity amount to obtain a sequence represented by multidimensional physiological parameter fusion information. Specifically, Figure 2 FIG. 1 is a block diagram of a multi-dimensional parameter fusion module in an intelligent status monitoring system for patients after radiotherapy according to an embodiment of the present application. Figure 2As shown, the multidimensional parameter fusion module 130 includes: a physiological parameter data time series feature extraction unit 131, which is used to obtain a sequence of heart rate time series feature vectors, a sequence of blood pressure time series feature vectors, a sequence of blood oxygen saturation time series feature vectors and a sequence of activity amount time series feature vectors through a physiological parameter time series feature extractor based on a 1D-FCN network; a physiological parameter data time series feature cascade unit 132, which is used to cascade each group of corresponding heart rate time series feature vectors, blood pressure time series feature vectors, blood oxygen saturation time series feature vectors and activity amount time series feature vectors in the sequence of heart rate time series feature vectors, the sequence of blood pressure time series feature vectors, the sequence of blood oxygen saturation time series feature vectors and the sequence of activity amount time series feature vectors to obtain a sequence of multidimensional physiological parameter fusion information representation vectors as the sequence of multidimensional physiological parameter fusion information representation.

[0033] In an embodiment of the present application, the physiological parameter data time series feature extraction unit 131 is used to pass the time queue of the heart rate value, the time queue of the blood pressure value, the time queue of the blood oxygen saturation and the time queue of the activity through a physiological parameter time series feature extractor based on a 1D-FCN network to obtain a sequence of heart rate time series feature vectors, a sequence of blood pressure time series feature vectors, a sequence of blood oxygen saturation time series feature vectors and a sequence of activity time series feature vectors. It should be understood that each physiological parameter has time series feature information in the time dimension, such as short-term heart rate fluctuations or long-term blood pressure trend information, etc. Therefore, in order to capture the instant information at each time point to reflect the dynamic change pattern between adjacent time points, thereby providing richer context information for subsequent analysis, in the technical solution of the present application, the time queue of the heart rate value, the time queue of the blood pressure value, the time queue of the blood oxygen saturation and the time queue of the activity are processed by a physiological parameter time series feature extractor based on a 1D-FCN network. It is understandable that 1D-FCN (one-dimensional fully convolutional network) is a deep learning model specially designed to process one-dimensional time series data. It mainly uses one-dimensional convolutional layers to directly process time series data. This architecture retains the time dimension of the time series and can achieve end-to-end learning without introducing pooling layers. Specifically, by sliding the convolution kernel at different scales, it can capture local transient changes (such as instantaneous heart rate acceleration) and perceive broader contextual information (such as persistent hypotension). This can automatically extract meaningful features from the time series, which may include short-term fluctuations, long-term trends, and correlations between different time points. For example, it can identify the changing pattern of heart rate throughout the day, or detect abnormal increases in blood pressure within a specific time period.

[0034] The following is a detailed description of an implementable method of "passing the time queue of the heart rate value, the time queue of the blood pressure value, the time queue of the blood oxygen saturation and the time queue of the activity amount through a physiological parameter time series feature extractor based on a 1D-FCN network to obtain a corresponding time series feature vector sequence": Before inputting the time queue data of different physiological parameters into the 1D-FCN network, a series of data preprocessing operations are required. The first is data format adjustment to ensure that the time queue data format of heart rate values, blood pressure values, blood oxygen saturation values, and activity values ​​is unified and accurately corresponds. This includes normalizing the data so that it is in a suitable numerical range, which not only helps the network to better learn data features, but also improves training efficiency and accuracy. At the same time, in order to adapt to the training and processing of the network, the data needs to be segmented and grouped. Usually, continuous time queue data is divided according to a certain time window, for example, 1 hour as a sample window. Such operations can effectively process data under limited computing resources while retaining time series information, and capture the characteristic changes of physiological parameters in different time periods through multiple sample learning.

[0035] Building a 1D-FCN network is a key step in the implementation process. Parameters such as the number of convolutional layers, the size and number of convolution kernels, and activation functions need to be set. When the network is initialized, weights and biases are usually set randomly, but in some cases, pre-trained model parameters can also be used for initialization to speed up network convergence and improve performance.

[0036] After the data and network are prepared, the training process begins. For the heart rate value time queue data, it is input into the 1D-FCN network, and the convolution kernel slides on the time series and performs convolution operations with the data to obtain a series of feature maps. These feature maps are then processed by activation functions (such as ReLU functions) to enhance the nonlinear expression ability and gradually extract more advanced heart rate time series features. In a 1D-FCN network containing multiple convolutional layers, the first layer of convolution may capture the short-term fluctuation characteristics of the heart rate data. The subsequent layers further combine and abstract on this basis to form a complex feature representation, and finally output a sequence of heart rate time series feature vectors.

[0037] Similarly, the time queue data of blood pressure values, blood oxygen saturation values, and activity values ​​are also processed by their respective 1D-FCN network branches. These branches can share part of the network structure or be designed independently, depending on the system requirements. They all continuously optimize network parameters through processes such as forward propagation, loss calculation, and back propagation. During the training process, a suitable loss function (such as mean square error MSE) is defined to measure the difference between the network prediction and the true label, and the network weights and biases are adjusted through the back propagation algorithm according to the loss value. Use a large amount of sample data for multiple iterative training until the predetermined stopping condition is reached, such as the loss value converges or the maximum number of training rounds is reached.

