Lung tumor motion trail tracking system in radiotherapy
By using multimodal information fusion and dynamic spatiotemporal fusion networks, a high-precision tumor trajectory tracking model was established, which solved the problems of insufficient accuracy in tumor motion trajectory tracking and lack of dose feedback in existing technologies, and realized personalized radiotherapy accuracy optimization and safety improvement.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-12-01
- Publication Date
- 2026-04-03
AI Technical Summary
Existing non-invasive lung tumor motion tracking systems have weak generalization ability, limited trajectory tracking accuracy, and lack a real-time dose-level feedback mechanism, which limits the further development of tumor tracking strategies in optimizing radiotherapy accuracy.
Using surface and in vivo information acquisition units, combined with a dynamic spatiotemporal fusion network, multimodal information is acquired through laser ranging and optical positioning to establish a tumor trajectory tracking model. A respiratory prompting unit is also introduced to provide real-time dose feedback and individualized respiratory guidance.
It significantly improves the accuracy of tumor trajectory tracking and the generalization ability of the model, realizes non-invasive high-precision tracking, provides real-time dose-level feedback, and improves the treatment accuracy and safety of radiotherapy.
Smart Images
Figure CN121775348A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of medical information monitoring technology, and in particular to a system for tracking the movement trajectory of lung tumors during radiotherapy. Background Technology
[0002] During lung cancer radiotherapy, respiratory movements significantly affect the spatial position of the tumor. Clinically, the irradiation trajectory of the treatment head is usually set based on the tumor movement trajectory measured by 4DCT scans during the simulated localization phase. However, during actual treatment, the patient's breathing pattern often changes, causing a mismatch between the actual tumor movement trajectory and the preset trajectory. This results in the radiation beam not accurately covering the tumor target area, reducing the treatment effect.
[0003] To achieve dynamic tracking of tumor movement, current clinical techniques mainly include X-ray fluoroscopic imaging tracking and image registration. While the former offers real-time tracking, it requires implanting metal markers within the patient, posing certain invasiveness and radiation risks. The latter constructs a time-continuous tumor movement trajectory through registration of 4DCT images, avoiding invasive procedures. However, its effectiveness is easily affected by dynamic respiratory behaviors such as coughing and sneezing, and it ignores surface motion information such as respiratory waveforms, limiting the model's dynamic adaptability.
[0004] In recent years, non-invasive tracking methods based on the motion of surface markers have gained attention. These methods rely on constructing a mapping model between surface motion and tumor motion within the body, offering good operability and safety. However, due to significant individual patient differences, particularly in body size, positional changes, and different respiratory patterns, existing mapping methods, which often employ linear modeling, struggle to fully characterize complex dynamic relationships, resulting in weak model generalization ability and limited trajectory tracking accuracy. Furthermore, most existing non-invasive tracking methods based on surface marker motion rely on single-modal information acquired by a single device, such as recording only the displacement of surface markers or single-channel respiratory waveform data. This type of information has limited dimensionality and cannot fully reflect the complexity of surface motion and its non-linear coupling relationship with tumor motion within the body. Therefore, constructing a mapping model based solely on single-modal data often lacks sufficient information to meet the dynamic tracking requirements of high-precision radiotherapy.
[0005] Furthermore, since the effectiveness of radiotherapy is highly dependent on precise dose coverage, existing technologies mainly focus on acquiring and tracking tumor motion trajectories, lacking a real-time feedback mechanism at the dose level, which limits the further development of tumor tracking strategies in optimizing radiotherapy accuracy. Summary of the Invention
[0006] The technical problem to be solved by this invention is that existing non-invasive lung tumor motion trajectory tracking systems have weak generalization ability and limited trajectory tracking accuracy.
[0007] The next technical problem to be solved is that the lack of a real-time dose-level feedback mechanism limits the further development of tumor tracking strategies in optimizing radiotherapy accuracy.
[0008] The technical solution adopted by this invention to solve its technical problem is: a lung tumor motion trajectory tracking system during radiotherapy, including a body surface information acquisition unit, an in vivo information acquisition unit, and a dynamic spatiotemporal fusion network. The body surface information acquisition unit is used to acquire the patient's body surface time series information and the patient's body surface spatial motion information. The in vivo information acquisition unit is used to acquire the tumor motion information in the patient's body. The dynamic spatiotemporal fusion network model is used to fuse the patient's body surface time series and body surface spatial motion information, and to establish a mapping relationship between the fused information and the tumor motion information in the patient's body to form a tumor trajectory tracking model. The tumor trajectory tracking model is used to obtain the tumor motion information in the patient's body based on the real-time acquired patient's body surface time series information and patient's body surface spatial motion information.
[0009] The body surface information acquisition unit includes a laser ranging unit and an optical positioning unit. The body surface information acquisition unit uses the laser ranging unit to collect the undulation changes of the chest and abdominal surface caused by respiratory movements and records them as the patient's body surface time series information.
