Pulmonary nodule real-time positioning system based on patient breathing fluctuation modeling

Abstract

The invention provides a pulmonary nodule real-time positioning system based on patient respiratory fluctuation modeling. The system comprises a multi-source sensing unit, a signal processing module, a respiratory dynamics modeling module, a regionalization sensitivity field module, a disturbance observer, a self-healing registration module, a fusion filtering module and an output and interaction module. A dynamic model with a hysteretic operator is established through multi-source signals such as thoracico-abdominal movement, laryngeal acoustics and vibration, end tidal carbon dioxide, respiratory flow, body position drifting and instrument states, millimeter-level positioning of pulmonary nodules along with respiration is achieved in combination with regionalized sensitivity mapping, errors caused by respiratory movement, local disturbance and body position drifting are effectively solved, and the positioning accuracy of the pulmonary nodules is improved. And the safety and effectiveness of clinical operation are obviously improved.

Description

Technical Field

[0001] This invention belongs to the field of medical navigation and treatment control technology, specifically a real-time positioning system for lung nodules based on patient respiratory fluctuation modeling. Background Technology

[0002] Pulmonary nodules are a common clinical imaging finding, and with the widespread use of low-dose spiral CT, the detection rate of nodules has significantly improved. Pathological examination or precise radiotherapy of suspicious nodules has become an important approach for the early diagnosis and treatment of lung cancer. However, in the aforementioned clinical procedures, the displacement caused by respiratory motion has always been a key challenge restricting the accuracy and safety of the procedure.

[0003] During spontaneous breathing, lung tissue and its internal nodules undergo periodic displacements ranging from millimeters to centimeters due to the movement of the chest wall and diaphragm. Studies have shown that nodules located in the lower lobes or near the diaphragm often exhibit greater respiratory movement amplitudes, making it difficult to stabilize their spatial position during radiotherapy, puncture, or bronchoscopic navigation. Existing localization methods mainly fall into the following categories: Respiratory-gated radiotherapy: This technique involves irradiating the nodule during a relatively stable respiratory phase, such as at the end of expiration, to reduce errors caused by tissue displacement. However, this method only works within a fixed window and cannot track the real-time position of the nodule; if the patient's breathing is irregular or if they sigh or cough, the gating strategy may fail.

[0004] Surface marking and tracking systems use cameras or infrared devices to track the movement of the patient's chest and abdomen surface and estimate the displacement of internal nodules. However, there is no linear correspondence between surface movement and intrapulmonary nodule movement, and the correlation between the two decreases significantly when there is respiratory asymmetry, alveolar recruitment, or changes in body position, leading to inaccurate predictions.

[0005] Preoperative image fusion and registration method: Before the interventional procedure, the preoperative CT images are registered with the intraoperative coordinate system in one go, serving as the reference for the nodule location. However, during long-term procedures, the patient's position may gradually slide down or rotate. Without real-time correction, significant cumulative deviations will eventually occur.

[0006] Existing technologies generally suffer from the following shortcomings: Most models treat the inspiratory and expiratory pathways as symmetrical, but in reality, the displacements differ for the same tidal volume. In particular, after sighing or alveolar recruitment events, the nodule trajectory exhibits a circular change, which existing methods cannot adapt to in a timely manner.

[0007] During puncture or bronchoscopic intervention, the entry of instruments into the lung segment alters the movement patterns of local tissues. Without compensation, systematic errors may occur in the predicted location.

[0008] During operation, bed vibration, patient slippage, or slight head movements can all cause global drift. Existing methods typically lack online correction mechanisms, making it difficult to guarantee long-term consistency.

[0009] Most systems use a single filtering model, which cannot remain robust to short-term events such as coughing and swallowing, and is prone to localization jumps or delays.

[0010] In summary, current technologies are still insufficient for achieving stable, real-time, millimeter-level precision localization of pulmonary nodules in complex clinical environments. Especially under conditions involving respiratory mode switching, device disturbances, and overall drift, existing methods are more prone to failure and cannot meet the dual clinical requirements of safety and effectiveness. Summary of the Invention

[0011] The purpose of this invention is to provide a real-time lung nodule localization system based on patient respiratory fluctuation modeling. This invention can effectively solve the errors caused by respiratory motion, local disturbances and body position drift, and significantly improve the safety and effectiveness of clinical operations.

[0012] The technical solution adopted in this invention is as follows: A real-time lung nodule localization system based on patient respiratory fluctuation modeling, the system comprising: The multi-source sensing unit is used to collect patient chest and abdominal surface motion signals, pre-larynx acoustic and vibration signals, end tidal carbon dioxide and respiratory flow signals, as well as body position and instrument status information. The signal processing module is used to denoise, extract features and synchronize time on the sensing signal, and generate low-dimensional state parameters related to respiration. The respiratory dynamics modeling module includes a state-space model with hysteresis operators, used to characterize the inconsistency between inhalation and exhalation pathways and mode switching caused by sighing / recruiting; The regional sensitivity field module constructs a mapping relationship between the lung segment where the nodule is located and respiratory status parameters based on preoperative imaging information; The perturbation observer is used to estimate and correct the local compliance parameters of the lung segment in real time when interventional devices enter the local airway; The self-healing registration module is used to perform rigid body or affine correction of global coordinate system drift using the inertial measurement unit on the bed, mask, or chest and external markers. The fusion filtering module uses a switching unscented Kalman filter or particle filter to fuse the outputs of the above modules to obtain the real-time three-dimensional position and uncertainty of the lung nodules. The output and interaction module is used to display nodule location, event status and quality control information, or to provide gating signals to external treatment / navigation devices.

[0013] The respiratory dynamics modeling module employs a Bouc-Wen type hysteresis operator to simulate the displacement difference between the inspiratory and expiratory pathways under the same tidal volume.

