A respiratory signal prediction method and device

Abstract

The present application relates to a respiratory signal prediction method and device. The method includes: dividing the acquired respiratory signal samples to obtain various respiratory signal sections; predicting each section to obtain the prediction results of each section, and determining the precision error and extreme error of the prediction results; based on the precision error and extreme error of each prediction result, the precision error weight and extreme error weight of each section, and the penalty loss of the prediction results of each section, determining the loss of the prediction results of the respiratory signal samples; based on the loss of the prediction results of the respiratory signal samples, training the respiratory signal prediction model to obtain a trained prediction model; and using the trained prediction model to predict the respiratory signal to be predicted. The use of this method can solve the problem that the trained respiratory signal model used in the existing respiratory signal prediction method has a large instantaneous error, which causes an instantaneous high dose of radiotherapy.

Description

Technical Field

[0001] The present application relates to the field of signal prediction technology, and in particular to a method and device for predicting a respiratory signal. Background Art

[0002] During radiotherapy, respiratory motion can cause dynamic changes in the target tumor's position. Intervention based solely on the target's real-time position can result in delays. Failure to effectively compensate for the dynamic displacement caused by this delay can lead to irradiation deviations, compromising radiotherapy precision and efficacy. The respiratory signal provides direct feedback on respiratory motion, making accurate prediction of the respiratory signal crucial.

[0003] Existing technologies primarily use neural network models, such as long short-term memory (LSTM) networks and multilayer perceptron (MLP) networks, to predict respiratory signals. Before using a neural network model for respiratory signal prediction, the respiratory signal model must be trained based on the accuracy of the overall respiratory signal trend. This trained model is then used for respiratory signal prediction.

[0004] Existing respiratory signal prediction methods only consider the fitting accuracy of the overall trend of the respiratory signal, resulting in large transient errors in the trained respiratory signal model. Large transient errors can lead to transient high doses during radiotherapy, which can cause irreversible damage to the body's cascade of critical organs (such as the spinal cord and small intestine).

[0005] The trained respiratory signal model used in the existing respiratory signal prediction method has a large instantaneous error, which causes the problem of instantaneous high dose of radiotherapy. No effective solution has been proposed so far. Summary of the Invention

[0006] Based on this, it is necessary to provide a respiratory signal prediction method and device to address the above technical problems.

[0007] In a first aspect, the present application provides a method for predicting a respiratory signal. The method comprises:

[0008] Acquire respiratory signal samples, and divide the acquired respiratory signal samples according to preset rules to obtain various respiratory signal sections;

[0009] Using the initial respiratory signal prediction model, predicting each respiratory signal section to obtain a prediction result of each respiratory signal section, and determining, based on the prediction result of each respiratory signal section, a precision error and an extreme error of the prediction result of each respiratory signal section; the precision error represents the overall accuracy of the prediction result of the respiratory signal; and the extreme error represents the maximum value of the error of the prediction result of the respiratory signal;

[0010] Based on the precision error and the extreme error of the prediction results of each respiratory signal section, the precision error weight and the extreme error weight of each respiratory signal section, and the penalty loss of the prediction results of each respiratory signal section, the loss of the prediction results of the respiratory signal samples is determined; the precision error weight represents the influence factor of the overall accuracy of the prediction result on the prediction result of the respiratory signal sample; the extreme error weight represents the influence factor of the maximum value of the error of the prediction result on the prediction result of the respiratory signal sample; the penalty loss is a clinical constraint function determined according to the relationship between the maximum value of the prediction error of the respiratory signal at the peak and valley and the preset clinical safety threshold;

[0011] Training the initial respiratory signal prediction model based on the loss of the prediction result of the respiratory signal sample to obtain a trained respiratory signal prediction model;

[0012] A respiratory signal to be predicted is acquired, and the trained respiratory signal prediction model is used to predict the respiratory signal to be predicted.

[0013] In one embodiment, the using the initial respiratory signal prediction model to predict each respiratory signal segment to obtain a prediction result for each respiratory signal segment, and determining the accuracy error and extreme error of the prediction result for each respiratory signal segment based on the prediction result for each respiratory signal segment, includes:

[0014] Using the initial respiratory signal prediction model, predicting each respiratory signal section to obtain a prediction result for each respiratory signal section;

[0015] Calculating the accuracy error of the prediction result of each respiratory signal section based on the difference between the prediction result of each respiratory signal section at each sampling time and the sampling point; the sampling time is the time point corresponding to each sampling point when the respiratory signal is digitally collected;

[0016] Based on the maximum value of the difference between the prediction result of each respiratory signal section at each sampling time and the sampling point, and the preset error level penalty coefficient, the extreme error of the prediction result of each respiratory signal section is calculated; the error level includes small error, medium error and out-of-limit error; the penalty coefficient represents the influence factor of different error levels on the prediction result of the respiratory signal.

[0017] In one embodiment, before determining the loss of the prediction result of the respiratory signal sample based on the precision error and extreme error of the prediction result of each respiratory signal section, the precision error weight and extreme error weight of each respiratory signal section, and the penalty loss of the prediction result of each respiratory signal section, the method includes:

[0018] The precision error weight and the extreme error weight of each respiratory signal section are determined based on the characteristics of each respiratory signal section; the characteristics include amplitude characteristics, frequency characteristics and stability characteristics.

[0019] In one embodiment, determining the precision error weight and the extreme error weight of each respiratory signal section based on the characteristics of each respiratory signal section includes:

[0020] Based on the features of each respiratory signal section, constructing a feature vector of each respiratory signal section; the feature vector is a vector formed by combining the normalized features of the respiratory signal;

[0021] Based on the feature vectors of each respiratory signal section and the sensitivity coefficient of each feature, the precision error weight and the extreme error weight of each respiratory signal section are determined; the sensitivity coefficient represents the influence factor of the feature change on the prediction result of the respiratory signal.

[0022] In one embodiment, before determining the precision error weight and the extreme error weight of each respiratory signal section based on the feature vector of each respiratory signal section and the sensitivity coefficient of each feature, the method includes:

[0023] Based on the application scenario corresponding to the respiratory signal sample, the sensitivity coefficient of each feature is determined.

[0024] In one embodiment, determining the precision error weight and the extreme error weight of each respiratory signal section based on the feature vector of each respiratory signal section and the sensitivity coefficient of each feature includes:

[0025] Calculating an initial precision error weight and an initial extreme error weight of a prediction result of each respiratory signal section based on the feature vector of each respiratory signal section and the sensitivity coefficient of each feature;

[0026] Based on the amplitude feature component in the feature vector of each respiratory signal section, the initial precision error weight and the initial extreme error weight of the prediction result of each respiratory signal section are adjusted to obtain the precision error weight and the extreme error weight of each respiratory signal section.

[0027] In one embodiment, determining the loss of the prediction result of the respiratory signal sample based on the precision error and extreme error of the prediction result of each respiratory signal section, the precision error weight and extreme error weight of each respiratory signal section, and the penalty loss of the prediction result of each respiratory signal section includes:

[0028] Calculate the penalty loss of the prediction results of each respiratory signal segment based on the peak safety threshold and the trough safety threshold corresponding to the current application scenario; the peak safety threshold is the maximum prediction error of the peak of the respiratory signal when the distance between the target area and the endangered organ is at a safe distance; the trough safety threshold is the maximum prediction error of the trough of the respiratory signal when the distance between the target area and the endangered organ is at a safe distance;

[0029] performing a weighted summation of the precision error and the extreme error of the prediction result of each respiratory signal section based on the precision error weight and the extreme error weight of each respiratory signal section, and determining the sum of the weighted summation result and the penalty loss of the prediction result of each respiratory signal section as the loss of the prediction result of each respiratory signal section;

[0030] The sum of the losses of the prediction results of the respective respiratory signal sections is determined as the loss of the prediction result of the respiratory signal sample.

[0031] In one embodiment, the calculation of the penalty loss of the prediction result of each respiratory signal section based on the peak safety threshold and the trough safety threshold corresponding to the current application scenario includes:

[0032] Determine a peak time point set of each respiratory signal section; the peak time point set is a collection of sampling times of all peaks of each respiratory signal section;

[0033] Determine a peak maximum prediction error of each respiratory signal section based on a difference between a sampling point at each sampling time of the peak time point set and a predicted value of each respiratory signal section; the peak maximum prediction error is the maximum value of the absolute value of the difference between a sampling point at each sampling time of the respiratory signal section and a predicted value;

[0034] Determine a trough time point set of each respiratory signal section; the trough time point set is a collection of sampling times of all troughs of each respiratory signal section;

[0035] Determine a maximum trough prediction error of each respiratory signal segment based on a difference between a sampling point at each sampling time of the trough time point set and a predicted value of each respiratory signal segment; the maximum trough prediction error is the maximum absolute value of the difference between a sampling point at each sampling time of the respiratory signal segment and a predicted value;

[0036] Based on the peak safety threshold and the trough safety threshold corresponding to the current application scenario, and the peak maximum prediction error and the trough maximum prediction error of each respiratory signal section, the penalty loss of the prediction result of each respiratory signal section is calculated.