[0010] The body surface information acquisition unit collects the three-dimensional motion coordinates of body surface markers corresponding to the location of the tumor in the body through the optical positioning unit, and records them as the patient's body surface spatial motion information.
[0011] In some embodiments, optionally, the lung tumor motion trajectory tracking system during radiotherapy further includes a breathing prompting unit, which includes a trajectory tracking difference acquisition unit and a gamma pass rate acquisition unit. The trajectory tracking difference acquisition unit is used to register and compare the real-time tumor motion information obtained by the tumor trajectory tracking model with the standard tumor motion information established in the simulation positioning stage, extract the positional deviation of the two tumor motion information at each time point, and obtain time-position deviation sequence information.
[0012] The gamma pass rate acquisition unit includes a dose distribution acquisition unit, which is used to acquire the real-time dose distribution, and the gamma pass rate acquisition unit is used to perform gamma analysis on the real-time dose distribution and the planned dose distribution to obtain time-gamma pass rate sequence information.
[0013] The respiratory prompting unit integrates time-position deviation sequence information and time-gamma pass rate sequence information to generate a dynamic score that reflects the stability of the patient's respiratory status and the consistency of treatment. This score is used to adjust the patient's respiratory rhythm based on the dynamic score.
[0014] In some embodiments, optionally, the breathing prompting unit converts the dynamic score into a guiding signal that is easy for the patient to understand, to guide the patient to adjust their breathing rhythm;
[0015] The guidance signals are divided into visual guidance signals and auditory guidance information. The breathing prompting unit uses both visual guidance signals and auditory guidance information simultaneously, or uses either visual guidance signals or auditory guidance information alone to guide the patient to adjust their breathing rhythm.
[0016] The visual guidance signals specifically include a breathing cue curve and cue information. The breathing cue curve is plotted with time on the horizontal axis and dynamic score on the vertical axis. The rise or fall of the breathing cue curve reflects the degree of deviation of the current breathing from the preset benchmark of the planning stage. If the deviation exceeds the preset score threshold, a cue signal is issued to prompt the patient to adjust the breathing rhythm.
[0017] The auditory guidance signal is specifically a prompting sound that prompts the patient to adjust their breathing.
[0018] In some embodiments, the prompting information may optionally include color and / or graphic prompts for the breathing prompt curve.
[0019] In some embodiments, the dynamic score is optionally smoothed to form a visual breathing cue curve.
[0020] In some embodiments, optionally, the dynamic spatiotemporal fusion network model uses body surface time series information and patient body surface spatial motion information as inputs to the network model, and uses tumor motion information in the patient's body as supervision labels to train the network model, and establishes a mapping relationship between the fused information and the tumor motion information in the patient's body to form a tumor trajectory tracking model.
[0021] In some embodiments, the dynamic spatiotemporal fusion network model may optionally include a dynamic spatiotemporal fusion module, a spatiotemporal feature encoder module, and a dynamic decoder module.
[0022] The dynamic spatiotemporal fusion module is used to fuse body surface time series information and body surface spatial motion information to construct body surface coupled information containing respiratory phase changes and spatial deformation information, which serves as the input to the network model.
[0023] The spatiotemporal feature encoder module is used to extract temporal and spatial features from the coupled information of the body surface;
[0024] The dynamic decoder module is used to extract temporal and spatial features based on the spatiotemporal feature encoder module, and introduces an attention mechanism based on historical tracking to obtain tumor movement information in the patient's body.
[0025] The beneficial effects of this invention are:
[0026] Advantages of Multimodal Information Fusion: This invention overcomes the limitations of existing technologies that rely on single-modal information. By coordinating the laser ranging unit and the optical positioning unit, it simultaneously acquires two-dimensional time-series information from the body surface and three-dimensional spatial motion information from the body surface, thereby constructing more comprehensive coupled information from the body surface. This multimodal fusion method significantly enhances the coupling expression ability between the body surface and the movement of tumors within the body, improving the information dimensionality and accuracy of the tumor trajectory tracking model.
[0027] Non-invasive high-precision tracking: Compared to X-ray fluoroscopic imaging tracking methods that require implantation of metal markers within the patient, this invention employs a completely non-contact tracking method, avoiding invasive procedures and additional radiation risks. Simultaneously, the nonlinear mapping relationship established through a dynamic spatiotemporal fusion network overcomes the shortcomings of traditional linear modeling methods in depicting complex dynamic relationships, significantly improving the accuracy of tumor trajectory tracking and the model's generalization ability.
[0028] Intelligent dynamic adaptability: The dynamic spatiotemporal fusion network of this invention introduces an attention mechanism based on historical tracking, which can adaptively capture the dynamic changes in body surface movement and effectively cope with the interference of individual patient differences, changes in body position, coughing, sneezing and other irregular breathing behaviors. Compared with traditional image registration technology, it has stronger dynamic adaptability and robustness.
