AI-based dynamic compensation system and method for thoracic and abdominal tumor target areas

Abstract

The present invention relates to the field of medical image processing technology, and more specifically to an AI-based dynamic compensation system and method for thoracic and abdominal tumor target areas. The method selects a portion of pending historical treatment samples as historical treatment samples based on the similarity of BMI values ​​between the samples, determines a position prediction accuracy index based on the CT image information of each historical treatment sample, then determines the tumor position similarity based on the CT image information of the current treatment sample and each historical treatment sample, and fuses the position prediction accuracy index and the tumor position similarity to obtain the radiotherapy process similarity. The tumor position of each historical treatment sample at the future irradiation time is determined, and the tumor position is weighted and summed using the radiotherapy process similarity to obtain the tumor position of the current treatment sample at the future irradiation time. The present invention can predict the tumor position of the current treatment sample at the future irradiation time, thereby triggering the irradiation equipment in advance for precise radiotherapy.

Description

Technical Field

[0001] The present invention relates to the technical field of medical image processing, and in particular to an AI-based dynamic compensation system and method for chest and abdominal tumor target areas. Background Art

[0002] Thoracic and abdominal tumors are tumors that occur within the chest and abdominal cavities. Respiratory motion is a key factor affecting treatment accuracy during radiotherapy for these tumors. This is because breathing periodically causes displacement of the tumor and surrounding organs. Failure to effectively manage tumor position can result in missed target areas, overirradiation of normal tissue, and distorted dose distribution during treatment, compromising treatment efficacy and patient safety.

[0003] To suppress the effects of respiratory motion, respiratory gating technology is used to dynamically compensate for tumor target volume. Its core principle is to control the timing of tumor irradiation during radiotherapy. Specifically, by monitoring the patient's respiratory cycle, treatment is restricted to specific respiratory phases, thereby reducing tumor movement during other respiratory phases and ensuring treatment accuracy.

[0004] However, existing respiratory gating technology, which restricts treatment to specific respiratory phases, can prolong treatment times. This means patients are exposed to radiation longer during treatment. Long-term or excessive radiation exposure can increase the risk of damage to normal tissue and even the likelihood of secondary cancers. Furthermore, during the implementation of respiratory gating technology, the correlation between external respiratory signals and the tumor's position and motion trajectory is low. This is because the tumor's morphology and volume can change during different respiratory cycles, especially when the patient coughs or moves. Traditional respiratory gating technology often struggles to adapt to these changes, potentially affecting the accuracy of treatment. Summary of the Invention

[0005] In order to solve the technical problem of excessively long treatment time in the existing dynamic compensation of thoracic and abdominal tumor target areas, the present invention aims to provide an AI-based dynamic compensation system and method for thoracic and abdominal tumor target areas. The technical solutions adopted are as follows:

[0006] One embodiment of the present invention provides an AI-based dynamic compensation method for thoracic and abdominal tumor target areas, the method comprising the following steps:

[0007] Obtaining the CT image time series and BMI values ​​of the current treatment sample and several pending historical treatment samples;

[0008] Based on the similarity of BMI values ​​between the current treatment sample and each pending historical treatment sample, the historical treatment samples are screened out from all pending historical treatment samples;

[0009] According to the CT image time series of each historical treatment sample, the position prediction accuracy index corresponding to each historical treatment sample is determined;

[0010] Determine the degree of similarity of tumor location of each historical treatment sample relative to the current treatment sample based on the CT image time series of the current treatment sample and the CT image time series of each historical treatment sample;

[0011] fusing the position prediction accuracy index and the tumor position similarity to determine the similarity of the radiotherapy process for each historical treatment sample;

[0012] Obtaining the tumor position of each historical treatment sample at a future irradiation time based on the CT image time series of each historical treatment sample; wherein the future irradiation time is the time obtained by adding the current imaging time to the irradiation time interval corresponding to the irradiation device;

[0013] The tumor position of each historical treatment sample at the future irradiation moment is weighted and summed using the similarity of the radiotherapy process to obtain the tumor position of the current treatment sample at the future irradiation moment, and perform dynamic compensation targeted radiotherapy.

[0014] Furthermore, the method of screening out historical treatment samples from all pending historical treatment samples based on the similarity of BMI values ​​between the current treatment sample and each pending historical treatment sample includes:

[0015] For each pending historical treatment sample, calculate the absolute value of the difference between the BMI value of the current treatment sample and the BMI value of the pending historical treatment sample;

[0016] performing negative correlation normalization processing on the absolute value of the difference to obtain a first normalized value, and determining the first normalized value as the degree of comparability of the breathing pattern of the pending historical treatment sample with respect to the current treatment sample;

[0017] A degree threshold is set, and historical treatment samples are obtained by comparing the comparable degrees of each breathing pattern with the degree threshold.

[0018] Furthermore, determining the position prediction accuracy index corresponding to each historical treatment sample based on the CT image time series of each historical treatment sample includes:

[0019] Based on the CT image time series of each historical treatment sample, the center position of the tumor at each historical shooting moment is determined and recorded as the actual tumor position; the center position of the tumor at each historical shooting moment is predicted and recorded as the false tumor position;

[0020] Calculating a position difference value between the actual tumor position and the false tumor position at the same historical shooting moment;

[0021] Perform fusion analysis on the position difference values ​​of all historical shooting moments corresponding to the same historical treatment sample to determine the position prediction accuracy index corresponding to each historical treatment sample;

[0022] The position difference value is negatively correlated with the position prediction accuracy index.

[0023] Furthermore, the fusion analysis of the position difference values ​​of all historical shooting moments corresponding to the same historical treatment sample to determine the position prediction accuracy index corresponding to each historical treatment sample includes:

[0024] For each historical shooting moment, performing negative correlation normalization processing on the position difference value to obtain a second normalized value, and determining the second normalized value as the prediction standard degree corresponding to the historical shooting moment;

[0025] The prediction standard degrees of all historical shooting moments corresponding to the same historical treatment sample are cumulatively multiplied and calculated, and a first cumulative multiplication value obtained is determined as the position prediction accuracy index corresponding to the corresponding historical treatment sample.