[0037] In one embodiment, the training of the initial respiratory signal prediction model based on the loss of the prediction result of the respiratory signal sample to obtain the trained respiratory signal prediction model includes:

[0038] When the loss of the prediction result of the respiratory signal sample to be predicted is greater than a preset loss threshold, the parameters of the initial respiratory signal prediction model are updated, and the loss of the prediction result of the respiratory signal sample to be predicted is recalculated using the updated prediction model. The process of updating the parameters of the initial respiratory signal prediction model and calculating the loss of the prediction result of the respiratory signal sample is repeated until the number of times the loss of the prediction result of the respiratory signal sample is calculated reaches a preset number requirement, or the loss of the prediction result of the respiratory signal sample reaches a preset loss threshold, thereby obtaining the trained respiratory signal prediction model.

[0039] In a second aspect, the present application further provides a respiratory signal prediction device. The device comprises:

[0040] The sample division module is used to obtain respiratory signal samples and divide the obtained respiratory signal samples according to preset rules to obtain various respiratory signal sections;

[0041] an error determination module, configured to predict each respiratory signal segment using the initial respiratory signal prediction model to obtain a prediction result for each respiratory signal segment, and determine, based on the prediction result for each respiratory signal segment, a precision error and an extreme error of the prediction result for each respiratory signal segment; the precision error represents the overall accuracy of the prediction result of the respiratory signal; and the extreme error represents the maximum value of the error of the prediction result of the respiratory signal;

[0042] A loss determination module is configured to determine the loss of the prediction result of the respiratory signal sample based on the precision error and the extreme error of the prediction result of each respiratory signal section, the precision error weight and the extreme error weight of each respiratory signal section, and the penalty loss of the prediction result of each respiratory signal section; the precision error weight represents the influence factor of the overall accuracy of the prediction result on the prediction result of the respiratory signal sample; the extreme error weight represents the influence factor of the maximum value of the error of the prediction result on the prediction result of the respiratory signal sample; the penalty loss is a clinical constraint function determined based on the relationship between the maximum value of the prediction error of the respiratory signal at the peak and valley and a preset clinical safety threshold;

[0043] a model training module, configured to train the initial respiratory signal prediction model based on the loss of the prediction result of the respiratory signal sample to obtain a trained respiratory signal prediction model;

[0044] and a prediction module, which is used to obtain a respiratory signal to be predicted and predict the respiratory signal to be predicted using the trained respiratory signal prediction model.

[0045] The above-mentioned respiratory signal prediction method and device divides the respiratory signal samples according to preset rules to obtain individual respiratory signal segments, and then uses the initial respiratory signal prediction model to predict each respiratory signal segment to obtain the prediction results for each respiratory signal segment. Then, based on the prediction results of each respiratory signal segment, the precision error and extreme error of each prediction result are determined. The loss of the prediction results of the respiratory signal samples is determined using the precision error and extreme error of the prediction results of each respiratory signal segment, the respective precision error weights and extreme error weights, and the penalty loss of the prediction results of each respiratory signal segment. The initial respiratory signal prediction model is trained using the loss of the prediction results of the respiratory signal samples, and the trained respiratory signal model is used to predict the respiratory signal to be predicted. In the process of calculating the loss of the prediction results of the respiratory signal samples, the relationship between the maximum error of the prediction results of the respiratory signal and the maximum error of the prediction of the respiratory signal at the peak and valley and the preset clinical safety threshold is used as a clinical constraint function to provide a calculation basis, so that the loss of the prediction results of the respiratory signal samples calculated in this way is used to train the initial respiratory signal prediction model, which can enable the trained respiratory signal prediction model to avoid the situation of large instantaneous errors, and solve the problem that the trained respiratory signal model used in the existing respiratory signal prediction method has large instantaneous errors and causes instantaneous high doses of radiotherapy.

[0046] The details of one or more embodiments of the present application are set forth in the following drawings and description to make other features, objects, and advantages of the present application more readily apparent. BRIEF DESCRIPTION OF THE DRAWINGS

[0047] The drawings described herein are used to provide a further understanding of the present application and constitute a part of the present application. The illustrative embodiments of the present application and their descriptions are used to explain the present application and do not constitute an improper limitation on the present application. In the drawings:

[0048] Figure 1 A hardware structure block diagram of a terminal for a respiratory signal prediction method provided in one embodiment of the present application;

[0049] Figure 2 A flowchart of a respiratory signal prediction method provided in one embodiment of the present application;

[0050] Figure 3 A flowchart of a respiratory signal prediction method provided in a preferred embodiment of the present application;

[0051] Figure 4 This is a structural block diagram of a respiratory signal prediction device provided in one embodiment of the present application. DETAILED DESCRIPTION

[0052] In order to more clearly understand the purpose, technical solutions and advantages of the present application, the present application is described and illustrated below in conjunction with the accompanying drawings and embodiments.

[0053] Unless otherwise defined, technical or scientific terms used in this application shall have the ordinary meanings as understood by persons of ordinary skill in the art to which this application belongs. The terms "a," "an," "the," "these," and similar expressions in this application do not denote limitations on quantity and may be singular or plural. The terms "comprise," "include," "have," and any variations thereof, as used in this application, are intended to cover non-exclusive inclusions. For example, a process, method, system, product, or device comprising a series of steps or modules (units) is not limited to the listed steps or modules (units) but may include unlisted steps or modules (units) or other steps or modules (units) inherent to the process, method, product, or device. The terms "connected," "connected," "coupled," and similar expressions used in this application are not limited to physical or mechanical connections but may include electrical connections, whether direct or indirect. As used in this application, "plurality" means two or more. "And / or" describes an association between associated objects, indicating that three possible relationships exist. For example, "A and / or B" can mean: A exists alone; A and B exist simultaneously; or B exists alone. Generally, the character " / " indicates that the objects in the preceding and following relationship are in an "or" relationship. The terms "first", "second", "third", etc. involved in this application are only used to distinguish similar objects and do not represent a specific ordering of the objects.

[0054] The method embodiment provided in this embodiment can be executed in a terminal, a computer or a similar computing device. For example, running on a terminal, Figure 1 FIG. 1 is a block diagram of the hardware structure of the terminal of the respiratory signal prediction method of this embodiment. Figure 1 As shown, the terminal may include one or more ( Figure 1 The processor 102 (only one is shown) and a memory 104 for storing data, wherein the processor 102 may include but is not limited to a processing device such as a microprocessor MCU or a programmable logic device FPGA. The terminal may also include a transmission device 106 for communication functions and an input / output device 108. It will be understood by those skilled in the art that Figure 1 The structure shown is only for illustration and does not limit the structure of the above terminal. Figure 1 More or fewer components than shown, or with Figure 1 Different configurations shown.

[0055] The memory 104 can be used to store computer programs, for example, software programs and modules of application software, such as the computer program corresponding to the respiratory signal prediction method in this embodiment. The processor 102 executes the computer program stored in the memory 104 to perform various functional applications and data processing, thereby implementing the above-mentioned method. The memory 104 may include a high-speed random access memory and may also include a non-volatile memory, such as one or more magnetic storage devices, flash memory, or other non-volatile solid-state memory. In some examples, the memory 104 may further include a memory remotely located relative to the processor 102, and these remote memories may be connected to the terminal via a network. Examples of the above-mentioned network include, but are not limited to, the Internet, an intranet, a local area network, a mobile communication network, and combinations thereof.

[0056] Transmission device 106 is used to receive or transmit data via a network. This network may include a wireless network provided by the terminal's communications provider. In one embodiment, transmission device 106 includes a network interface controller (NIC), which can be connected to other network devices via a base station to enable communication with the Internet. In another embodiment, transmission device 106 may be a radio frequency (RF) module, which is used to communicate with the Internet wirelessly.

[0057] In this embodiment, a respiratory signal prediction method is provided. Figure 2 is a flow chart of the respiratory signal prediction method of this embodiment, as shown in FIG. Figure 2 As shown, the process includes the following steps:

[0058] Step S210 , obtaining respiratory signal samples, and dividing the obtained respiratory signal samples according to a preset rule to obtain respiratory signal sections.

[0059] The above-mentioned breathing signal samples can be breathing signals of a certain length of time of the same user used to train the breathing signal prediction model. It should be noted that in the process of training the breathing signal prediction model, in order to ensure that the trained breathing signal prediction model is more universal, a sample library can be constructed in advance, and breathing signal samples can be selected one by one from the sample library. The breathing signal prediction model can be trained using the selected breathing signal samples respectively. After the breathing signal prediction model is trained with multiple breathing signal samples, the final trained breathing signal prediction model is obtained. This embodiment only introduces the process of using one breathing signal sample to train the initial breathing signal prediction model and obtain the trained breathing signal prediction model. In fact, when each breathing signal sample is used to train the breathing signal prediction model, the trained breathing signal prediction model obtained by the previous breathing signal sample can be used as the initial breathing signal prediction model for this model training.

[0060] Therefore, obtaining a respiratory signal sample can involve selecting a respiratory signal sample from a pre-constructed sample library. It should be noted that to ensure the versatility of the trained respiratory signal prediction model, the pre-constructed respiratory signal sample library must be highly generalizable, representing respiratory signals under different conditions. Furthermore, the respiratory signal samples in the respiratory signal sample library must be diverse. Therefore, to obtain diverse respiratory signal samples, it is necessary to ensure diversity in sources, conditions, environments, and methods used in obtaining respiratory signal samples. This source diversity can include data from people of varying ages, genders, ethnicities, health conditions, and regions, to cover as many individual differences as possible. This state diversity can include data from various respiratory states, such as normal breathing, abnormal breathing (e.g., breathing caused by asthma, chronic obstructive pulmonary disease, etc.), breathing after exercise, and breathing after eating. This environmental diversity can include respiratory data from different environmental conditions, such as breathing data in quiet environments, noisy environments, and outdoor environments. The diversity of the above methods can be that the sources of the acquired respiratory signal samples include multiple respiratory signal acquisition methods. For example, the sources of the acquired respiratory signal samples include respiratory data collected by various methods such as a respiratory flow meter, a chest impedance sensor, an abdominal belt sensor or an RPM system.