[0029] Real-time dose-level feedback: This invention is the first to incorporate feedback information on the actual dose distribution in vivo into a tumor tracking system. The dose distribution acquisition unit collects the transmission flux distribution in real time through an electronic detector, thereby obtaining the real-time dose distribution. By calculating the gamma pass rate, a real-time dose-level feedback mechanism is formed, which overcomes the shortcomings of existing technologies that only focus on trajectory tracking while ignoring the quality of dose coverage.
[0030] Personalized breathing guidance function: This invention provides patients with intuitive visual or auditory guidance by fusing trajectory tracking differences and gamma pass rate information to generate a personalized breathing prompt curve in real time. This helps patients actively adjust their breathing state during treatment, making the actual tumor movement trajectory closer to the standard trajectory, thereby improving the precise coverage of the tumor target area by the X-ray beam.
[0031] Significant clinical application value: This invention solves the problem of tumor position changes caused by respiratory motion during lung cancer radiotherapy, improving the precision and safety of radiotherapy. Through precise dynamic tracking and real-time dose-level feedback, it can maximize the dose coverage of the tumor target area while minimizing damage to surrounding normal tissues, providing lung cancer patients with more precise and personalized radiotherapy treatment plans. Attached Figure Description
[0032] The present invention will be further described below with reference to the accompanying drawings and embodiments;
[0033] Figure 1 This is a system principle block diagram of the present invention;
[0034] Figure 2 This is a schematic diagram of the structure of the body surface information acquisition unit of the present invention;
[0035] In the diagram, 1. CT positioning machine, 2. laser rangefinder, 3. controller, 4. fixed bracket, 5. binocular infrared camera, 6. infrared LED light, 7. optical positioning unit, and 8. body surface marker. Detailed Implementation
[0036] like Figure 1 As shown, a lung tumor motion trajectory tracking system during radiotherapy includes a body surface information acquisition unit, an in vivo information acquisition unit, and a dynamic spatiotemporal fusion network. The body surface information acquisition unit is used to collect the patient's body surface time-series information and the patient's body surface spatial motion information. The in vivo information acquisition unit is used to collect the tumor motion information inside the patient's body. The dynamic spatiotemporal fusion network model is used to fuse the patient's body surface time-series information and body surface spatial motion information, and establish a mapping relationship between the fused information and the tumor motion information inside the patient's body to form a tumor trajectory tracking model. The tumor trajectory tracking model is used to obtain the tumor motion information inside the patient's body based on the real-time collected patient's body surface time-series information and patient's body surface spatial motion information.
[0037] The body surface information acquisition unit includes a laser ranging unit and an optical positioning unit 7. The laser ranging unit acquires the undulations of the chest and abdominal surfaces caused by respiratory movements, recording this as the patient's body surface time-series information. The optical positioning unit 7 acquires the three-dimensional motion coordinates of body surface markers 8 corresponding to the location of the tumor within the body, recording this as the patient's body surface spatial motion information.
[0038] The lung tumor motion trajectory tracking system during radiotherapy also includes a breathing prompting unit, which includes a trajectory tracking difference acquisition unit and a gamma pass rate acquisition unit.
[0039] The tumor motion information is specifically the tumor motion trajectory. The trajectory tracking difference acquisition unit is used to register and compare the real-time tumor motion trajectory obtained by the tumor trajectory tracking model with the standard trajectory established in the simulation positioning stage, extract the positional deviation of the two trajectories at each time point, and obtain the time-positional deviation sequence information.
[0040] The gamma pass rate acquisition unit includes a dose distribution acquisition unit, which is used to acquire the real-time dose distribution, and the gamma pass rate acquisition unit is used to perform gamma analysis on the real-time dose distribution and the planned dose distribution to obtain time-gamma pass rate sequence information.
[0041] The dose distribution acquisition unit acquires the flux distribution of radiation rays passing through the patient's lungs in real time using an electronic detector positioned below the lungs. This flux distribution is called the transmission flux distribution. The electronic detector is an electronic radiation field imaging device.
[0042] The dose distribution acquisition unit uses ray tracing technology to reconstruct the specific path of the radiotherapy rays from the radiation source to the patient's body surface based on the input transmission flux distribution. It then uses lung tissue flux-dose conversion factor and correction coefficient combined with 4DCT images to convert the transmission flux distribution into the real-time dose distribution of the patient's lungs.
[0043] The dose distribution acquisition unit compares the real-time dose distribution with the planned dose distribution calculated based on the treatment planning system. It also considers two dimensions: spatial location deviation and dose difference. The unit uses gamma analysis to perform quantitative evaluation. Under the preset tolerance criteria, it calculates the proportion of voxels that pass the gamma evaluation out of the total voxels, which is denoted as the gamma pass rate.