[0026] Furthermore, determining the degree of similarity of the tumor location of each historical treatment sample relative to the current treatment sample based on the CT image time series of the current treatment sample and the CT image time series of each historical treatment sample includes:

[0027] Using the CT image time series of the current treatment sample as input data, obtaining the predicted tumor position of the CT image at the next moment;

[0028] For each historical treatment sample, selecting a CT image at a target historical shooting moment from the CT image time series of the historical treatment sample, and determining a target historical tumor location of the selected CT image; wherein the target historical shooting moment has the same image sequence number as the image at the next moment;

[0029] The degree of similarity of the tumor position of each historical treatment sample relative to the current treatment sample is determined based on the positional similarity between the predicted tumor position and each of the target historical tumor positions, and between the current tumor position and the historical tumor position of the same image sequence number; wherein the current tumor position is the tumor position in the CT image of the current treatment sample.

[0030] Furthermore, determining the degree of similarity of the tumor position of each historical treatment sample relative to the current treatment sample based on the positional similarity between the predicted tumor position and each of the target historical tumor positions, and between the current tumor position and the historical tumor position for the same image sequence number, includes:

[0031] For any historical treatment sample, calculate the distance between the predicted tumor position and the historical tumor position of the historical treatment sample, and calculate the distance between the current tumor position and the historical tumor position of the same image sequence number;

[0032] The second cumulative product values ​​of all distance values ​​corresponding to the historical treatment sample are obtained, and negative correlation normalization processing is performed on the second cumulative product values ​​of all distance values ​​to obtain a third normalized value, and the third normalized value is used as the tumor location similarity.

[0033] Furthermore, the fusing of the position prediction accuracy index and the tumor position similarity to determine the similarity of the radiotherapy process of each historical treatment sample includes:

[0034] For each historical treatment sample, the position prediction accuracy index of the historical treatment sample and the tumor position similarity are multiplied to obtain a product; and the product is used as the radiotherapy process similarity of the corresponding historical treatment sample.

[0035] Furthermore, obtaining the tumor position of each historical treatment sample at the future irradiation moment based on the CT image time series of each historical treatment sample includes:

[0036] For each historical treatment sample, the historical tumor position at each historical shooting moment is obtained based on the CT image time series of the historical treatment sample;

[0037] According to each historical shooting moment and its historical tumor position, a position acquisition equation is constructed;

[0038] The future irradiation time is used as the independent variable of the position acquisition equation to obtain the tumor position of the corresponding historical treatment sample at the future irradiation time.

[0039] Furthermore, the step of performing weighted summation processing on the tumor position of each historical treatment sample at the future irradiation moment using the similarity of the radiotherapy process to obtain the tumor position of the current treatment sample at the future irradiation moment includes:

[0040] For each historical treatment sample, the ratio of the radiotherapy process similarity of the historical treatment sample to the cumulative value of the similarity of all radiotherapy processes is used as the position weight of the corresponding historical treatment sample;

[0041] Calculate the third product of the position weight of each historical treatment sample and the tumor position of the corresponding historical treatment sample at the future irradiation moment as the tumor sub-position of the current treatment sample at the future irradiation moment;

[0042] All tumor sub-positions are fused to obtain a fused position, which is used as the tumor position of the current treatment sample at a future irradiation moment.

[0043] Another embodiment of the present invention provides an AI-based dynamic compensation system for thoracic and abdominal tumor target areas, the system comprising a processor and a memory, the processor being used to process instructions stored in the memory to implement an AI-based dynamic compensation method for thoracic and abdominal tumor target areas.

[0044] The present invention has the following beneficial effects:

[0045] Compared with existing respiratory gating technology, its extended treatment time may lead to excessive radiation, and the insufficient correlation between external signals and tumor movement may lead to irradiation deviation. The present invention uses the CT image time series obtained during the targeted treatment of historical treatment samples to analyze the similarities of the treatment process between the current treatment samples to predict the tumor position of the current treatment sample at the next irradiation moment, so as to trigger the irradiation equipment in advance and keep the irradiation equipment at the highest frequency irradiation state as much as possible, thereby shortening the treatment time; through historical group data drive and real-time dynamic fusion, it breaks through the limitation of traditional gating technology that relies solely on the patient's respiratory signal, effectively overcomes the problem of tumor displacement caused by respiratory periodicity, and thus realizes dynamic compensation targeted treatment.

[0046] First, historical treatment samples are screened from all pending historical treatment samples based on the similarity in BMI values ​​between the current treatment sample and each pending historical treatment sample. By screening historical treatment samples, breathing patterns of treatment samples with large body size differences are different, so the screened historical treatment samples have comparable breathing patterns with the current treatment samples, which is conducive to subsequent more accurate tumor location prediction, and is also conducive to reducing the analysis of unnecessary historical treatment samples and saving computational analysis time.

[0047] Secondly, the degree of similarity of the radiotherapy process of each historical treatment sample relative to the current treatment sample is analyzed from two perspectives, namely, the position prediction accuracy index and the tumor position similarity. The position prediction accuracy index can characterize the accuracy of the tumor position prediction of the historical treatment sample itself, while the tumor position similarity indicates the similarity of the tumor position of the historical treatment sample relative to the current treatment sample. The numerical accuracy of the radiotherapy process similarity determined by the two perspectives is stronger, which is conducive to accurately analyzing the importance of each historical treatment sample in the predictive analysis, and improving the numerical accuracy of the tumor position of the current treatment sample determined subsequently at the future irradiation moment.