[0014] The regionalized sensitivity field module divides a lung lobe or segment into multiple sub-regions and establishes main coupling parameters only for the sub-region where the nodule is located, thereby reducing the computational dimension and improving the robustness of localization.

[0015] The disturbance observer estimates the degree of decrease in local compliance of the lung segment by combining instrument insertion mechanical information, end-expiratory carbon dioxide waveform changes and pre-larynx high-frequency friction sound characteristics, and completes parameter convergence within 3 to 5 respiratory cycles.

[0016] The self-healing registration module automatically performs rigid body correction on the nodule position when it detects an overall drift of more than 1 to 2 millimeters or a rotation of more than 0.5 degrees, and temporarily increases the uncertainty to ensure the safety of the positioning result.

[0017] The fusion filtering module adopts an interactive multi-model (IMM) structure, which switches filtering parameters when a sighing, recruiting, coughing, or swallowing event is detected.

[0018] The output and interaction module provides the nodule location with a 95% confidence ellipsoid, and outputs a gating stop signal to link the radiotherapy or puncture equipment when the uncertainty exceeds the threshold.

[0019] A real-time lung nodule localization method based on a patient respiratory fluctuation modeling system includes the following steps: S1. Acquire multi-source signals from the thoracic and abdominal surface, pre-larynx, CO2 / flow rate, and IMU; S2. Feature extraction and generation of low-dimensional state parameters; S3. Predicting nodule location based on hysteresis dynamics model and regionalized sensitivity field; S4. When instrument entry is detected, adjust the local compliance parameters; S5. When global drift is detected, perform self-healing registration; S6. The outputs of each module are fused using a switching filtering method to obtain the three-dimensional position and uncertainty of the nodule; S7. Output the results for display and linkage with clinical equipment.

[0020] The method executes a short-window expectation-maximization (EM) algorithm for 10 to 20 respiratory cycles after each mode switching event to re-estimate the sensitivity field parameters.

[0021] The method temporarily freezes the location output or increases the uncertainty when a swallowing or coughing event is detected, in order to avoid misleading clinical operations.

[0022] The beneficial effects of this invention are as follows: This invention introduces a Bouc–Wen type hysteresis operator, which can accurately describe the displacement differences in the inspiratory / expiratory pathways under the same tidal volume. It switches the parameter set upon detecting a sigh or alveolar recruitment, and achieves rapid convergence by combining a short window reestimation over 10–20 respiratory cycles. This avoids systematic errors caused by changes in lung compliance, maintaining millimeter-level accuracy even in complex respiratory patterns.

[0023] By combining instrument depth signals, end-expiratory carbon dioxide waveforms, and pre-larynx high-frequency acoustic features, the mechanism estimates local compliance decline in real time and updates parameters within 3–5 cycles. This mechanism effectively addresses the problem of existing technologies neglecting interventional disturbances, ensuring that the predicted location of nodules no longer exhibits systematic shifts during puncture or bronchoscopy.

[0024] The system utilizes an IMU in conjunction with external optical / ToF markers for monitoring. When the overall drift exceeds 1–2 mm or 0.5°, it triggers rigid body / affine transformation correction. Unlike traditional static registration, this invention can maintain coordinate consistency continuously during long-term operation and prevent misjudgment of respiratory motion by temporarily increasing uncertainty, thereby enhancing the safety of clinical operations.

[0025] This invention employs an IMM (Integrated Microwave Mechanism) structure in its fusion filtering module, automatically switching filtering parameters under different states such as steady breathing, sighing, coughing, and swallowing. Compared with the single filtering algorithm in existing technologies, this scheme can effectively avoid abnormal fluctuations or distortions in output under abnormal events, ensuring the continuity and reliability of the results.

[0026] The system outputs the nodule location along with a 95% confidence ellipsoid and automatically sends a stop signal to external devices when the uncertainty exceeds the threshold. This closed-loop safety design is superior to existing methods that rely on manual judgment and can significantly reduce the treatment risks caused by inaccurate positioning.

[0027] In summary, by organically combining hysteresis modeling, local perturbation compensation, self-healing registration, and multi-model fusion filtering, this invention not only overcomes the shortcomings of existing technologies in handling respiratory asymmetry, interventional perturbations, and body position drift, but also establishes an event-driven dynamic adaptive mechanism. Therefore, in actual clinical environments, this invention can maintain a positioning accuracy of 2–3 mm even under complex conditions, significantly improving the system's reliability and clinical safety. Attached Figure Description

[0028] Figure 1 This is a schematic diagram of the method flow of the present invention; Figure 2 This is a schematic diagram of the positioning process of the present invention; Figure 3 This is a schematic diagram of the system architecture of the present invention. Detailed Implementation

[0029] This invention provides a real-time lung nodule localization system based on patient respiratory fluctuation modeling, which can achieve closed-loop precise control of the arterial infusion process while ensuring safety. The invention is further described below with reference to the accompanying drawings and embodiments, but this does not limit the scope of protection of the invention.

[0030] like Figures 1 to 3 As shown, this invention proposes a real-time lung nodule localization system and method based on patient respiratory fluctuation modeling, which is used to solve the nodule localization error problem caused by respiratory motion, instrument disturbance and body position drift in clinical scenarios such as radiotherapy, percutaneous puncture or bronchoscopic navigation, and achieve millimeter-level real-time tracking of lung nodules.

