Artificial intelligence driven chest and abdomen movement simulation die body and simulation method

Abstract

The invention relates to the technical field of medical simulation devices, in particular to an artificial intelligence driven thoracic and abdominal movement simulation die body which comprises internal soft tissue equivalent to a human tissue CT value, the simulation die body can simulate movement of the heart, the lung, the liver, the intestinal tract and the tumor at the same time, surface movement simulation of the die body is considered, and the simulation effect is good. Simulation movement is achieved through multiple maneuvering modes, and movement of thoracic and abdominal organs, tumors and body surfaces is restored to the maximum extent. The model body is made of a simulation material equivalent to a human tissue CT value, it is ensured that the physical property of the model body is similar to that of the human tissue, the simulation authenticity is improved, and organ movement control modes including fixed frequency input, collected patient frequency input and movement frequency predicted through an artificial intelligence algorithm are flexible and diversified. And guarantee is provided for dosimetry verification by shooting 4D CT and position error verification by means of a body surface optical positioning system.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of medical simulation devices, in particular to an artificial intelligence driven chest and abdomen motion simulation phantom and a simulation method. BACKGROUND

[0002] Chest and abdominal malignant tumors are diseases that seriously threaten the life and health of patients. According to statistics, lung cancer ranks first in the number of deaths from malignant tumors in the United States in 2022, and chest and abdominal malignant tumors account for two-thirds of the top three in the number of deaths from cancer in women. The latest data from the China National Cancer Center shows that lung cancer, liver cancer, gastric cancer, colorectal cancer and esophageal cancer are the top five in the number of cancer deaths in China, accounting for 69.25% of the total number of cancer deaths. At present, the treatment methods for chest and abdominal malignant tumors mainly include surgery, chemotherapy and radiotherapy.

[0003] With the advancement of radiotherapy technology, especially the application of image-guided radiotherapy (IGRT), the accuracy and efficiency of tumor radiotherapy have been significantly improved, and the side effects have also been reduced accordingly. In IGRT, through the shooting of positioning CT, oncologists delineate the target area, radiation physicists develop radiotherapy plans, and finally radiation therapists execute. However, the organ motion and tumor motion of chest and abdominal tumor patients can cause differences between planned dose and actual treatment dose, which seriously affects the accuracy and safety of radiotherapy.

[0004] In order to evaluate this dose difference, a simulation phantom is usually used for dose verification. Through four-dimensional (4D) CT, the motion of chest and abdominal organs and tumors is simulated and incorporated into the radiotherapy plan. During radiotherapy, the dosimetric difference is evaluated by measuring the gamma passing rate in three-dimensional space. However, most existing chest and abdominal phantoms only simulate part of the organs and cannot fully simulate the motion of multiple organs and tumors in the chest and abdomen, while ignoring the simulation of body surface motion. For example, the chest and abdominal phantom researched by the Second Affiliated Hospital of the Army Medical University only simulates muscle tissue, spinal bone tissue and lung tissue; the RSD ART dosimetry verification phantom of Universal Medical Company covers from the head to the pelvic cavity, but does not pay attention to the details of the internal chest and abdominal organs at risk; the various chest dynamic phantoms developed by CIRS company are only suitable for quality control and radiotherapy dosimetry verification, and the range is limited to the chest.

[0005] Therefore, there is currently no research to produce a chest and abdominal motion simulation phantom that can fully simulate the heart, lungs, liver, intestines and tumors, and at the same time take into account the simulation of the motion of the phantom surface. In order to accurately evaluate the impact of the motion of multiple organs and tumors in the chest and abdomen on the radiotherapy dose and verify the position difference brought by the motion, there is an urgent need in the clinic to research a new type of chest and abdominal motion simulation phantom. SUMMARY

[0006] The purpose of this invention is to address the shortcomings of existing technologies, such as the lack of research and development of a chest and abdominal movement simulation model that can comprehensively simulate the heart, lungs, liver, intestines, tumors, and body surface. Therefore, this invention proposes an artificial intelligence-driven chest and abdominal movement simulation model and simulation method.