[0044] The respiratory prompting unit integrates time-position deviation sequence information and time-gamma pass rate sequence information to generate a dynamic score that reflects the stability of the patient's respiratory status and the consistency of treatment. This score is used to adjust the patient's respiratory rhythm based on the dynamic score.
[0045] The formula for generating dynamic scores by the breathing prompting unit is shown in equation (1):
[0046] (1)
[0047] In equation (1), Score(t) represents the dynamic score at time t; S(t) represents the severity of the positional deviation (a larger value indicates poorer spatial stability); T(t) represents the trend of gamma pass rate (a larger value indicates poorer treatment consistency); I p (t) represents the stability score based on positional deviation; Ig(t) represents the consistency score based on gamma pass rate.
[0048] To facilitate patient adjustment of breathing rhythm, the breathing prompting unit converts dynamic scores into easily understandable guidance signals to guide patients in adjusting their breathing rhythm.
[0049] The guidance signals are divided into visual guidance signals and auditory guidance information. The system uses both visual and auditory guidance signals simultaneously, or uses either visual or auditory guidance signals alone for breathing guidance.
[0050] The visual guidance signals specifically include a breathing cue curve, a horizontal baseline, horizontal threshold lines, and cue information. The breathing cue curve is plotted on the horizontal axis with time and the dynamic score on the vertical axis. The horizontal baseline is plotted on the horizontal axis with time and the dynamic score corresponding to the ideal breathing state preset in the planning stage on the vertical axis. The horizontal baseline represents the ideal breathing state in the planning stage. There are two horizontal threshold lines, located on the upper and lower sides of the horizontal baseline, respectively. The horizontal threshold lines are plotted on the horizontal axis with time, and their distance from the horizontal baseline in the vertical direction is equal to the preset score threshold.
[0051] More specifically, the breathing prompting unit converts dynamic scores into breathing prompting curves as follows: after smoothing, the dynamic scores form a visual breathing prompting curve. The rise or fall of the curve reflects the degree of deviation of the current breathing from the preset baseline of the planning stage. If the deviation exceeds the preset scoring threshold, a prompting message is issued to prompt the patient to adjust the breathing rhythm as close as possible to the horizontal baseline, thereby achieving individualized breathing guidance.
[0052] The smoothing process specifically involves using a moving average algorithm or a Kalman filter algorithm to eliminate high-frequency noise in the breathing prompt curve, thereby improving the stability and readability of the curve.
[0053] To better guide the visuals, the prompts are provided by color and / or graphic cues on the breathing cue curve.
[0054] When the respiratory indicator curve is above or below the horizontal baseline and exceeds the horizontal threshold, the respiratory indicator curve will turn red or a graphic prompt will pop up to guide the patient to reduce or increase the respiratory amplitude and frequency. The graphic prompt will specifically say "Please slow down", "Breathe steadily", or "Breathe faster". When the respiratory indicator curve does not exceed the horizontal threshold and is close to the horizontal baseline, it indicates that the respiratory status is good. The color of the respiratory indicator curve will remain green or an encouraging graphic prompt will be displayed to guide the patient to maintain the current respiratory rhythm.
[0055] The visual guidance signals are displayed on a screen in the treatment room.
[0056] Auditory guidance signals are specifically prompts for patients to adjust their breathing. These prompts include a soft but continuous "beep" sound, the frequency of which is positively correlated with the degree of over-limit, as well as a quiet or gentle background sound.
[0057] When the dynamic score is higher or lower than the baseline and exceeds the threshold, a soft but continuous "beep" sound is emitted, the frequency of which is positively correlated with the degree of exceedance, indicating that the patient needs to adjust. When breathing returns to the ideal breathing state, that is, when the breathing prompt curve does not exceed the threshold and is in line with the baseline, the prompt sound stops or becomes a gentle background sound.
[0058] The breathing prompting unit provides patients with real-time, intuitive, and personalized breathing guidance through its generated dynamic scores and guidance signals, forming a closed-loop feedback system of "monitoring-calculation-prompting-adjustment." The process of breathing guidance through the breathing prompting unit aims to train and assist patients to reproduce, as closely as possible, the ideal breathing pattern corresponding to the standard trajectory during the simulation positioning phase.
[0059] The process of respiratory guidance via a respiratory prompting unit in the lung tumor motion trajectory tracking system during radiotherapy includes the following steps:
[0060] Step 1. Real-time monitoring and calculation: The system runs in real time. The trajectory tracking difference acquisition unit continuously acquires the time-position deviation sequence, the gamma pass rate acquisition unit continuously acquires the time-gamma pass rate sequence, and the breathing prompting unit integrates these two types of information according to the above formula (1) to calculate a dynamic score that reflects the current breathing quality, i.e., the stability of the patient's breathing status and the consistency of treatment.
[0061] Step 2. Signal Conversion and Cue Generation: The breathing cue unit converts the dynamic score into visual and auditory guidance signals that are easy for patients to understand.