[0048] Finally, the tumor position of each historical treatment sample at the future irradiation moment is obtained, which is the benchmark data for determining the tumor position of the current treatment sample at the future irradiation moment. The tumor position of each historical treatment sample at the future irradiation moment is weighted and summed using the similarity of the radiotherapy process. This helps to obtain the tumor position of the current treatment sample at the future irradiation moment with higher accuracy and reliability, so as to facilitate dynamic compensation targeted radiotherapy. BRIEF DESCRIPTION OF THE DRAWINGS

[0049] In order to more clearly illustrate the technical solutions and advantages of the embodiments of the present invention or the prior art, the following briefly introduces the drawings required for use in the embodiments or the prior art descriptions. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0050] Figure 1 This is a flowchart of an implementation method of an AI-based dynamic compensation method for thoracic and abdominal tumor target areas according to an embodiment of the present invention;

[0051] Figure 2 Flowchart for implementing step S3 in an embodiment of the present invention;

[0052] Figure 3 This is a flowchart for implementing step S4 in an embodiment of the present invention. DETAILED DESCRIPTION

[0053] To further illustrate the technical means and effects employed by the present invention to achieve its intended objectives, the following, in conjunction with the accompanying drawings and preferred embodiments, describes in detail the specific implementations, structures, features, and effects of the technical solutions proposed by the present invention. In the following description, references to "one embodiment" or "another embodiment" do not necessarily refer to the same embodiment. Furthermore, specific features, structures, or characteristics of one or more embodiments may be combined in any suitable manner.

[0054] Unless defined otherwise, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention belongs.

[0055] The application scenarios targeted by the present invention may be:

[0056] While respiratory gating can improve the precision of radiotherapy, its extended treatment duration also means that patients are exposed to radiation for a longer period of time. If treatment duration is too long, especially with high radiation doses, normal tissues (such as the lungs and heart) may be exposed to unnecessary radiation, leading to side effects such as pulmonary fibrosis and heart problems. Furthermore, the poor correlation between respiratory signals and the tumor's position and motion trajectory during respiratory gating can lead to irradiation deviations, compromising treatment effectiveness.

[0057] In order to overcome the defects of the existing respiratory gating technology used in radiotherapy, an embodiment of the present invention provides an AI-based dynamic compensation method for chest and abdominal tumor target areas, such as Figure 1 As shown, the following steps are included:

[0058] S1, obtaining the CT image time series and BMI values ​​of the current treatment sample and several pending historical treatment samples.

[0059] Here, the CT image time series refers to a collection of time-ordered images obtained from multiple consecutive CT scans of the same examined area. It has clear temporal properties, and the time interval between adjacent CT images is fixed; the BMI value (Body Mass Index) refers to the patient's height and weight status; the current treatment sample is patients with thoracic and abdominal tumors currently undergoing radiotherapy, while the pending historical treatment sample is patients with thoracic and abdominal tumors in the past who have completed radiotherapy.

[0060] In order to analyze the tumor motion trajectory of patients with thoracic and abdominal tumors during targeted treatment and determine the future irradiation time for advance irradiation, it is necessary to obtain CT image time series; the CT image time series of the current treatment sample can be used as input data for prediction. During the prediction process, due to the lack of historical data related to the current treatment sample, it is necessary to use the CT image time series of the pending historical treatment sample for supplementary analysis.

[0061] First, the CT image time series of the current treatment sample and several pending historical treatment samples are obtained.

[0062] In this embodiment, during radiotherapy, a patient acquires a time series of CT images using imaging equipment. The time series of CT images for the current treatment sample consists of N CT images captured at each time instant, while the time series of CT images for the pending historical treatment sample consists of M CT images captured at each time instant, where both N and M are positive integers, and M is greater than N. The number of pending historical treatment samples can be set by the implementer based on actual circumstances and is not specifically limited here.

[0063] Among them, each CT image in the CT image time series has an image serial number sorted by the shooting time; the CT image shooting frequency of the current treatment sample and different pending historical treatment samples is the same, so the number of CT images corresponding to different pending historical treatment samples is the same, and the CT image of the current treatment sample collected at each shooting time has its corresponding CT image relative to different pending historical treatment samples.

[0064] Furthermore, tumor motion trajectory data sequences of the current treatment sample and several pending historical treatment samples are obtained based on the CT image time series.

[0065] In this embodiment, for the current treatment sample and each pending historical treatment sample, the tumor center position of each CT image in the CT image time series is determined as the tumor position; the tumor position determined by the first CT image is used as the coordinate origin, that is, (0,0), to construct a position coordinate system, and all tumor positions corresponding to the same patient are mapped to the position coordinate system, and all tumor positions are connected to obtain the tumor motion trajectory.

[0066] The tumor motion trajectory data sequence of the current treatment sample is shown in Table 1 below:

[0067] Table 1

[0068]

[0069] The tumor motion trajectory data sequence of each pending historical treatment sample is shown in Table 2:

[0070] Table 2

[0071]

[0072] Among them, the horizontal axis of the position coordinate system is the horizontal coordinate of the tumor position, and the vertical axis is the vertical coordinate of the tumor position; the tumor motion trajectory data sequence is composed of the tumor positions of the same patient at different shooting times; each CT image has its corresponding tumor position, and the process of determining the tumor position is existing technology and is not within the scope of protection of the present invention, and will not be elaborated here.

[0073] It should be noted that the subsequent steps of determining the tumor location can directly obtain the corresponding data here without repeating the calculation and analysis of the tumor location.

[0074] Second, obtain the BMI values ​​of the current treatment sample and each pending historical treatment sample.

[0075] During radiotherapy for thoracic and abdominal tumors, obese patients face difficulties with abdominal breathing due to abdominal fat compressing the diaphragm. Anesthesia and pain in the abdomen and chest can make it even more difficult for the patient to breathe effectively, leading to significant changes in breathing patterns. Thin patients may have lower lung capacity, which can also lead to significant changes in breathing patterns, but these changes differ between obese and thin patients. Therefore, to analyze the similarities in breathing patterns between the pending historical treatment samples and the current treatment sample, it is necessary to obtain the BMI values ​​for the current treatment sample and each pending historical treatment sample.

[0076] Among them, the BMI value can be obtained by height and weight. The method of obtaining the BMI value is a prior art and is not within the scope of protection of the present invention, so it will not be elaborated here.