[0038] After being processed by the trained 1D-FCN network, the corresponding time series feature vector sequence is output at the last layer of the network. Each vector in the sequence of heart rate time series feature vectors integrates the fluctuation pattern and trend of the heart rate in a specific time window, and the same is true for the time series feature vector sequence of blood pressure, blood oxygen saturation and activity.

[0039] These valuable time series feature information extracted from the original physiological parameter time queue can effectively reflect the changing patterns of the patient's physiological state.

[0040] In an embodiment of the present application, the physiological parameter data timing feature cascade unit 132 is used to cascade each group of corresponding heart rate timing feature vectors, blood pressure timing feature vectors, blood oxygen saturation timing feature vectors and activity amount timing feature vectors in the sequence of the heart rate timing feature vectors, the sequence of the blood pressure timing feature vectors, the sequence of the blood oxygen saturation timing feature vectors and the sequence of the activity amount timing feature vectors to obtain a sequence of multi-dimensional physiological parameter fusion information representation vectors as the sequence of the multi-dimensional physiological parameter fusion information representation. It should be understood that each individual physiological parameter (such as heart rate, blood pressure, blood oxygen saturation and activity) can only provide one aspect of the patient's health status. Therefore, in order to construct a more comprehensive representation of the patient's health status and capture the overall pattern that a single parameter cannot reveal, in the technical solution of the present application, each group of corresponding heart rate timing feature vectors, blood pressure timing feature vectors, blood oxygen saturation timing feature vectors and activity timing feature vectors in the sequence of the heart rate timing feature vectors, the sequence of the blood pressure timing feature vectors, the sequence of the blood oxygen saturation timing feature vectors and the sequence of the activity timing feature vectors are cascaded to obtain a sequence of multi-dimensional physiological parameter fusion information representation vectors. In this way, the multi-dimensional fused information representation provides a richer feature space, which helps the machine learning model to better understand the patient's health status and provide a more comprehensive health assessment.

[0041] In the embodiment of the present application, the multi-dimensional physiological parameter dynamic propagation module 140 is used to perform gated writing of feature dynamic propagation on the sequence represented by the multi-dimensional physiological parameter fusion information to obtain the dynamic propagation feature of the multi-dimensional physiological parameter fusion information. It should be understood that the sequence of the vector representing the multi-dimensional physiological parameter fusion information not only contains the overall time trend, but also contains the short-term time series feature information, and different time points have different significance in the entire time series. Based on this, in the technical solution of the present application, the gated writing feature dynamic propagation mode is introduced to process the sequence represented by the multi-dimensional physiological parameter fusion information to obtain the dynamic propagation feature of the multi-dimensional physiological parameter fusion information. That is, the mechanism captures the overall trend (such as the long-term heart rate change pattern) and local details (such as instantaneous blood pressure fluctuations) when processing the vector sequence, which is crucial to understanding the patient's health status because it not only reflects the long-term trend, but also reveals important changes in the short term. At the same time, through the gating mechanism, the network realizes the effective screening and strengthening of information, can identify and enhance those features that are crucial to health assessment, and reduce the influence of noise or irrelevant information, so that the system can make more accurate judgments in a complex data environment, and improve the reliability and accuracy of the prediction.

[0042] Specifically, Figure 3 FIG. 1 is a block diagram of a multi-dimensional physiological parameter dynamic propagation module in an intelligent status monitoring system for patients after radiotherapy according to an embodiment of the present application. Figure 3 As shown, the multidimensional physiological parameter dynamic propagation module 140 includes: a multidimensional physiological parameter initial propagation unit 141, which is used to perform initial dynamic propagation on the sequence of the multidimensional physiological parameter fusion information representation vector to obtain an initial multidimensional physiological parameter fusion information feature dynamic propagation coding vector; a multidimensional physiological parameter private gating feature calculation unit 142, which is used to calculate the private gating features of each multidimensional physiological parameter fusion information representation vector in the sequence of the multidimensional physiological parameter fusion information representation vector based on the initial multidimensional physiological parameter fusion information feature dynamic propagation coding vector to obtain a sequence of gated-written multidimensional physiological parameter private feature coding vectors; a multidimensional physiological parameter time series dynamic aggregation representation generation unit 143, which is used to fuse the sequence of gated-written multidimensional physiological parameter private feature coding vectors and the initial multidimensional physiological parameter fusion information feature dynamic propagation coding vector to obtain a multidimensional physiological parameter fusion information dynamic propagation coding vector as a multidimensional physiological parameter fusion information dynamic propagation feature.

[0043] Specifically, in the embodiment of the present application, the multi-dimensional physiological parameter initial propagation unit is used to: input the sequence of the multi-dimensional physiological parameter fusion information representation vector into the sequence reasoning module based on the forward LSTM model to obtain the initial multi-dimensional physiological parameter fusion information feature dynamic propagation coding vector. The above process can be expressed as: ; ; in, is a sequence of multi-dimensional physiological parameter fusion information representation vectors, , , and are the first, second, and third vectors in the sequence of the multi-dimensional physiological parameter fusion information representation vector. and A multi-dimensional physiological parameter fusion information representation vector, is the forward LSTM encoding, It is the dynamic propagation coding vector of the initial multi-dimensional physiological parameter fusion information features.