[0007] To achieve the above objectives, the present invention adopts the following technical solution: Design an artificial intelligence-driven simulation model of chest and abdominal movement, the model including internal soft tissue equivalent to the CT value of human tissue, and heart tissue, lung tissue, liver tissue, intestinal tissue and bone tissue distributed in the internal soft tissue according to the location of adult organs; The internal soft tissue is provided with a first movable insertion rod, which is sequentially connected to the lung tissue and the liver tissue. The first movable insertion rod is also provided with at least one tumor tissue. The interior of the soft tissue also contains a wash-free film for measuring radiotherapy dose.

[0008] Furthermore, a second motion rod is installed in the internal soft tissue, and a third motion rod is installed in the bone tissue; The first, second, and third motion rods are all connected to an external drive device for movement within space. The internal soft tissue also has a reserved hole for a moving rod.

[0009] Furthermore, the third motion insert is an elastic flexible tube, which is driven by a drive device to simulate the forward tilting, backward tilting, or bending movements of the vertebrae.

[0010] Furthermore, the lung tissue is connected to an external inflation / deflation pump via a flexible lung tube, which is inserted into the oral cavity of the phantom. The heart tissue is connected to an external mechanical air pump via a flexible heart tube, which is inserted into the navel of the model. The intestinal tissue is connected to a peristaltic pump via an intestinal elastic tubing, which is inserted through the perineum of the phantom.

[0011] Furthermore, the tumor tissue is made of a soft tissue equivalent material, and its HU value is greater than that of lung tissue and / or liver tissue.

[0012] Furthermore, the liver tissue is made of soft tissue simulation material, and the liver tissue is set with a spot contrast level that matches that of a healthy liver to simulate real liver imaging.

[0013] Furthermore, this invention also proposes an artificial intelligence-driven method for simulating chest and abdominal movements, which employs the aforementioned phantom. Specifically, this method includes: The drive device controls the first motion rod to perform translation and rotation operations, simulating the movement of internal soft tissue and tumor tissue; The heart tissue is connected to an external mechanical air pump, which is controlled to set a motor frequency according to the heart rate of an adult or patient, in order to simulate the pumping action of the human heart. The intestinal tissue is connected to an external peristaltic pump, which is controlled and the peristaltic frequency of normal adults or patients is set to simulate changes in intestinal contents and intestinal movement. The lung tissue is connected to an external inflation / deflation pump. Based on a deep learning algorithm model, the frequency of lung breathing movements is predicted. The deep learning algorithm model is used to randomly generate breathing movement signals, which are transmitted to a motor drive device to control the inflation / deflation frequency of the external inflation / deflation pump, so as to simulate the lung breathing movements of different people.

[0014] Furthermore, the respiratory signal data includes respiratory flow rate, airway pressure, and respiratory state at corresponding time points in a time series. The respiratory state includes inspiratory trigger point, intermediate state point, and expiratory trigger point. The inhalation trigger point is the time point at which the patient transitions from an exhalation state to an inhalation state. The intermediate state point is the time point at which the patient does not experience any change in exhalation or inhalation state. The exhalation trigger point is the time point at which the patient transitions from an inhalation state to an exhalation state.

[0015] Furthermore, the method for generating the respiratory motion signal includes: The respiratory signal dataset was divided into training, validation, and test sets in a 7:2:1 ratio. The training set is input into the deep learning algorithm model for training, and parameters such as the initial learning rate and batch size are set for the network. The Adam optimizer's automatic optimization strategy is used to automatically decay the learning rate of the network. Set up a dedicated loss function and evaluation metric for the model; During each training batch, the network model is evaluated using a validation set to monitor the quality of the network's predictions of respiratory signals. Select the best training model to predict the test set and obtain randomly generated respiratory signal data.

[0016] Furthermore, the deep learning model algorithm framework employs Generative Adversarial Networks (GANs), including: The generator iteratively learns the data features of the input training set through transposed convolutional layers and normalization layers, and generates predicted respiratory signal data. The discriminator uses a loss function to distinguish between real and fake samples and continuously optimizes the network loss value to achieve the best network performance. The loss functions for the discriminator and generator in GAN are defined as follows: L_D=max{log(D(x))+log(1-D(G(y)))} L_G=min{log(D(x))+log(1-D(G(y)))} Where D(x) represents the probability that the discriminator network D predicts the original data x, y represents the input data of the generator network G, the discriminator's role is to classify real and fake samples, so it is defined by the combination and maximum function, while the generator loss is defined by the minimum function.

