ICU patient physiological information detection method and system

Abstract

The invention discloses an ICU patient physiological information detection method and system. The multiple patient time physiological index sets cut by the patient are used for finding the physiological indexes of the patient in the common state, and the physical states of the patient changing at multiple time points are judged according to the overall physical state, composed of the multiple physiological indexes, of the patient by adopting the constructed matrix. And according to the physiological indexes of a plurality of time points segmented into a certain class, converting the physiological indexes into Fourier images so as to judge the change condition of the physiological indexes of a single class. The patient change network detected in real time, the time convolution network for calculating the common state and the three-dimensional reconstruction network are trained and detected through multiple types of physiological indexes of multiple time points of the user. The method is used for calculating the physiological indexes in a common state for specific diseases. And the patient can be trained to change the network, so that the technical effect of more accurately obtaining the improvement, deterioration and common states of the user compared with the user under the condition of reducing data is achieved.

Description

Technical Field

[0001] The present invention relates to the field of computer technology, and in particular to a method and system for detecting physiological information of ICU patients. Background Art

[0002] At present, in the field of critical care, monitoring of patients' physiological index parameters is a necessary means to guide clinical medical staff to quantitatively evaluate, diagnose and treat critically ill patients. After the patient enters the operating room or intensive care unit (ICU), medical staff will connect a monitor that can monitor physiological indicators such as heart rate, blood pressure, blood oxygen, respiration and body temperature to the human body, and monitor the patient according to the patient's physiological indicators, so that medical staff can observe the patient's condition.

[0003] At present, all existing ICU alarms are based on threshold diagnosis alarms, which use the patient's physiological indicators to reach a certain threshold to sound an alarm. However, because the physiological indicators of patients in ICU are generally different from those of normal people, and they are different due to different symptoms. Even if each patient is in a normal state, his or her physiological indicators will be different from normal physiological indicators due to his or her own symptoms. Therefore, it is necessary to judge the physiological indicators of the normal state plus the changes in their physiological indicators for each patient with a specific disease according to the condition of the disease they have. Summary of the invention

[0004] The purpose of the present invention is to provide a method and system for detecting physiological information of ICU patients to solve the above-mentioned problems existing in the prior art.

[0005] In a first aspect, an embodiment of the present invention provides a method for detecting physiological information of an ICU patient, comprising:

[0006] Obtaining a set of physiological indicators of a patient at multiple time points; the set of physiological indicators of a patient includes multiple types of physiological indicators of a patient detected by a monitor connected to a human body; the set of physiological indicators of a patient at multiple time points is multiple types of physiological indicators of a patient detected at multiple time points before and including the current time point;

[0007] The patient's physiological indicators at the multiple time points are divided according to a fixed time length, and the time step is 1, to obtain multiple patient time physiological index sets and a current physiological indicator set; the patient time physiological index set includes multiple types of physiological indicators of the patient detected at multiple time points of a fixed time length; the current physiological indicator set is a patient time physiological index set including the physiological indicators detected at the current time point in the multiple patient time physiological index sets;

[0008] Based on the multiple patient time physiological index sets, determine the physiological indexes of the patient's normal state when sick, and obtain a normal physiological index set, a current physiological Fourier image and a normal Fourier image;

[0009] Obtaining a trained patient change network; the patient change network comprises a storage structure and a detection structure;

[0010] Storing the common physiological index set and the common Fourier image in a storage structure of a patient change network;

[0011] Based on the current set of physiological indicators and the current physiological Fourier image, the changes of the patient's physiological indicators are judged through the patient change network to obtain the patient's physiological state; the patient's physiological state includes three states: improvement, deterioration and normal;

[0012] If the patient's physiological state is improving, an improvement signal is sent; if the patient's physiological state is deteriorating, a deterioration signal is sent.

[0013] Optionally, the detection structure of the patient change network includes a first neural network, a first convolutional network, a physiological characteristic discrimination network and a Fourier discriminant network.

[0014] Optionally, the patient change network is a trained patient change network obtained by training with a plurality of training data;

[0015] The training data is to use one of the multiple patient time physiological index sets as the training current patient time physiological index set, and to use the Fourier image transformed from the training current patient physiological index set as the training current Fourier image;

[0016] Based on the set of current training patient physiological indicators and the current training Fourier image, the changes of the patient's physiological indicators are judged through the patient change network to obtain the training patient's physiological state;

[0017] Obtaining a marked physiological state; the marked physiological state indicates a marked improvement, deterioration, or normal state;

[0018] Calculate the loss of the training patient's physiological state and the marked physiological state to obtain a physiological state loss value;

[0019] If the physiological state loss value is less than the loss threshold, backpropagation is used to train the patient change network.

