A multiple model approach to respiratory motion prediction for real-time IGRT

The interacting multiple model (IMM) algorithm is a widely accepted state estimation scheme for solving maneuvering target tracking problems, which are generally nonlinear. During the IMM filtering process, serious errors can arise when a Gaussian mixture of posterior probability density functions is approximated by a single Gaussian. Particle filters (PFs) are effective in dealing with nonlinearity and non-Gaussianity. This work considers an IMM algorithm that includes a constant velocity model, a constant acceleration model and a 3D turning rate (3DTR) model for tracking three-dimensional (3D) target motion, using various combinations of nonlinear filters. In existing literature on combining IMM and particle filtering techniques to tackle difficult target maneuvers, a PF is usually used in every model In comparison, simulation results show that by using a computationally economical PF in the 3DTR model and Kalman filters in the remaining models, superior performance can be achieved with significant reduction in computational costs.

Adaptive Cruise Control (ACC) is an important part of automotive autonomy. To have a robust Adaptive Cruise Control system, a well-defined sensor fusion system is required. This paper focuses on feature-level sensor fusion system where three fusion techniques are explored: independent Kalman filtering, interacting multiple model (IMM) filtering, and a novel hybrid technique using a combination of both an IMM Filter and deep neural network (DNN). Kalman Filtering is a popular approach for automotive sensor fusion, but heavily relies on the given kinematic models to make state estimations. We explore the IMM filter to capture multi-model motion as well as a DNN approach to build a data-driven model of the system. Our IMM filter is built with several Kalman Filters featuring different kinematic models. The proposed DNN used is a Long Short-Term Memory (LSTM) network trained on radar and camera data to forecast a vehicle's state. To overcome the weaknesses in both IMM filtering and LSTM networks, we propose a hybrid technique of consisting of both the IMM filter and the LSTM network. The proposed hybrid system uses a single filter for each sensor. The filtered sensor data is synchronized and used as the input for a trained LSTM network. Our LSTM network was trained on over 100 simulated highway driving scenarios that might occur in an ACC application. Simulations and code were developed using MATLAB and the Autonomous Driving Toolbox. Our hybrid IMM-LSTM system outperformed the independent Kalman Filtering approach in longitudinal and lateral tracking accuracy (RMSE) by 23 percent and had promising results compared to the IMM filter where the tracking error was decreased by nearly 50 percent in certain driving scenarios.

Background. In radiotherapy, the magnitude of respiratory-induced tumor motion is often measured using a single four-dimensional computed tomography (4DCT). This magnitude is required to determine the internal target volume. The aim of this study was to compare the magnitude of respiratory-induced motion of pancreatic tumors on a single 4DCT with the motion on daily cone beam CT (CBCT) scans during a 3–5-week fractionated radiotherapy scheme. In addition, we investigated changes in the respiratory motion during the treatment course.Material and methods. The mean peak-to-peak motion (i.e. magnitude of motion) of pancreatic tumors was measured for 18 patients using intratumoral gold fiducials visible on CBCT scans made prior to each treatment fraction (10–27 CBCTs per patient; 401 CBCTs in total). For each patient, these magnitudes were compared to the magnitude measured on 4DCT. Possible time trends were investigated by applying linear fits to the tumor motion determined from daily CBCTs as a function of treatment day.Results. We found a significant (p ≤ 0.01) difference between motion magnitude on 4DCT and on CBCT in superior-inferior, anterior-posterior and left-right direction, in 13, 9 and 12 of 18 patients, respectively. In the anterior- posterior and left-right direction no fractions had a difference ≥ 5 mm. In the superior-inferior direction the difference was ≥ 5 mm for 17% of the 401 fractions. In this direction, a significant (p ≤ 0.05) time trend in tumor motion was observed in 4 of 18 patients, but all trends were small (− 0.17–0.10 mm/day) and did not explain the large differences in motion magnitude between 4DCT and CBCT.Conclusion. A single measurement of the respiratory-induced motion magnitude of pancreatic tumors using 4DCT is often not representative for the magnitude during daily treatment over a 3–5-week radiotherapy scheme. For this patient group it may be beneficial to introduce breath-hold to eliminate respiratory-induced tumor motion.