[0031] The system mainly includes a multi-source sensing unit, a signal processing module, a respiratory dynamics modeling module, a regionalized sensitivity field module, a disturbance observer, a self-healing registration module, a fusion filtering module, and an output and interaction module. Under the control of the processor, each module works collaboratively to form a complete technical chain from data acquisition, feature extraction, modeling and prediction, disturbance compensation, registration correction to result output. Specifically: Multi-source sensing unit: In the real-time localization of pulmonary nodules, a single signal often cannot fully reflect the patient's true condition. Respiratory movements are not only manifested as the up-and-down movement of the chest wall and diaphragm, but also accompanied by airflow dynamics, gas exchange, body positional drift, and interference from external instrument manipulation. If relying on only one sensor, the system can easily confuse these factors, leading to localization errors. Therefore, this invention introduces a multi-source sensing design, acquiring information from multiple dimensions through different types of sensors to form a complementary relationship.

[0032] First, the surface sensors on the chest and abdomen directly monitor the movement of the chest and abdominal walls. During respiration, the expansion of the thoracic cavity and the rise and fall of the abdomen are most pronounced; flexible resistance strain gauges or fiber optic micro-bending sensors placed at these locations can accurately capture these deformations. By deploying 8 to 12 sensing points in different regions of the chest and abdomen, the system can not only obtain the overall respiratory amplitude but also distinguish different movement patterns of the left and right chest, upper chest, and lower abdomen. This is crucial for determining differences in the activity of the left and right diaphragms and identifying abnormal breathing patterns.

[0033] Secondly, the prelarynx sensor is used to capture airway flow and laryngeal vibrations. As air passes through the larynx and trachea, it generates unique acoustic signals and subtle mechanical vibrations. These signals are relatively regular during normal breathing, but change significantly during coughing, swallowing, or sighing. By attaching a miniature microphone and a triaxial accelerometer to the prelarynx, the system can acquire both acoustic spectral characteristics and sense localized mechanical vibrations, thus providing additional evidence for event detection and respiratory phase determination.

[0034] Third, the end-tidal carbon dioxide sensor and flow meter are responsible for reflecting gas exchange and respiratory dynamics. The flow meter directly measures the incoming and outgoing airflow and can calculate tidal volume and respiratory cycle; the carbon dioxide sensor measures the end-expiratory gas concentration and waveform characteristics, which can reveal alveolar ventilation efficiency and gas distribution uniformity. These data are particularly crucial for determining respiratory depth and identifying sighs or alveolar recruitment, as these events often manifest as rapid changes in tidal volume and abrupt changes in carbon dioxide waveform.

[0035] In addition, inertial measurement units (IMUs) are primarily used to monitor overall body positional drift. In clinical settings, patients may experience displacements ranging from millimeters to centimeters due to bed vibrations, positional adjustments, or involuntary movements. IMUs can capture acceleration and angular velocity in real time, and calculate drift vectors and angular changes through attitude analysis. This type of information does not directly reflect respiratory motion, but it is crucial for maintaining coordinate system consistency and avoiding long-term cumulative errors.

[0036] Finally, in interventional procedures, the movement of the instrument itself directly affects nodule localization. When the puncture needle or bronchoscope enters the lung tissue, local compliance decreases, altering the movement patterns of the lung segment. Without monitoring of the instrument's status, the system may mistakenly interpret this local change as displacement caused by respiration. This invention, by installing an IMU or depth encoder on the instrument, acquires real-time information on the instrument's insertion depth, displacement, or angle. In this way, the system can determine whether the instrument has entered the target area and promptly trigger disturbance compensation.

[0037] In summary, the respiratory process is influenced by a combination of multiple factors, and a single signal is insufficient to distinguish motion from different sources. By introducing chest and abdominal sensors, pre-larynx sensors, gas exchange monitoring, body position drift monitoring, and device status monitoring, the system can simultaneously acquire information from five dimensions: body surface motion, airway dynamics, gas exchange, whole-body drift, and external intervention. After comprehensive processing, these signals complement each other, improving both the model's accuracy and its ability to identify abnormalities, providing a solid data foundation for real-time localization of pulmonary nodules.

[0038] Signal processing module: In the system of this invention, the signal processing module transforms the raw multi-source sensor signals into concise, low-dimensional respiratory state parameters suitable for modeling. Directly using the raw data often results in problems such as high noise, asynchrony, and excessively high dimensionality, which are detrimental to subsequent modeling and prone to misjudgment. Therefore, this module needs to perform filtering, denoising, and time synchronization, and then extract key features based on these processes.

[0039] First, regarding the surface signals of the chest and abdomen, since the movement frequency of the human chest and abdominal walls is highly correlated with respiration, other body movements, heartbeats, or environmental interference may be mixed in. The system uses bandpass filtering to retain only the 0.05–1.2 Hz component, which precisely covers most of the human respiratory frequency range from quiet breathing to deep breathing. The filtered signal may still contain motion components in multiple directions, so principal component analysis is needed to decompose these complex waveforms into several main modes, such as the movement of the upper chest, lower chest, left diaphragm, and right diaphragm. The purpose of this is to allow the system to distinguish the contribution of different parts to the nodule movement, rather than simply treating the chest and abdomen as a uniform indicator.

[0040] Secondly, regarding prelarynx signals, respiration produces continuous airflow sounds and slight vibrations in the larynx, while specific events such as coughing, swallowing, and sighing alter the frequency distribution and energy level of the sound. To extract valuable information, the system employs a short-time Fourier transform to convert the original time-domain acoustic signal into a time-spectrum. On this spectrum, different events correspond to different energy peaks and frequency band characteristics. For example, a sigh causes a sudden increase in low-frequency energy, a cough causes a sudden increase in mid-to-high-frequency energy, and swallowing results in transient, irregular frequency band energy. By extracting these frequency band energy characteristics, the system can promptly detect abnormal events, avoiding misinterpreting them as normal respiratory movements.