[0044] It should be understood that the forward LSTM can effectively learn and remember the key features in the long time series through its internal gating mechanism (input gate, forget gate and output gate), and thus can capture the time dependency of the input sequence data. Here, by inputting the sequence of the multi-dimensional physiological parameter fusion information representation vector into the sequence reasoning module based on the forward LSTM model, it is possible to more comprehensively and accurately understand the changes of multiple physiological parameters over time, which helps to extract a more representative and connotative initial multi-dimensional physiological parameter fusion information feature dynamic propagation encoding vector, so as to fully reflect the complex evolution characteristics of the multi-dimensional physiological parameters in the time series.

[0045] Specifically, in the embodiment of the present application, the multidimensional physiological parameter private gated feature calculation unit is used to: calculate the private features of each multidimensional physiological parameter fusion information representation vector in the sequence of the multidimensional physiological parameter fusion information representation vector relative to the initial multidimensional physiological parameter fusion information feature dynamic propagation coding vector to obtain a sequence of multidimensional physiological parameter private feature coding vectors. This process can be expressed as: ; in, is the initial multi-dimensional physiological parameter fusion information feature dynamic propagation coding vector, is the first in the sequence of the multi-dimensional physiological parameter fusion information representation vector A multi-dimensional physiological parameter fusion information representation vector, It is subtracted by position point. Indicates taking the absolute value, is the point convolutional coding, is the weight matrix, is the activation function, yes The corresponding multidimensional physiological parameter weight vector, is the first in the sequence of multidimensional physiological parameter private feature encoding vectors A multi-dimensional physiological parameter private feature encoding vector; Each multidimensional physiological parameter private feature encoding vector in the sequence of the multidimensional physiological parameter private feature encoding vector is input into the gated writing unit based on the Softmax function to obtain the sequence of the gated writing multidimensional physiological parameter private feature encoding vector. The process can be expressed as: ; ; ; in, is the first in the sequence of multidimensional physiological parameter private feature encoding vectors A multi-dimensional physiological parameter private feature encoding vector, is matrix multiplication, yes The corresponding multi-dimensional physiological parameter private feature weight matrix, yes The corresponding multi-dimensional physiological parameter private feature bias vector, yes The node position in the sequence of the multi-dimensional physiological parameter private feature encoding vector, yes function, is the first in the sequence of multi-dimensional physiological parameter private feature activation vectors Multidimensional physiological parameter private feature activation vectors, For Perform mask operations. yes The eigenvalues ​​at each position in , is the preset threshold, is the first in the sequence of multi-dimensional physiological parameter private feature mask vectors A multi-dimensional physiological parameter private feature mask vector, , and are the first, second and third in the sequence of the multi-dimensional physiological parameter private feature encoding vector A multi-dimensional physiological parameter private feature encoding vector, , and are the first, second and third in the sequence of the multi-dimensional physiological parameter private feature mask vectors. A multi-dimensional physiological parameter private feature mask vector, It is a sequence of gated writes into the multi-dimensional physiological parameter private feature encoding vector.

[0046] It should be understood that in the sequence of multidimensional physiological parameter fusion information representation vectors, each vector corresponds to the comprehensive situation of physiological parameters at a specific time point and has its own unique characteristic performance. The initial multidimensional physiological parameter fusion information feature dynamic propagation coding vector is a holistic expression extracted by integrating the previous information of the entire sequence, reflecting relatively macroscopic and common characteristics. In order to mine the unique part at each time point that is different from the overall common characteristics, so as to more finely distinguish the difference in physiological state at different time points, it is necessary to calculate the private features of each multidimensional physiological parameter fusion information representation vector in the sequence of the multidimensional physiological parameter fusion information representation vector relative to the initial multidimensional physiological parameter fusion information feature dynamic propagation coding vector. By calculating the private features, it is possible to capture the content that was not focused on or even omitted in the previous propagation process, which helps to understand the patient's physiological state more comprehensively and accurately. Next, it is necessary to input each multidimensional physiological parameter private feature coding vector in the sequence of the multidimensional physiological parameter private feature coding vector into the gated write unit based on the Softmax function. Specifically, the gated write unit will determine that the features corresponding to the multi-dimensional physiological parameter private feature coding vectors that are assigned higher probability values ​​(that is, higher importance scores) are more representative and relevant, and thus decide to retain more of them and write them into the subsequent sequence of gated write multi-dimensional physiological parameter private feature coding vectors; while for features with lower probability values ​​(low importance scores), their influence will be weakened accordingly, or even directly ignored, and they will not be allowed to enter the subsequent feature sequence. That is, the gated write unit based on the Softmax function is used to determine whether each multi-dimensional physiological parameter private feature coding vector needs to be written into the initial propagation inference information of the feature sequence. In this way, it can be ensured that the sequence of gated write multi-dimensional physiological parameter private feature coding vectors obtained mainly contains those features that best reflect the key patient physiological information.

