Target Centroid Research Articles

Patient-specific deep learning tracking framework for real-time 2D target localization in MRI-guided radiotherapy

Purpose: We propose a tumor tracking framework for 2D cine MRI based on a pair of deep learning (DL) models relying on patient-specific (PS) training.Methods and materials: The chosen DL models are: 1) an image registration transformer and 2) an auto-segmentation convolutional neural network (CNN). We collected over 1,400,000 cine MRI frames from 219 patients treated on a 0.35 T MRI-linac plus 7,500 frames from additional 35 patients which were manually labelled and subdivided into fine-tuning, validation, and testing sets. The transformer was first trained on the unlabeled data (without segmentations). We then continued training (with segmentations) either on the fine-tuning set or for PS models based on eight randomly selected frames from the first 5 s of each patient's cine MRI. The PS auto-segmentation CNN was trained from scratch with the same eight frames for each patient, without pre-training. Furthermore, we implemented B-spline image registration as a conventional model, as well as different baselines. Output segmentations of all models were compared on the testing set using the Dice similarity coefficient (DSC), the 50% and 95% Hausdorff distance (HD50%/HD95%), and the root-mean-square-error of the target centroid in superior-inferior direction (RMSESI).Results: The PS transformer and CNN significantly outperformed all other models, achieving a median (inter-quartile range) DSC of 0.92 (0.03)/0.90 (0.04), HD50% of 1.0 (0.1)/1.0 (0.4) mm, HD95% of 3.1 (1.9)/3.8 (2.0) mm and RMSESI of 0.7 (0.4)/0.9 (1.0) mm on the testing set. Their inference time was about 36/8 ms per frame and PS fine-tuning required 3 min for labelling and 8/4 min for training. The transformer was better than the CNN in 9/12 patients, the CNN better in 1/12 patients and the two PS models achieved the same performance on the remaining 2/12 testing patients.Conclusion: For targets in the thorax, abdomen and pelvis, we found two PS DL models to provide accurate real-time target localization during MRI-guided radiotherapy.

A global reweighting approach for cross-domain semantic segmentation

Unsupervised domain adaptation semantic segmentation attracts much research attention due to the expensive pixel-level annotation cost. Since the adaptation difficulty of samples is different, the weight of samples should be set independently, which is called reweighting. However, existing reweighting methods only calculate local reweighting information from predicted results or context information in batch images of two domains, which may lead to over-alignment or under-alignment problems. To handle this issue, we propose a global reweighting approach. Specifically, we first define the target centroid distance, which describes the distance between the source batch data and the target centroid. Then, we employ a Fréchet Inception Distance metric to evaluate the domain divergence and embed it into the target centroid distance. Finally, a global reweighting strategy is proposed to enhance the knowledge transferability in the source domain supervision. Extensive experiments demonstrate that our approach achieves competitive performance and helps to improve performance in other methods.

Experimental comparison of linear regression and LSTM motion prediction models for MLC-tracking on an MRI-linac.