[0041] Furthermore, the end-tidal carbon dioxide signal directly reflects the efficiency of alveolar gas exchange and respiratory dynamics. The system focuses on extracting several morphological features: the α and β angles represent gas mixing during expiration, the terminal slope reflects the uniformity of ventilation at end-expiration, and the end-expiratory carbon dioxide value is a commonly used clinical ventilation indicator. Combining these features with tidal volume and inspiratory-to-expiratory ratio (IPR) allows for a low-dimensional description of lung function. For example, when a patient sighs, the end-expiratory carbon dioxide concentration decreases, and the slope becomes gentler, suggesting a sudden change in alveolar ventilation.

[0042] Finally, the data from the inertial measurement unit (IMU) is primarily used to detect changes in body position or overall drift. Since the IMU itself is prone to accumulating errors, the system uses attitude calculation to convert the raw acceleration, gyroscope, and magnetometer signals into drift vectors and angular offsets. This information is not directly used to predict nodule location but rather provides a "global correction basis" for subsequent registration modules.

[0043] Through the above processing steps, the originally high-dimensional and noisy raw signal is transformed into a small number of stable and representative parameters, such as tidal volume, diaphragmatic height, respiratory phase, acoustic spectral energy, final CO2 value, and drift vector. These parameters significantly reduce computational complexity, enabling the model to run in real time; they also provide the system with reliable evidence to distinguish respiratory movements, postural drift, and abnormal events.

[0044] Respiratory dynamics modeling module: In actual breathing, the displacement trajectory of the nodules during inspiration and expiration is not a straight line, but rather resembles a clockwise or counterclockwise loop. This is because the lungs and chest wall are viscoelastic; many alveoli partially close during shallow breathing and only open during deeper inspiration. Once opened, they do not immediately return along the same path during expiration. This phenomenon of "going out one path and coming back another" is called hysteresis. The respiratory dynamics modeling module of this invention first compresses multi-source signals into several key "respiratory driving forces" (such as tidal volume, the height of the left and right diaphragms, abdominal phase, etc.) during a short baseline phase after the patient is put on the machine. Then, it uses the correspondence between these driving forces and nodule displacement to fit a patient-specific "respiratory loop." This loop is not just a graphic; it contains several readable shape parameters, such as the curvature of the inspiratory branch, the curvature of the expiratory branch, the area between the two branches, and the phase difference between inspiration and expiration. After that, the system used the shape of this "ring" as the basis for daily predictions: for each breathing sample, it first determined whether it was inhaling or exhaling, and then predicted the location of the nodule based on which segment of the ring it was on.

[0045] Clinically, this isn't always "routine." Patients may occasionally sigh, or alveolar recruitment may occur in certain positions or depths of anesthesia: a previously "closed" portion of the alveoli suddenly opens, and in the following breaths, the compliance of the entire lung changes. This manifests at the signal level as follows: first, a significant increase in tidal volume over one or two cycles (reaching one and a half times or even more than baseline); then, changes in the slope and value of the final segment of the end-expiratory carbon dioxide waveform; and pre-larynx sensors may also detect a low-frequency energy rise or brief crackle during a "deep inhalation." When these three types of evidence occur simultaneously, the system no longer treats it as a "normal fluctuation" but rather identifies it as a sigh / recruitment event.

[0046] Once an event is detected, the model will not continue to force the same "old loop" but will instead enter the "switching" process in the mode manager. The first step of the switch is to temporarily increase the output uncertainty, displaying a "parameter re-evaluation in progress" message on the interface. Simultaneously, external gating becomes more conservative to prevent radiotherapy or puncture equipment from relying on an unstable prediction during the shape-changing transition period. The second step involves loading a set of initial values ​​reserved for "higher lung compliance"—which can be understood as a preliminary new loop shape after the old loop has been moderately "enlarged, shifted outward, and slightly rearranged." These initial values ​​are derived from statistical ranges of healthy subjects and historical cases to ensure they are not wildly inaccurate. The third step involves the model increasing its "flexibility": in the following tens of seconds, the prediction is allowed to be more sensitive to new data, facilitating a faster approximation of the actual loop shape.

[0047] Switching doesn't mean immediately locking in the new parameters. To ensure both speed and stability, the system initiates a short-window reassessment process. Starting from the "event moment," it captures approximately ten to twenty respiratory cycles, storing the key points of each cycle (inspiratory onset, peak inspiration, end of expiration, etc.) and the corresponding nodal position residuals in a sliding window. New samples in the window have higher weights, while old samples have lower weights; if coughing or swallowing occurs during this period, these cycles are automatically removed to avoid contaminating the parameters. During this time, the system focuses on adjusting three types of things: the overall size and offset of the loop (corresponding to changes in overall compliance), the relative shape of the inspiratory and expiratory branches (corresponding to inspiratory-expiratory asymmetry), and the phase difference between inspiration and expiration (corresponding to the rhythmic coordination between the diaphragm and chest wall). After each update, the latest few cycles are used to verify whether the average deviation between the loop and the actual data continues to decrease and whether the difference between two adjacent updates has become smaller. If the deviation remains within millimeters for several consecutive breaths and the update magnitude is very small, the system considers the convergence complete and solidifies the new parameters as the patient's new baseline; the uncertainty on the interface returns to a normal level, and the external gating strategy is restored to its original state.

[0048] During the switchover, the system simultaneously maintains a shadow trajectory of the "old loop." If the new loop not only fails to reduce the residuals but actually increases them, or if tidal volume, carbon dioxide, and prelarynx characteristics quickly revert to their pre-event state, the pattern manager automatically rolls back: undoing the previous parameter changes, reactivating the old loop, and the entire event is labeled as a brief "deep breath rather than recruitment." This ensures sensitivity to genuine structural changes while avoiding mistaking short bursts of noise for physiological remodeling.