[0017] This invention proposes an AI-driven simulation phantom and method for thoracic and abdominal movement. The advantages are as follows: The simulation phantom can simultaneously simulate the movement of the heart, lungs, liver, intestines, tumors, and body surface, while also considering surface movement simulation. It achieves simulated movement through various maneuvering methods, maximizing the reproduction of the movement of thoracic and abdominal organs, tumors, and the body surface. The phantom uses simulation materials equivalent to the CT values ​​of human tissue, ensuring that the physical properties of the phantom are similar to human tissue and improving the realism of the simulation. The control methods for organ movement are flexible and diverse, including fixed frequency input, acquired patient frequency input, and movement frequency predicted by AI algorithms. This provides assurance for dosimetric verification during 4D CT scans and positional error verification using a surface optical positioning system, ensuring precise radiation irradiation of the target area, reducing irradiation of organs at risk, and improving the effectiveness of radiotherapy. Attached Figure Description

[0018] Figure 1 This is a front view of the mold of the present invention; Figure 2 This is a schematic diagram of the cross-section of the model at the chest position; Figure 3 This is a schematic diagram of the cross-section of the model at the abdominal position; Figure 4 A block diagram illustrating the motor braking principle of the internal organs and tissues of the phantom. Figure 5 A block diagram illustrating the principle of simulating respiratory movements in the lungs; Figure 6 This is a schematic diagram illustrating the principle of deep learning-based respiratory motion signal generation. Figure 7 This is a schematic diagram of a deep learning model algorithm.

[0019] In the diagram: 1. Phantom body; 11. Internal soft tissue; 12. Heart tissue; 13. Lung tissue; 14. Liver tissue; 15. Intestinal tissue; 16. Bone tissue; 17. First moving rod; 18. Tumor tissue; 19. Second moving rod; 20. Third moving rod; 21. Moving rod hole. Detailed Implementation

[0020] The technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments.

[0021] Reference Figures 1-4 In one embodiment of the present invention, an artificial intelligence-driven chest and abdominal motion simulation phantom is disclosed. This simulation phantom aims to simulate the movement of multiple organs, tumors, and body surface in the chest and abdomen. By using simulation materials equivalent to human tissue, multiple organs and tumors in the chest and abdomen are restored. 4D CT scans are taken of the motion simulation phantom and dosimetric verification is performed to ensure that the radiation accurately irradiates the target area, minimizes the irradiation of organs at risk, and improves the radiotherapy effect. At the same time, by using the changes in organ filling and the movement of organs to drive the movement changes of the simulation phantom's body surface, the position verification of the body surface optical system can also be performed. Reference Figure 2 , Figure 3 Specifically, the simulation model described in this invention includes the model 1 comprising internal soft tissue 11 equivalent to the CT value of human tissue, and heart tissue 12, lung tissue 13, liver tissue 14, intestinal tissue 15 and bone tissue 16 distributed in the internal soft tissue 11 according to the location of adult organs; The internal soft tissue 11 is provided with a first motion insert 17, which is sequentially connected to the lung tissue 13 and the liver tissue 14. The first motion insert 17 is also provided with at least one tumor tissue 18. Two tumor tissues 18 can be provided, corresponding to the tumor simulation at the liver tissue 14 and the tumor simulation at the lung tissue 13, respectively. Specifically, in this invention, the internal soft tissue 11 is composed of grease and a filling material equivalent to human soft tissue. The equivalent filling material is a proprietary epoxy resin soft tissue simulation material. The soft tissue filling density ranges from 80% to 100%, with a target HU value of 0-40 HUs for the tissue equivalence density. The internal soft tissue 11 is filled with a serrated structure for tissue equivalence, reducing micro-air gaps that occur in a solid body and significantly lowering the mass density. The spaces between organs are filled with grease and a filling material equivalent to human soft tissue, with a filling density range of 80% to 100%. The lung tissue 13 is filled with lung simulation material, which is epoxy resin. The filling density of the peripheral lung tissue ranges from 80% to 100%, and the filling density of the lung tissue filler is 5% to 30%. The target HU value of the soft tissue equivalent density is 0-40 HUs, and the target HU value of the lung equivalent density is approximately -650 HUs. The lung is connected to an external mechanical inflation / deflation pump, which is controlled by a motor drive device to simulate breathing motion. The breathing motion mode is controlled by a computer algorithm. Considering that breathing motion has a significant impact on the verification of radiotherapy dose for lung tumors, it is necessary to realistically reproduce human breathing motion as much as possible. It is proposed to use a deep learning model to randomly generate breathing motion signals to simulate the breathing motion of different people. The principle can be referred to the simulation method described below, and will not be elaborated here. Reference Figure 1 The lung tissue 13 is connected to an external inflation / deflation pump via a flexible lung tube, which is inserted into the oral cavity of the phantom 1.