[0062] The process of visual guidance using visual guidance signals is as follows: On the display screen in the treatment room, a breathing prompt curve is presented to the patient. Simultaneously, a horizontal baseline representing the "ideal breathing state" and two horizontal threshold lines are also displayed on the screen. When the patient's real-time breathing prompt curve is higher or lower than the horizontal baseline and exceeds the horizontal threshold lines, it indicates that the current breathing is too fast, too deep, unstable, or too slow. The system guides the patient to reduce or increase their breathing amplitude and frequency by changing the breathing prompt curve to red or displaying a visual prompt such as "Please slow down or steady your breathing or speed up your breathing." When the breathing prompt curve does not exceed the horizontal threshold lines and is aligned with the horizontal baseline, it indicates a good breathing state. The system guides the patient to maintain the current breathing rhythm by keeping the horizontal baseline green or displaying an encouraging indicator.
[0063] The process of auditory guidance, either simultaneously or alternatively, is as follows: when the dynamic score is higher or lower than the preset scoring threshold, a soft but continuous "beep" sound is emitted, the frequency of which is positively correlated with the degree of exceeding the limit, prompting the patient to make adjustments; when breathing returns to the ideal state, i.e., the dynamic score is within the preset scoring threshold, the prompt sound stops or becomes a gentle background sound.
[0064] Step 3. Patient Response and Closed-Loop Regulation: Based on the received visual guidance signals and / or auditory guidance information, the patient actively and consciously adjusts their breathing, such as attempting more stable diaphragmatic breathing or exhaling and inhaling in accordance with the rhythm of the auditory guidance information. This adjustment by the patient immediately changes their body surface information, including the patient's body surface time-series information and the patient's body surface spatial motion information, which is then captured by the body surface information acquisition unit.
[0065] Step 4. System Feedback and Optimization: New surface information is input into the tumor trajectory tracking model to predict new tumor movement trajectories, and the cycle of steps 1-4 is repeated.
[0066] Through this real-time, high-frequency closed-loop feedback, patients can quickly learn and continuously optimize their breathing patterns, making the actual tumor movement trajectory infinitely close to the planned standard trajectory, thereby ensuring the precise delivery of radiotherapy doses.
[0067] In this embodiment, the laser ranging unit includes a laser rangefinder 2, a controller 3, and a mounting bracket 4. The laser rangefinder 2 is a high-precision laser rangefinder, which is vertically mounted above the chest and abdomen region of the human body via the mounting bracket 4. The mounting bracket 4 is mounted on a CT positioning machine 1 used for 4DCT scanning, and the CT positioning machine 1 is a Brilliance Big Bore CT positioning machine.
[0068] The specific process by which the body surface information acquisition unit acquires the patient's body surface time-series information through the laser ranging unit is as follows:
[0069] The laser rangefinder 2 of the laser ranging unit continuously sends light pulses at a frequency of 60Hz to the patient's chest and abdomen to perform continuous ranging. The controller 3 automatically collects the ranging data. This ranging data, arranged in chronological order, represents the undulations of the chest and abdominal surface caused by respiratory movements, and is recorded as the patient's body surface time-series information. The body surface time-series information is represented as a body surface time-series information matrix X. tepmoral .
[0070] In this embodiment, the optical positioning unit 7 is equipped with two sets of high-precision infrared LED light sources and a binocular infrared camera 5. The surface marker 8 is a spherical optical reflective marker affixed to the anatomical region corresponding to the location of the tumor in the body, also known as an optical positioning marker sphere. The infrared LED light source illuminates the surface marker 8, and the reflected light information is synchronously captured by the binocular infrared camera 5. The three-dimensional motion coordinates of the surface marker 8 are calculated in real time using triangulation. These three-dimensional motion coordinates are the three-dimensional spatial coordinate information of the tumor movement projection onto the body surface, and are recorded as the patient's surface spatial motion information.
[0071] More specifically, the optical positioning unit 7 is the AimPosition OP-M620 optical positioning system, and the surface marker 8 is an 8 mm spherical optical reflective marker with a highly reflective coating. The AimPosition OP-M620 optical positioning system uses a high-precision infrared LED light 6 with a wavelength of 850 nm, and a binocular infrared camera 5 with a resolution of 1280×1024 and a pixel size of 12 μm. The AimPosition OP-M620 optical positioning system continuously captures the three-dimensional spatial coordinate information of the tumor motion projection onto the body surface at a frame rate of 60 frames per second.