[0049] In terms of engineering implementation, this module is completed collaboratively by three resident sub-units: the first is the "loop tracker," which is responsible for continuously updating the current inspiratory / exhalation phase, the position on the loop, and the distance to the loop; the second is the "event discriminator," which performs time alignment and cross-validation of three pieces of evidence—tidal volume, carbon dioxide form, and prelarynx spectrum—and only issues a "sigh / recruitment" signal if these conditions are consistently met within one to two cycles; the third is the "short-window recalculator," which maintains a circular buffer for one respiratory cycle, automatically filters out abnormal cycles, and uses a recursive weight update to smoothly push the parameters to a stable point on a timescale of about half a minute. These three components together ensure that the system is stable under normal conditions, flexible when events occur, and quickly returns to stability after an event.

[0050] The ultimate result is that the system no longer treats all breaths as "the same route back and forth," but instead acknowledges and characterizes the difference between the "outbound" and "outbound" paths. Once a patient's overall breathing loop changes, the system can learn the "new loop" within ten to twenty breaths. Clinically, this means that when sighing or recruitment occur—situations that should exist but are often overlooked—nodule localization will not experience long-term shifts, and the safety boundaries of gating and navigation will always be well-defined.

[0051] Regionalized sensitivity field module: During respiration, lung tissue does not move uniformly as a whole. Different lung segments respond differently to the same respiratory drive: the region near the diaphragm changes the most with tidal volume and diaphragmatic amplitude, while the region near the lung apex is relatively stable; the left and right lungs also differ in compliance and airflow distribution. If the entire lung is treated as a whole during modeling, the system must process dozens or even hundreds of parameters simultaneously, which not only involves a large computational load but also introduces many disturbances unrelated to the target nodule, often resulting in slow and unstable results.

[0052] To address this issue, this invention proposes a "regionalized sensitivity field" design. A sensitivity field is essentially a "mapping table" that describes the correspondence between low-dimensional respiratory state parameters (such as tidal volume, diaphragmatic dome height, and abdominal phase) and lung tissue displacement. Different parameters contribute different weights to different lung tissue regions: tidal volume affects overall expansion, diaphragmatic dome height affects basal region movement, and abdominal phase primarily affects the lower lobes involved in abdominal breathing.

[0053] In practice, the system first uses preoperative CT images to automatically segment the lungs, dividing the entire lung into lobes and smaller segments. Next, the doctor or algorithm identifies the segment in which the nodule is located. This way, the system only needs to establish "primary coupling parameters" within that segment, rather than performing full-coverage modeling of the entire lung. For example, when the nodule is located in the right lower lobe, the sensitivity field primarily focuses on the influence of right diaphragmatic height and tidal volume on that area, while giving minimal weight to left diaphragmatic movement or upper chest undulations.

[0054] This approach has three advantages: Reduced dimensionality and computational burden: The system only needs to maintain the coupling relationship between the target lung segment and a few key parameters, without having to deal with the global complexity of the entire lung.

[0055] Improved prediction accuracy: Avoids motion signals from irrelevant regions being "incorrectly" projected onto nodule location predictions, thus making the model more focused and stable.

[0056] Enhanced robustness: Even if the patient has uneven chest wall movement or airflow distribution during breathing, the system can automatically ignore fluctuations that are irrelevant to the target lung segment.

[0057] During operation, the sensitivity field module continuously receives low-dimensional respiratory parameters generated by the signal processing module. Combining this with preoperative segmentation information, it maps these parameters to predicted displacements of the target lung segment according to predefined weights. For example, a 20% increase in tidal volume corresponds to a downward displacement of 3 mm and an outward displacement of 2 mm for the corresponding lung segment. If the right diaphragm height changes more significantly than the left diaphragm, weights are assigned to the right diaphragm parameters, resulting in a prediction biased towards the lower right. These displacement predictions are then combined with the hysteresis path output by the respiratory dynamics modeling module to form a preliminary estimate of the nodule location.

[0058] When interventional devices enter the airway or puncture path, the compliance of the local lung segment changes, which can lead to positioning errors if not compensated for. To address this, this invention incorporates a perturbation observer that estimates the degree of decrease in local compliance in real time by combining the device's depth signal or IMU signal, changes in end-expiratory carbon dioxide waveform, and high-frequency friction rub characteristics from the prelarynx. The perturbation observer employs a recursive least squares algorithm, achieving parameter convergence within 3 to 5 respiratory cycles and feeding the correction results back to the sensitivity field module.

[0059] Self-healing registration module: In clinical settings, patients are not completely still. Even when fixed to a treatment bed or wearing a mask, some overall drift may still occur over time: for example, the bed may vibrate slightly when the equipment is running, the patient's body may slowly slide down due to gravity, or the head may turn slightly after maintaining a position for a long time. Unlike respiratory movements, these drifts do not occur periodically but accumulate slowly at the millimeter or angular level. If the system relies solely on respiratory models and local sensitivity mapping without drift correction, the predicted location of the nodule will gradually "deviate," eventually creating a stable discrepancy with the actual location. If this discrepancy is not controlled, it can pose significant risks during puncture or radiotherapy.

[0060] To avoid such errors, this invention incorporates a self-healing registration module. Its first step is drift detection. An IMU (Intraluminal Unit) is deployed on the patient's chest, face mask, or treatment bed to monitor overall acceleration and angular velocity in real time. While IMUs excel at capturing sudden, small displacements, they are prone to accumulating biases over extended periods. Therefore, the system also utilizes external optical markers or time-of-flight (ToF) sensors to provide independent references, such as attaching reflective dots to the chest wall or bed, or using a ToF camera to track the marker's position. By fusing these two types of information, the system can reliably determine whether the patient has experienced drift beyond the normal breathing range.