This paper presents Track-LITE, a novel target-tracking framework that synergistically integrates an Interacting Multiple Model (IMM) filter with a Long Short-Term Memory (LSTM) network, governed by a dynamic on–off trigger mechanism. The IMM filter provides a reliable baseline for gradual motion, while the LSTM is selectively activated to refine estimates following abrupt maneuvers, a key contribution enabling adaptive accuracy. Quickest Change Detection (QCD) serves as the mechanism to detect these motion changes, collecting evidence of significant shifts and triggering the powerful LSTM to enhance prediction results. A bespoke feedback loop, another significant innovation, leverages LSTM residual errors to iteratively adjust IMM model weights, ensuring continuous adaptability across diverse motion profiles. The activation threshold, a critical derived design contribution, is optimized by jointly balancing false-alarm rate and latency, enabling precise coordination between the IMM and LSTM modules. With LSTM active only 30% of the time, Track-LITE achieves an RMSE comparable to continuous LSTM, while reducing computational load by 70%, a testament to its efficiency. This performance trade-off makes Track-LITE exceptionally well-suited for real-time, resource-constrained applications such as autonomous navigation and drone tracking.

Video synthetic aperture radar (VideoSAR) combines SAR imaging and video techniques together, enabling continuous surveillance of a specific area, which has the advantages of high frame rate, high resolution, the ability of real-time tracking, etc. For VideoSAR systems, detecting and tracking the shadows of moving targets can effectively avoid the complex algorithms needed for moving target location and refocusing. In this paper, an interactive multiple model (IMM) filter is designed for VideoSAR moving target tracking task. Considering of the maneuverability of ground moving target is more complex than aerial target, we define three different target motion models according to the motion characteristics of ground moving targets, including a stop model and a rotated model. The experimental results show that the proposed IMM filter outperforms conventional filter for ground moving target tracking application.

Accurate prediction of tumour motion - over a prescribed time - is essential for enabling adaptive radiotherapy. The prediction time horizon is determined by measurement processing time, predictor algorithm processing time and the time-to-adapt radiation delivery. A trade off between the predictor algorithm complexity and the required prediction time horizon, therefore, has to be made. This paper proposes an interacting multiple model (IMM) filter and two Kalman filters to predict 0.2 s ahead respiratory tumour motions. The performance of the filters is evaluated using 333 traces of 4 minutes respiratory motions for 24 adult patients. The average RMSE of the IMM filter and the best Kalman filter with 5 Hz measurements rate are 0.98 mm and 1.1 mm, which are improvements of 38% and 30% compared to use of measurements only. (4 pages)

<div class="htmlview paragraph">In a CAS system, the distance and relative velocity between front and host vehicles are estimated to calculate time-to-collision (TTC). The distance estimates by different methods will certainly include noise which should be removed to ensure the accuracy of TTC calculations. Kalman filter is a good tool to filter such type of noise. Nevertheless, Kalman filter is a model based filter, which means a correct model is important to get the good filtering results. Usually, a vehicle is either moving with a constant velocity (CV) or constant acceleration (CA) maneuvers. This means the distance data between front and host vehicles can be described by either constant velocity or constant acceleration model. In this paper, first, CV and CA models are used to design two Kalman filters and an interacting multiple model (IMM) is used to dynamically combine the outputs from two filters. In detail, IMM technique is used to estimate the mode probabilities for CV and CA based Kalman filters and mix the two filter results based on the mode probabilities. Then the motion equation is used to calculate the TTC based on the distance and velocity estimates obtained from the IMM based filter. The experimental results on both simulated and real estimated distance data are found to be satisfactory and indicate that the proposed algorithm does improve the signal-to-noise ratio of distance data.</div>