[0047] Finally, the sequence of the gated multidimensional physiological parameter private feature coding vector and the initial multidimensional physiological parameter fusion information feature dynamic propagation coding vector are fused to obtain the multidimensional physiological parameter temporal dynamic aggregation representation vector of the fusion symptom semantic information as the multidimensional physiological parameter temporal dynamic aggregation representation of the fusion symptom semantic information. The above process can be expressed as: ; in, is the initial multi-dimensional physiological parameter fusion information feature dynamic propagation coding vector, is a sequence of gated writes into the multi-dimensional physiological parameter private feature encoding vector, For cascade processing, It is the dynamic propagation coding vector of the multi-dimensional physiological parameter fusion information.

[0048] It should be understood that the initial multidimensional physiological parameter fusion information feature dynamic propagation coding vector focuses on reflecting the dynamic interaction mode of multidimensional physiological parameters at each time point, and specifically captures the information at the interaction level such as the correlation between physiological parameters at different time points and the overall trend of change over time. The sequence of gated writing of multidimensional physiological parameter private feature coding vectors emphasizes the unique attributes of multidimensional physiological parameters at each time point. After gated screening, it retains the most representative and individualized features, that is, it highlights the uniqueness of each time point from other time points. The multidimensional physiological parameter time series dynamic aggregation representation vector of fusion symptom semantic information obtained by fusing the two not only includes the dynamic interaction mode of interaction and mutual influence between multidimensional physiological parameters at different time points, but also accurately captures the unique attributes of each time point itself, so as to achieve a comprehensive and detailed characterization of the patient's physiological parameters.

[0049] In an embodiment of the present application, the symptom report semantic encoding module 150 is used to semantically encode the symptom report text description to obtain the semantic encoding features of the symptom report text. Specifically, in an embodiment of the present application, the symptom report semantic encoding module is used to: pass the symptom report text description through a semantic encoder including a Bert model to obtain a symptom report text semantic encoding feature vector as the symptom report text semantic encoding feature. It should be understood that the symptom report text description contains a large amount of text information, and there is a semantic correlation between the contexts between these text information. Therefore, in the technical solution of the present application, the symptom report text description is passed through a semantic encoder including a Bert model to capture the contextual relationship and semantic meaning between each word in the symptom description, and obtain a symptom report text semantic encoding feature vector. In particular, the Bert model is a pre-trained language model, which is based on the Transformer architecture and uses a bidirectional encoder structure to process natural language tasks. Unlike traditional unidirectional language models, BERT understands the meaning of each word in a sentence by simultaneously considering contextual information, thereby more accurately capturing the semantic relationship in the text. Therefore, the Bert model can learn deeper features and rich semantic information patterns in the text through a multi-layer Transformer architecture, which helps to discover the complex associations implicit in the symptom descriptions, such as the causal relationship or concomitant phenomena between different symptoms, so as to better understand and analyze the patient's symptom information.

[0050] In an embodiment of the present application, the multidimensional physiological parameter time series dynamic aggregation module 160 is used to perform attention feature interactive fusion on the multidimensional physiological parameter fusion information dynamic propagation feature and the symptom report text semantic coding feature to obtain a multidimensional physiological parameter time series dynamic aggregation representation of the fusion symptom semantic information. Specifically, in an embodiment of the present application, the multidimensional physiological parameter time series dynamic aggregation module is used to: pass the multidimensional physiological parameter fusion information dynamic propagation coding vector and the symptom report text semantic coding feature vector through a feature interactive fusion device based on an X-linear attention network to obtain a multidimensional physiological parameter time series dynamic aggregation representation vector of the fusion symptom semantic information as the multidimensional physiological parameter time series dynamic aggregation representation of the fusion symptom semantic information. It should be understood that the multidimensional physiological parameter fusion information dynamic propagation feature provides objective quantitative data on the patient's health status, while the symptom report text semantic coding feature captures the patient's symptom semantic description. Therefore, in order to capture the complex interactions between different physiological parameters and between them and symptoms, and provide a more detailed health assessment, in the technical solution of the present application, the multi-dimensional physiological parameter fusion information dynamic propagation encoding vector and the symptom report text semantic encoding feature vector are passed through a feature interaction fusion device based on an X-linear attention network to obtain a multi-dimensional physiological parameter time series dynamic aggregation representation vector that integrates symptom semantic information. Specifically, the proposed X-linear attention network adopts a bilinear pooling method to effectively capture the nonlinear relationship between physiological parameters and symptom semantic descriptions by calculating the outer product of two sets of feature vectors and flattening them into a one-dimensional vector. This design enables the model to recognize more complex patterns. In addition, by analyzing the interaction of different-order features (such as multi-dimensional physiological parameters and symptom descriptions), the generated high-dimensional feature representation not only retains the data of the original features, but also shows the complex relationship between them, thereby enhancing the expression of the output features. In actual operation, X-linear attention can be integrated into a component of the Transformer structure, or integrated into an existing model as an additional layer. It usually receives one or more sets of input features (such as multi-dimensional physiological parameter features and symptom description features), and then generates new feature expressions through bilinear pooling operations to capture the complex interactions between the two, and integrates these new features into the subsequent processing flow. By designing appropriate fusion strategies, such as weighted averaging or further processing through a fully connected layer after concatenation, it ensures that the two types of feature vectors can be effectively fused while retaining their respective core information.