[0061] In order to facilitate the training of the respiratory signal prediction model, save computing resources, and improve computing efficiency, before training the respiratory signal prediction model, the embodiment of the present application divides the data of the respiratory signal sample into sections to reduce the amount of calculation. The above-mentioned division of the acquired respiratory signal samples according to a preset rule to obtain each respiratory signal section (hereinafter referred to as a section) can be based on a preset time length as the time length of a respiratory signal section, and the acquired respiratory signal samples are divided into sections in chronological order until the time length of the last remaining section is less than or equal to the preset time length, thereby obtaining each respiratory signal section. Alternatively, the number of sections can be pre-set, and the respiratory signal samples are evenly divided into a preset number of sections in chronological order to obtain each respiratory signal section.

[0062] Taking the division according to the preset time length as an example, the specific division process is:

[0063] ;

[0064] Among them, N batch is the number of respiratory signal sections divided by the respiratory signal sample, T is the time length of the respiratory signal sample (corresponding to the index of the last sampling point), N B The preset time length.

[0065] When acquiring a respiratory signal, digital acquisition is typically used to obtain sampling points at preset time intervals. For example, if the sampling frequency is 25 Hz, 1560 respiratory signal data are collected in one minute, i.e., 1560 sampling points are collected in one minute, and each sampling point represents a respiratory signal data value. The preset time interval can be the sampling time interval. The time point corresponding to each sampling point is the sampling time of the signal. When dividing the respiratory signal samples, the different sampling points are also divided, for example, according to 100 sampling points per section.

[0066] The respiratory signal samples are divided into N batch Sections, first N batch -1 section is divided equally, and the respiratory signal data of each section can be expressed as:

[0067] , ;

[0068] in, represents the respiratory signal data of the i-th section of the respiratory signal sample, Represents the index number of the first sampling point in the respiratory signal data of the i-th section of the respiratory signal sample, Represents the first sampling point in the respiratory signal data of the i-th section of the respiratory signal sample, Represents the index number of the last sampling point in the respiratory signal data of the i-th section of the respiratory signal sample, Represents the last sampling point in the respiratory signal data of the i-th section of the respiratory signal sample.

[0069] The first N samples of the breathing signal batch -1 section, the index set B of all sampling points in section i i It can be expressed as:

[0070] ;

[0071] Nth batch The respiratory signal data of a respiratory signal section can be expressed as:

[0072] , ;

[0073] Among them, T is the Nth batch The index of the last sampling point of the respiratory signal section, S(T) is the index of the Nth batch The last sampling point of the respiratory signal section.

[0074] In the last section of the respiratory signal sample, the index set B of all sampling points i It can be expressed as:

[0075] , ;

[0076] In this step, the respiratory signal samples are divided into respiratory signal sections according to preset rules. The division into sections allows calculations to be performed based on the data volume of each section, thus saving computing resources of the computing device and improving computing efficiency.

[0077] In step S220, the initial respiratory signal prediction model is used to predict each respiratory signal section to obtain the prediction results of each respiratory signal section, and the precision error and extreme error of the prediction results of each respiratory signal section are determined based on the prediction results of each respiratory signal section; the precision error represents the overall accuracy of the prediction results of the respiratory signal; the extreme error represents the maximum value of the error of the prediction results of the respiratory signal.

[0078] In this step, the initial respiratory signal prediction model can be a model based on a neural network architecture, such as an LSTM (Long Short-term Memory) or MLP (Multilayer Perceptron), and after initializing the model parameters. It should be noted that the respiratory signal prediction model can also be other neural network models suitable for respiratory signal prediction, and this embodiment does not specifically limit this.

[0079] The above-mentioned prediction results of each respiratory signal section are used to determine the precision error and extreme error of the prediction results of each respiratory signal section. The precision error of the prediction results of each respiratory signal section can be calculated based on the difference between the prediction results of each respiratory signal section at each sampling time and the sampling point. The extreme error of the prediction results of each respiratory signal section is calculated based on the maximum absolute value of the difference between the prediction results of each respiratory signal section at each sampling time and the sampling point, and a penalty coefficient of a preset error level. The error of the above-mentioned prediction result can be the error between the prediction result and the sampling point at the same sampling time. The above-mentioned precision error can be represented by the average value of the root mean square error between the prediction results and the sampling point at each sampling time, and can also be represented by the average value of the absolute error between the prediction results and the sampling point at each sampling time. The calculation method of the above-mentioned precision error is not specifically limited in this embodiment, as long as the overall accuracy of the prediction result of the respiratory signal can be characterized by the precision error.

[0080] Step S230, based on the precision error and extreme error of the prediction results of each respiratory signal section, the precision error weight and extreme error weight of each respiratory signal section, and the penalty loss of the prediction results of each respiratory signal section, determine the loss of the prediction results of the respiratory signal samples; the precision error weight represents the influence factor of the overall accuracy of the prediction result on the prediction result of the respiratory signal sample; the extreme error weight represents the influence factor of the maximum value of the error of the prediction result on the prediction result of the respiratory signal sample; the penalty loss is a clinical constraint function determined according to the relationship between the maximum value of the prediction error of the respiratory signal at the peak and valley and the preset clinical safety threshold.

[0081] The above-mentioned determination of the loss of the prediction result of the respiratory signal sample based on the precision error and extreme error of the prediction result of each respiratory signal section, the precision error weight and extreme error weight of each respiratory signal section, and the penalty loss of the prediction result of each respiratory signal section can be performed by weighted summing the precision error and extreme error of the prediction result of each respiratory signal section based on the precision error weight and extreme error weight of each respiratory signal section, and determining the sum of the weighted summation result and the penalty loss of the prediction result of each respiratory signal section as the loss of the prediction result of each respiratory signal section, and then determining the sum of the losses of the prediction results of each respiratory signal section as the loss of the prediction result of the respiratory signal sample. The above-mentioned loss of the prediction result of the respiratory signal sample can be a quantitative representation of the difference between the predicted value of the respiratory signal sample and the sampling point.

[0082] Step S240 : training the initial respiratory signal prediction model based on the loss of the prediction result of the respiratory signal sample to obtain a trained respiratory signal prediction model.

[0083] The aforementioned training of the initial respiratory signal prediction model based on the loss of the prediction results of the respiratory signal samples to obtain the trained respiratory signal prediction model may involve updating the parameters of the initial respiratory signal prediction model when the loss of the prediction results of the respiratory signal sample to be predicted is greater than a preset loss threshold, and recalculating the loss of the prediction results of the respiratory signal sample to be predicted using the updated prediction model. The process of updating the parameters of the initial respiratory signal prediction model and calculating the loss of the prediction results of the respiratory signal samples is repeated until the number of times the loss of the prediction results of the respiratory signal samples is calculated reaches a preset number requirement, or the loss of the prediction results of the respiratory signal samples reaches a preset loss threshold, thereby obtaining the trained respiratory signal prediction model. The number of times the loss of the prediction results of the respiratory signal samples is calculated can be specifically set according to the specific application scenario and is not specifically limited in this embodiment. The aforementioned preset loss threshold can also be specifically set according to the specific application scenario and is not specifically limited in this embodiment. In the aforementioned application scenario, each irradiated area can be considered as an application scenario. For example, if the area irradiated by radiotherapy is a certain section of the small intestine, the scenario of irradiating this area can be considered as the same application scenario.

[0084] Step S250 , obtaining a respiratory signal to be predicted, and using the trained respiratory signal prediction model to predict the respiratory signal to be predicted.

[0085] In steps S210 to S250, the respiratory signal samples are divided according to a preset rule to obtain respiratory signal segments, and the initial respiratory signal prediction model is used to predict each respiratory signal segment to obtain a prediction result for each respiratory signal segment. Furthermore, the precision error and extreme error of each prediction result are determined based on the prediction result of each respiratory signal segment. The loss of the prediction result of each respiratory signal segment is determined using the precision error and extreme error of the prediction result of each respiratory signal segment, the precision error weights and extreme error weights, and the penalty loss of the prediction result of each respiratory signal segment. The initial respiratory signal prediction model is trained using the loss of the prediction result of the respiratory signal sample, and the trained respiratory signal model is used to predict the respiratory signal to be predicted. In the process of calculating the loss of the prediction results of the respiratory signal samples, the relationship between the maximum error of the prediction results of the respiratory signal and the maximum error of the prediction of the respiratory signal at the peak and valley and the preset clinical safety threshold is used as a clinical constraint function to provide a calculation basis, so that the loss of the prediction results of the respiratory signal samples calculated in this way is used to train the initial respiratory signal prediction model, which can enable the trained respiratory signal prediction model to avoid the situation of large instantaneous errors, and solve the problem that the trained respiratory signal model used in the existing respiratory signal prediction method has large instantaneous errors and causes instantaneous high doses of radiotherapy.

[0086] In one embodiment, step S220 predicts each respiratory signal segment using the initial respiratory signal prediction model to obtain a prediction result for each respiratory signal segment, and determines the accuracy error and extreme error of the prediction result for each respiratory signal segment based on the prediction result of each respiratory signal segment, including:

[0087] Step S222 : using the initial respiratory signal prediction model, predict each respiratory signal segment to obtain a prediction result for each respiratory signal segment.