[0020] Optionally, the determining the changes of the patient's physiological indicators through a patient change network based on the current physiological indicator set and the current physiological Fourier image to obtain the patient's physiological state includes:

[0021] Obtaining normal physiological indicators; the normal physiological indicators refer to physiological indicators without any disease;

[0022] Inputting the values ​​in the current physiological indicator set into the first neural network, extracting the features of the physiological indicators at the current time point, and obtaining the current physiological indicator features;

[0023] Inputting the values ​​in the common physiological index set in the storage structure into the first neural network, extracting the features of the physiological indexes of the patient's common state, and obtaining the features of the common physiological indexes;

[0024] Inputting the normal physiological index into a first neural network, extracting the characteristics of the normal physiological index, and obtaining the characteristics of the normal physiological index;

[0025] Inputting the current physiological index feature and the common physiological index feature into a physiological feature discrimination network, extracting the feature of the physiological index change, and obtaining the physiological index change feature;

[0026] Inputting the current physiological Fourier image into a first convolutional network, extracting the features of the physiological indicators represented in the Fourier image at the current time point, and obtaining the current Fourier features;

[0027] Inputting the common Fourier image in the storage structure into the second convolutional network, extracting the features of the physiological indicators of the patient's common state in the Fourier image, and obtaining the common Fourier features;

[0028] Inputting the current Fourier feature and the common Fourier feature into a Fourier discriminant network, extracting the feature of the change of the physiological index, and obtaining the Fourier change feature;

[0029] The physiological index change characteristics, normal physiological index characteristics and Fourier change characteristics are input into the discriminant neural network, and judged as one of the three states of improvement, deterioration and normal to obtain the patient's physiological state.

[0030] Optionally, judging the physiological indexes of the patient's normal state when sick based on the multiple patient time physiological index sets to obtain the normal physiological index set, the current physiological Fourier image and the normal Fourier image includes:

[0031] Based on the multiple patient physiological index sets, a current physiological Fourier image and a common Fourier image are obtained through a time convolution network and a three-dimensional reconstruction network;

[0032] Based on the plurality of sets of physiological indicators of the patients, determining the changes of the physiological characteristics of the user jointly constructed by the plurality of types of physiological indicators, and obtaining a common change matrix;

[0033] Based on the common change matrix, a set of common physiological indicators is obtained.

[0034] Optionally, the obtaining of the current physiological Fourier image and the ordinary Fourier image based on the multiple patient physiological indicator sets through a time convolution network and a three-dimensional reconstruction network includes:

[0035] According to the patient's physiological indicator set, the set is divided into a plurality of time physiological category sets according to the categories of the physiological indicators;

[0036] The temporal physiological category set is subjected to Fourier transformation to determine the change of the category corresponding to a physiological indicator at multiple time points, and obtain a physiological category Fourier image; multiple temporal physiological category sets correspond to multiple physiological category Fourier images;

[0037] The physiological category Fourier image corresponding to the current time point is used as the current physiological Fourier image.

[0038] The multiple physiological category Fourier images are superimposed to obtain a three-dimensional physiological category image; multiple patient physiological indicator sets are correspondingly obtained to obtain multiple three-dimensional physiological category images;

[0039] Inputting the multiple three-dimensional physiological category images into a temporal convolutional network, extracting features of changes between the multiple three-dimensional physiological category images, and obtaining a three-dimensional feature map;

[0040] The three-dimensional feature map is input into a three-dimensional reconstruction network to obtain a common Fourier image.

[0041] Optionally, the training method of the temporal convolutional network and the three-dimensional reconstruction network is:

[0042] Obtaining a plurality of training images; the plurality of training images are three-dimensional physiological category images of historical changes in physiological indicators of patients;

[0043] Obtaining a labeled image; the labeled image is a training image of a patient in a normal physiological state among a plurality of training images;

[0044] A temporal convolutional network and a three-dimensional reconstruction network are jointly trained according to the multiple training images and the annotated images.

[0045] Optionally, the rows of the common change matrix represent time points from morning to night; the rows of the common change matrix represent categories of physiological indicators.

[0046] Optionally, obtaining a set of common physiological indicators based on the common change matrix includes:

[0047] Obtaining a first two-dimensional convolution kernel; the number of rows of the first two-dimensional convolution kernel is 1, and the number of columns is n; n is a natural number greater than 0; n represents the number of categories of physiological indicators;

[0048] Convolving the first two-dimensional convolution kernel with a row in the common change matrix with a step size of 1 to extract features of multiple categories of physiological indicators to obtain a first convolution feature;

[0049] Inputting the first convolution feature into a matrix change network, extracting the change features of multiple types of physiological indicators at multiple time points, and obtaining a second convolution feature;

[0050] The second convolution feature is input into the reconstructed neural network, the feature is converted into data, and a set of common physiological indicators is obtained.

[0051] In a second aspect, an embodiment of the present invention provides an ICU patient physiological information detection system, comprising:

[0052] Data acquisition module: obtains a set of physiological indicators of a patient at multiple time points; the set of physiological indicators of a patient includes multiple types of physiological indicators of a patient detected by a monitor connected to a human body; the set of physiological indicators of a patient at multiple time points is multiple types of physiological indicators of a patient detected at multiple time points before and including the current time point;

[0053] Segmentation module: segment the patient's physiological indicators at the multiple time points according to a fixed time length, and the time step is 1, to obtain multiple patient time physiological index sets and a current physiological index set; the patient time physiological index set includes multiple types of physiological indicators of the patient detected at multiple time points of a fixed time length; the current physiological index set is a patient time physiological index set including the physiological indicators detected at the current time point in the multiple patient time physiological index sets;