[0051] In the embodiment of the present application, the abnormal state judgment module 170 is used to judge whether the state of the patient object after radiotherapy is abnormal based on the multi-dimensional physiological parameter time series dynamic aggregation representation of the fused symptom semantic information. Specifically, in the embodiment of the present application, the abnormal state judgment module is used to: pass the multi-dimensional physiological parameter time series dynamic aggregation representation vector of the fused symptom semantic information through a patient state monitor based on a classifier to obtain a monitoring result, and the monitoring result is used to indicate whether the state of the patient object after radiotherapy is abnormal. It should be understood that the classifier is trained based on a large amount of existing annotated data (multi-dimensional physiological parameter time series data corresponding to normal and abnormal states and symptom report text descriptions), and has the ability to distinguish different categories of feature patterns, so that it can accurately judge whether the patient is more likely to be in a normal or abnormal state based on the multi-dimensional physiological parameter time series dynamic aggregation representation vector of the fused symptom semantic information currently input. Once an abnormality is found in the patient's condition, it is equivalent to issuing a risk warning signal, and medical personnel can intervene in time. For example, for patients who have just experienced slight fluctuations in physiological indicators and accompanied by corresponding symptoms, they can adjust the rehabilitation plan after radiotherapy in advance, add necessary examination items, etc. This will help to take effective measures when the abnormal situation is still in the early stages and relatively easy to control, so as to avoid further deterioration of the disease and ensure the health and safety of patients. In particular, in a specific embodiment of the present application, the multi-dimensional physiological parameter time series dynamic aggregation representation vector of the fused symptom semantic information is passed through a patient status monitor based on a classifier to obtain a monitoring result, and the monitoring result is used to indicate whether the state of the patient object after radiotherapy is abnormal, including: using the fully connected layer of the classifier to fully connect the multi-dimensional physiological parameter time series dynamic aggregation representation vector of the fused symptom semantic information to obtain a multi-dimensional physiological parameter time series dynamic aggregation fully connected encoding feature vector of the fused symptom semantic information; inputting the multi-dimensional physiological parameter time series dynamic aggregation fully connected encoding feature vector of the fused symptom semantic information into the Softmax classification function of the classifier to obtain the probability value of the multi-dimensional physiological parameter time series dynamic aggregation representation vector of the fused symptom semantic information belonging to each classification label, and the classification labels include those for indicating that the state of the patient object after radiotherapy is abnormal and those for indicating that the state of the patient object after radiotherapy is not abnormal; and determining the classification label corresponding to the largest of the probability values ​​as the monitoring result.

[0052] In a preferred example, considering that the multidimensional physiological parameter fusion information dynamic propagation coding vector and the symptom report text semantic coding feature vector respectively represent the time-domain dynamic propagation temporal correlation characteristics of the patient's multidimensional physiological parameters and the coding semantic characteristics of the patient's symptom report text description, when performing feature interaction based on the X-linear attention network, due to the bilinear attention difference caused by the heteromodal feature distribution difference of the source data, the multidimensional physiological parameter temporal dynamic aggregation representation vector of the fused symptom semantic information will also have a probabilistic convergence and divergence based on different attention enhancement interactivity, thereby affecting the accuracy of the monitoring results obtained by the classifier-based patient status monitor.