[0088] The above-mentioned initial respiratory signal prediction model is used to predict each respiratory signal section to obtain the prediction results of each respiratory signal section. Part of the respiratory signal data of each respiratory signal section can be used as a training set and part as a verification set. The data of the training set is before the data of the verification set in time sequence. The data of the training set is input into the initial respiratory signal prediction model to obtain the prediction results of the respiratory signal data of each verification set part.

[0089] Step S224, calculating the accuracy error of the prediction result of each respiratory signal segment based on the difference between the prediction result of each respiratory signal segment at each sampling time and the sampling point; the sampling time is the time point corresponding to each sampling point when the respiratory signal is digitally collected.

[0090] The calculation process of the accuracy error of the prediction result of each respiratory signal section based on the difference between the prediction result of each respiratory signal section at each sampling time and the sampling point is as follows:

[0091] ;

[0092] in, represents the accuracy error of the prediction result of the i-th section of the respiratory signal sample, t represents the index number of the t-th sampling point, Indicates the prediction result of the sampling time corresponding to the t-th sampling point, It represents the t-th sampling point, and can also be expressed as the actual value of the respiratory signal collected at the sampling time corresponding to the t-th sampling point.

[0093] Step S226, based on the maximum absolute value of the difference between the prediction result of each respiratory signal segment at each sampling time and the sampling point, and the penalty coefficient of the preset error level, calculate the extreme error of the prediction result of each respiratory signal segment; the error level includes small error, medium error and out-of-limit error; the penalty coefficient represents the influence factor of different error levels on the prediction result of the respiratory signal.

[0094] In this step, the process of calculating the extreme error of the prediction result of each respiratory signal segment based on the maximum absolute value of the difference between the prediction result of each respiratory signal segment at each sampling time and the sampling point, and the preset error level penalty coefficient is as follows:

[0095] ;

[0096] in, Represents the extreme error of the prediction result of the i-th section of the respiratory signal sample, λ1 is the penalty coefficient for small errors, which is a linear penalty for small errors. The value of λ1 can be set according to the specific situation, for example, λ1 can be 0.5. λ2 is the penalty coefficient for medium errors, which is a linear penalty for medium errors. The value of λ2 can be set according to the specific situation, for example, λ2 can be 1. λ3 is the penalty coefficient for out-of-limit errors, which is a linear penalty for out-of-limit errors. The value of λ3 can be set according to the specific situation, for example, λ3 can be 1.5. MAE i represents the maximum prediction error of the i-th section of the respiratory signal sample, that is, the maximum absolute value of the difference between the prediction result and the sampling point of the i-th section of the respiratory signal sample at each sampling time, , δ1 is the small error threshold, which is the limit value for distinguishing small errors from medium errors, δ2 is the safety threshold, which is the limit value for triggering high penalties, and is the limit value for distinguishing medium errors from over-limit errors. When MAEi For a small error, When MAE i For medium error, When MAE i The values ​​of δ1 and δ2 can be determined based on specific application scenarios or experience and can be adjusted based on specific needs.

[0097] In the above steps S222 to S226, the initial breathing signal prediction model is used to predict each breathing signal section to obtain the prediction results of each breathing signal section, and the precision error and extreme error of the prediction results of each breathing signal section are determined based on the prediction results of each breathing signal section. The determination of the precision error and extreme error of the prediction results of each breathing signal section facilitates the subsequent calculation of the loss of the prediction results of the breathing signal samples using the precision error and extreme error of the prediction results of each breathing signal section.

[0098] Specifically, in one embodiment, before step S230, the following steps are included:

[0099] Step S228, based on the characteristics of each respiratory signal section, determine the accuracy error weight and extreme error weight of each respiratory signal section; the characteristics include amplitude characteristics, frequency characteristics and stability characteristics.

[0100] The above-mentioned determination of the precision error weight and extreme error weight of each respiratory signal section based on the characteristics of each respiratory signal section can be performed by constructing a feature vector of each respiratory signal section based on the characteristics of each respiratory signal section, and then determining the precision error weight and extreme error weight of each respiratory signal section based on the feature vector of each respiratory signal section and the sensitivity coefficient of each feature.

[0101] In addition, in one embodiment, step S228, determining the accuracy error weight and extreme error weight of each respiratory signal segment based on the characteristics of each respiratory signal segment, includes:

[0102] Step S2282: construct a feature vector of each respiratory signal segment based on the features of each respiratory signal segment; the feature vector is a vector formed by combining the normalized features of the respiratory signal.

[0103] Before constructing the feature vector of each respiratory signal section based on the features of each respiratory signal section, the features of each respiratory signal section need to be determined.

[0104] Specifically, the calculation process for determining the amplitude characteristics of each respiratory signal segment is as follows:

[0105] ;

[0106] in, is the amplitude feature of the i-th section of the respiratory signal sample.

[0107] The calculation process for determining the frequency characteristics of each respiratory signal section is:

[0108] ;

[0109] in, is the frequency feature of the i-th section of the respiratory signal sample, N zero To count the number of times the signal crosses the mean line within the window, f s is the sampling frequency.

[0110] The calculation process for determining the stability characteristics of each respiratory signal segment is:

[0111] ;

[0112] in, is the stability feature of the i-th section of the respiratory signal sample, is the mean value of the signal amplitude in the i-th section of the respiratory signal sample.

[0113] Preferably, before determining the characteristics of each respiratory signal segment, each respiratory signal segment may be preprocessed, such as using Dobesi 5 wavelet to remove high-frequency noise in each respiratory signal segment and retain the effective frequency band in the respiratory signal.

[0114] The above-mentioned construction of the feature vector of each respiratory signal section based on the features of each respiratory signal section can be performed by normalizing the features of each respiratory signal section to obtain the normalized features of each respiratory signal section, and constructing the feature vector of each respiratory signal section based on the normalized features of each respiratory signal section.

[0115] The process of normalizing the features of each respiratory signal segment to obtain the normalized features of each respiratory signal segment is as follows:

[0116] The amplitude characteristics of each respiratory signal section are normalized to obtain the normalized amplitude characteristics of each respiratory signal section as follows:

[0117] ;

[0118] in, is the normalized amplitude feature of the i-th section of the respiratory signal sample.

[0119] The frequency characteristics of each respiratory signal section are normalized to obtain the normalized frequency characteristics of each respiratory signal section as follows:

[0120] ;

[0121] in, is the normalized frequency feature of the i-th section of the respiratory signal sample.

[0122] The stability characteristics of each respiratory signal section are normalized to obtain the normalized stability characteristics of each respiratory signal section as follows:

[0123] ;

[0124] in, is the normalized stability feature of the i-th section of the respiratory signal sample.

[0125] Furthermore, based on the normalized features of each respiratory signal segment, the process of constructing the feature vector F(t) of each respiratory signal segment is as follows:

[0126] ;

[0127] Step S2284, based on the feature vectors of each respiratory signal segment and the sensitivity coefficient of each feature, determine the precision error weight and extreme error weight of each respiratory signal segment; the sensitivity coefficient represents the influence factor of the feature change on the prediction result of the respiratory signal.

[0128] The above-mentioned determination of the precision error weight and extreme error weight of each respiratory signal section based on the characteristic vector of each respiratory signal section and the sensitivity coefficient of each feature can be based on the characteristic vector of each respiratory signal section and the sensitivity coefficient of each feature to calculate the initial precision error weight and initial extreme error weight of each respiratory signal section, and then, based on the amplitude characteristic component in the characteristic vector of each respiratory signal section, adjust the initial precision error weight and initial extreme error weight of each respiratory signal section to obtain the precision error weight and extreme error weight of each respiratory signal section.

[0129] In the above steps S2282 to S2284, a feature vector of each respiratory signal section is constructed based on the features of each respiratory signal section, and the precision error weight and extreme error weight of each respiratory signal section are determined based on the feature vector of each respiratory signal section and the sensitivity coefficient of each feature. The determination of the precision error weight and extreme error weight of each respiratory signal section facilitates the subsequent use of the precision error weight and extreme error weight of each respiratory signal section to determine the loss of the prediction results of the respiratory signal samples.

[0130] In addition, in one embodiment, before step S2284, the following steps are included:

[0131] Step S2283: Determine the sensitivity coefficient of each feature based on the application scenario corresponding to the respiratory signal sample.

[0132] The sensitivity coefficients of each of the above-mentioned features include the amplitude feature sensitivity coefficient corresponding to the amplitude feature, the frequency feature sensitivity coefficient corresponding to the frequency feature, and the stability feature sensitivity coefficient corresponding to the stability feature. The amplitude feature sensitivity coefficient represents the influence of changes in the amplitude feature on the prediction results of the respiratory signal. A larger amplitude feature sensitivity coefficient indicates a greater influence of changes in the amplitude feature on the prediction results of the respiratory signal, and vice versa. The frequency feature sensitivity coefficient represents the influence of changes in the frequency feature on the prediction results of the respiratory signal. A larger frequency feature sensitivity coefficient indicates a greater influence of changes in the frequency feature on the prediction results of the respiratory signal, and vice versa. The stability feature sensitivity coefficient represents the influence of changes in the stability feature on the prediction results of the respiratory signal. A larger stability feature sensitivity coefficient indicates a greater influence of changes in the stability feature on the prediction results of the respiratory signal, and vice versa. In general, the sensitivity coefficients of each feature can be determined based on the application scenario corresponding to the respiratory signal sample. For example, if the patient's target area is close to a serially endangered organ and urgently needs protection, the amplitude feature sensitivity coefficient and the stability feature sensitivity coefficient can be appropriately increased to better reduce the maximum error of the respiratory signal prediction results. If the patient does not have any critical organs, the frequency feature sensitivity coefficient can be significantly increased to reduce the average error of the respiratory signal prediction results. It should be noted that the sensitivity coefficients of the features of the same respiratory signal sample are the same.