[0054] Normal state physiological index detection module: based on the multiple patient time physiological index sets, determine the patient's normal state physiological index when sick, and obtain the normal physiological index set, the current physiological Fourier image and the normal Fourier image;

[0055] Network acquisition module: obtains a trained patient change network; the patient change network includes a storage structure and a detection structure;

[0056] Storage module: storing the common physiological index set and the common Fourier image in the storage structure of the patient change network;

[0057] Detection module: Based on the current physiological index set and the current physiological Fourier image, the patient change network is used to determine the changes in the patient's physiological indexes to obtain the patient's physiological state; the patient's physiological state includes three states: improvement, deterioration and normal;

[0058] Early warning module: If the patient's physiological state improves, an improvement signal is sent; if the patient's physiological state deteriorates, a deterioration signal is sent.

[0059] Compared with the prior art, the embodiments of the present invention achieve the following beneficial effects:

[0060] The embodiment of the present invention also provides a method and system for detecting physiological information of ICU patients, the method comprising: obtaining a set of patient physiological indicators at multiple time points. The set of patient physiological indicators includes multiple types of physiological indicators of patients detected by a monitor connected to the human body. The set of patient physiological indicators at multiple time points is multiple types of physiological indicators of patients detected at multiple time points before and including the current time point. The patient physiological indicators at the multiple time points are divided according to a fixed time length, and the time step is 1, to obtain multiple patient time physiological index sets and a current physiological index set. The patient time physiological index set includes multiple types of physiological indicators of patients detected at multiple time points of a fixed time length. The current physiological index set is a patient time physiological index set including the physiological indicators detected at the current time point in multiple patient time physiological index sets. Based on the multiple patient time physiological index sets, the physiological indicators of the patient's normal state when sick are judged to obtain a normal physiological index set, a current physiological Fourier image and a normal Fourier image. A trained patient change network is obtained. The patient change network includes a storage structure and a detection structure. The normal physiological index set and the normal Fourier image are stored in the storage structure of the patient change network. Based on the current set of physiological indicators and the current physiological Fourier image, the changes of the patient's physiological indicators are judged through the patient change network to obtain the patient's physiological state. The patient's physiological state includes three states: improvement, deterioration and normal. If the patient's physiological state is improved, an improvement signal is sent. If the patient's physiological state is deteriorated, a deterioration signal is sent.

[0061] In the present invention, multiple types of physiological indicators detected at multiple time points of the patient are obtained. The multiple types of physiological indicators detected at multiple time points are cut according to a fixed time length to obtain a set of patient time physiological indexes corresponding to multiple types of physiological indicators at multiple time points containing a fixed time length. And a current set of physiological indicators corresponding to multiple types of physiological indicators at a time point containing a current time point and a fixed time length is obtained. The multiple sets of patient time physiological indexes cut by the patient are used to find the physiological indicators of the patient in a normal state, and a matrix is ​​constructed to judge the patient's physical state of changes at multiple time points according to the patient's overall physical state composed of multiple physiological indicators, thereby obtaining a set of ordinary physiological indicators. And according to the physiological indicators of multiple time points divided into a certain category, they are converted into Fourier images, so as to judge the changes of physiological indicators of a single category, and obtain the current physiological Fourier image and the ordinary Fourier image. Adding a storage structure in the patient change network can make the previous set of physiological indicators and the current physiological Fourier image input into the patient change network for judgment in real time without recalculating the ordinary physiological indicator set and the ordinary Fourier image. The patient is only replaced when the patient is replaced.

[0062] The patient change network for real-time detection and the temporal convolution network and three-dimensional reconstruction network for calculating the normal state are trained and tested through multiple types of physiological indicators at multiple time points of the user. It is used to calculate physiological indicators in the normal state for specific diseases. It can also train the patient change network, achieving the technical effect of more accurately obtaining the user's improvement, deterioration and normal state compared to himself while reducing data. BRIEF DESCRIPTION OF THE DRAWINGS

[0063] Figure 1 It is a flow chart of a method for detecting physiological information of ICU patients provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0064] The present invention will be described in detail below in conjunction with the accompanying drawings.

[0065] Example 1

[0066] like Figure 1 As shown, an embodiment of the present invention provides a method for detecting physiological information of an ICU patient, the method comprising:

[0067] S101: Obtain a set of physiological indicators of a patient at multiple time points. The set of physiological indicators of a patient includes multiple types of physiological indicators of a patient detected by a monitor connected to a human body. The set of physiological indicators of a patient at multiple time points is multiple types of physiological indicators of a patient detected at multiple time points before and including the current time point.

[0068] The patient's physiological index set includes physiological indexes such as heart rate, blood pressure, blood oxygen, respiration and body temperature.

[0069] Among them, a set of patient physiological indicators corresponds to a disease category.

[0070] The current time point is the time point of the patient's current real-time detection.

[0071] S102: Segment the patient's physiological indicators at the multiple time points according to a fixed time length, with a time step of 1, to obtain multiple patient time physiological index sets and a current physiological indicator set. The patient time physiological index set includes multiple types of physiological indicators of the patient detected at multiple time points of a fixed time length. The current physiological indicator set is a patient time physiological index set including the physiological indicators detected at the current time point among the multiple patient time physiological index sets.