[0072] The specific process by which the body surface information acquisition unit collects the patient's body surface spatial motion information through the optical positioning unit 7 is as follows: Before the 4DCT scan, a body surface marker 8 with a diameter of 8 mm is affixed to the anatomical region on the patient's body surface corresponding to the location of the tumor inside the body. This body surface marker 8 employs a standardized counterweight design to ensure relative stability during patient breathing or changes in body position. During the 4DCT scan, the infrared light emitted by the two sets of high-precision infrared LED light sources of the optical positioning unit 7 illuminates the body surface marker 8 at a downward angle of 75°, a height of 2 m from the ground, and a distance of 1.2 m directly in front of the bed. The reflected light information is simultaneously captured by the binocular infrared camera 5, and the three-dimensional motion coordinates of the body surface marker 8 are calculated in real time using triangulation. These three-dimensional motion coordinates represent the three-dimensional spatial coordinates of the tumor's motion projection onto the body surface, denoted as the patient's body surface spatial motion information, and represented as the body surface spatial motion information matrix X. spatial .
[0073] The in vivo information acquisition unit acquires tumor motion information within the patient's body using 4DCT images. The specific method is as follows: Step 1. Segment the tumor target region in the CT image for each respiratory phase; Step 2. Based on the pixel spacing and slice thickness information in the CT image, convert the voxel index coordinates within the tumor target region into actual three-dimensional coordinates in physical space, thereby obtaining the tumor centroid coordinates for each phase; Step 3. Use a Kalman filter algorithm to smooth the discrete tumor centroid coordinates and model their trajectory to obtain the tumor motion information within the patient's body.
[0074] In this embodiment, the specific process by which the in vivo information acquisition unit acquires tumor movement information from the patient's body using 4DCT images is as follows:
[0075] Step 1. Invite two radiation oncologists with more than 5 years of clinical experience to perform tumor target region segmentation on the CT images of ten phases in the 4DCT image using the 3D Slicer platform. The segmented CT images are GTV images.
[0076] Step 2. Physical calibration of voxel coordinates: Based on the pixel spacing and slice thickness information in the CT image, the voxel index coordinates in the tumor target area are converted into actual three-dimensional coordinates in physical space, thereby obtaining the tumor centroid coordinates in each phase.
[0077] Step 2 specifically involves: reading the pixel spacing and layer thickness in the GTV image, converting the voxel index coordinates in the GTV image into actual three-dimensional coordinates in physical space, defining the origin as (x0, y0, z0), the pixel spacing as (dx, dy), and the layer thickness as dz. Then, the actual three-dimensional coordinates of the voxels in the GTV image can be expressed as (x0+i×dx, y0+j×dy, z0+k×dz), where (i, j, k) are the voxel indices in the GTV image.
[0078] The formula for calculating the centroid is shown in equation (2):
[0079] (2)
[0080] In equation (2), The coordinates of the three-dimensional centroid of the GTV image are represented by N; N represents the total number of voxels in the GTV image.
[0081] Step 3. Trajectory Modeling: The tumor centroid coordinates extracted from the 4DCT images are discrete and discontinuous, making it difficult to directly construct a continuous motion trajectory. Therefore, the Kalman filter algorithm is used to smooth the discrete tumor centroid coordinates and model the trajectory to obtain the tumor motion information in the patient's body.
[0082] Step 3 is specifically divided into:
[0083] First, establish the state equations, as shown in equation (3):
[0084] (3)
[0085] In equation (3), X(k) represents the state vector at the current time k, which contains the position and velocity information of the tumor centroid; A is the state transition matrix; B is the control matrix; u(k) is the control vector; and w(k) is the process noise.
[0086] Next, the observation equation is established, as shown in equation (4):
[0087] (4)
[0088] In equation (4), Z(k) is the observation vector, i.e., the coordinates of the tumor centroid extracted from 4DCT; M is the observation matrix; and v(k) is the observation noise.
[0089] The Kalman filter algorithm iterates through tracking and updating steps, continuously updating the state vector X(k) and observation vector Z(k) to obtain a series of continuous estimates of the tumor centroid coordinates. These estimates can be expressed as a function of time, i.e., the equation of continuous trajectory motion, reflecting the continuous motion state of the tumor in three-dimensional space, denoted as the in vivo tumor motion information, which is represented by the in vivo information matrix X. invivo .
[0090] The process of acquiring real-time dose distribution by the dose distribution acquisition unit is as follows: A) The flux distribution of the radiotherapy rays after passing through the patient's lungs is acquired in real time by an electronic detector below the patient. This flux distribution is called the fluoroscopic flux distribution; B) The fluoroscopic flux distribution is mapped onto a three-dimensional model of the patient's lung surface captured in real time to obtain the three-dimensional transmission flux distribution. The specific path of the radiotherapy rays from the radiation source to the patient's lung surface is reconstructed using ray tracing technology. Based on the three-dimensional transmission flux distribution and the specific path, the flux distribution of the radiotherapy rays before entering the patient's lungs is calculated according to the energy deposition per unit mass of the original radiation and the flux distribution kernel. This flux distribution is called the incident flux distribution P1; C) Based on the exponential decay law of the radiotherapy rays, the incident flux distribution P2 is directly derived by reverse calculation from the incident flux distribution P1; D) The incident flux distribution P1 and the incident flux distribution P2 are iteratively optimized multiple times to obtain the optimized incident flux distribution Pn; E) The incident flux distribution Pn is dynamically corrected in real time using a biodeformation registration algorithm. The corrected measurement results are mapped to a unified reference coordinate system. Then, the Monte Carlo algorithm is used in the unified reference coordinate system to simulate the transmission process of photons in the radiotherapy beam in a non-uniform medium. Specific biological tissue-dose conversion factors and correction coefficients are applied in combination with 4DCT images of the patient's lungs to convert the incident flux distribution Pn into the actual absorbed dose distribution, thus reconstructing the three-dimensional real-time dose distribution in the patient's lungs.