[0133] Furthermore, in one embodiment, step S2284, determining the precision error weight and extreme error weight of each respiratory signal segment based on the feature vector of each respiratory signal segment and the sensitivity coefficient of each feature, includes:

[0134] Step S1 : Calculate the initial precision error weight and initial extreme error weight of each respiratory signal segment based on the feature vector of each respiratory signal segment and the sensitivity coefficient of each feature.

[0135] The calculation process of the initial accuracy error weight of each respiratory signal section based on the feature vector of each respiratory signal section and the sensitivity coefficient of each feature is as follows:

[0136] ;

[0137] in, is the initial accuracy error weight of the i-th section of the respiratory signal sample, is the amplitude characteristic sensitivity coefficient, is the frequency characteristic sensitivity coefficient, is the stability characteristic sensitivity coefficient.

[0138] The calculation process of the initial extreme error weight of each respiratory signal section is:

[0139] ;

[0140] in, is the initial extreme error weight of the i-th section of the respiratory signal sample.

[0141] Step S2: Based on the amplitude characteristic component in the characteristic vector of each respiratory signal segment, the initial precision error weight and the initial extreme error weight of each respiratory signal segment are adjusted to obtain the precision error weight and the extreme error weight of each respiratory signal segment.

[0142] The calculation process of step S1 can determine the initial precision error weight and initial extreme error weight of each respiratory signal section. At this time, it is necessary to consider whether the respiratory signal sample is deep breathing or shallow breathing. Different breathing patterns need to set different weight thresholds to avoid the risk of extreme errors. The amplitude characteristic component in the eigenvector of each respiratory signal section can reflect whether the breathing pattern of the respiratory signal is deep breathing or shallow breathing. An amplitude breathing pattern threshold can be set based on experience. When the amplitude characteristic component in the eigenvector of the respiratory signal section is less than or equal to the breathing pattern threshold, the breathing pattern is shallow breathing. When the amplitude characteristic component in the eigenvector of the respiratory signal section is greater than the breathing pattern threshold, the breathing pattern is deep breathing. When the breathing pattern is shallow breathing, it is necessary to limit the minimum threshold of the precision error weight of the respiratory signal to ensure the precision error weight. For example, the minimum threshold of the precision error weight of the respiratory signal can be limited to 0.9. That is, when the initial precision error weight obtained is less than 0.9, the precision error weight is directly adjusted to 0.9. At this time, the extreme error weight is 0.1. If the initial precision error weight obtained at this time is greater than or equal to 0.9, no adjustment is required. When the breathing mode is deep breathing, the extreme error weight value can be directly set to suppress the extreme displacement risk. For example, when taking a deep breath, the extreme error weight is set to 0.2 to suppress the extreme displacement risk. At this time, the precision error weight of the respiratory signal section is 0.8. It should be noted that the minimum threshold of the precision error weight of the respiratory signal during shallow breathing and the set value of the extreme error weight during deep breathing can be set according to the specific situation. This embodiment does not make specific restrictions here. As long as it can ensure that the overall accuracy of the prediction result is the main influencing factor of the prediction result of the respiratory signal sample during shallow breathing, and the extreme displacement risk is suppressed during deep breathing, it can be used.

[0143] In steps S1 and S2, the initial precision error weight and initial extreme error weight for each respiratory signal segment are calculated based on the feature vectors of each respiratory signal segment and the sensitivity coefficients of each feature. The initial precision error weight and initial extreme error weight for each respiratory signal segment are adjusted based on the amplitude feature components in the feature vectors of each respiratory signal segment to obtain the precision error weight and extreme error weight for each respiratory signal segment. Determining the precision error weight and extreme error weight for each respiratory signal segment facilitates determining the loss of the prediction results for each respiratory signal sample based on the precision error weight and extreme error weight for each respiratory signal segment.

[0144] In one embodiment, step S230 determines the loss of the prediction result of the respiratory signal sample based on the precision error and extreme error of the prediction result of each respiratory signal section, the precision error weight and extreme error weight of each respiratory signal section, and the penalty loss of the prediction result of each respiratory signal section, including:

[0145] In step S232, based on the peak safety threshold and trough safety threshold corresponding to the current application scenario, the penalty loss of the prediction results of each respiratory signal segment is calculated; the peak safety threshold is the maximum prediction error of the peak of the respiratory signal when the distance between the target area and the endangered organ is at a safe distance; the trough safety threshold is the maximum prediction error of the trough of the respiratory signal when the distance between the target area and the endangered organ is at a safe distance.

[0146] The above-mentioned calculation of the penalty loss of the prediction result of each respiratory signal segment based on the peak safety threshold and trough safety threshold corresponding to the current application scenario can be performed by determining the peak time point set of each respiratory signal segment, determining the maximum peak prediction error of each respiratory signal segment based on the difference between the sampling point of each respiratory signal segment at each sampling time in the peak time point set and the predicted value, determining the trough time point set of each respiratory signal segment, determining the maximum trough prediction error of each respiratory signal segment based on the difference between the sampling point of each respiratory signal segment at each sampling time in the trough time point set and the predicted value, and calculating the penalty loss of the prediction result of each respiratory signal segment based on the peak safety threshold and trough safety threshold corresponding to the current application scenario, as well as the maximum peak prediction error and the maximum trough prediction error of each respiratory signal segment. The above-mentioned peak safety threshold and trough safety threshold can be extracted from the radiotherapy plan to check the clinical safety threshold corresponding to the serial organs at risk at the peak and trough of the respiratory signal. For example, for the spinal cord, the 45Gy dose line can be obtained by using the images corresponding to the peaks and troughs in 4D-C and the dose contour map generated by the planned design. This allows the minimum distance of this dose line from the spinal cord at the peaks and troughs of respiratory waves to be determined. Constraints are implemented by embedding a clinical penalty mechanism in the loss function based on peak and trough safety thresholds.

[0147] In step S234, based on the precision error weight and extreme error weight of each respiratory signal section, the precision error and extreme error of the prediction results of each respiratory signal section are weightedly summed, and the sum of the weighted summation result and the penalty loss of the prediction results of each respiratory signal section is determined as the loss of the prediction results of each respiratory signal section.

[0148] The above-mentioned precision error weight and extreme error weight of each respiratory signal section are weightedly summed for the precision error and extreme error of the prediction result of each respiratory signal section, and the sum of the weighted summation result and the penalty loss of the prediction result of each respiratory signal section is determined as the calculation process of the loss of the prediction result of each respiratory signal section as follows:

[0149] ;

[0150] in, is the loss of the prediction result of the i-th section of the respiratory signal sample, is the penalty loss for the prediction result of the i-th section of the respiratory signal sample.

[0151] Step S236 : Determine the sum of the losses of the prediction results of the respiratory signal sections as the loss of the prediction result of the respiratory signal sample.

[0152] In this step, the loss of the prediction result of the breathing signal sample is calculated The process is:

[0153] ;

[0154] In the above steps S232 to S236, the penalty loss of the prediction results of each respiratory signal section is calculated based on the peak safety threshold and the trough safety threshold corresponding to the current application scenario. Then, based on the precision error weight and the extreme error weight of each respiratory signal section, the precision error and the extreme error of the prediction results of each respiratory signal section are weightedly summed. The sum of the weighted summation result and the penalty loss of the prediction results of each respiratory signal section is determined as the loss of the prediction results of each respiratory signal section. The sum of the losses of the prediction results of each respiratory signal section is determined as the loss of the prediction results of the respiratory signal samples. The calculation of the loss of the prediction results of the respiratory signal samples facilitates the subsequent training of the initial respiratory signal prediction model based on the loss of the prediction results of the respiratory signal samples to obtain the trained respiratory signal prediction model.

[0155] Specifically, in one embodiment, step S232 calculates the penalty loss of the prediction result of each respiratory signal segment based on the peak safety threshold and the trough safety threshold corresponding to the current application scenario, including:

[0156] Step S2321 , determining a peak time point set of each respiratory signal section; the peak time point set is a collection of sampling times of all peaks of each respiratory signal section.

[0157] The above-mentioned determination of the peak time point set of each respiratory signal section can be performed by traversing each sampling point starting from the second sampling point of the respiratory signal section. If the respiratory signal data of the current sampling point and the respiratory signal data of the previous sampling point and the next sampling point are all positive, and the respiratory signal data value of the current sampling point is the largest, then the sampling time corresponding to the current sampling point is determined as the peak time point. After traversing the respiratory signal section, all the peak time points of the respiratory signal section are obtained, and all the peak time points of the respiratory signal section are constructed as the peak time point set of the current respiratory signal section.