[0022] In addition, the heart tissue 12 is made of heart simulation material to form a heart simulation software. The heart is connected to an external mechanical air pump through a thin tube. The air pump is controlled by a motor drive device. The motor frequency is randomly set according to the heart rate of a normal adult or according to the heart rate collected by the patient to simulate the human body's blood pumping action. Reference Figure 1 The heart tissue 12 is connected to an external mechanical air pump through a flexible heart tube, which is inserted from the navel of the phantom 1.

[0023] Furthermore, the liver tissue 14 is made of soft tissue simulation material, specifically a sponge-like soft tissue material to make a liver simulation soft body. The target HU value of the liver equivalent density is about 60 HUs. The liver tissue 14 is set with a spot contrast level that matches that of a healthy liver to simulate real liver imaging. The volume change of the liver is simulated by the different water content of the soft tissue.

[0024] The intestinal tissue 15 uses an elastic tube with equivalent tissue density to simulate real intestinal tissue. An external peristaltic pump is connected to the elastic tube, and a motor-driven device controls the pump. The peristaltic frequency is randomly set for normal adults or patients to simulate changes in intestinal contents and intestinal movement. Reference Figure 1 The intestinal tissue 15 is externally connected to a peristaltic pump via an intestinal elastic tube, which is inserted from the perineum of the phantom 1.

[0025] It should be noted that the motor drive devices described above in this embodiment can all be servo motors. By adjusting and controlling the speed of the servo motor, the speed and other parameters of the air pump, mechanical air pump and peristaltic pump can be controlled. The specific control method is a conventional technical means for those skilled in the art and will not be elaborated here.

[0026] Furthermore, the bone tissue 16 described in this invention is composed of a barium sulfate suspension of varying weight percentages mixed with an epoxy resin potting mixture. The epoxy resin itself has a density close to that of the tissue equivalent. Before curing, the epoxy resin and hardener are added to the two parts of the potting mixture in a 1:1 ratio. The reaction between the epoxy resin and the hardener is exothermic; therefore, care must be taken to ensure that the generated heat does not cause deformation of the phantom. During the first hour after mixing, the potting mixture is thoroughly stirred to reduce the amount of high-density barium sulfate deposited at the bottom of the phantom. The weight percentage of barium sulfate relative to the epoxy resin potting mixture varies between 2% and 10%, typically 6%. The target HU value for bone equivalent density is approximately 700 HUs. By adding different amounts of barium sulfate to simulate the radiation attenuation effect produced by bone tissue, various different bone equivalent densities are formed, such as cancellous bone (approximately 300 HUs) or cortical bone (approximately 1250 HUs).

[0027] More specifically, the cross-sectional shape of the simulation phantom of the present invention is approximately elliptical, and it is similar to a typical adult in terms of size proportions, density, and spatial structure. In order to meet the actual clinical application needs for tumor radiotherapy dose verification, a no-wash film for measuring radiotherapy dose is also placed inside the internal soft tissue 11. The no-wash film can acquire the three-dimensional dose distribution of the location of interest according to actual needs. Combined with the gamma pass rate index during radiotherapy, the dosimetric verification results are more accurate. This feature can provide a phantom simulation experimental tool for dosimetric verification of thoracic and abdominal organs and tumor movement.