[0061] When the detected drift exceeds a preset threshold (translation greater than 1–2 mm, or rotation greater than 0.5 degrees), the system triggers registration correction. Instead of fine-tuning each element of the internal model, it directly applies a rigid body or affine transformation at the coordinate level. This can be understood as "realigning" the entire predicted coordinate system to ensure it remains consistent with the patient's true anatomical coordinates. This correction is often a one-time action, not dependent on continuous iteration, and therefore can be completed in a very short time.

[0062] However, drift correction is not entirely harmless. Because drift and respiratory motion may partially overlap in signal, treating drift correction as an absolutely accurate true value could potentially eliminate genuine respiratory fluctuations. Therefore, after registration, the system temporarily increases the uncertainty of the output position. This is equivalent to alerting the doctor and equipment that the localization result has been corrected, but its reliability is slightly reduced for a short period due to the recent global adjustment. During this time, the system will still output the nodule location, but its gating strategy will be more conservative to ensure that radiotherapy or puncture procedures are not misled. As data accumulates over several subsequent respiratory cycles, the filtering module will converge again, and the uncertainty will gradually decrease back to normal levels.

[0063] Through this mechanism, the self-healing registration module distinguishes between "drift" and "respiratory motion": once the drift exceeds the threshold, it is quickly pulled back using an external reference point; after being pulled back, the uncertainty is temporarily increased to avoid misjudgment; finally, the system re-stabilizes on the new reference coordinates. The reason for this approach is that slow drift is unavoidable in clinical settings; the solution is to use IMU and external markers for joint detection, and to use rigid body / affine transformation for overall correction after exceeding the threshold; the final effect is that even if the operation lasts for tens of minutes, the system can maintain the same coordinate accuracy as the preoperative images, while avoiding model collapse caused by mistaking drift for breathing.

[0064] Fusion filtering module: In the entire system, each front-end module performs its specific function: the respiratory dynamics model provides predicted trajectories based on the respiratory loop; the regionalized sensitivity field provides corrections based on parameter mapping of local lung segments; the perturbation observer identifies and compensates for the decrease in local compliance caused by device entry; and the self-healing registration module is responsible for aligning the overall coordinate system. However, these outputs are not entirely consistent: some emphasize long-term trends (such as the hysteresis model), some reflect short-term perturbations (such as device compensation), and some make abrupt adjustments to the coordinates (such as drift correction). If any one of these is directly used as the final result, bias or instability can easily occur.

[0065] This is precisely the reason for the existence of the fusion filtering module. Its goal is to organically fuse results from different sources, preserving the value of each module while offsetting their limitations. To this end, the system employs two types of filtering strategies: unscented Kalman filtering (UKF) and particle filtering. UKF is suitable for handling continuous, smooth dynamic changes, providing nuanced trajectory predictions within the respiratory cycle; particle filtering is better suited for maintaining robustness in nonlinear, abrupt scenarios, such as sudden instrument insertion or a patient's sudden cough. The system selects one of these as the core algorithm based on the specific circumstances.

[0066] The first step in the fusion process is to map the outputs of all modules into a unified state space. For example, the hysteresis model outputs predicted displacement based on tidal volume and diaphragmatic motion, the sensitivity field outputs displacement corrections weighted by lung segments, the perturbation observer provides amplitude adjustments due to compliance changes, and the self-healing criterion provides an overall transformation of the coordinate system. The fusion filtering module synchronizes these results and uses them for different observations of the "true location of the nodule." Then, it dynamically weighs the reliability of each observation through a filtering algorithm: when the respiratory rhythm is stable, the weights of the hysteresis model and sensitivity field are greater; when the device entry signal is strong, the correction ratio of the perturbation observer is increased; when drift exceeding the threshold is detected, the correction from the registration module is prioritized. In this way, the final output location is a compromise of multi-source information, rather than blindly relying on any one source.

[0067] The second key point is event-driven parameter switching. Breathing doesn't always occur in a stable, shallow, and rapid pattern; sighing, alveolar recruitment, coughing, and swallowing can all disrupt the rhythm for short periods. If the filter continues to operate with the original parameters, it can easily "fall behind," resulting in delays or oscillations. To address this issue, the system employs an interactive multi-model (IMM) architecture. Simply put, multiple filtering models are prepared in advance: one for steady breathing, one for the sudden increase in tidal volume after a sigh, one for short, large-amplitude disturbances like coughing, and one for transient, ineffective signals like swallowing. Once the event detection module identifies an anomaly, it notifies the IMM to switch to the appropriate sub-model; after the event ends, it gradually returns to the normal model. The entire switching process is continuous, preventing sudden jumps in output position, and instead smoothly transitions by increasing uncertainty.

[0068] Under this mechanism, the fusion filtering module can not only provide smooth millimeter-level positioning under normal breathing, but also maintain robustness in the event of special events. For example, when a patient suddenly coughs, the filter will reduce the reliability of the position signal and amplify the uncertainty to prevent the radiotherapy equipment from being driven incorrectly; after the patient takes a deep breath, the filter will automatically adjust its parameters so that the new breathing trajectory converges again within a dozen cycles.

[0069] Ultimately, the output of the fusion filtering module is not only a set of three-dimensional coordinate points, but also includes a 95% confidence ellipsoid, clearly indicating the prediction range. This allows clinicians or external devices to intuitively see the system's confidence in its results and receive timely safety alerts when uncertainty increases.

[0070] The output and interaction module displays the nodule location and a 95% confidence ellipsoid on the 3D lung model, while also indicating the event status and quality control score. When the uncertainty exceeds a threshold, the system automatically outputs a gating stop signal, which is linked to the radiotherapy accelerator or puncture navigation system to avoid medical risks caused by inaccurate positioning.