[0158] In addition, the process of determining the peak time point set of each respiratory signal section can also be:

[0159] ;

[0160] in, is the peak time point set of the i-th section of the respiratory signal sample, ζ is the amplitude threshold, m is a positive integer, and is the half-width of the window for peak or trough detection, indicating that m adjacent points are taken around time point t to form a 2m+1 window. For example, when m=2, the window for peak or trough detection contains 5 sampling points, namely t-2, t-1, t, t+1, and t+2. By setting m, it is ensured that t is a local extreme value within the window (rather than an isolated noise point). The size of m can be set according to specific needs and is not specifically limited in this embodiment. k is the offset within the window, which is used to traverse all adjacent points within the window to verify whether t meets the local maximum or local minimum conditions. For example, when detecting a peak, if m=2, then when k=1, points t-1 and t+1 are checked, and when k=2, points t-2 and t+2 are checked; all the adjacent points corresponding to k must meet the requirements before t is considered a peak.

[0161] Step S2322, based on the difference between the sampling points of each respiratory signal segment at each sampling time in the peak time point set and the predicted value, determine the maximum peak prediction error of each respiratory signal segment; the maximum peak prediction error is the maximum value of the absolute value of the difference between the sampling points of the respiratory signal segment at each sampling time in the peak time point set and the predicted value.

[0162] The calculation process for determining the maximum prediction error of the peak of each respiratory signal segment based on the difference between the sampling point at each sampling time of the peak time point set and the predicted value of each respiratory signal segment is as follows:

[0163] ;

[0164] in, is the maximum prediction error of the peak of the i-th section of the respiratory signal sample.

[0165] Step S2323 , determining a trough time point set of each respiratory signal segment; the trough time point set is a collection of sampling times of all troughs of each respiratory signal segment.

[0166] The above-mentioned determination of the trough time point set of each respiratory signal section can be performed by traversing each sampling point starting from the second sampling point of the respiratory signal section. If the respiratory signal data of the current sampling point and the respiratory signal data of the previous sampling point and the next sampling point are all negative, and the absolute value of the respiratory signal data of the current sampling point is the largest, then the sampling time corresponding to the current sampling point is determined as the trough time point. After traversing the respiratory signal section, all trough time points of the respiratory signal section are obtained. Based on all the trough time points of the respiratory signal section, the trough time point set of the current respiratory signal section is obtained.

[0167] In addition, the process of determining the trough time point set of each respiratory signal section may also be:

[0168] ;

[0169] in, is the set of trough time points of the i-th section of the respiratory signal sample.

[0170] Step S2324, based on the difference between the sampling point of each respiratory signal segment at each sampling time in the trough time point set and the predicted value, determine the maximum trough prediction error of each respiratory signal segment; the maximum trough prediction error is the maximum value of the absolute value of the difference between the sampling point of each respiratory signal segment at each sampling time in the trough time point set and the predicted value.

[0171] The process of determining the maximum prediction error of the trough of each respiratory signal segment based on the difference between the sampling point and the predicted value at each sampling time of the trough time point set of each respiratory signal segment is as follows:

[0172] ;

[0173] in, is the maximum prediction error of the trough of the i-th section of the respiratory signal sample.

[0174] Step S2325 , based on the peak safety threshold and trough safety threshold corresponding to the current application scenario, and the peak maximum prediction error and trough maximum prediction error of each respiratory signal segment, calculate the penalty loss of the prediction result of each respiratory signal segment.

[0175] Specifically, based on the peak safety threshold and trough safety threshold corresponding to the current application scenario, as well as the peak maximum prediction error and trough maximum prediction error of each respiratory signal segment, the penalty loss of the prediction result of each respiratory signal segment is calculated as follows:

[0176] ;

[0177] in, is the peak safety threshold corresponding to the current application scenario, It is the trough safety threshold corresponding to the current application scenario.

[0178] The above steps S2321 to S2325 calculate the penalty loss of the prediction results of each respiratory signal section through the peak safety threshold and the trough safety threshold corresponding to the current application scenario. The determination of the penalty loss of the prediction results of each respiratory signal section facilitates the subsequent determination of the loss of the prediction results of each respiratory signal section based on the penalty loss of the prediction results of each respiratory signal section.

[0179] In one embodiment, step S240, based on the loss of the prediction result of the respiratory signal sample, the initial respiratory signal prediction model is trained to obtain the trained respiratory signal prediction model, including:

[0180] Step S242: When the loss of the prediction result of the respiratory signal sample to be predicted is greater than a preset loss threshold, the parameters of the initial respiratory signal prediction model are updated, and the loss of the prediction result of the respiratory signal sample to be predicted is recalculated using the updated prediction model. The process of updating the parameters of the initial respiratory signal prediction model and calculating the loss of the prediction result of the respiratory signal sample is repeated until the number of times the loss of the prediction result of the respiratory signal sample is calculated reaches a preset number requirement, or the loss of the prediction result of the respiratory signal sample reaches a preset loss threshold, thereby obtaining a trained respiratory signal prediction model.

[0181] The present embodiment is described and illustrated below through preferred embodiments.

[0182] Figure 3 This is a flow chart of a respiratory signal prediction method provided by a preferred embodiment of the present application. Figure 3 As shown, the respiratory signal prediction method includes the following steps:

[0183] Step S301, obtaining respiratory signal samples, and dividing the obtained respiratory signal samples according to a preset rule to obtain various respiratory signal sections;

[0184] Step S302 , calculating the accuracy error of the prediction result of each respiratory signal segment based on the difference between the prediction result of each respiratory signal segment at each sampling time and the sampling point;

[0185] Step S303, calculating the extreme error of the prediction result of each respiratory signal segment based on the maximum value of the difference between the prediction result of each respiratory signal segment at each sampling time and the sampling point, and a preset error level penalty coefficient;

[0186] Step S304, constructing a feature vector of each respiratory signal segment based on the features of each respiratory signal segment;

[0187] Step S305: determining the sensitivity coefficient of each feature based on the application scenario corresponding to the respiratory signal sample;

[0188] Step S306 , determining the precision error weight and extreme error weight of each respiratory signal segment based on the feature vector of each respiratory signal segment and the sensitivity coefficient of each feature;

[0189] Step S307 , determining the loss of the prediction result of the respiratory signal sample based on the precision error and extreme error of the prediction result of each respiratory signal segment, the precision error weight and extreme error weight of each respiratory signal segment, and the penalty loss of the prediction result of each respiratory signal segment;

[0190] Step S308 , training the initial respiratory signal prediction model based on the loss of the prediction result of the respiratory signal sample to obtain a trained respiratory signal prediction model;

[0191] Step S309 : obtaining a respiratory signal to be predicted, and using the trained respiratory signal prediction model to predict the respiratory signal to be predicted.

[0192] In steps S301 to S309, the respiratory signal samples are divided according to a preset rule to obtain respiratory signal segments, and the initial respiratory signal prediction model is used to predict each respiratory signal segment to obtain a prediction result for each respiratory signal segment. Furthermore, the precision error and extreme error of each prediction result are determined based on the prediction result of each respiratory signal segment. The loss of the prediction result of each respiratory signal segment is determined using the precision error and extreme error of the prediction result of each respiratory signal segment, the precision error weights and extreme error weights, and the penalty loss of the prediction result of each respiratory signal segment. The initial respiratory signal prediction model is trained using the loss of the prediction result of the respiratory signal sample, and the trained respiratory signal model is used to predict the respiratory signal to be predicted. In the process of calculating the loss of the prediction results of the respiratory signal samples, the relationship between the maximum error of the prediction results of the respiratory signal and the maximum error of the prediction of the respiratory signal at the peak and valley and the preset clinical safety threshold is used as a clinical constraint function to provide a calculation basis, so that the loss of the prediction results of the respiratory signal samples calculated in this way is used to train the initial respiratory signal prediction model, which can enable the trained respiratory signal prediction model to avoid the situation of large instantaneous errors, and solve the problem that the trained respiratory signal model used in the existing respiratory signal prediction method has large instantaneous errors and causes instantaneous high doses of radiotherapy.

[0193] It should be understood that, although the various steps in the flowcharts involved in the various embodiments described above are displayed in sequence according to the instructions of the arrows, these steps are not necessarily executed in sequence in the order indicated by the arrows. Unless otherwise specified herein, there is no strict order restriction on the execution of these steps, and these steps can be executed in other orders. Moreover, at least a portion of the steps in the flowcharts involved in the various embodiments described above can include multiple steps or multiple stages, and these steps or stages are not necessarily executed and completed at the same time, but can be executed at different times, and the execution order of these steps or stages is not necessarily to be carried out in sequence, but can be executed in turn or alternately with other steps or at least a portion of steps or stages in other steps.

[0194] Based on the same inventive concept, this embodiment also provides a respiratory signal prediction device, which is used to implement the above-mentioned embodiments and preferred embodiments. Details already described will not be repeated here. The terms "module," "unit," "subunit," etc. used below may refer to a combination of software and / or hardware that implements a predetermined function. Although the devices described in the following embodiments are preferably implemented in software, implementation using hardware, or a combination of software and hardware, is also possible and contemplated.