Magnetic resonance imaging (MRI)-guided radiotherapy with multileaf collimator (MLC)-tracking is a promising technique for intra-fractional motion management, achieving high dose conformality without prolonging treatment times. To improve beam-target alignment, the geometric error due to system latency should be reduced by using temporal prediction. To experimentally compare linear regression (LR) and long-short-term memory (LSTM) motion prediction models for MLC-tracking on an MRI-linac using multiple patient-derived traces with different complexities. Experiments were performed on a prototype 1.0T MRI-linac capable of MLC-tracking. A motion phantom was programmed to move a target in superior-inferior (SI) direction according to eight lung cancer patient respiratory motion traces. Target centroid positions were localized from sagittal 2D cine MRIs acquired at 4Hz using a template matching algorithm. The centroid positions were input to one of four motion prediction models. We used (1) a LSTM network which had been optimized in a previous study on patient data from another cohort (offline LSTM). We also used (2) the same LSTM model as a starting point for continuous re-optimization of its weights during the experiment based on recent motion (offline+online LSTM). Furthermore, we implemented (3) a continuously updated LR model, which was solely based on recent motion (online LR). Finally, we used (4) the last available target centroid without any changes as a baseline (no-predictor). The predictions of the models were used to shift the MLC aperture in real-time. An electronic portal imaging device (EPID) was used to visualize the target and MLC aperture during the experiments. Based on the EPID frames, the root-mean-square error (RMSE) between the target and the MLC aperture positions was used to assess the performance of the different motion predictors. Each combination of motion trace and prediction model was repeated twice to test stability, for a total of 64 experiments. The end-to-end latency of the system was measured to be (389 ± 15)ms and was successfully mitigated by both LR and LSTM models. The offline+online LSTM was found to outperform the other models for all investigated motion traces. It obtained a median RMSE over all traces of (2.8 ± 1.3)mm, compared to the (3.2 ± 1.9)mm of the offline LSTM, the (3.3 ± 1.4)mm of the online LR and the (4.4 ± 2.4)mm when using the no-predictor. According to statistical tests, differences were significant (p-value <0.05) among all models in a pair-wise comparison, but for the offline LSTM and online LR pair. The offline+online LSTM was found to be more reproducible than the offline LSTM and the online LR with a maximum deviation in RMSE between two measurements of 10%. This study represents the first experimental comparison of different prediction models for MRI-guided MLC-tracking using several patient-derived respiratory motion traces. We have shown that among the investigated models, continuously re-optimized LSTM networks are the most promising to account for the end-to-end system latency in MRI-guided radiotherapy withMLC-tracking.

Dim and Small Space-Target Detection and Centroid Positioning Based on Motion Feature Learning

The detection of dim and small space-targets is crucial in space situational awareness missions; however, low signal-to-noise ratio (SNR) targets and complex backgrounds pose significant challenges to such detection. This paper proposes a space-target detection framework comprising a space-target detection network and a k-means clustering target centroid positioning method. The space-target detection network performs a three-dimensional convolution of an input star image sequence to learn the motion features of the target, reduces the interference of noise using a soft thresholding module, and outputs the target detection result after positioning via the offsetting branch. The k-means centroid positioning method enables further high-precision subpixel-level centroid positioning of the detection network output. Experiments were conducted using simulated data containing various dim and small space-targets, multiple noises, and complex backgrounds; semi-real data with simulated space-targets added to the real star image; and fully real data. Experiments on the simulated data demonstrate the superior detection performance of the proposed method for multiple SNR conditions (particularly with very low false alarm rates), robustness regarding targets of varying numbers and speeds, and complex backgrounds (such as those containing stray light and slow motion). Experiments performed with semi-real and real data both demonstrate the excellent detection performance of the proposed method and its generalization capability.

Mel Spectrogram-based advanced deep temporal clustering model with unsupervised data for fault diagnosis

Fault diagnosis of mechanical equipment using data-driven machine learning methods has been developed recently as a promising technique for improving the reliability of industrial systems. However, these methods suffer from data sparsity due to the difficulty in data collection, which limits the feature extraction of anomalies. To solve this problem, we propose the mel spectrogram-based advanced deep temporal clustering (ADTC) model, which can extract and verify the features of unlabeled data through an unsupervised learning based autoencoder and the K-means. In addition, the ADTC model uses the proposed centroid based learning to obtain calibrated unsupervised learning data by minimizing the data point and target centroid distances for misclustered encoder output features in ensemble-based unsupervised learning. The classifier of the ADTC model uses a supervised learning based deep support vector machine network model, which is robust to nonlinear data, to diagnose the faults of the mechanical equipment. The proposed ADTC model was validated using mechanical equipment dataset with data augmentation to address the imbalanced dataset problem. During experiments, the mel spectrogram-based ADTC model exhibited the best performance in the various industrial environment with a prediction accuracy as high as 98.06%, outperforming other compared algorithms.

Precision detection of crop diseases based on improved YOLOv5 model.