[0071] Real-time localization method for lung nodules: In this invention, the real-time localization method for pulmonary nodules first requires acquiring multi-source signals that reflect the patient's respiratory status and positional changes. Therefore, in step S1, simultaneous data acquisition is performed using a chest and abdominal surface sensor, a pre-larynx acoustic and vibration sensor, an end-tidal carbon dioxide sensor, a flow meter, and an inertial measurement unit. In interventional procedures, instrument status signals from the puncture needle or bronchoscope are also acquired simultaneously. These signals reflect chest wall movement, airflow dynamics, gas exchange, positional drift, and external operational disturbances during respiration from different perspectives, forming the basis for subsequent modeling and correction. Relying solely on a single signal often makes it difficult to distinguish between displacement caused by respiration and instrument disturbances or posture changes; therefore, simultaneous multi-source acquisition is a prerequisite for stable localization.

[0072] After data acquisition, feature extraction is performed on the signals in step S2. The system filters and performs principal component analysis on the chest and abdominal signals to obtain low-dimensional parameters representing the movement of the left and right diaphragmatic vaults; it performs time-frequency analysis on the prelarynx signals to extract energy in specific frequency bands for identifying coughing, swallowing, or sighing events; it extracts end-expiratory concentration, slope, and angle from the end-tidal carbon dioxide waveform to estimate alveolar ventilation status; it calculates parameters such as tidal volume and inspiratory-expiratory time ratio from the flow signal; and the IMU signal provides body posture and drift information. Through these processes, the originally complex and high-dimensional raw sensor data is transformed into low-dimensional state parameters such as tidal volume, diaphragmatic height, respiratory phase, and acoustic spectral energy, enabling the model to characterize respiratory dynamics with lower computational load.

[0073] In step S3, the system predicts the initial location of the nodule based on a hysteresis dynamics model and a regionalized sensitivity field. The hysteresis model reflects the asymmetry of tissue displacement during inspiration and expiration; for example, under the same tidal volume, the inspiratory and expiratory pathways will exhibit different positional relationships. The regionalized sensitivity field, derived from preoperative imaging data, determines the degree of response of the lung segment containing the nodule to different respiratory parameters. Combining these two methods, the theoretical displacement location of the nodule can be output under a given respiratory state, forming an initial prediction.

[0074] However, in clinical practice, the entry of instruments into the lungs can lead to a decrease in the compliance of the local lung segment. If this factor is ignored, the predicted location of the nodule will be systematically biased. Therefore, in step S4, when instrument entry is detected, the system activates the perturbation compensation module. This module uses information such as the instrument depth signal, changes in the end-tidal carbon dioxide waveform, and high-frequency friction rub in the prelarynx to estimate the degree of decreased compliance in the lung segment where the nodule is located, and corrects the sensitivity field in real time, so that the model can continue to accurately reflect actual tissue movement.

[0075] Meanwhile, the patient's overall position may also experience slight drift during the procedure, such as bed vibration, patient slippage, or slight head rotation. Without correction, the nodule's location will gradually shift relative to the preoperative coordinate system. Therefore, a self-healing registration mechanism is introduced in step S5. When the IMU detects a drift exceeding a preset threshold (e.g., 1 to 2 mm or 0.5 degrees of rotation), the system automatically performs a rigid body or affine transformation to correct the nodule's position and temporarily increases the output uncertainty to alert the operator to potential errors. This measure ensures coordinate consistency throughout long-term procedures.

[0076] In step S6, all the aforementioned results are integrated through a fusion filtering module. This module employs a switched unscented Kalman filter or particle filter, taking the predictions from the hysteresis model, the mapping of the sensitivity field, the corrections from the perturbation observer, and the corrections from the self-healing registration as inputs, and outputting the nodule location updated in real time. Since special events such as sighing, recruitment, coughing, or swallowing may occur during respiration, the system adopts an interactive multi-model structure. When the event detection module is triggered, the filtering parameters automatically switch to ensure that prediction accuracy is maintained even under atypical respiratory conditions. The final output includes not only the three-dimensional coordinates of the nodule but also the corresponding 95% confidence ellipsoid, used to quantify uncertainty.

[0077] Finally, in step S7, the results are sent to the output and interaction module. The nodule location and uncertainty are presented through a 3D visualization interface, with event status and signal quality scores displayed simultaneously, facilitating real-time assessment of system reliability by the operator. If the system confidence level reaches a preset standard, a gating enable signal is output, linking with the radiotherapy accelerator, ablation device, or puncture navigation system; if the confidence level decreases or drifts excessively, a stop signal is automatically output to prevent unsafe operations. This closed loop ensures the safety and practicality of clinical applications.

[0078] In summary, S1 to S7 form a logically progressive chain: first, multi-source signals are acquired, then simplified features are extracted; preliminary predictions are made based on hysteresis models and sensitivity fields; corresponding compensation and registration are performed when encountering local instrument disturbances and overall drift; finally, all information is fused through multi-model filtering to output a stable and reliable real-time nodule position. This layered protection and step-by-step correction approach ensures millimeter-level positioning accuracy even in complex clinical environments.

[0079] In an optional implementation, whenever a mode-switching event (such as a sigh or alveolar recruitment) is detected, the system executes a short-window expectancy-maximization algorithm for 10 to 20 respiratory cycles to re-estimate the sensitivity field parameters to restore prediction accuracy. Upon detection of swallowing or coughing, the system temporarily freezes the position output or increases the uncertainty to prevent clinical misinterpretation.

[0080] Through the above technical solution, the present invention can achieve real-time localization of pulmonary nodules under actual clinical conditions. Even under chest wall drift, instrument disturbance, and respiratory mode switching, it can still maintain a localization accuracy of 2 to 3 millimeters and complete parameter convergence within 10 to 20 respiratory cycles after mode switching, which significantly improves the safety and effectiveness of clinical operation.