[0195] In one embodiment, Figure 4 This is a structural block diagram of a respiratory signal prediction device provided by an embodiment of the present application. Figure 4 As shown, the respiratory signal prediction device includes:

[0196] The sample division module 41 is used to obtain respiratory signal samples and divide the obtained respiratory signal samples according to preset rules to obtain various respiratory signal sections;

[0197] The error determination module 42 is configured to use the initial respiratory signal prediction model to predict each respiratory signal segment to obtain a prediction result for each respiratory signal segment, and determine the precision error and extreme error of the prediction result for each respiratory signal segment based on the prediction result of each respiratory signal segment; the precision error represents the overall accuracy of the respiratory signal prediction result; the extreme error represents the maximum error of the respiratory signal prediction result;

[0198] a loss determination module 43 for determining the loss of the prediction result of the respiratory signal sample based on the precision error and extreme error of the prediction result of each respiratory signal segment, the precision error weight and extreme error weight of each respiratory signal segment, and the penalty loss of the prediction result of each respiratory signal segment; the precision error weight represents the influence factor of the overall accuracy of the prediction result on the prediction result of the respiratory signal sample; the extreme error weight represents the influence factor of the maximum value of the error of the prediction result on the prediction result of the respiratory signal sample; the penalty loss is a clinical constraint function determined based on the relationship between the maximum value of the prediction error at the peak and valley of the respiratory signal and a preset clinical safety threshold;

[0199] A model training module 44 is configured to train an initial respiratory signal prediction model based on the loss of the prediction results of the respiratory signal samples to obtain a trained respiratory signal prediction model;

[0200] The prediction module 45 is configured to obtain a respiratory signal to be predicted and predict the respiratory signal to be predicted using the trained respiratory signal prediction model.

[0201] The above-mentioned breathing signal prediction device divides the breathing signal samples according to a preset rule to obtain individual breathing signal segments, and uses the initial breathing signal prediction model to predict each breathing signal segment to obtain prediction results for each breathing signal segment. Then, based on the prediction results of each breathing signal segment, the precision error and extreme error of each prediction result are determined. The loss of the prediction result of each breathing signal segment is determined using the precision error and extreme error of the prediction results of each breathing signal segment, the respective precision error weights and extreme error weights, and the penalty loss of the prediction results of each breathing signal segment. The initial breathing signal prediction model is trained using the loss of the prediction results of the breathing signal samples, and the trained breathing signal model is used to predict the breathing signal to be predicted. In the process of calculating the loss of the prediction results of the respiratory signal samples, the relationship between the maximum error of the prediction results of the respiratory signal and the maximum error of the prediction of the respiratory signal at the peak and valley and the preset clinical safety threshold is used as a clinical constraint function to provide a calculation basis, so that the loss of the prediction results of the respiratory signal samples calculated in this way is used to train the initial respiratory signal prediction model, which can enable the trained respiratory signal prediction model to avoid the situation of large instantaneous errors, and solve the problem that the trained respiratory signal model used in the existing respiratory signal prediction method has large instantaneous errors and causes instantaneous high doses of radiotherapy.

[0202] In one embodiment, the error determination module 42 further includes:

[0203] A prediction module, configured to predict each respiratory signal segment using an initial respiratory signal prediction model to obtain a prediction result for each respiratory signal segment;

[0204] an accuracy error calculation module, configured to calculate the accuracy error of the prediction result of each respiratory signal segment based on the difference between the prediction result of each respiratory signal segment at each sampling time and the sampling point; the sampling time is the time point corresponding to each sampling point when the respiratory signal is digitally collected;

[0205] And an extreme error calculation module, which is used to calculate the extreme error of the prediction result of each respiratory signal section based on the maximum value of the difference between the prediction result of each respiratory signal section at each sampling time and the sampling point, and the penalty coefficient of the preset error level; the error level includes small error, medium error and out-of-limit error; the penalty coefficient represents the influence factor of different error levels on the prediction result of the respiratory signal.

[0206] In one embodiment, the respiratory signal prediction device further includes: a weight determination module for determining the precision error weight and the extreme error weight of each respiratory signal section based on the characteristics of each respiratory signal section; the characteristics include amplitude characteristics, frequency characteristics and stability characteristics.

[0207] In one embodiment, the weight determination module further includes:

[0208] A feature vector construction unit is used to construct a feature vector of each respiratory signal segment based on the features of each respiratory signal segment; the feature vector is a vector formed by combining the normalized features of the respiratory signal;

[0209] The weight determination unit is also used to determine the precision error weight and the extreme error weight of each respiratory signal section based on the feature vector of each respiratory signal section and the sensitivity coefficient of each feature; the sensitivity coefficient represents the influence factor of the change of the feature on the prediction result of the respiratory signal.

[0210] In one embodiment, the respiratory signal prediction device further includes: a sensitivity coefficient determination module, configured to determine the sensitivity coefficient of each of the features based on an application scenario corresponding to the respiratory signal sample.

[0211] In one embodiment, the respiratory signal prediction device further includes:

[0212] The weight determination unit is further configured to calculate an initial precision error weight and an initial extreme error weight of a prediction result of each respiratory signal segment based on the feature vector of each respiratory signal segment and the sensitivity coefficient of each feature;

[0213] The weight determination unit is also used to adjust the initial precision error weight and the initial extreme error weight of the prediction results of each respiratory signal section based on the amplitude characteristic component in the characteristic vector of each respiratory signal section, so as to obtain the precision error weight and the extreme error weight of each respiratory signal section.

[0214] In one embodiment, the loss determination module includes:

[0215] a penalty loss determination unit, configured to calculate a penalty loss for the prediction results of each respiratory signal segment based on a peak safety threshold and a trough safety threshold corresponding to the current application scenario; the peak safety threshold being a maximum prediction error of a peak of the respiratory signal when the distance between the target area and the organ at risk is within a safe distance; and the trough safety threshold being a maximum prediction error of a trough of the respiratory signal when the distance between the target area and the organ at risk is within a safe distance;

[0216] a measure loss determining unit, configured to perform a weighted summation of the precision error and the extreme error of the prediction result of each respiratory signal measure based on the precision error weight and the extreme error weight of each respiratory signal measure, and determine the sum of the weighted summation result and the penalty loss of the prediction result of each respiratory signal measure as the loss of the prediction result of each respiratory signal measure;

[0217] The overall loss determining unit is configured to determine the sum of the losses of the prediction results of the respective respiratory signal sections as the loss of the prediction result of the respiratory signal sample.

[0218] In one embodiment, the penalty loss determination unit includes:

[0219] The first point set determination subunit is configured to determine a peak time point set of each respiratory signal segment; the peak time point set is a collection of sampling times of all peaks of each respiratory signal segment;

[0220] a first error determination subunit, configured to determine a peak maximum prediction error of each respiratory signal segment based on a difference between a sampling point of each respiratory signal segment at each sampling time in the peak time point set and a predicted value; the peak maximum prediction error being the maximum absolute value of the difference between a sampling point of each respiratory signal segment at each sampling time in the peak time point set and a predicted value;

[0221] The second point set determination subunit is used to determine the trough time point set of each respiratory signal section; the trough time point set is the collection of sampling times of all troughs of each respiratory signal section;

[0222] a second error determination subunit, configured to determine a maximum trough prediction error of each respiratory signal segment based on a difference between a sampling point of each respiratory signal segment at each sampling time in the trough time point set and a predicted value; the maximum trough prediction error being the maximum absolute value of the difference between a sampling point of each respiratory signal segment at each sampling time in the trough time point set and a predicted value;

[0223] And a penalty loss determination subunit, which is used to calculate the penalty loss of the prediction results of each respiratory signal section based on the peak safety threshold and the trough safety threshold corresponding to the current application scenario, and the peak maximum prediction error and the trough maximum prediction error of each respiratory signal section.

[0224] In one embodiment, the model training module includes: a model training unit, which is used to update the parameters of the initial respiratory signal prediction model when the loss of the prediction result of the respiratory signal sample to be predicted is greater than a preset loss threshold, and use the updated prediction model to recalculate the loss of the prediction result of the respiratory signal sample to be predicted, and repeat the process of updating the parameters of the initial respiratory signal prediction model and calculating the loss of the prediction result of the respiratory signal sample until the number of times the loss of the prediction result of the respiratory signal sample is calculated reaches a preset number requirement, or the loss of the prediction result of the respiratory signal sample reaches a preset loss threshold, thereby obtaining the trained respiratory signal prediction model.

[0225] It should be noted that the above modules can be functional modules or program modules, and can be implemented through software or hardware. For modules implemented through hardware, the above modules can be located in the same processor; or the above modules can be located in different processors in any combination.

[0226] It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, stored data, displayed data, etc.) involved in this application are all information and data authorized by the user or fully authorized by all parties.

[0227] Those skilled in the art will appreciate that all or part of the processes in the above-mentioned embodiments can be implemented by instructing the relevant hardware through a computer program. The computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the above-mentioned embodiments. In particular, any reference to memory, database, or other media used in the embodiments provided in this application can include at least one of non-volatile and volatile memory. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory, optical memory, high-density embedded non-volatile memory, resistive random access memory (ReRAM), magnetic random access memory (MRAM), ferroelectric random access memory (FRAM), phase change memory (PCM), graphene memory, etc. Volatile memory can include random access memory (RAM) or external cache memory, etc. By way of illustration and not limitation, RAM can take various forms, such as static random access memory (SRAM) or dynamic random access memory (DRAM). The databases involved in the various embodiments provided herein may include at least one of a relational database and a non-relational database. Non-relational databases may include, but are not limited to, distributed databases based on blockchains. The processors involved in the various embodiments provided herein may be, but are not limited to, general-purpose processors, central processing units (CPUs), graphics processing units (GPUs), digital signal processors (DSPs), programmable logic devices (PLDs), data processing logic devices based on quantum computing, and the like.