[0053] Therefore, when the multi-dimensional physiological parameter time series dynamic aggregation representation vector of the fusion symptom semantic information passes through the classifier-based patient status monitor, the present application first optimizes the multi-dimensional physiological parameter time series dynamic aggregation representation vector of the fusion symptom semantic information, including the following steps: Determine the multi-dimensional physiological parameter time series dynamic aggregation representation vector of the fusion symptom semantic information through the multi-dimensional physiological parameter time series dynamic aggregation distribution probability value of the fusion symptom semantic information of the patient status monitor based on the classifier , where the multidimensional physiological parameter time series dynamic aggregation distribution probability value of the fusion symptom semantic information Indicates the probability that the patient's status after radiotherapy is abnormal; The multi-dimensional physiological parameter time series of the fusion symptom semantic information is dynamically aggregated to represent each eigenvalue in the vector Perform maximum normalization to obtain the time series dynamic aggregation probability feature value of multi-dimensional physiological parameters that integrate symptom semantic information ; The probability value of the multi-dimensional physiological parameter time series dynamic aggregation distribution of the fusion symptom semantic information is subtracted from one Calculate the time series dynamic aggregation probability feature value of each multi-dimensional physiological parameter of the fusion symptom semantic information for the index The power function of is used to obtain the convergence characteristic value of the multi-dimensional physiological parameter time series dynamic aggregation probability of the first fusion symptom semantic information. The process can be expressed as: ; in, The characteristic values ​​of each position in the multi-dimensional physiological parameter time series dynamic aggregation representation vector of the fusion symptom semantic information are as follows, for The multi-dimensional physiological parameter time series dynamic aggregation probability characteristic value of the fused symptom semantic information obtained by maximum normalization, is the time series dynamic aggregation distribution probability value of the multi-dimensional physiological parameter fused with symptom semantic information, for The corresponding multi-dimensional physiological parameter time series dynamic aggregation probability convergence eigenvalue of the first fusion symptom semantic information; The probability value of the multi-dimensional physiological parameter time series dynamic aggregation distribution of the fusion symptom semantic information Calculate a time series dynamic aggregation probability feature value of each multi-dimensional physiological parameter of the fusion symptom semantic information for the index The power function of the difference is used to obtain the multi-dimensional physiological parameter time series dynamic aggregation probability convergence eigenvalue of the second fusion symptom semantic information. The process can be expressed as: ; in, for The corresponding multi-dimensional physiological parameter time series dynamic aggregation probability convergence characteristic value of the second fusion symptom semantic information; The multi-dimensional physiological parameter time series of each fusion symptom semantic information is dynamically aggregated into probability feature values. The probability value of the multi-dimensional physiological parameter time series dynamic aggregation distribution minus the fusion symptom semantic information The difference is multiplied to obtain the multi-dimensional physiological parameter time series dynamic aggregation probability limiting feature value of the first fusion symptom semantic information. The process can be expressed as: ; in, for The corresponding multi-dimensional physiological parameter time series dynamic aggregation probability limiting characteristic value of the first fusion symptom semantic information; A multi-dimensional physiological parameter time series dynamic aggregation probability feature value is combined with each of the fusion symptom semantic information The difference between the time series dynamic aggregation distribution probability value of the multi-dimensional physiological parameter of the fusion symptom semantic information Multiply them to obtain the time series dynamic aggregation probability limiting feature value of the multi-dimensional physiological parameter of the second fusion symptom semantic information. The process can be expressed as: ; in, for The corresponding multi-dimensional physiological parameter time series dynamic aggregation probability defined characteristic value of the second fusion symptom semantic information; After multiplying the convergence eigenvalue of the probability of dynamic aggregation of multidimensional physiological parameters of the first fused symptom semantic information with the convergence eigenvalue of the probability of dynamic aggregation of multidimensional physiological parameters of the second fused symptom semantic information, further adding the limited eigenvalue of the probability of dynamic aggregation of multidimensional physiological parameters of the first fused symptom semantic information and the limited eigenvalue of the probability of dynamic aggregation of multidimensional physiological parameters of the second fused symptom semantic information to obtain the optimized eigenvalue of the representation vector of the dynamic aggregation of multidimensional physiological parameters of the fused symptom semantic information. This process can be expressed as: ; in, for The corresponding optimized eigenvalues.

[0054] That is, by using the cross entropy form power series of the regression probability of the multi-dimensional physiological parameter time series dynamic aggregation representation vector that integrates the symptom semantic information as the probability distribution convergence limit, the regression probability convergence approximation is performed on the basis of the combination of the feature set distribution and the probability density distribution of the multi-dimensional physiological parameter time series dynamic aggregation representation vector that integrates the symptom semantic information, so as to avoid the fatal potential divergence of the multi-dimensional physiological parameter time series dynamic aggregation representation vector of the symptom semantic information. Under the premise of conservative convergence control, the probability distribution cross entropy is further used as the objective function to guide the limited recovery strategy, so as to achieve the common agility of the convergence of the probability convergence and divergence of the feature set of the multi-dimensional physiological parameter time series dynamic aggregation representation vector of the symptom semantic information to the probability density distribution space, thereby improving the accuracy of the monitoring results obtained by the classifier-based patient status monitor.

[0055] In summary, the intelligent state monitoring system 100 for patients after radiotherapy based on the embodiment of the present application is explained, which uses data analysis and coding technology based on machine learning to extract time series features and fuse parameters of various multidimensional physiological parameters of patients after radiotherapy, and then dynamically propagates gated features of the fusion information representation of each multidimensional physiological parameter, and semantically encodes the text description of the patient's symptom report, so as to realize intelligent judgment of the state of the patient after radiotherapy based on the interactive fusion representation between the dynamic propagation features of the multidimensional physiological parameter fusion information and the semantic coding features of the symptom report text. In this way, the dynamic changes of the patient's state can be captured more comprehensively and meticulously, and the internal connection and interaction between the various indicators are fully considered, which helps to more accurately evaluate the overall health status of the patient.

[0056] Figure 4 FIG. 1 is a flow chart of a method for monitoring the intelligent status of a patient after radiotherapy according to an embodiment of the present application. Figure 4As shown, in the method for intelligent status monitoring of patients after radiotherapy, the method includes: S110, collecting multi-dimensional physiological parameter time series data of the patient object after radiotherapy from the sensor group, the multi-dimensional physiological parameter time series data including the time queue of heart rate value, the time queue of blood pressure value, the time queue of blood oxygen saturation and the time queue of activity; S120, obtaining the text description of the patient's symptom report; S130, performing time series feature analysis and multi-dimensional parameter fusion on the time queue of heart rate value, the time queue of blood pressure value, the time queue of blood oxygen saturation and the time queue of activity to obtain a sequence represented by multi-dimensional physiological parameter fusion information; S140, S150, performing gated writing of feature dynamic propagation of the sequence of multi-dimensional physiological parameter fusion information representation to obtain multi-dimensional physiological parameter fusion information dynamic propagation features; S150, performing semantic encoding of the symptom report text description to obtain symptom report text semantic encoding features; S160, performing attention feature interactive fusion of the multi-dimensional physiological parameter fusion information dynamic propagation features and the symptom report text semantic encoding features to obtain a multi-dimensional physiological parameter time series dynamic aggregation representation of fused symptom semantic information; S170, judging whether there is abnormality in the state of the patient object after radiotherapy based on the multi-dimensional physiological parameter time series dynamic aggregation representation of the fused symptom semantic information.