[0072] Wherein, the time step of 1 indicates that the difference in the number is 1. For example, the previous time point is the first second of the test, and the number is 1. The next adjacent time point is the second second of the test, and the number is 2. The time point is the third second of the test, and the number is 3. The time point is the fourth second of the test, and the number is 4. The time point is the fifth second of the test, and the number is 5.

[0073] Wherein, the fixed time length is 4. That is, multiple patient time physiological index sets include {{a1, b1, c1, d1, e1}, {a2, b2, c2, d2, e2}, {a3, b3, c3, d3, e3}, {a4, b4, c4, d4, e4}} and {{a2, b2, c2, d2, e2}, {a3, b3, c3, d3, e3}, {a4, b4, c4, d4, e4}, {a5, b5, c5, d5, e5}}. The current physiological index set is {{a2, b2, c2, d2, e2}, {a3, b3, c3, d3, e3}, {a4, b4, c4, d4, e4}, {a5, b5, c5, d5, e5}}. Among them, a1 represents the heart rate at the first second of detection, b1 represents the blood pressure at the first second of detection, c1 represents the blood oxygen at the first second of detection, d1 represents the respiration at the first second of detection, and e1 represents the body temperature at the first second of detection. a2 represents the heart rate at the second second of detection, b2 represents the blood pressure at the second second of detection, c2 represents the blood oxygen at the second second of detection, d2 represents the respiration at the first second of detection, and e2 represents the body temperature at the first second of detection.

[0074] S103: Based on the multiple patient time physiological index sets, determine the physiological indicators of the patient's normal state when sick, and obtain a normal physiological indicator set, a current physiological Fourier image and a normal Fourier image.

[0075] The normal state is the state of physiological indicators that generally appear when the patient is ill. The normal state is neither a state where the body condition is getting better nor a state where the body condition is getting worse. The good or bad state of the body condition changes according to the condition of the physiological indicators. For example, after catching a cold, the body temperature rises to 37 degrees, and the normal state is 37 degrees. If the temperature drops from 37 degrees to 36.5 degrees, it means that the body condition is getting better, and if the temperature rises from 37 degrees to 37.5 degrees, it means that the body condition is getting worse.

[0076] S104: Obtaining a trained patient change network. The patient change network includes a storage structure and a detection structure.

[0077] S105: Storing the common physiological indicator set and the common Fourier image in the storage structure of the patient change network.

[0078] S106: Based on the current set of physiological indicators and the current physiological Fourier image, the patient's physiological indicator changes are determined through the patient change network to obtain the patient's physiological state, which includes three states: improvement, deterioration and normal.

[0079] During the period, the patient's physiological state indicates the deviation of the patient's physiological indicators in the normal state toward the normal physiological indicators. There is a difference with the patient's physiological indicators in the normal state, but the physiological indicators are close to the normal physiological indicators, indicating improvement. There is a difference with the patient's physiological indicators in the normal state, but the physiological indicators are far away from the normal physiological indicators, indicating deterioration.

[0080] S107: If the patient's physiological state is improving, an improvement signal is sent. If the patient's physiological state is deteriorating, a deterioration signal is sent.

[0081] Among them, whether it is a signal of improvement or a signal of deterioration, medical staff will be reminded to conduct physical examinations on patients.

[0082] Optionally, the detection structure of the patient change network includes a first neural network, a first convolutional network, a physiological characteristic discrimination network and a Fourier discriminant network.

[0083] Optionally, the patient change network is a trained patient change network obtained by training with multiple training data.

[0084] The training data is to use one of the multiple patient time physiological index sets as the training current patient time physiological index set, and to use the Fourier image transformed from the training current patient physiological index set as the training current Fourier image.

[0085] Based on the current training patient physiological index set and the current training Fourier image, the changes of the patient's physiological indexes are judged through the patient change network to obtain the training patient's physiological state.

[0086] A marked physiological state is obtained, wherein the marked physiological state indicates a marked improved, deteriorated, or normal state.

[0087] The loss of the training patient's physiological state and the marked physiological state is calculated to obtain a physiological state loss value.

[0088] In this embodiment, the loss is calculated by a cross entropy loss function.

[0089] If the physiological state loss value is less than the loss threshold, backpropagation is used to train the patient change network.

[0090] In this embodiment, the loss threshold is 0.2.

[0091] Through the above method, multiple segmented patient time physiological index sets are used to calculate physiological indicators under normal conditions for specific diseases. Patient change networks can also be trained to obtain the user's status more accurately while reducing data.

[0092] Optionally, the determining the changes of the patient's physiological indicators through a patient change network based on the current physiological indicator set and the current physiological Fourier image to obtain the patient's physiological state includes:

[0093] Normal physiological indicators are obtained; the normal physiological indicators refer to physiological indicators in the absence of any disease.

[0094] The values ​​in the current physiological indicator set are input into the first neural network, and the features of the physiological indicators at the current time point are extracted to obtain the current physiological indicator features.

[0095] Among them, the first neural network is a fully connected neural network (Fully Connected Neural Network, FCNN).