[0077] So far, this embodiment has obtained the CT image time series and BMI value of the current treatment sample and each pending historical treatment sample in the data acquisition phase.

[0078] S2, based on the similarity of the BMI values ​​between the current treatment sample and each pending historical treatment sample, screen out historical treatment samples from all pending historical treatment samples.

[0079] Here, the similarity of BMI values ​​refers to the reliability of the pending historical treatment sample as a comparison object for the current treatment sample analyzed from the perspective of breathing patterns. The more similar the breathing patterns are, the higher the credibility of the pending historical treatment sample as a comparison object.

[0080] When the size difference between the pending historical treatment sample and the current treatment sample is too large, even if the tumor location and motion trajectories of the two are highly consistent, the treatment process of the current treatment sample will be biased, and the respiratory trajectory cannot be directly compared and corrected. Therefore, it is necessary to screen out some pending historical treatment samples with similar breathing patterns from all pending historical treatment samples, or to determine whether the degree of similarity between the breathing patterns of each pending historical treatment sample and the current treatment sample meets the preset requirements and select the pending historical treatment samples that meet the requirements.

[0081] As an exemplary implementation, the above step S2 can be implemented by the following steps:

[0082] In the first step, for each pending historical treatment sample, the absolute value of the difference between the BMI value of the current treatment sample and the BMI value of the pending historical treatment sample is calculated.

[0083] In the second step, a negative correlation normalization process is performed on the absolute value of the difference to obtain a first normalized value, and the first normalized value is determined as the degree of comparability of the breathing pattern of the pending historical treatment sample relative to the current treatment sample.

[0084] As an example, the calculation formula for the degree of comparability of the breathing pattern of the ath pending historical treatment sample with respect to the current treatment sample may be:

[0085] Where, It indicates the degree of comparability of the breathing pattern of the ath undetermined historical treatment sample relative to the current treatment sample, and exp represents the exponential function with a natural constant as the base. represents the BMI value of the current treatment sample, represents the BMI value of the a-th pending historical treatment sample, Represents the absolute value function, exp (-) is used to realize the normalization of the negative correlation of the data. Represents the first normalized value.

[0086] In the calculation formula of the degree of comparability of breathing patterns, the greater the difference in body size between the pending historical treatment sample and the current treatment sample, i.e. The larger the value is, the greater the difference in the breathing patterns between the two treatment samples is, the worse the comparability of the breathing pattern of the corresponding historical treatment sample is with respect to the current treatment sample, and the lower the degree of comparability of the breathing pattern, the greater the possibility of it being screened out.

[0087] With reference to the degree of comparability of the breathing pattern of the ath pending historical treatment sample relative to the current treatment sample, the degree of comparability of the breathing pattern of each pending historical treatment sample relative to the current treatment sample can be obtained.

[0088] The third step is to set a degree threshold and obtain historical treatment samples by comparing the comparable degrees of each breathing pattern with the degree threshold.

[0089] In this embodiment, a degree threshold of 0.7 is set. Undetermined historical treatment samples with a respiratory pattern comparability greater than or equal to the threshold are selected as historical treatment samples for subsequent data analysis. This, to a certain extent, overcomes the impact of low respiratory pattern comparability. The degree threshold can be set by the implementer based on specific circumstances and is not specifically defined here.

[0090] It should be noted that when analyzing the comparability of breathing patterns, not all pending historical treatment samples can be used as comparison objects for the current treatment samples. Instead, pending historical treatment samples with smaller body size differences and comparable breathing patterns are screened out. This can improve the numerical reliability of the similarity of the radiotherapy process, and can also reduce the subsequent computational analysis of pending historical treatment samples with less correlation, thereby improving the efficiency of computational analysis.

[0091] So far, this embodiment has screened out historical treatment samples having a breathing pattern similar to the current treatment sample from all pending historical treatment samples.

[0092] S3, determining the position prediction accuracy index corresponding to each historical treatment sample based on the CT image time series of each historical treatment sample.

[0093] Here, the position prediction accuracy index refers to the degree of similarity between the predicted tumor position and the actual tumor position at each historical shooting moment of the historical treatment sample. The more similar the predicted tumor position is to the actual tumor position, the more accurate the tumor position prediction result of the corresponding historical treatment sample is, and the higher its importance in analyzing the current treatment sample.

[0094] As an exemplary embodiment, the above step S3 can be performed by Figure 2 The steps S31 to S33 shown implement:

[0095] S31, based on the CT image time series of each historical treatment sample, determine the center position of the tumor at each historical shooting moment, which is recorded as the actual tumor position; and predict the center position of the tumor at each historical shooting moment, which is recorded as the false tumor position.

[0096] In this embodiment, only the historical shooting moments that can be predicted are analyzed, so each historical shooting moment has its corresponding actual tumor position and false tumor position.

[0097] The motion trajectory data of historical treatment samples at each historical recording moment during radiotherapy can be predicted and analyzed using an LSTM (Long Short-Term Memory) neural network. If the deviation between the predicted and actual motion trajectory data at the same historical recording moment is small—that is, the change between the predicted and actual values ​​on both the x-axis and y-axis is small—then the motion trajectory data for the corresponding historical recording moment is standard and predictable. The LSTM neural network prediction process is prior art and falls outside the scope of this invention, so it will not be elaborated on here.

[0098] S32, calculating the position difference between the actual tumor position and the false tumor position at the same historical shooting moment.

[0099] In this embodiment, the position difference value refers to the distance between two position points. The larger the distance between the two position points, the larger the position difference value between the actual tumor position and the false tumor position; conversely, the smaller the position difference value between the actual tumor position and the false tumor position.

[0100] S33, performing a fusion analysis on the position difference values ​​of all historical shooting moments corresponding to the same historical treatment sample to determine the position prediction accuracy index corresponding to each historical treatment sample.

[0101] Here, the position difference value is negatively correlated with the position prediction accuracy index, that is, the larger the position difference value, the smaller the position prediction accuracy index.