[0228] The technical features of the above embodiments can be combined arbitrarily. To make the description concise, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0229] The above-described embodiments merely represent several implementation methods of the present application. While the descriptions are relatively specific and detailed, they should not be construed as limiting the scope of the present application. It should be noted that a person of ordinary skill in the art may make various modifications and improvements without departing from the spirit of the present application, and these modifications and improvements fall within the scope of protection of the present application. Therefore, the scope of protection of the present application shall be determined by the appended claims.

Claims

1. A respiratory signal prediction method, characterized in that: The method comprises: Acquire respiratory signal samples, and divide the acquired respiratory signal samples according to preset rules to obtain various respiratory signal sections; Using the initial respiratory signal prediction model, predicting each respiratory signal section to obtain a prediction result of each respiratory signal section, and determining, based on the prediction result of each respiratory signal section, a precision error and an extreme error of the prediction result of each respiratory signal section; the precision error represents the overall accuracy of the prediction result of the respiratory signal; and the extreme error represents the maximum value of the error of the prediction result of the respiratory signal; Based on the precision error and the extreme error of the prediction results of each respiratory signal section, the precision error weight and the extreme error weight of each respiratory signal section, and the penalty loss of the prediction results of each respiratory signal section, the loss of the prediction results of the respiratory signal samples is determined; the precision error weight represents the influence factor of the overall accuracy of the prediction result on the prediction result of the respiratory signal sample; the extreme error weight represents the influence factor of the maximum value of the error of the prediction result on the prediction result of the respiratory signal sample; the penalty loss is a clinical constraint function determined according to the relationship between the maximum value of the prediction error of the respiratory signal at the peak and valley and the preset clinical safety threshold; Training the initial respiratory signal prediction model based on the loss of the prediction result of the respiratory signal sample to obtain a trained respiratory signal prediction model; A respiratory signal to be predicted is acquired, and the trained respiratory signal prediction model is used to predict the respiratory signal to be predicted.

2. The respiratory signal prediction method according to claim 1, wherein The method of using the initial respiratory signal prediction model to predict each respiratory signal segment to obtain a prediction result of each respiratory signal segment, and determining the accuracy error and extreme error of the prediction result of each respiratory signal segment based on the prediction result of each respiratory signal segment, includes: Using the initial respiratory signal prediction model, predicting each respiratory signal section to obtain a prediction result for each respiratory signal section; Calculating the accuracy error of the prediction result of each respiratory signal section based on the difference between the prediction result of each respiratory signal section at each sampling time and the sampling point; the sampling time is the time point corresponding to each sampling point when the respiratory signal is digitally collected; Based on the maximum absolute value of the difference between the prediction result of each respiratory signal section at each sampling time and the sampling point, and the penalty coefficient of the preset error level, the extreme error of the prediction result of each respiratory signal section is calculated; the error level includes small error, medium error and out-of-limit error; the penalty coefficient represents the influence factor of different error levels on the prediction result of the respiratory signal.

3. The respiratory signal prediction method according to claim 1, wherein Before determining the loss of the prediction result of the respiratory signal sample based on the precision error and the extreme error of the prediction result of each respiratory signal section, the precision error weight and the extreme error weight of each respiratory signal section, and the penalty loss of the prediction result of each respiratory signal section, the method includes: The precision error weight and the extreme error weight of each respiratory signal section are determined based on the characteristics of each respiratory signal section; the characteristics include amplitude characteristics, frequency characteristics and stability characteristics.

4. The respiratory signal prediction method according to claim 3, wherein The determining, based on the features of the respective respiratory signal sections, the precision error weight and the extreme error weight of the respective respiratory signal sections, comprises: Based on the features of each respiratory signal section, constructing a feature vector of each respiratory signal section; the feature vector is a vector formed by combining the normalized features of the respiratory signal; Based on the feature vectors of each respiratory signal section and the sensitivity coefficient of each feature, the precision error weight and the extreme error weight of each respiratory signal section are determined; the sensitivity coefficient represents the influence factor of the feature change on the prediction result of the respiratory signal.

5. The respiratory signal prediction method according to claim 4, wherein Before determining the precision error weight and the extreme error weight of each respiratory signal section based on the feature vector of each respiratory signal section and the sensitivity coefficient of each feature, the method includes: Based on the application scenario corresponding to the respiratory signal sample, the sensitivity coefficient of each feature is determined.

6. The respiratory signal prediction method according to claim 4, wherein: The method of determining the precision error weight and the extreme error weight of each respiratory signal section based on the feature vector of each respiratory signal section and the sensitivity coefficient of each feature includes: Calculating an initial precision error weight and an initial extreme error weight of each respiratory signal section based on the feature vector of each respiratory signal section and the sensitivity coefficient of each feature; Based on the amplitude characteristic component in the characteristic vector of each respiratory signal section, the initial precision error weight and the initial extreme error weight of each respiratory signal section are adjusted to obtain the precision error weight and the extreme error weight of each respiratory signal section.

7. The respiratory signal prediction method according to claim 1, wherein Determining the loss of the prediction result of the respiratory signal sample based on the precision error and the extreme error of the prediction result of each respiratory signal section, the precision error weight and the extreme error weight of each respiratory signal section, and the penalty loss of the prediction result of each respiratory signal section includes: Calculate the penalty loss of the prediction results of each respiratory signal segment based on the peak safety threshold and the trough safety threshold corresponding to the current application scenario; the peak safety threshold is the maximum prediction error of the peak of the respiratory signal when the distance between the target area and the endangered organ is at a safe distance; the trough safety threshold is the maximum prediction error of the trough of the respiratory signal when the distance between the target area and the endangered organ is at a safe distance; performing a weighted summation of the precision error and the extreme error of the prediction result of each respiratory signal section based on the precision error weight and the extreme error weight of each respiratory signal section, and determining the sum of the weighted summation result and the penalty loss of the prediction result of each respiratory signal section as the loss of the prediction result of each respiratory signal section; The sum of the losses of the prediction results of the respective respiratory signal sections is determined as the loss of the prediction result of the respiratory signal sample.

8. The respiratory signal prediction method according to claim 7, wherein: The calculation of the penalty loss of the prediction results of each respiratory signal section based on the peak safety threshold and the trough safety threshold corresponding to the current application scenario includes: Determine a peak time point set of each respiratory signal section; the peak time point set is a collection of sampling times of all peaks of each respiratory signal section; Determine a peak maximum prediction error of each respiratory signal section based on a difference between a sampling point at each sampling time of the peak time point set and a predicted value of each respiratory signal section; the peak maximum prediction error is the maximum value of the absolute value of the difference between a sampling point at each sampling time of the respiratory signal section and a predicted value; Determine a trough time point set of each respiratory signal section; the trough time point set is a collection of sampling times of all troughs of each respiratory signal section; Determine a maximum trough prediction error of each respiratory signal segment based on a difference between a sampling point at each sampling time of the trough time point set and a predicted value of each respiratory signal segment; the maximum trough prediction error is the maximum absolute value of the difference between a sampling point at each sampling time of the respiratory signal segment and a predicted value; Based on the peak safety threshold and the trough safety threshold corresponding to the current application scenario, and the peak maximum prediction error and the trough maximum prediction error of each respiratory signal section, the penalty loss of the prediction result of each respiratory signal section is calculated.

9. The respiratory signal prediction method according to any one of claims 1 to 8, characterized in that: The loss of the prediction result of the respiratory signal sample is used to train the initial respiratory signal prediction model to obtain a trained respiratory signal prediction model, including: When the loss of the prediction result of the respiratory signal sample to be predicted is greater than a preset loss threshold, the parameters of the initial respiratory signal prediction model are updated, and the loss of the prediction result of the respiratory signal sample to be predicted is recalculated using the updated prediction model. The process of updating the parameters of the initial respiratory signal prediction model and calculating the loss of the prediction result of the respiratory signal sample is repeated until the number of times the loss of the prediction result of the respiratory signal sample is calculated reaches a preset number requirement, or the loss of the prediction result of the respiratory signal sample reaches a preset loss threshold, thereby obtaining the trained respiratory signal prediction model.

10. A respiratory signal prediction device, characterized in that: The device comprises: The sample division module is used to obtain respiratory signal samples and divide the obtained respiratory signal samples according to preset rules to obtain various respiratory signal sections; an error determination module, configured to predict each respiratory signal segment using the initial respiratory signal prediction model to obtain a prediction result for each respiratory signal segment, and determine, based on the prediction result for each respiratory signal segment, a precision error and an extreme error of the prediction result for each respiratory signal segment; the precision error represents the overall accuracy of the prediction result of the respiratory signal; and the extreme error represents the maximum value of the error of the prediction result of the respiratory signal; A loss determination module is configured to determine the loss of the prediction result of the respiratory signal sample based on the precision error and the extreme error of the prediction result of each respiratory signal section, the precision error weight and the extreme error weight of each respiratory signal section, and the penalty loss of the prediction result of each respiratory signal section; the precision error weight represents the influence factor of the overall accuracy of the prediction result on the prediction result of the respiratory signal sample; the extreme error weight represents the influence factor of the maximum value of the error of the prediction result on the prediction result of the respiratory signal sample; the penalty loss is a clinical constraint function determined based on the relationship between the maximum value of the prediction error of the respiratory signal at the peak and valley and a preset clinical safety threshold; a model training module, configured to train the initial respiratory signal prediction model based on the loss of the prediction result of the respiratory signal sample to obtain a trained respiratory signal prediction model; and a prediction module, which is used to obtain a respiratory signal to be predicted and predict the respiratory signal to be predicted using the trained respiratory signal prediction model.

Images