[0096] The values ​​in the common physiological index set in the storage structure are input into the second neural network, and the features of the physiological indexes of the patient's common state are extracted to obtain the features of the common physiological indexes.

[0097] The normal physiological index is input into the first neural network, and the characteristics of the normal physiological index are extracted to obtain the characteristics of the normal physiological index.

[0098] The current physiological index feature and the common physiological index feature are input into a physiological feature discrimination network, and the feature of the physiological index change is extracted to obtain the physiological index change feature.

[0099] Among them, the physiological feature discrimination network is a discriminator network (DiscriminatorNetwork).

[0100] The current physiological Fourier image is input into the first convolutional network, and the features of the physiological indicators shown in the Fourier image at the current time point are extracted to obtain the current Fourier features.

[0101] Among them, the first convolutional network is a convolutional neural network (CNN).

[0102] The common Fourier image in the storage structure is input into the first convolutional network, and the characteristics of the physiological indicators of the patient's common state expressed in the Fourier image are extracted to obtain the common Fourier features.

[0103] The current Fourier feature and the common Fourier feature are input into a Fourier discriminant network to extract the features of changes in physiological indicators and obtain Fourier change features.

[0104] The Fourier discriminant network is a discriminator network (DiscriminatorNetwork) whose number of neurons and structure are different from those of the physiological feature discriminant network.

[0105] The physiological index change characteristics, normal physiological index characteristics and Fourier change characteristics are input into the discriminant neural network, and judged as one of the three states of improvement, deterioration and normal to obtain the patient's physiological state.

[0106] The discriminant neural network is a discriminator network (Discriminator Network) whose number of neurons and structure are different from those of the physiological feature discriminant network and the Fourier discriminant network.

[0107] Optionally, judging the physiological indexes of the patient's normal state when sick based on the multiple patient time physiological index sets to obtain the normal physiological index set, the current physiological Fourier image and the normal Fourier image includes:

[0108] Based on the multiple patient physiological indicator sets, a current physiological Fourier image and a common Fourier image are obtained through a temporal convolution network and a three-dimensional reconstruction network.

[0109] Based on the multiple patient physiological indicator sets, changes in the user's physiological characteristics jointly constructed by the multiple types of physiological indicators are determined to obtain a common change matrix.

[0110] Based on the common change matrix, a set of common physiological indicators is obtained.

[0111] Optionally, the obtaining of the current physiological Fourier image and the ordinary Fourier image based on the multiple patient physiological indicator sets through a time convolution network and a three-dimensional reconstruction network includes:

[0112] According to a plurality of patient physiological indicator sets, the sets are divided into a plurality of time physiological category sets according to categories of the physiological indicators.

[0113] Among them, a time physiological category set corresponds to a category of a physiological indicator. For example, in this embodiment, heart rate corresponds to a set of physiological indicators that only include heart rate at multiple time points. Blood pressure corresponds to a set of physiological indicators that only include blood pressure at multiple time points. Blood oxygen corresponds to a set of physiological indicators that only include blood oxygen at multiple time points. Respiration corresponds to a set of physiological indicators that only include respiration at multiple time points. Body temperature corresponds to a set of physiological indicators that only include body temperature at multiple time points.

[0114] The temporal physiological category set is subjected to Fourier transformation to determine the change of the category corresponding to a physiological indicator at multiple time points, and obtain a physiological category Fourier image. Multiple physiological category Fourier images are obtained corresponding to multiple temporal physiological category sets.

[0115] Among them, a temporal physiological category set corresponds to a category of physiological indicators and corresponds to a physiological category Fourier image.

[0116] In this embodiment, fast Fourier transform (FFT) is used to perform Fourier transform.

[0117] The physiological category Fourier image corresponding to the current time point is used as the current physiological Fourier image.

[0118] The multiple physiological category Fourier images are superimposed to obtain a three-dimensional physiological category image. Multiple three-dimensional physiological category images are obtained corresponding to multiple patient physiological indicator sets.

[0119] The three-dimensional physiological category image is related to the number of time points at which the patient's physiological index set is obtained. For example, a patient's physiological index set at m time points corresponds to m-1 patient's time physiological index sets corresponding to m three-dimensional physiological category images. m is a natural number greater than 0.

[0120] The multiple three-dimensional physiological category images are input into a temporal convolutional network, and features of changes between the multiple three-dimensional physiological category images are extracted to obtain a three-dimensional feature map.

[0121] Among them, in this embodiment, the temporal convolutional network (Temporal Convolutional Network, TCN).

[0122] The three-dimensional feature map is input into a three-dimensional reconstruction network to obtain a common Fourier image.

[0123] Wherein, the three-dimensional reconstruction network is a three-dimensional convolutional neural network (3D Convolutional Neural Networks, 3DCNN).

[0124] Optionally, the training method of the temporal convolutional network and the three-dimensional reconstruction network is:

[0125] A plurality of training images are obtained, wherein the plurality of training images are three-dimensional physiological category images of the patient during the change process of the physiological index in history.

[0126] Obtain a marked image, wherein the marked image is a training image of a patient in a normal physiological state among the plurality of training images.

[0127] A temporal convolutional network and a three-dimensional reconstruction network are jointly trained according to the multiple training images and the annotated images.