[0102] Specifically, for each historical shooting moment, the position difference value is negatively correlated and normalized to obtain a second normalized value, and the second normalized value is determined as the prediction standard degree of the corresponding historical shooting moment; the prediction standard degrees of all historical shooting moments corresponding to the same historical treatment sample are cumulatively multiplied, and the obtained first cumulative value is determined as the position prediction accuracy index corresponding to the corresponding historical treatment sample.

[0103] It should be noted that respiratory movement is a continuous process. The accuracy of position prediction at a single historical shooting moment cannot represent the position prediction situation of the entire treatment process. Therefore, it is necessary to perform a fusion analysis of the position difference values ​​of all historical shooting moments corresponding to the same historical treatment sample to determine the position prediction accuracy index corresponding to the historical treatment sample.

[0104] As an example, the calculation formula for the position prediction accuracy index corresponding to the j-th historical treatment sample can be:

[0105] Where, represents the degree of prediction standardization of the j-th historical treatment sample at the t-th historical shooting moment, represents the horizontal coordinate of the false tumor position of the jth historical treatment sample at the tth historical shooting moment, represents the abscissa of the actual tumor position of the jth historical treatment sample at the tth historical shooting moment, represents the ordinate of the false tumor position of the jth historical treatment sample at the tth historical shooting moment, represents the ordinate of the actual tumor position of the jth historical treatment sample at the tth historical shooting moment, represents the absolute value function, and The added value represents the position difference value, exp represents the exponential function with the natural constant as the base, and exp (-) is used to achieve negative correlation processing of data. Represents the second normalized value.

[0106] Where, represents the position prediction accuracy index corresponding to the j-th historical treatment sample, Indicates the number of historical shooting moments, represents the degree of prediction standardization of the j-th historical treatment sample at the t-th historical shooting moment.

[0107] In the calculation formula of the position prediction accuracy index, the prediction standard degree for each historical shooting moment is first calculated. The prediction standard degree is determined by the similarity between the actual tumor position and the false tumor position on the x-axis and y-axis at the same historical shooting moment. The greater the position similarity, the higher the prediction standard degree corresponding to the historical shooting moment.

[0108] It should be noted that since the false tumor position at a certain historical shooting moment is predicted by the actual tumor position before the historical shooting moment, there is no false tumor position for the first historical shooting moment, so it is not analyzed.

[0109] Referring to the calculation process of the position prediction accuracy index corresponding to the j-th historical treatment sample, the position prediction accuracy index corresponding to each historical treatment sample is obtained.

[0110] At this point, this embodiment has obtained the position prediction accuracy index corresponding to each historical treatment sample.

[0111] S4, determining the degree of similarity of the tumor position of each historical treatment sample with respect to the current treatment sample based on the CT image time series of the current treatment sample and the CT image time series of each historical treatment sample.

[0112] Here, the degree of similarity of tumor positions refers to the similarity of the tumor position motion trajectories between the historical treatment sample and the current treatment sample. The greater the degree of similarity of the tumor positions, the smaller the distance between the two tumor position motion trajectories on the x-axis and the y-axis.

[0113] As an exemplary embodiment, the above step S4 can be performed by Figure 3 The steps S41 to S43 shown implement:

[0114] S41 , using the CT image time series of the current treatment sample as input data, and obtaining the predicted tumor position of the CT image at the next moment.

[0115] In this embodiment, the CT image time series of the current treatment sample is acquired in the current shooting period, and the next moment is the next shooting moment of the current shooting period obtained according to the shooting frequency, that is, the future shooting moment.

[0116] Specifically, the current tumor locations corresponding to the time series of the CT images of the current treatment sample are used as input data for the LSTM neural network to predict the tumor location of the current treatment sample at the next moment, which is recorded as the predicted tumor location. The current tumor location is the tumor location in the CT image of the current treatment sample.

[0117] For example, if the last capture moment in the time series of CT images of the current treatment sample is the Nth capture moment, then the next capture moment is the N+1th capture moment. The time interval between the Nth and N+1th capture moments is still the time interval when the CT images were acquired in step S1 above. N is a positive integer and also represents the image sequence number.

[0118] S42 , for each historical treatment sample, selecting a CT image at a target historical shooting moment from the CT image time series of the historical treatment sample, and determining a target historical tumor position of the selected CT image.

[0119] In this embodiment, the target historical shooting moment has the same image sequence number as the next moment. That is, the target historical tumor position corresponding to the image sequence number of the CT image at the N+1th historical shooting moment is selected from the CT image time series of each historical treatment sample. The target historical tumor position is used for comparative analysis with the predicted tumor position.

[0120] It should be noted that the predicted tumor position and the target historical tumor position belong to the position points on the corresponding tumor position motion trajectory. In order to improve the reliability of the similarity analysis of the tumor position motion trajectory between the current treatment sample and the historical treatment sample, it is also necessary to analyze the position similarity between the predicted tumor position and each target historical tumor position.

[0121] S43 , determining the degree of similarity of the tumor position of each historical treatment sample relative to the current treatment sample based on the positional similarity between the predicted tumor position and each target historical tumor position, and between the current tumor position and the historical tumor position of the same image sequence number.

[0122] In this embodiment, the method for determining the similarity of the tumor location of each historical treatment sample relative to the current treatment sample is the same. For the sake of ease of description, any historical treatment sample is taken as an example to determine the similarity of the tumor location of a single historical treatment sample relative to the current treatment sample.

[0123] As an exemplary embodiment, determining the degree of similarity of tumor location between historical treatment samples and current treatment samples includes:

[0124] The distance between the predicted tumor position and the historical tumor position of the historical treatment samples is calculated, and the distance between the current tumor position and the historical tumor position of the same image sequence number is calculated.

[0125] The second cumulative product values ​​of all distance values ​​corresponding to the historical treatment samples are obtained, and the second cumulative product values ​​of all distance values ​​are negatively normalized to obtain a third normalized value, which is used as the tumor location similarity.