[0091] This embodiment uses a laser ranging unit to emit and receive light pulses and three-dimensional point clouds to capture the deformation of the patient's lung surface in real time, constructing a three-dimensional model of the patient's lung surface, thus realizing the function of capturing a three-dimensional model of the patient's lung surface in real time.
[0092] In this embodiment, a single laser ranging unit is used to capture the three-dimensional model of the patient's lung surface and the real-time time-series information of the patient's body surface. Alternatively, two laser ranging units can be used to capture the three-dimensional model of the patient's lung surface and the real-time time-series information of the patient's body surface, respectively.
[0093] The dynamic spatiotemporal fusion network model uses time-series information from the body surface and spatial motion information from the patient's body surface as inputs to the network model, and tumor motion information within the patient's body as a supervisory label to train the network model. The fused information is then mapped to the tumor motion information within the patient's body to form a tumor trajectory tracking model.
[0094] The dynamic spatiotemporal fusion network model includes a dynamic spatiotemporal fusion module, a spatiotemporal feature encoder module, and a dynamic decoder module. The body surface time series information is represented as a body surface time series information matrix X. tepmoral The body surface spatial motion information is represented as the body surface spatial motion information matrix X. spatial ;
[0095] The dynamic spatiotemporal fusion module is used to fuse the time series information matrix X of the body surface. tepmoral and the surface space motion information matrix X spatial Construct a body surface coupled information matrix X containing information on respiratory phase changes and spatial deformation. fuse , as input to the network model;
[0096] Body surface coupling information matrix X fuse It reflects the correlation between respiratory waveform information and spatial motion information at different time points.
[0097] The time series information matrix X of the fusion body of the dynamic spatiotemporal fusion module tepmoral and the surface space motion information matrix X spatial The specific steps are as follows:
[0098] First, X tepmoral With X spatial The min-max standardization is performed, and the calculation process is shown in equation (5):
[0099] (5)
[0100] In equation (5), X stand-temporal-ij X represents the element in the i-th row and j-th column of the standardized body surface time series information matrix. stand-spatial-ij X represents the element in the i-th row and j-th column of the standardized body surface spatial motion information matrix. temporal-min and X temporal-max Let X represent the minimum and maximum values of all elements in the body surface time series information matrix, respectively. spatial-min and X spatial-max These represent the minimum and maximum values of all elements in the body surface spatial motion information matrix, respectively.
[0101] Secondly, split X stand-temporal and X stand-spatial The elements in X are used to slide the window method. stand-temporalDecomposed into a body surface time series information submatrix X stand-temporal-sub1 X stand-temporal-sub2 X stand-temporal-sub3 X stand-spatial Decomposed into vertical motion information matrix X in body surface space stand-spatial-ud The front-to-back motion information matrix X on the body surface space stand-spatial-ba X, the left and right motion information matrix on the body surface space stand-spatial-lr .
[0102] Finally, X is obtained by fusing the covariance minimization principle. fuse As shown in equation (6):
[0103] (6)
[0104] In equation (6), X fuse-sub1 X fuse-sub2 X fuse-sub3 X represents the multimodal body surface coupling information matrix. fuse The specific calculation process for the submatrix in the equation is shown in equation (7):
[0105] (7)
[0106] In equation (7), X stand-temporal-sub1-ij X stand-temporal-sub2-ij X stand-temporal-sub3-ij X represents the elements in the i-th row and j-th column of the submatrix of body surface time series information, respectively. stand-spatial-ud-ij X stand-spatial-ba-ij X stand-spatial-lr-ij These represent the elements in the i-th row and j-th column of the vertical motion information matrix, the forward-backward motion information matrix, and the left-right motion information matrix in the body surface space, respectively. w is the weight, and w = w m The condition is met such that the covariance of the elements in both matrices is minimized. m The value range is 0.01 to 0.99, with a step size of 0.01, and corresponds one-to-one with the values of m = 1, 2, ..., 99, i.e., w1 = 0.01, w2 = 0.02, ..., w 99 = 0.99.