[0128] The multiple training images are input into a temporal convolutional network, and features of changes between the multiple training images are extracted to obtain a training three-dimensional feature map.

[0129] The training three-dimensional feature map is input into a three-dimensional reconstruction network to obtain a training ordinary Fourier image.

[0130] The training common Fourier image and the annotated image are passed through a loss function to obtain a loss value, and the temporal convolutional network and the three-dimensional reconstruction network are trained by backpropagation according to the loss value.

[0131] Among them, in this embodiment, the loss function is a cross entropy loss function.

[0132] Optionally, the rows of the common change matrix represent time points from morning to night. The rows of the common change matrix represent categories of physiological indicators.

[0133] Optionally, obtaining a set of common physiological indicators based on the common change matrix includes:

[0134] A first two-dimensional convolution kernel is obtained. The number of rows of the first two-dimensional convolution kernel is 1, and the number of columns is n. The n is a natural number greater than 0. The n represents the number of categories of the physiological indicator.

[0135] The first two-dimensional convolution kernel is convolved with a row in the common change matrix with a step size of 1 to extract features of multiple categories of physiological indicators and obtain a first convolution feature.

[0136] The first convolution feature is input into a matrix change network to extract the change features of multiple types of physiological indicators at multiple time points to obtain a second convolution feature.

[0137] The second convolution feature is input into the reconstructed neural network, the feature is converted into data, and a set of common physiological indicators is obtained.

[0138] Example 2

[0139] Based on the above-mentioned ICU patient physiological information detection method, an embodiment of the present invention also provides an ICU patient physiological information detection system, which includes a data acquisition module, a segmentation module, a normal state physiological indicator detection module, a network acquisition module, a storage module, a detection module and an early warning module.

[0140] The data acquisition module is used to obtain a set of physiological indicators of a patient at multiple time points. The set of physiological indicators of a patient includes multiple types of physiological indicators of a patient detected by a monitor connected to a human body. The set of physiological indicators of a patient at multiple time points is multiple types of physiological indicators of a patient detected at multiple time points before and including the current time point.

[0141] The segmentation module is used to segment the patient's physiological indicators at the multiple time points according to a fixed time length, and the time step is 1, to obtain multiple patient time physiological index sets and a current physiological index set. The patient time physiological index set includes multiple types of physiological indicators of the patient detected at multiple time points of a fixed time length. The current physiological index set is a patient time physiological index set including the physiological indicators detected at the current time point in the multiple patient time physiological index sets.

[0142] The normal state physiological index detection module is used to determine the physiological indexes of the patient's normal state when he is sick based on the multiple patient time physiological index sets, and obtain the normal physiological index set, the current physiological Fourier image and the normal Fourier image.

[0143] The network acquisition module is used to obtain a trained patient change network. The patient change network includes a storage structure and a detection structure.

[0144] The storage module is used to store the common physiological indicator set and the common Fourier image in the storage structure of the patient change network.

[0145] The detection module is used to determine the changes of the patient's physiological indicators through the patient change network based on the current physiological indicator set and the current physiological Fourier image, and obtain the patient's physiological state. The patient's physiological state includes three states: improvement, deterioration and normal.

[0146] The early warning module is used to send an improvement signal if the patient's physiological state improves, and to send a deterioration signal if the patient's physiological state deteriorates.

[0147] The algorithm and display provided herein are not inherently related to any particular computer, virtual system or other device. Various general purpose systems can also be used together with the teachings based on this. According to the above description, it is obvious that the structure required for constructing such systems. In addition, the present invention is not directed to any specific programming language either. It should be understood that various programming languages ​​can be utilized to realize the content of the present invention described herein, and the description of the above specific languages ​​is for disclosing the best mode of the present invention.

[0148] In the description provided herein, a large number of specific details are described. However, it is understood that embodiments of the present invention can be practiced without these specific details. In some instances, well-known methods, structures and techniques are not shown in detail so as not to obscure the understanding of this description.

[0149] Similarly, it should be understood that in order to streamline the present disclosure and aid in understanding one or more of the various inventive aspects, in the above description of exemplary embodiments of the present invention, various features of the present invention are sometimes grouped together into a single embodiment, figure, or description thereof. However, this disclosed method should not be interpreted as reflecting the intention that the claimed invention requires more features than those explicitly recited in each claim. More specifically, as reflected in the claims below, inventive aspects lie in less than all the features of the individual embodiments disclosed above. Therefore, the claims that follow the specific embodiment are hereby expressly incorporated into the specific embodiment, with each claim itself serving as a separate embodiment of the present invention.

[0150] Those skilled in the art will appreciate that the modules in the devices in the embodiments may be adaptively changed and arranged in one or more devices different from the embodiments. The modules or units or components in the embodiments may be combined into one module or unit or component, and in addition they may be divided into a plurality of submodules or subunits or subcomponents. Except that at least some of such features and / or processes or units are mutually exclusive, all features disclosed in this specification (including the accompanying claims, abstracts and drawings) and all processes or units of any method or device disclosed in this manner may be combined in any combination. Unless otherwise expressly stated, each feature disclosed in this specification (including the accompanying claims, abstracts and drawings) may be replaced by an alternative feature providing the same, equivalent or similar purpose.