[0057] Here, those skilled in the art can understand that the specific operations of each step in the above-mentioned method for monitoring the intelligent status of patients after radiotherapy have been described in the above reference. Figures 1 to 3 The invention has been introduced in detail in the description of the intelligent status monitoring system for patients after radiotherapy, and therefore, its repeated description will be omitted.

[0058] In summary, the intelligent state monitoring method for patients after radiotherapy based on the embodiment of the present application is explained, which uses data analysis and coding technology based on machine learning to extract time series features and fuse parameters of various multidimensional physiological parameters of patients after radiotherapy, and then dynamically propagate gated features of the fusion information representation of each multidimensional physiological parameter, and semantically encode the text description of the patient's symptom report, so as to realize intelligent judgment of the state of the patient after radiotherapy based on the interactive fusion representation between the dynamic propagation characteristics of the multidimensional physiological parameter fusion information and the semantic coding characteristics of the symptom report text. In this way, the dynamic changes of the patient's state can be captured more comprehensively and meticulously, and the intrinsic connection and interaction between the various indicators are fully considered, which helps to more accurately evaluate the overall health status of the patient.

Claims

1. An intelligent status monitoring system for patients after radiotherapy, characterized in that: include: A physiological parameter data acquisition module, used to collect multi-dimensional physiological parameter time series data of a patient subject after radiotherapy from a sensor group, wherein the multi-dimensional physiological parameter time series data includes a time queue of a heart rate value, a time queue of a blood pressure value, a time queue of blood oxygen saturation, and a time queue of activity amount; A symptom report description acquisition module is used to obtain a text description of a patient's symptom report; A multi-dimensional parameter fusion module, used for performing time series feature analysis and multi-dimensional parameter fusion on the time queue of the heart rate value, the time queue of the blood pressure value, the time queue of the blood oxygen saturation and the time queue of the activity amount to obtain a sequence represented by multi-dimensional physiological parameter fusion information; A multi-dimensional physiological parameter dynamic propagation module, used for performing gated writing feature dynamic propagation on the sequence represented by the multi-dimensional physiological parameter fusion information to obtain the multi-dimensional physiological parameter fusion information dynamic propagation feature; A symptom report semantic encoding module, used for semantically encoding the symptom report text description to obtain semantic encoding features of the symptom report text; A multi-dimensional physiological parameter time series dynamic aggregation module is used to perform attention feature interactive fusion on the dynamic propagation characteristics of the multi-dimensional physiological parameter fusion information and the semantic encoding characteristics of the symptom report text to obtain a multi-dimensional physiological parameter time series dynamic aggregation representation of the fusion symptom semantic information; The abnormal state judgment module is used to judge whether the state of the patient object after radiotherapy is abnormal based on the multi-dimensional physiological parameter time series dynamic aggregation representation fused with the symptom semantic information.

2. The intelligent status monitoring system for patients after radiotherapy according to claim 1, characterized in that: The multi-dimensional parameter fusion module comprises: A physiological parameter data time series feature extraction unit, used for passing the time queue of the heart rate value, the time queue of the blood pressure value, the time queue of the blood oxygen saturation and the time queue of the activity amount through a physiological parameter time series feature extractor based on a 1D-FCN network to obtain a sequence of heart rate time series feature vectors, a sequence of blood pressure time series feature vectors, a sequence of blood oxygen saturation time series feature vectors and a sequence of activity amount time series feature vectors; The physiological parameter data timing feature cascade unit is used to cascade each group of corresponding heart rate timing feature vectors, blood pressure timing feature vectors, blood oxygen saturation timing feature vectors and activity amount timing feature vectors in the sequence of the heart rate timing feature vectors, the sequence of the blood pressure timing feature vectors, the sequence of the blood oxygen saturation timing feature vectors and the sequence of the activity amount timing feature vectors to obtain a sequence of multi-dimensional physiological parameter fusion information representation vectors as the sequence of the multi-dimensional physiological parameter fusion information representation.

3. The intelligent status monitoring system for patients after radiotherapy according to claim 2, characterized in that: The multi-dimensional physiological parameter dynamic propagation module comprises: A multi-dimensional physiological parameter initial propagation unit, used for performing initial dynamic propagation on the sequence of the multi-dimensional physiological parameter fusion information representation vector to obtain an initial multi-dimensional physiological parameter fusion information feature dynamic propagation coding vector; A multi-dimensional physiological parameter private gating feature calculation unit, used to calculate the private gating features of each multi-dimensional physiological parameter fusion information representation vector in the sequence of multi-dimensional physiological parameter fusion information representation vectors based on the initial multi-dimensional physiological parameter fusion information feature dynamic propagation coding vector to obtain a sequence of gated multi-dimensional physiological parameter private feature coding vectors; A multi-dimensional physiological parameter time series dynamic aggregation representation generation unit is used to fuse the sequence of the gated-written multi-dimensional physiological parameter private feature coding vector and the initial multi-dimensional physiological parameter fusion information feature dynamic propagation coding vector to obtain a multi-dimensional physiological parameter fusion information dynamic propagation coding vector as a multi-dimensional physiological parameter fusion information dynamic propagation feature.