[0028] The phantom is filled with soft tissue. Movement of organs such as the lungs, heart, liver, and intestines compresses this soft tissue, thus inducing movement on the phantom's surface. Simultaneously, specific materials, such as silicone and polyurethane, are used as the phantom's outer surface material to facilitate position tracking by the surface optical system. The surface optical positioning system employs a structured light surface imaging algorithm, using a high-precision depth camera and projector to acquire images of the moving phantom's surface and reconstruct real-time 3D surface point cloud images. Therefore, the moving phantom can be used for position verification by the surface optical positioning system.

[0029] Reference Figure 2 , Figure 3 In some embodiments, a second motion insert 19 is also installed in the internal soft tissue 11 of the present invention, and a third motion insert 20 is installed in the bone tissue; The first motion plug 17, the second motion plug 19, and the third motion plug 20 are all connected to an external drive device for use in spatial movement. The internal soft tissue 11 is also provided with a motion insertion hole 21. In this preferred embodiment, the size of the motion insertion hole 21 is adapted to the size of the first motion insertion rod 17. That is, in use, the internal soft tissue 11 can be driven to move by inserting another first motion insertion rod 17.

[0030] Furthermore, the third motion insert 20 described in this invention is an elastic flexible tube, which is driven by a driving device to simulate the forward tilting, backward tilting, or bending movements of the vertebrae.

[0031] It should be noted that the driving device described in this invention can use a servo motor and / or a telescopic motor to control movement and rotation. Of course, for the third moving rod 20, multiple telescopic motors can be connected, and the forward tilting, backward tilting and bending movements of the third moving rod 20 can be driven by controlling the position of the shaft ends of multiple telescopic motors. Of course, the driving device has been disclosed in the prior art, and will not be described in detail here.

[0032] The movement inside the mold 1 is simulated by the first moving rod 17, the second moving rod 19, and the third moving rod 20. The rods can perform translational and rotational operations in the up, down, left, right, and forward and backward directions, and all operations are controlled by a motor. In addition, in this invention, the diameter of the first moving rod 17 is larger than the diameter of the second moving rod 19, and the diameter of the second moving rod 19 is larger than the diameter of the third moving rod 20.

[0033] In a further embodiment, the tumor tissue 18 in this invention uses a soft tissue equivalent material, specifically a mixture of agar and water, and its HU value is greater than that of the lung tissue 13 and / or liver tissue 14. The lung tumor 18 is placed inside the first motion insert 17, and the first motion insert 17 simulates the movement of the tumor. The HU value of the tumor tissue 18 is approximately -400 HUs. Specifically, in this invention, the tumor tissue 18 is used to simulate the movement state of the tumor in the lung tissue 13 and liver tissue 14, maximally restoring the movement of the thoracic and abdominal organs and tumors. It covers the thoracic and abdominal movement simulation phantom of the heart, lungs, liver, intestines and tumors, and also takes into account the surface movement simulation of the phantom to ensure that the radiation accurately irradiates the target area, minimizes the irradiation of organs at risk, and improves the radiotherapy effect.

[0034] Reference Figure 5 , Figure 6 as well as Figure 7Furthermore, this invention also proposes an artificial intelligence-driven method for simulating chest and abdominal movements, using the aforementioned phantom. Specifically, the method includes the following steps: The drive device controls the first motion insert 17 to perform translation and rotation operations to simulate the movement of internal soft tissue 11 and tumor tissue 18. Specifically, the first motion insert 17 can perform translation and rotation operations in the up, down, left, right, and front and back directions to simulate the movement of human soft tissue and lung tumors and liver tumors. It is worth noting that the tumor tissue 18 corresponding to lung tumors and liver tumors is placed inside the first motion insert 17. The displacement of the first motion insert 17 drives the movement of the tumor. The second motion insert 19 can simulate the movement of human soft tissue through similar mechanical movements. The third motion insert 20 can be made into an elastic hose and simulates the forward tilting, backward tilting or bending movements of the spine through the motor drive device. The heart tissue 12 is connected to an external mechanical air pump. The mechanical air pump is controlled to set the motor frequency according to the heart rate of an adult or patient to simulate the pumping action of the human heart. Specifically, in this embodiment, the mechanical air pump is also controlled by a motor drive device. Of course, the motor drive device can be set as a servo motor. By changing its speed and other parameters, the motor frequency of the air pump can be controlled. The intestinal tissue 15 is connected to an external peristaltic pump, which controls the peristaltic pump and sets the peristaltic frequency of normal adults or patients to simulate changes in intestinal contents and intestinal movement. Similarly, the peristaltic pump is also controlled by a motor drive device to change the peristaltic frequency of the peristaltic pump. Reference Figure 5 Lung tissue 13 is connected to an external inflation and deflation pump. Based on a deep learning algorithm model, the frequency of lung respiratory movements is predicted. The deep learning algorithm model is used to randomly generate respiratory movement signals and transmit the respiratory signal data to the motor drive device to control the inflation and deflation frequency of the external inflation and deflation pump in order to simulate the lung respiratory movements of different people. By taking 4D CT scans of the motion simulation phantom and performing dosimetric verification, or by using a surface optical positioning system to verify positional errors, we can ensure that the radiation accurately irradiates the target area, minimize the irradiation of organs at risk, and improve the effectiveness of radiotherapy.