[0126] As an example, the calculation formula for the similarity of the tumor location of the jth historical treatment sample with respect to the current treatment sample can be:

[0127] Where, Indicates the similarity of the tumor location of the jth historical treatment sample relative to the current treatment sample, N indicates the number of images in the CT image sequence of the current treatment sample, and also indicates the number of locations of the current tumor location. Indicates that the image number is The distance value between the current tumor position and the historical tumor position of the jth historical treatment sample, It represents the distance between the predicted tumor position when the image number is N+1 and the historical tumor position of the jth historical treatment sample, It represents the second cumulative value of all distance values ​​corresponding to the j-th historical treatment sample, exp represents the exponential function with the natural constant as the base, and exp(-) represents the normalization processing to achieve the negative correlation of the data.

[0128] The image number is The calculation formula for the distance between the current tumor position and the historical tumor position of the jth historical treatment sample can be:

[0129] Where, Indicates that the image number is The horizontal coordinate of the current tumor position, Indicates that the image number is The ordinate of the current tumor position, Indicates that the image number is The horizontal coordinate of the historical tumor position of the j-th historical treatment sample, Indicates that the image number is The ordinate of the historical tumor position of the jth historical treatment sample, Represents the absolute value function.

[0130] Where, Indicates the horizontal coordinate of the predicted tumor position when the image number is N+1, Indicates the vertical coordinate of the predicted tumor position when the image number is N+1, The horizontal coordinate of the historical tumor position of the j-th historical treatment sample with image sequence number N+1, The vertical coordinate represents the historical tumor position of the j-th historical treatment sample with image sequence number N+1.

[0131] With reference to the above-mentioned calculation process of the similarity degree of the tumor location of the j-th historical treatment sample relative to the current treatment sample, the similarity degree of the tumor location of each historical treatment sample relative to the current treatment sample can be obtained.

[0132] So far, this embodiment has obtained the similarity of the tumor location of each historical treatment sample with respect to the current treatment sample.

[0133] S5, the fusion location prediction accuracy index and the tumor location similarity, determine the similarity of the radiotherapy process for each historical treatment sample.

[0134] Here, the degree of similarity of the radiotherapy process refers to the similarity between the radiotherapy process of the historical treatment sample and the radiotherapy process of the current treatment sample. The greater the degree of similarity of the radiotherapy process, the more important the corresponding historical treatment sample is in determining the tumor position of the current treatment sample at the future irradiation moment.

[0135] In this embodiment, the degree of similarity in the radiotherapy process is analyzed from two perspectives: the position prediction accuracy index and the tumor position similarity. A larger position prediction accuracy index indicates a higher accuracy in tumor position prediction for the corresponding historical treatment sample. That is, the distance between the actual tumor position and the predicted false tumor position is smaller, and the similarity is higher. This indicates that the corresponding historical treatment sample is more reliable in achieving tumor position prediction and is used to correct the degree of tumor position similarity. The tumor position similarity indicates the similarity in the tumor position trajectory between the current treatment sample and the historical treatment sample. The greater the tumor position similarity, the higher the similarity in the treatment process between the current treatment sample and the corresponding historical treatment sample during radiotherapy. When both the position prediction accuracy index and the tumor position similarity are high, it further indicates that the corresponding historical treatment sample is more important in subsequently determining the tumor position of the current treatment sample at the time of future irradiation.

[0136] Specifically, for each historical treatment sample, the location prediction accuracy index of the historical treatment sample and the tumor location similarity are multiplied to obtain a product; the product is used as the radiotherapy process similarity of the corresponding historical treatment sample.

[0137] So far, this embodiment has obtained the radiotherapy process similarity level corresponding to each historical treatment sample, which is used to indicate the importance of the corresponding historical treatment sample.

[0138] S6, according to the CT image time series of each historical treatment sample, obtaining the tumor position of each historical treatment sample at the future irradiation time.

[0139] Here, the future irradiation time is the sum of the current capture time and the irradiation interval corresponding to the irradiation device. The future irradiation time is not equivalent to any specific capture time, cannot be directly obtained from the CT image time series, and is much smaller than the irradiation interval used by respiratory gating techniques. For example, if the current capture time is 10:20:06, and the irradiation interval corresponding to the irradiation device is 1 second, the future irradiation time will be 10:20:07.

[0140] As an exemplary embodiment, the above step S6 is implemented by the following steps:

[0141] In the first step, for each historical treatment sample, the historical tumor position at each historical shooting moment is obtained based on the CT image time series of the historical treatment sample.

[0142] In this embodiment, in the CT image time series of the historical treatment samples, the center position of the tumor in the CT image at each historical shooting moment is determined and recorded as the historical tumor position.

[0143] In the second step, a position acquisition equation is constructed based on each historical shooting moment and its historical tumor position.

[0144] In this embodiment, the historical shooting time is used as the independent variable, and the historical tumor position corresponding to the historical shooting time is used as the dependent variable. A two-variable linear equation is constructed as the position acquisition equation corresponding to the historical treatment sample.

[0145] In the third step, the future irradiation time is used as the independent variable of the position acquisition equation to obtain the tumor position of the corresponding historical treatment sample at the future irradiation time.

[0146] In this embodiment, the future irradiation time is used as the independent variable of the position acquisition equation, and the dependent variable corresponding to the future irradiation time can be obtained through calculation, that is, the tumor position of the historical treatment sample at the future irradiation time.

[0147] It should be noted that the tumor position of each historical treatment sample at the future irradiation time is the benchmark data for the subsequent prediction of the tumor position of the current treatment sample at the future irradiation time.

[0148] In another exemplary embodiment, the calculation formula for the abscissa of the tumor position of the jth historical treatment sample at the future irradiation time may be:

[0149] Where, represents the abscissa of the tumor position of the jth historical treatment sample at the future irradiation moment, represents the horizontal coordinate of the historical tumor position of the jth historical treatment sample at the Nth historical shooting moment, represents the future irradiation time of the jth historical treatment sample, represents the Nth historical shooting moment of the jth historical treatment sample, represents the N+1th historical shooting moment of the jth historical treatment sample.