4. The intelligent status monitoring system for patients after radiotherapy according to claim 3, characterized in that: The multi-dimensional physiological parameter initial propagation unit is used to: input the sequence of the multi-dimensional physiological parameter fusion information representation vector into the sequence reasoning module based on the forward LSTM model to obtain the initial multi-dimensional physiological parameter fusion information feature dynamic propagation coding vector.

5. The intelligent status monitoring system for patients after radiotherapy according to claim 4, characterized in that: The multi-dimensional physiological parameter private gated feature calculation unit is used to: Calculating the private features of each multidimensional physiological parameter fusion information representation vector in the sequence of multidimensional physiological parameter fusion information representation vectors relative to the initial multidimensional physiological parameter fusion information feature dynamic propagation coding vector to obtain a sequence of multidimensional physiological parameter private feature coding vectors; Each multidimensional physiological parameter private feature coding vector in the sequence of multidimensional physiological parameter private feature coding vectors is input into a gated writing unit based on a Softmax function to obtain the sequence of gated writing multidimensional physiological parameter private feature coding vectors.

6. The intelligent status monitoring system for patients after radiotherapy according to claim 5, characterized in that: The symptom report semantic encoding module is used to: pass the symptom report text description through a semantic encoder including a Bert model to obtain a symptom report text semantic encoding feature vector as the symptom report text semantic encoding feature.

7. The intelligent status monitoring system for patients after radiotherapy according to claim 6, characterized in that: The multidimensional physiological parameter time series dynamic aggregation module is used to: pass the multidimensional physiological parameter fusion information dynamic propagation coding vector and the symptom report text semantic coding feature vector through a feature interaction fusion device based on an X-linear attention network to obtain a multidimensional physiological parameter time series dynamic aggregation representation vector of fused symptom semantic information as the multidimensional physiological parameter time series dynamic aggregation representation of the fused symptom semantic information.

8. The intelligent status monitoring system for patients after radiotherapy according to claim 7, characterized in that: The abnormal state judgment module is used to: pass the multi-dimensional physiological parameter time series dynamic aggregation representation vector that integrates symptom semantic information through a classifier-based patient state monitor to obtain a monitoring result, and the monitoring result is used to indicate whether the state of the patient object after radiotherapy is abnormal.

9. A method for monitoring the intelligent status of a patient after radiotherapy, characterized in that: include: Collecting multi-dimensional physiological parameter time series data of the patient subject after radiotherapy from the sensor group, wherein the multi-dimensional physiological parameter time series data includes a time queue of heart rate value, a time queue of blood pressure value, a time queue of blood oxygen saturation and a time queue of activity amount; Obtain text descriptions of patients' symptom reports; Performing time series feature analysis and multi-dimensional parameter fusion on the time queue of the heart rate value, the time queue of the blood pressure value, the time queue of the blood oxygen saturation, and the time queue of the activity amount to obtain a sequence represented by multi-dimensional physiological parameter fusion information; Perform gated writing of feature dynamic propagation on the sequence represented by the multi-dimensional physiological parameter fusion information to obtain the multi-dimensional physiological parameter fusion information dynamic propagation feature; Performing semantic coding on the symptom report text description to obtain semantic coding features of the symptom report text; Performing attention feature interactive fusion on the dynamic propagation features of the multi-dimensional physiological parameter fusion information and the semantic encoding features of the symptom report text to obtain a multi-dimensional physiological parameter temporal dynamic aggregation representation of the fusion symptom semantic information; Based on the multi-dimensional physiological parameter time series dynamic aggregation representation fused with symptom semantic information, it is judged whether the state of the patient object after radiotherapy is abnormal.

10. The method for intelligent status monitoring of patients after radiotherapy according to claim 9, characterized in that: Performing time series feature analysis and multi-dimensional parameter fusion on the time queue of the heart rate value, the time queue of the blood pressure value, the time queue of the blood oxygen saturation, and the time queue of the activity amount to obtain a sequence represented by multi-dimensional physiological parameter fusion information, including: The time queue of the heart rate value, the time queue of the blood pressure value, the time queue of the blood oxygen saturation and the time queue of the activity amount are passed through a physiological parameter time series feature extractor based on a 1D-FCN network to obtain a sequence of heart rate time series feature vectors, a sequence of blood pressure time series feature vectors, a sequence of blood oxygen saturation time series feature vectors and a sequence of activity amount time series feature vectors; Each group of corresponding heart rate timing feature vectors, blood pressure timing feature vectors, blood oxygen saturation timing feature vectors and activity amount timing feature vectors in the sequence of heart rate timing feature vectors, the sequence of blood pressure timing feature vectors, the sequence of blood oxygen saturation timing feature vectors and the sequence of activity amount timing feature vectors are cascaded to obtain a sequence of multi-dimensional physiological parameter fusion information representation vectors as the sequence of multi-dimensional physiological parameter fusion information representation.

Images