[0107] The spatiotemporal feature encoder module is used to extract the body surface coupling information matrix X. fuse Temporal and spatial characteristics;
[0108] The dynamic decoder module is used to extract temporal and spatial features based on the spatiotemporal feature encoder module, and introduces an attention mechanism based on historical tracking to obtain tumor movement information in the patient's body.
Claims
1. A lung tumor motion trajectory tracking system during radiotherapy, characterized in that: It includes a body surface information acquisition unit, an internal body information acquisition unit, and a dynamic spatiotemporal fusion network. The body surface information acquisition unit is used to collect the patient's body surface time series information and the patient's body surface spatial motion information. The internal body information acquisition unit is used to collect the tumor motion information inside the patient's body. The dynamic spatiotemporal fusion network model is used to fuse the patient's body surface time series and body surface spatial motion information, and establish a mapping relationship between the fused information and the tumor motion information inside the patient's body to form a tumor trajectory tracking model. The tumor trajectory tracking model is used to obtain the tumor motion information inside the patient's body based on the real-time collected patient's body surface time series information and patient's body surface spatial motion information. The body surface information acquisition unit includes a laser ranging unit and an optical positioning unit. The body surface information acquisition unit uses the laser ranging unit to collect the undulation changes of the chest and abdominal surface caused by respiratory movements and records them as the patient's body surface time series information. The body surface information acquisition unit collects the three-dimensional motion coordinates of body surface markers corresponding to the location of the tumor in the body through the optical positioning unit, and records them as the patient's body surface spatial motion information.
2. The lung tumor motion trajectory tracking system during radiotherapy according to claim 1, characterized in that: It also includes a breathing prompting unit, which includes a trajectory tracking difference acquisition unit and a gamma pass rate acquisition unit. The trajectory tracking difference acquisition unit is used to register and compare the real-time tumor motion information obtained by the tumor trajectory tracking model with the standard tumor motion information established in the simulation positioning stage, extract the positional deviation of the two tumor motion information at each time point, and obtain time-positional deviation sequence information. The gamma pass rate acquisition unit includes a dose distribution acquisition unit, which is used to acquire the real-time dose distribution, and the gamma pass rate acquisition unit is used to perform gamma analysis on the real-time dose distribution and the planned dose distribution to obtain time-gamma pass rate sequence information. The respiratory prompting unit integrates time-position deviation sequence information and time-gamma pass rate sequence information to generate a dynamic score that reflects the stability of the patient's respiratory status and the consistency of treatment. This score is used to adjust the patient's respiratory rhythm based on the dynamic score.
3. The lung tumor motion trajectory tracking system during radiotherapy according to claim 2, characterized in that: The breathing prompting unit converts dynamic scores into guidance signals that are easy for patients to understand, guiding patients to adjust their breathing rhythm; The guidance signals are divided into visual guidance signals and auditory guidance information. The breathing prompting unit uses both visual guidance signals and auditory guidance information simultaneously, or uses either visual guidance signals or auditory guidance information alone to guide the patient to adjust their breathing rhythm. The visual guidance signals specifically include a breathing cue curve and cue information. The breathing cue curve is plotted with time on the horizontal axis and dynamic score on the vertical axis. The rise or fall of the breathing cue curve reflects the degree of deviation of the current breathing from the preset benchmark of the planning stage. If the deviation exceeds the preset score threshold, a cue signal is issued to prompt the patient to adjust the breathing rhythm. The auditory guidance signal is specifically a prompting sound that prompts the patient to adjust their breathing.
4. The lung tumor motion trajectory tracking system during radiotherapy according to claim 3, characterized in that: The prompt information includes the color and / or graphic prompts of the breathing prompt curve.
5. The lung tumor motion trajectory tracking system during radiotherapy according to claim 3, characterized in that: The dynamic score is smoothed to form a visual breathing cue curve.
6. The lung tumor motion trajectory tracking system during radiotherapy according to claim 1, characterized in that: The dynamic spatiotemporal fusion network model uses body surface time series information and patient body surface spatial motion information as inputs to the network model, and uses tumor motion information in the patient's body as supervision labels to train the network model. The fused information is then mapped to the tumor motion information in the patient's body to form a tumor trajectory tracking model.
7. The lung tumor motion trajectory tracking system during radiotherapy according to claim 1 or 6, characterized in that: The dynamic spatiotemporal fusion network model includes a dynamic spatiotemporal fusion module, a spatiotemporal feature encoder module, and a dynamic decoder module; The dynamic spatiotemporal fusion module is used to fuse body surface time series information and body surface spatial motion information to construct body surface coupled information containing respiratory phase changes and spatial deformation information, which serves as the input to the network model. The spatiotemporal feature encoder module is used to extract temporal and spatial features from the coupled information of the body surface; The dynamic decoder module is used to extract temporal and spatial features based on the spatiotemporal feature encoder module, and introduces an attention mechanism based on historical tracking to obtain tumor movement information in the patient's body.