This paper presents a simple yet efficient adaptive Kalman filter for tracking targets expected to perform varying turn maneuvers. The process noise covariance level of a second order Kalman filter is adjusted at each time step according to the calculated turn rate. The turning rate is estimated from the magnitude of the calculated acceleration divided by the estimated speed of the target. At each scan the previous and current velocity estimates are used to calculate the acceleration. The comparison of the performance of the proposed algorithm is made with that of an interacting multiple model (IMM) algorithm, employing three models with different levels of process noise covariance and also to that of a second order Kalman filter. Two different assumptions have been made for selecting the process noise values for the the IMM and Kalman filter algorithms, in the first case it was assumed that there was no prior information about the target motion whereas in the second case it was assumed that the largest turn rate that the target of interest could perform was known. The IMM algorithm utilizing three models gives slightly better estimates during the nonmaneuvering periods, but the proposed algorithm is superior to the IMM algorithm during maneuvering periods in terms of estimation errors. Also the proposed algorithm requires 78% less computation, almost the same number of calculations required by a single fixed process noise Kalman filter, than the IMM algorithm.

This study proposes an airborne behaviour monitoring methodology of ground vehicles based on a statistical learning approach with domain knowledge given by road map information. To monitor and track the moving ground target using unmanned aerial vehicle aboard a moving target indicator, an interactive multiple model (IMM) filter is firstly applied. The IMM filter consists of an on‐road moving mode using a road‐constrained filter and an off‐road moving mode using a conventional filter. Mode probability is also calculated from the IMM filter, and it provides deviation of the vehicle from the road. Then, a novel hybrid algorithm for anomalous behaviour recognition is developed using a Gaussian process regression on velocity profile along the one‐dimensionalised position of the vehicle, as well as the deviation of the vehicle. To verify the feasibility and benefits of the proposed approach, a numerical simulation is performed using realistic car trajectory data in a city traffic.

The International Civil Aviation Organization (ICAO) has decided to adopt Communications, Navigation, and Surveillance/Air Traffic Management (CNS/ATM) as the 21st century standard for navigation. Accordingly, ICAO members have provided an impetus to develop related technology and build sufficient infrastructure. For aviation surveillance with CNS/ATM, Ground-Based Augmentation System (GBAS), Automatic Dependent Surveillance-Broadcast (ADS-B), multilateration (MLAT) and wide-area multilateration (WAM) systems are being established. These sensors can track aircraft positions more accurately than existing radar and can compensate for the blind spots in aircraft surveillance. In this paper, we applied a novel sensor fusion method with Interacting Multiple Model (IMM) filter to GBAS, ADS-B, MLAT, and WAM data in order to improve the reliability of the aircraft position. Results of performance analysis show that the position accuracy is improved by the proposed sensor fusion method with the IMM filter.

PurposeWe evaluated the effects of respiratory gating on treatment accuracy in lung cancer patients undergoing lung stereotactic body radiotherapy by using electronic portal imaging device (EPID) images.Materials and MethodsOur study population consisted of 30 lung cancer patients treated with stereotactic body radiotherapy (48 Gy/4 fractions/4 to 9 days). Of these, 14 were treated with- (group A) and 16 without gating (group B); typically the patients whose tumors showed three-dimensional respiratory motion ≧5 mm were selected for gating. Tumor respiratory motion was estimated using four-dimensional computed tomography images acquired during treatment simulation. Tumor position variability during all treatment sessions was assessed by measuring the standard deviation (SD) and range of tumor displacement on EPID images. The two groups were compared for tumor respiratory motion and position variability using the Mann-Whitney U test.ResultsThe median three-dimensional tumor motion during simulation was greater in group A than group B (9 mm, range 3–30 mm vs. 2 mm, range 0–4 mm; p<0.001). In groups A and B the median SD of the tumor position was 1.1 mm and 0.9 mm in the craniocaudal- (p = 0.24) and 0.7 mm and 0.6 mm in the mediolateral direction (p = 0.89), respectively. The median range of the tumor position was 4.0 mm and 3.0 mm in the craniocaudal- (p = 0.21) and 2.0 mm and 1.5 mm in the mediolateral direction (p = 0.20), respectively.ConclusionsAlthough patients treated with respiratory gating exhibited greater respiratory tumor motion during treatment simulation, tumor position variability in the EPID images was low and comparable to patients treated without gating. This demonstrates the benefit of respiratory gating.