[0035] Based on the above embodiments, the respiratory signal data in this invention includes respiratory flow rate, airway pressure and respiratory state at corresponding time points in a time series, and the respiratory state includes inspiratory trigger point, intermediate state point and expiratory trigger point. The inhalation trigger point is the time point at which the patient transitions from an exhalation state to an inhalation state. The intermediate state point is the time point at which the patient does not experience any change in exhalation or inhalation state. The exhalation trigger point is the time point at which the patient transitions from an inhalation state to an exhalation state.

[0036] Reference Figure 6 Furthermore, in a further embodiment, the method for generating respiratory motion signals according to the present invention includes: The respiratory signal dataset was divided into training, validation, and test sets in a 7:2:1 ratio. The training set is input into the deep learning algorithm model for training, and parameters such as the initial learning rate and batch size are set for the network. The Adam optimizer's automatic optimization strategy is used to automatically decay the learning rate of the network. Set up a dedicated loss function and evaluation metric for the model; During each training batch, the network model is evaluated using a validation set to monitor the quality of the network's predictions of respiratory signals. Select the best training model to predict the test set and obtain randomly generated respiratory signal data.

[0037] Reference Figure 7 It should be noted that the deep learning model algorithm framework in this invention adopts a Generative Adversarial Network (GAN), including: The generator iteratively learns the data features of the input training set through transposed convolutional layers and normalization layers, and generates predicted respiratory signal data. The discriminator uses a loss function to distinguish between real and fake samples and continuously optimizes the network loss value to achieve the best network performance. The loss functions for the discriminator and generator in GAN are defined as follows: L_D=max{log(D(x))+log(1-D(G(y)))} L_G=min{log(D(x))+log(1-D(G(y)))} Where D(x) represents the probability that the discriminator network D predicts the original data x, y represents the input data of the generator network G, the discriminator's role is to classify real and fake samples, so it is defined by the combination and maximum function, while the generator loss is defined by the minimum function.

[0038] In summary, the simulation phantom of this invention can simultaneously simulate the movement of the heart, lungs, liver, intestines, tumors, and body surface, and achieves simulated movement through multiple maneuvering methods, maximizing the reproduction of the movement of thoracic and abdominal organs, tumors, and body surface. The phantom uses simulation materials equivalent to the CT values ​​of human tissue, ensuring that the physical properties of the phantom are similar to human tissue and improving the realism of the simulation. The control methods for organ movement are flexible and diverse, including fixed frequency input, acquired patient frequency input, and movement frequency predicted by artificial intelligence algorithms. This provides assurance for dosimetric verification during 4D CT scans and positional error verification using a body surface optical positioning system, ensuring precise radiation irradiation of the target area, reducing irradiation of organs at risk, and improving the effectiveness of radiotherapy.

[0039] The above description is only a preferred embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any equivalent substitutions or modifications made by those skilled in the art within the scope of the technology disclosed in the present invention, based on the technical solution and inventive concept of the present invention, should be covered within the scope of protection of the present invention.

Claims

1. An artificial intelligence-driven simulation model of chest and abdominal movement, characterized in that, The phantom includes internal soft tissue equivalent to the CT value of human tissue, and heart tissue, lung tissue, liver tissue, intestinal tissue and bone tissue distributed in the internal soft tissue according to the location of adult organs; The internal soft tissue is provided with a first movable insertion rod, which is sequentially connected to the lung tissue and the liver tissue. The first movable insertion rod is also provided with at least one tumor tissue. The interior of the soft tissue also contains a wash-free film for measuring radiotherapy dose.