[0151] In addition, those skilled in the art will appreciate that, although some embodiments herein include certain features included in other embodiments but not other features, the combination of features of different embodiments is meant to be within the scope of the present invention and form different embodiments. For example, in the claims below, any one of the claimed embodiments may be used in any combination.

[0152] The various component embodiments of the present invention may be implemented in hardware, or in software modules running on one or more processors, or in a combination thereof. It should be understood by those skilled in the art that a microprocessor or digital signal processor (DSP) may be used in practice to implement some or all of the functions of some or all of the components in the apparatus according to an embodiment of the present invention. The present invention may also be implemented as a device or device program (e.g., a computer program and a computer program product) for executing part or all of the methods described herein. Such a program implementing the present invention may be stored on a computer-readable medium, or may have the form of one or more signals. Such a signal may be downloaded from an Internet website, or provided on a carrier signal, or provided in any other form.

[0153] It should be noted that the above embodiments illustrate the present invention rather than limit it, and that those skilled in the art may devise alternative embodiments without departing from the scope of the appended claims. In the claims, any reference symbol between brackets shall not be construed as a limitation on the claims. The word "comprising" does not exclude the presence of elements or steps not listed in the claims. The word "one" or "an" preceding an element does not exclude the presence of a plurality of such elements. The present invention may be implemented by means of hardware comprising a number of different elements and by means of a suitably programmed computer. In a unit claim enumerating a number of devices, several of these devices may be embodied by the same hardware item. The use of the words first, second, and third, etc., does not indicate any order. These words may be interpreted as names.

Claims

1. A method for detecting physiological information of ICU patients, characterized in that: include: Obtaining a set of physiological indicators of a patient at multiple time points; the set of physiological indicators of a patient includes multiple types of physiological indicators of a patient detected by a monitor connected to a human body; The set of physiological indicators of the patient at multiple time points is multiple types of physiological indicators of the patient detected at multiple time points before and including the current time point; The patient's physiological indicators at the multiple time points are divided according to a fixed time length, and the time step is 1, to obtain multiple patient time physiological index sets and a current physiological indicator set; the patient time physiological index set includes multiple types of physiological indicators of the patient detected at multiple time points of a fixed time length; the current physiological indicator set is a patient time physiological index set including the physiological indicators detected at the current time point in the multiple patient time physiological index sets; Based on the multiple patient time physiological index sets, determine the physiological indexes of the patient's normal state when sick, and obtain a normal physiological index set, a current physiological Fourier image and a normal Fourier image; Obtaining a trained patient change network; the patient change network comprises a storage structure and a detection structure; Storing the common physiological index set and the common Fourier image in a storage structure of a patient change network; Based on the current set of physiological indicators and the current physiological Fourier image, the patient's physiological state is obtained by judging the changes of the patient's physiological indicators through a patient change network; The patient's physiological state includes three states: improvement, deterioration and normal; If the patient's physiological state is improving, an improvement signal is sent; if the patient's physiological state is deteriorating, a deterioration signal is sent.

2. A method for detecting physiological information of ICU patients according to claim 1, characterized in that: The detection structure of the patient change network includes a first neural network, a first convolutional network, a physiological characteristic discrimination network and a Fourier discrimination network.

3. A method for detecting physiological information of ICU patients according to claim 2, characterized in that: The patient change network is a trained patient change network obtained by training with a plurality of training data; The training data is to use one of the multiple patient time physiological index sets as the training current patient time physiological index set, and to use the Fourier image transformed from the training current patient physiological index set as the training current Fourier image; Based on the set of current training patient physiological indicators and the current training Fourier image, the changes of the patient's physiological indicators are judged through the patient change network to obtain the training patient's physiological state; Obtain annotated physiological status; The marked physiological state represents a marked improved, worsened or normal state; Calculate the loss of the training patient's physiological state and the marked physiological state to obtain a physiological state loss value; If the physiological state loss value is less than the loss threshold, backpropagation is used to train the patient change network.

4. A method for detecting physiological information of ICU patients according to claim 2, characterized in that: The method of determining changes in the patient's physiological indicators based on the current set of physiological indicators and the current physiological Fourier image through a patient change network to obtain the patient's physiological state includes: Obtaining normal physiological indicators; the normal physiological indicators refer to physiological indicators without any disease; Inputting the values ​​in the current physiological indicator set into the first neural network, extracting the features of the physiological indicators at the current time point, and obtaining the current physiological indicator features; Inputting the values ​​in the common physiological index set in the storage structure into the first neural network, extracting the features of the physiological indexes of the patient's common state, and obtaining the features of the common physiological indexes; Inputting the normal physiological index into a first neural network, extracting the characteristics of the normal physiological index, and obtaining the characteristics of the normal physiological index; Inputting the current physiological index feature and the common physiological index feature into a physiological feature discrimination network, extracting the feature of the physiological index change, and obtaining the physiological index change feature; Inputting the current physiological Fourier image into a first convolutional network, extracting the features of the physiological indicators represented in the Fourier image at the current time point, and obtaining the current Fourier features; Inputting the common Fourier image in the storage structure into the second convolutional network, extracting the features of the physiological indicators of the patient's common state in the Fourier image, and obtaining the common Fourier features; Inputting the current Fourier feature and the common Fourier feature into a Fourier discriminant network, extracting the feature of the change of the physiological index, and obtaining the Fourier change feature; The physiological index change characteristics, normal physiological index characteristics and Fourier change characteristics are input into the discriminant neural network, and judged as one of the three states of improvement, deterioration and normal to obtain the patient's physiological state.