[0150] In the calculation formula of the horizontal coordinate of the tumor position at the future irradiation time, The horizontal coordinate represents the historical tumor position at the moment before the future irradiation moment, express The proportion of the entire current time interval is greater, and the horizontal coordinate distance between adjacent shooting moments is greater. The bigger, The larger the abscissa of the tumor position at the future irradiation time.

[0151] The ordinate of the tumor position of each historical treatment sample at the future irradiation time can be obtained by referring to the abscissa of the tumor position of each historical treatment sample at the future irradiation time.

[0152] So far, this embodiment has obtained the tumor position of each historical treatment sample at the future irradiation time.

[0153] S7, using the similarity of the radiotherapy process to perform weighted summation processing on the tumor position of each historical treatment sample at the future irradiation moment, obtain the tumor position of the current treatment sample at the future irradiation moment, and perform dynamic compensation targeted radiotherapy.

[0154] The core of respiratory gating technology is to control the timing of tumor irradiation during radiotherapy. By monitoring the respiratory cycle, treatment is limited to a specific respiratory phase, namely the end of exhalation or end of inspiration, thereby reducing tumor displacement and improving treatment accuracy. However, this will prolong the treatment time and increase the patient's radiation exposure, which may cause damage to normal tissues or the risk of secondary cancers. In particular, tissues such as the lungs and heart may be exposed to unnecessary radiation. To reduce the above risks, the present invention obtains the tumor position of the current treatment sample at the time of future irradiation, so as to facilitate radiotherapy at the highest frequency of the irradiation equipment.

[0155] As an exemplary embodiment, the above step S7 can be implemented by the following steps:

[0156] In the first step, for each historical treatment sample, the ratio of the similarity of the radiotherapy process of the historical treatment sample to the cumulative similarity of all radiotherapy processes is used as the position weight of the corresponding historical treatment sample.

[0157] In this embodiment, the value range of the position weight is limited to between 0 and 1, and the cumulative value of the position weight of each historical treatment sample is 1.

[0158] In the second step, the third product of the position weight of each historical treatment sample and the tumor position of the corresponding historical treatment sample at the future irradiation moment is calculated as the tumor sub-position of the current treatment sample at the future irradiation moment.

[0159] In the third step, all tumor sub-positions are fused to obtain the fused position, which is used as the tumor position of the current treatment sample at the future irradiation moment.

[0160] In this embodiment, all tumor sub-positions are added together. Specifically, the sum of the horizontal coordinates of the tumor sub-positions is used as the horizontal coordinate of the tumor position of the current treatment sample at the future irradiation time, and the sum of the vertical coordinates of the tumor sub-positions is used as the vertical coordinate of the tumor position of the current treatment sample at the future irradiation time.

[0161] As an example, the calculation formula for the horizontal coordinate of the tumor position of the current treatment sample at the future irradiation time can be:

[0162] Where, represents the horizontal coordinate of the tumor position of the current treatment sample at the future irradiation time, J represents the number of historical treatment samples, Indicates the similarity of the tumor location of the jth historical treatment sample relative to the current treatment sample, It represents the cumulative value of the similarity of tumor location of all historical treatment samples relative to the current treatment sample, The horizontal coordinate represents the tumor position of the jth historical treatment sample at the future irradiation time.

[0163] The ordinate of the tumor position of the current treatment sample at the future irradiation time can be obtained by referring to the calculation process of the abscissa of the tumor position of the current treatment sample at the future irradiation time.

[0164] It should be noted that the time interval for determining the future irradiation moment is much smaller than the irradiation interval of the respiratory gating technology. Before the irradiation moment arrives, the irradiation equipment can be triggered in advance to irradiate the predicted tumor location, thereby effectively shortening the targeted treatment time.

[0165] After obtaining the tumor location of the current treatment sample at the future irradiation time, the irradiation device can be triggered in advance at the next irradiation time to perform precise radiotherapy. The above process of determining the tumor location of the current treatment sample at the future irradiation time is repeated continuously. After each irradiation, the tumor location at the next irradiation time with the highest irradiation device frequency is obtained to trigger the irradiation device in advance and continue precise radiotherapy until the treatment is completed, ultimately achieving dynamic compensation targeted radiotherapy for the current chest and abdominal tumor patient. The irradiation frequency of the irradiation device used to perform radiotherapy can be collected by the implementer.

[0166] Another embodiment of the present invention provides an AI-based dynamic compensation system for thoracic and abdominal tumor target areas, the system comprising a processor and a memory, the processor being used to process instructions stored in the memory to implement an AI-based dynamic compensation method for thoracic and abdominal tumor target areas.

[0167] The embodiments described above are only used to illustrate the technical solutions of the present invention, rather than to limit the same. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. These modifications or replacements do not deviate the essence of the corresponding technical solutions from the scope of the technical solutions of the embodiments of the present invention, and should all be included in the protection scope of the present invention.

Claims

1. An AI-based dynamic compensation method for chest and abdominal tumor target areas, characterized in that: The following steps are involved: Obtaining the CT image time series and BMI values ​​of the current treatment sample and several pending historical treatment samples; Based on the similarity of BMI values ​​between the current treatment sample and each pending historical treatment sample, the historical treatment samples are screened out from all pending historical treatment samples; According to the CT image time series of each historical treatment sample, the position prediction accuracy index corresponding to each historical treatment sample is determined; Determine the degree of similarity of tumor location of each historical treatment sample relative to the current treatment sample based on the CT image time series of the current treatment sample and the CT image time series of each historical treatment sample; fusing the position prediction accuracy index and the tumor position similarity to determine the similarity of the radiotherapy process for each historical treatment sample; Obtaining the tumor position of each historical treatment sample at a future irradiation time based on the CT image time series of each historical treatment sample; wherein the future irradiation time is the time obtained by adding the current imaging time to the irradiation time interval corresponding to the irradiation device; The tumor position of each historical treatment sample at the future irradiation moment is weighted and summed using the similarity of the radiotherapy process to obtain the tumor position of the current treatment sample at the future irradiation moment, and perform dynamic compensation targeted radiotherapy.