This paper is concerned with modeling and prediction of lung tumor motion for a new adaptive tumor tracking system in radiotherapy. Information of the tumor motion is used to control a robotic treatment couch, which moves in six degrees of freedom and maintains the tumor in a certain fixed spatial position. With this new technology it is possible to avoid gating during the radiotherapy as well as active breathing control, such that duration and cost of the therapy as well as trauma of the patient can be reduced. The tumor position is detected by an electronic portal imaging device (EPID) such that internal markers for the tumor tracking are no longer necessary. An infrared (IR) camera is introduced to detect abdominal motion, which provides a reference to the respiratory signal and can be used to predict the tumor position over a complete breathing period. This paper presents the effort of tumor motion modeling and prediction. Main challenges with the proposed method in this context are: lack of breathing control causes irregular respiration, low data rate of the EPID device delivers too few information about the tumor position. Two methods, namely an adaptive filtering and a nonlinear method based on Takens Theorem, have been investigated and compared with respect to diverse criteria including prediction accuracy, model complexity and computational effort. Performance analysis with clinical data has shown that both methods can well describe correlation between the EPID and the IR signals, while the adaptive filter possesses advantage with respect of computational effort.

For multi-target tracking, the IMM (interactive multiple model) algorithm has been proposed. The IMM is expected to reduce tracking errors for both nonmaneuvering and maneuvering targets. However, the IMM requires a heavy computational burden, because it utilizes multiple Kalman filters in parallel. On the other hand, the Kalman filter with a turning acceleration estimator, which can adapt a maneuvering target, has been proposed. The Kalman filter with a turning acceleration estimator is a single two-stage type filter and has a problem of setting threshold for the maneuver detector. In this paper, we propose the hybrid filter with a constant-velocity (CV) filter and a turning acceleration estimation filter. The proposed filter does not require a maneuver detector, because it integrates the outputs of two filters using the likelihood of each filter. And its computational requirement is smaller than the IMM, since it consists of only two Kalman based filters. The proposed method can prevent deteriorating tracking accuracy by reducing the risk of maneuver misdetection when a target maneuvers. We evaluate the performance of the proposed filter by computer simulation, and show the effectiveness of the proposed filter, comparing with the conventional Kalman filter and the two-stage Kalman filter.

Space systems are characterized by a low-intensity process noise resulting from uncertain forces and moments. In many cases, their scalar measurement channels can be assumed to be independent, with one-dimensional internal dynamics. The nominal operation of these systems can be severely damaged by faults in the sensors. A natural method that can be used to yield fault tolerant estimates of such systems is the interacting multiple model (IMM) filtering algorithm, which is known to provide very accurate results. However, having been derived for a general class of systems with switching parameters, the IMM filter does not utilize the independence of the measurement errors in different channels, nor does it exploit the fact that the process noise is of low intensity. Thus, the implementation of the IMM in this case is computationally expensive. A new estimation technique is proposed herein, that explicitly utilizes the aforementioned properties. In the resulting estimation scheme separate measurement channels are handled separately, thus reducing the computational complexity. It is shown that, whereas the IMM complexity is exponential in the number of fault-prone measurements, the complexity of the proposed technique is polynomial. A simulation study involving spacecraft attitude estimation is carried out. This study shows that the proposed technique closely approximates the full-blown IMM algorithm, while requiring only a modest fraction of the computational cost.

Previous papers (1997, 1999, 2000) have described a tracking approach which utilized a Combined Kalman Filter (CKF), adaptive tracking for maneuver tracking, and the JVC association algorithm for reports to tracks, for use in Airborne Early Warning (AEW) applications. In this paper we present our incorporation of Interacting Multiple Model (IMM) tracking. First our previous AEW tracking approach is briefly reviewed as most of this approach is still utilized and forms our baseline. The new IMM approach and equations are then described. Then the two IMM tracking approaches used are discussed. One involves a two model IMM containing two constant velocity models, one a low process noise and the other a high process noise model. The other approach involves three IMM filter models, a coordinated turn filter model, and the same two constant velocity filter models as in the two model IMM approach. Results for both IMM approaches and the baseline tracker are shown. The results presented involve a 120 target scenario with two second update time with simulated radar data. In addition computer timing results are presented. These results indicate that while the three model IMM approach provides the best tracking results, it does so at a substantial computational cost. The two model IMM provides comparable tracking improvement but at a far less computational cost.