5. The method for detecting physiological information of ICU patients according to claim 1, characterized in that: The method of judging the physiological indexes of the patient's normal state when sick based on the multiple patient time physiological index sets, and obtaining the normal physiological index set, the current physiological Fourier image and the normal Fourier image, comprises: Based on the multiple patient physiological index sets, a current physiological Fourier image and a common Fourier image are obtained through a time convolution network and a three-dimensional reconstruction network; Based on the plurality of sets of physiological indicators of the patients, determining the changes of the physiological characteristics of the user jointly constructed by the plurality of types of physiological indicators, and obtaining a common change matrix; Based on the common change matrix, a set of common physiological indicators is obtained.

6. A method for detecting physiological information of ICU patients according to claim 1, characterized in that: The method of obtaining a current physiological Fourier image and a common Fourier image based on the multiple sets of physiological indicators of the patients through a time convolution network and a three-dimensional reconstruction network includes: According to the patient's physiological indicator set, the set is divided into a plurality of time physiological category sets according to the categories of the physiological indicators; The temporal physiological category set is subjected to Fourier transformation to determine the change of the category corresponding to a physiological indicator at multiple time points, and obtain a physiological category Fourier image; multiple temporal physiological category sets correspond to multiple physiological category Fourier images; The physiological category Fourier image corresponding to the current time point is used as the current physiological Fourier image; The multiple physiological category Fourier images are superimposed to obtain a three-dimensional physiological category image; multiple patient physiological indicator sets are correspondingly obtained to obtain multiple three-dimensional physiological category images; Inputting the multiple three-dimensional physiological category images into a temporal convolutional network, extracting features of changes between the multiple three-dimensional physiological category images, and obtaining a three-dimensional feature map; The three-dimensional feature map is input into a three-dimensional reconstruction network to obtain a common Fourier image.

7. A method for detecting physiological information of ICU patients according to claim 6, characterized in that: The training method of the temporal convolutional network and the three-dimensional reconstruction network: Obtaining a plurality of training images; the plurality of training images are three-dimensional physiological category images of historical changes in physiological indicators of patients; Obtain annotated images; The annotated image is a training image of a patient in a normal physiological state among the plurality of training images; A temporal convolutional network and a three-dimensional reconstruction network are jointly trained according to the multiple training images and the annotated images.

8. A method for detecting physiological information of ICU patients according to claim 5, characterized in that: The rows of the common change matrix represent time points from morning to night; the rows of the common change matrix represent categories of physiological indicators.

9. A method for detecting physiological information of ICU patients according to claim 5, characterized in that: The method of obtaining a set of common physiological indicators based on the common change matrix includes: Obtaining a first two-dimensional convolution kernel; the number of rows of the first two-dimensional convolution kernel is 1, and the number of columns is n; n is a natural number greater than 0; n represents the number of categories of physiological indicators; Convolving the first two-dimensional convolution kernel with a row in the common change matrix with a step size of 1 to extract features of multiple categories of physiological indicators to obtain a first convolution feature; Inputting the first convolution feature into a matrix change network, extracting the change features of multiple types of physiological indicators at multiple time points, and obtaining a second convolution feature; The second convolution feature is input into the reconstructed neural network, the feature is converted into data, and a set of common physiological indicators is obtained.

10. An ICU patient physiological information detection system, characterized in that: include: Data acquisition module: obtains a set of physiological indicators of a patient at multiple time points; the set of physiological indicators of a patient includes multiple types of physiological indicators of a patient detected by a monitor connected to the human body; The set of physiological indicators of the patient at multiple time points is multiple types of physiological indicators of the patient detected at multiple time points before and including the current time point; Segmentation module: segment the patient's physiological indicators at the multiple time points according to a fixed time length, and the time step is 1, to obtain multiple patient time physiological index sets and a current physiological index set; the patient time physiological index set includes multiple types of physiological indicators of the patient detected at multiple time points of a fixed time length; the current physiological index set is a patient time physiological index set including the physiological indicators detected at the current time point in the multiple patient time physiological index sets; Normal state physiological index detection module: based on the multiple patient time physiological index sets, determine the patient's normal state physiological index when sick, and obtain the normal physiological index set, the current physiological Fourier image and the normal Fourier image; Network acquisition module: obtains a trained patient change network; the patient change network includes a storage structure and a detection structure; Storage module: storing the common physiological index set and the common Fourier image in the storage structure of the patient change network; Detection module: based on the current physiological index set and the current physiological Fourier image, through the patient change network, judge the changes of the patient's physiological indexes to obtain the patient's physiological state; The patient's physiological state includes three states: improvement, deterioration and normal; Early warning module: If the patient's physiological state improves, an improvement signal is sent; if the patient's physiological state deteriorates, a deterioration signal is sent.

Images