2. The AI-based dynamic compensation method for thoracic and abdominal tumor target areas according to claim 1, characterized in that: According to the similarity of BMI values ​​between the current treatment sample and each pending historical treatment sample, the historical treatment samples are screened from all pending historical treatment samples, including: For each pending historical treatment sample, calculate the absolute value of the difference between the BMI value of the current treatment sample and the BMI value of the pending historical treatment sample; performing negative correlation normalization processing on the absolute value of the difference to obtain a first normalized value, and determining the first normalized value as the degree of comparability of the breathing pattern of the pending historical treatment sample with respect to the current treatment sample; A degree threshold is set, and historical treatment samples are obtained by comparing the comparable degrees of each breathing pattern with the degree threshold.

3. The AI-based dynamic compensation method for thoracic and abdominal tumor target areas according to claim 1, characterized in that: Determining the position prediction accuracy index corresponding to each historical treatment sample based on the CT image time series of each historical treatment sample includes: Based on the CT image time series of each historical treatment sample, the center position of the tumor at each historical shooting moment is determined and recorded as the actual tumor position; the center position of the tumor at each historical shooting moment is predicted and recorded as the false tumor position; Calculating a position difference value between the actual tumor position and the false tumor position at the same historical shooting moment; Perform fusion analysis on the position difference values ​​of all historical shooting moments corresponding to the same historical treatment sample to determine the position prediction accuracy index corresponding to each historical treatment sample; The position difference value is negatively correlated with the position prediction accuracy index.

4. The AI-based dynamic compensation method for thoracic and abdominal tumor target areas according to claim 3, characterized in that: The fusion analysis of the position difference values ​​of all historical shooting moments corresponding to the same historical treatment sample to determine the position prediction accuracy index corresponding to each historical treatment sample includes: For each historical shooting moment, performing negative correlation normalization processing on the position difference value to obtain a second normalized value, and determining the second normalized value as the prediction standard degree corresponding to the historical shooting moment; The prediction standard degrees of all historical shooting moments corresponding to the same historical treatment sample are cumulatively multiplied and calculated, and a first cumulative multiplication value obtained is determined as the position prediction accuracy index corresponding to the corresponding historical treatment sample.

5. The AI-based dynamic compensation method for thoracic and abdominal tumor target areas according to claim 1, characterized in that: Determining the degree of similarity of the tumor position of each historical treatment sample relative to the current treatment sample based on the CT image time series of the current treatment sample and the CT image time series of each historical treatment sample includes: Using the CT image time series of the current treatment sample as input data, obtaining the predicted tumor position of the CT image at the next moment; For each historical treatment sample, selecting a CT image at a target historical shooting moment from the CT image time series of the historical treatment sample, and determining a target historical tumor location of the selected CT image; wherein the target historical shooting moment has the same image sequence number as the image at the next moment; The degree of similarity of the tumor position of each historical treatment sample relative to the current treatment sample is determined based on the positional similarity between the predicted tumor position and each of the target historical tumor positions, and between the current tumor position and the historical tumor position of the same image sequence number; wherein the current tumor position is the tumor position in the CT image of the current treatment sample.

6. The AI-based dynamic compensation method for thoracic and abdominal tumor target areas according to claim 5, characterized in that: Determining the degree of similarity of the tumor position of each historical treatment sample relative to the current treatment sample based on the positional similarity between the predicted tumor position and each of the target historical tumor positions, and between the current tumor position and the historical tumor position of the same image sequence number, includes: For any historical treatment sample, calculate the distance between the predicted tumor position and the historical tumor position of the historical treatment sample, and calculate the distance between the current tumor position and the historical tumor position of the same image sequence number; The second cumulative product values ​​of all distance values ​​corresponding to the historical treatment sample are obtained, and negative correlation normalization processing is performed on the second cumulative product values ​​of all distance values ​​to obtain a third normalized value, and the third normalized value is used as the tumor location similarity.

7. The AI-based dynamic compensation method for thoracic and abdominal tumor target areas according to claim 1, characterized in that: The fusing of the position prediction accuracy index and the tumor position similarity to determine the similarity of the radiotherapy process of each historical treatment sample includes: For each historical treatment sample, the position prediction accuracy index of the historical treatment sample and the tumor position similarity are multiplied to obtain a product; and the product is used as the radiotherapy process similarity of the corresponding historical treatment sample.

8. The AI-based dynamic compensation method for thoracic and abdominal tumor target areas according to claim 1, characterized in that: The method of obtaining the tumor position of each historical treatment sample at a future irradiation moment based on the CT image time series of each historical treatment sample includes: For each historical treatment sample, the historical tumor position at each historical shooting moment is obtained based on the CT image time series of the historical treatment sample; According to each historical shooting moment and its historical tumor position, a position acquisition equation is constructed; The future irradiation time is used as the independent variable of the position acquisition equation to obtain the tumor position of the corresponding historical treatment sample at the future irradiation time.

9. The AI-based dynamic compensation method for thoracic and abdominal tumor target areas according to claim 1, characterized in that: The step of performing weighted summation processing on the tumor position of each historical treatment sample at the future irradiation moment by utilizing the similarity of the radiotherapy process to obtain the tumor position of the current treatment sample at the future irradiation moment includes: For each historical treatment sample, the ratio of the radiotherapy process similarity of the historical treatment sample to the cumulative value of the similarity of all radiotherapy processes is used as the position weight of the corresponding historical treatment sample; Calculate the third product of the position weight of each historical treatment sample and the tumor position of the corresponding historical treatment sample at the future irradiation moment as the tumor sub-position of the current treatment sample at the future irradiation moment; All tumor sub-positions are fused to obtain a fused position, which is used as the tumor position of the current treatment sample at a future irradiation moment.

10. An AI-based dynamic compensation system for chest and abdominal tumor target areas, characterized in that: The system includes a processor and a memory, and the processor is used to process instructions stored in the memory to implement an AI-based dynamic compensation method for thoracic and abdominal tumor target areas as described in any one of claims 1 to 9.

Images