[0081] The system and method of this invention can be used in clinical practice simultaneously with existing methods, ensuring patient safety and allowing for comparison. Representative cases are summarized in the table below:

[0082] Existing methods lack adaptability to sudden events such as sighing, coughing, and swallowing. This invention can recover within 10–20 respiratory cycles through mode switching and short-window reestimation, significantly shortening the error convergence time.

[0083] In interventional scenarios such as puncture / endoscopy, traditional single-signal methods cannot distinguish between respiratory displacement and instrument disturbance. The disturbance observer of this invention can correct compliance parameters within 3–5 cycles, effectively avoiding systematic bias.

[0084] Existing systems based on preoperative CT will accumulate drift during long-term operation. This invention uses a self-healing registration module to achieve millimeter-level real-time correction and ensure the consistency of the preoperative coordinate system.

[0085] Traditional breathing band or surface marking methods are not robust to large fluctuations in tidal volume or irregular breathing. This invention maintains high-precision prediction through hysteresis loop dynamics and regionalized sensitivity fields.

[0086] In five typical clinical scenarios, the present invention is significantly superior to the existing system, with improved positioning accuracy by about 2–4 mm, reduced event recovery time by about 50%, and significantly enhanced overall robustness and safety.

[0087] In summary, this invention can effectively solve the positioning errors caused by respiratory motion, instrument disturbance and body position drift in actual clinical applications, and achieve stable and reliable millimeter-level real-time lung nodule positioning, which is superior to existing systems.

[0088] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A lung nodule real-time positioning system based on patient respiratory fluctuation modeling, characterized in that, The system comprises: a multi-source sensing unit for collecting patient chest and abdominal surface motion signals, pre-laryngeal acoustic and vibration signals, end-tidal carbon dioxide and respiratory flow rate signals, and body position and instrument status information; a signal processing module for denoising, feature extraction, and time synchronization of the sensing signals, and generating low-dimensional state parameters related to respiration; a respiratory dynamics modeling module including a state-space model with a hysteresis operator for characterizing the inconsistent inhalation and exhalation paths and the mode switching caused by sighing / recruitment; a regionalized sensitivity field module for constructing the mapping relationship between the lung segment where the nodule is located and the respiratory state parameters based on preoperative imaging information; a perturbation observer for real-time estimation and correction of the local compliance parameters of the lung segment when the interventional instrument enters the local airway; a self-healing registration module for rigid or affine correction of the overall coordinate system drift using the inertial measurement unit and external markers on the bed, mask, or chest; a fusion filtering module for fusing the outputs of the above modules using switched unscented Kalman filtering or particle filtering to obtain the real-time three-dimensional position and uncertainty of the lung nodule; an output and interaction module for displaying nodule position, event status, and quality control information, or providing a gating signal to external treatment / navigation equipment.

2. The real-time lung nodule positioning system based on patient respiration fluctuation modeling of claim 1, wherein: The respiratory dynamics modeling module uses a Bouc-Wen type hysteresis operator to simulate the displacement difference between the inhalation and exhalation paths under the same tidal volume.

3. The real-time lung nodule positioning system based on patient respiration fluctuation modeling of claim 1, wherein: The regionalized sensitivity field module divides the lung lobe or segment into multiple sub-regions, and only establishes the main coupling parameters for the sub-region where the nodule is located, thereby reducing the computational dimension and improving the positioning robustness.

4. The real-time lung nodule positioning system based on patient respiration fluctuation modeling of claim 1, wherein: The perturbation observer estimates the degree of local compliance reduction of the lung segment by combining instrument insertion mechanics information, end-tidal carbon dioxide waveform changes, and pre-laryngeal high-frequency friction sound characteristics, and completes parameter convergence within 3 to 5 breathing cycles.

5. The real-time lung nodule positioning system based on patient respiration fluctuation modeling of claim 1, wherein: The self-healing registration module automatically corrects the nodule position in a rigid body when the overall drift exceeds 1 to 2 millimeters or the rotation exceeds 0.5 degrees, and temporarily increases the uncertainty to ensure the safety of the positioning result.

6. The real-time lung nodule positioning system based on patient respiration fluctuation modeling of claim 1, wherein: The fusion filtering module uses an interactive multiple model (IMM) structure to switch filtering parameters when sighing, recruitment, coughing, or swallowing events are detected.

7. The real-time lung nodule positioning system based on modeling of patient respiratory fluctuations of claim 1, wherein: The output and interaction module provides nodule position with a 95% confidence ellipsoid, and outputs a gating stop signal when the uncertainty exceeds the threshold to link with radiotherapy or puncture equipment.

8. A lung nodule real-time positioning method based on the lung nodule real-time positioning system of any one of claims 1 to 7 based on modeling of patient respiratory fluctuation, characterized in that, The method comprises the following steps: S1. Collecting multi-source signals such as chest and abdominal surface, pre-laryngeal, CO2 / flow rate, and IMU; S2. Feature extraction and generation of low-dimensional state parameters; S3. Prediction of nodule position based on hysteresis dynamics model and regionalized sensitivity field; S4. Correction of local compliance parameters when instrument entry is detected; S5. Self-healing registration when overall drift is detected; S6. Fusing the outputs of each module using switched filtering method to obtain the three-dimensional position and uncertainty of the nodule; S7. Outputting the results for display and linking with clinical equipment.

9. The lung nodule real-time localization method of claim 8, wherein: The method performs a short window expectation maximization (EM) algorithm of 10 to 20 breathing cycles after each mode switching event to re-estimate the sensitivity field parameters.

10. The lung nodule real-time localization method of claim 8, wherein: The method temporarily freezes the position output or increases the uncertainty upon detection of a swallow or cough event to avoid misleading the clinical procedure.

Images