2. The phantom body according to claim 1, characterized in that: A second motion rod is also installed in the internal soft tissue, and a third motion rod is installed in the bone tissue; The first, second, and third motion rods are all connected to an external drive device for movement within space. The internal soft tissue also has a reserved hole for a moving rod.

3. The phantom body according to claim 2, characterized in that: The third motion insert is an elastic flexible tube, which is driven by a drive device to simulate the forward tilting, backward tilting or bending movements of the spine.

4. The phantom according to claim 1, characterized in that: The lung tissue is connected to an external inflation / deflation pump via a flexible lung tube, which is inserted into the phantom's oral cavity. The heart tissue is connected to an external mechanical air pump via a flexible heart tube, which is inserted into the navel of the model. The intestinal tissue is connected to a peristaltic pump via an intestinal elastic tubing, which is inserted through the perineum of the phantom.

5. The phantom body according to claim 1, characterized in that: The tumor tissue is made of soft tissue equivalent material, and its HU value is greater than that of lung tissue and / or liver tissue.

6. The mold body according to any one of claims 1-5, characterized in that: The liver tissue is made of soft tissue simulation material, and the liver tissue is set with spot contrast levels that match those of a healthy liver to simulate real liver imaging.

7. An artificial intelligence-driven method for simulating chest and abdominal movements, using a phantom as described in any one of claims 1-6, characterized in that, include: The drive device controls the first motion rod to perform translation and rotation operations, simulating the movement of internal soft tissue and tumor tissue; The heart tissue is connected to an external mechanical air pump, which is controlled to set a motor frequency according to the heart rate of an adult or patient, in order to simulate the pumping action of the human heart. The intestinal tissue is connected to an external peristaltic pump, which is controlled and the peristaltic frequency of normal adults or patients is set to simulate changes in intestinal contents and intestinal movement. The lung tissue is connected to an external inflation / deflation pump. Based on a deep learning algorithm model, the frequency of lung breathing movements is predicted. The deep learning algorithm model is used to randomly generate breathing movement signals, which are transmitted to a motor drive device to control the inflation / deflation frequency of the external inflation / deflation pump, so as to simulate the lung breathing movements of different people.

8. The method according to claim 7, characterized in that: The respiratory signal data includes respiratory flow rate, airway pressure and respiratory state at corresponding time points in a time series. The respiratory state includes inspiratory trigger point, intermediate state point and expiratory trigger point. The inhalation trigger point is the time point at which the patient transitions from an exhalation state to an inhalation state. The intermediate state point is the time point at which the patient does not experience any change in exhalation or inhalation state. The exhalation trigger point is the time point at which the patient transitions from an inhalation state to an exhalation state.

9. The method according to claim 7, characterized in that: The method for generating the respiratory motion signal includes: The respiratory signal dataset was divided into training, validation, and test sets in a 7:2:1 ratio. The training set is input into the deep learning algorithm model for training, and parameters such as the initial learning rate and batch size are set for the network. The Adam optimizer's automatic optimization strategy is used to automatically decay the learning rate of the network. Set up a dedicated loss function and evaluation metric for the model; During each training batch, the network model is evaluated using a validation set to monitor the quality of the network's predictions of respiratory signals. Select the best training model to predict the test set and obtain randomly generated respiratory signal data.

10. The method according to any one of claims 7-9, characterized in that: The deep learning model algorithm framework adopts Generative Adversarial Network (GAN), which includes: The generator iteratively learns the data features of the input training set through transposed convolutional layers and normalization layers, and generates predicted respiratory signal data. The discriminator uses a loss function to distinguish between real and fake samples and continuously optimizes the network loss value to achieve the best network performance. The loss functions for the discriminator and generator in GAN are defined as follows: L_D=max{log(D(x))+log(1-D(G(y)))} L_G=min{log(D(x))+log(1-D(G(y)))} Where D(x) represents the probability that the discriminator network D predicts the original data x, y represents the input data of the generator network G, the discriminator's role is to classify real and fake samples, so it is defined by the combination and maximum function, while the generator loss is defined by the minimum function.

Images