Accurate identification of crop diseases can effectively improve crop yield. Most current crop diseases present small targets, dense numbers, occlusions and similar appearance of different diseases, and the current target detection algorithms are not effective in identifying similar crop diseases. Therefore, in this paper, an improved model based on YOLOv5s was proposed to improve the detection of crop diseases. First, the CSP structure of the original model in the feature fusion stage was improved, and a lightweight structure was used in the improved CSP structure to reduce the model parameters, while the feature information of different layers was extracted in the form of multiple branches. A structure named CAM was proposed, which can extract global and local features of each network layer separately, and the CAM structure can better fuse semantic and scale inconsistent features to enhance the extraction of global information of the network. In order to increase the number of positive samples in the model training process, one more grid was added to the original model with three grids to predict the target, and the formula for the prediction frame centroid offset was modified to obtain the better prediction frame centroid offset when the target centroid falled on the special point of the grid. To solve the problem of the prediction frame being scaled incorrectly during model training, an improved DIoU loss function was used to replace the GIoU loss function used in the original YOLOv5s. Finally, the improved model was trained using transfer learning, the results showed that the improved model had the best mean average precision (mAP) performance compared to the Faster R-CNN, SSD, YOLOv3, YOLOv4, YOLOv4-tiny, and YOLOv5s models, and the mAP, F1 score, and recall of the improved model were 95.92%, 0.91, and 87.89%, respectively. Compared with YOLOv5s, they improved by 4.58%, 5%, and 4.78%, respectively. The detection speed of the improved model was 40.01 FPS, which can meet the requirement of real-time detection. The results showed that the improved model outperformed the original model in several aspects, had stronger robustness and higher accuracy, and can provide better detection for crop diseases.

Detection method for table grape ears and stems based on a far-close-range combined vision system and hand-eye-coordinated picking test

To improve the picking success rate of table grapes in the process of automatic picking, it is important to accurately locate the point picking of grape ears and the cutting point of grape stems. In this research, a set of far-close-range stereoscopic vision systems was constructed to detect grape ears and grape stems. First, identification algorithms for grape ear centroid were proposed. A series of image operations were performed to complete the recognition and feature extraction of far-range grapes, including median filtering, threshold segmentation, morphological operation, executable region marking, and the extraction of the target centroid. Second, a calculation model of the cutting point of grape stems was established. The selected region of interest (ROI) of the close-range grape stem was determined to carry out edge detection and use a cumulative probability Hough transform. The straight line detection and cutting point positioning of the close-range grape stem were completed. Finally, based on LabVIEW software, the measurement and control system of the grape-picking robot was developed. Experiments verified that the identification of grape stems and localization algorithm of the cutting point had higher reliability, with success rates of 92% on sunny days, 82% during sunny backlight and 86% for overcast conditions. The grape-picking tests showed that the grape-picking robot can complete the table grape-picking task within an average time of 53.4 s under the given computer configuration.

Deep metric learning for accurate protein secondary structure prediction

Predicting the secondary structure of a protein from its amino acid sequence alone is a challenging prediction task for each residue in bioinformatics. Recent work has mainly used deep models based on the profile feature derived from multiple sequence alignments to make predictions. However, the existing state-of-the-art predictors usually have higher computational costs due to their large model sizes and complex network architectures. Here, we propose a simple yet effective deep centroid model for sequence-to-sequence secondary structure prediction based on deep metric learning. The proposed model adopts a lightweight embedding network with multibranch topology to map each residue in a protein chain into an embedding space. The goal of embedding learning is to maximize the similarity of each residue to its target centroid while minimizing its similarity to nontarget centroids. By assigning secondary structure types based on the learned centroids, we bypass the need for a time-consuming k-nearest neighbor search. Experimental results on six test sets demonstrate that our method achieves state-of-the-art performance with a simple architecture and smaller model size than existing models. Moreover, we also experimentally show that the embedding feature from the pretrained protein language model ProtT5-XL-U50 is superior to the profile feature in terms of prediction accuracy and feature generation speed. Code and datasets are available at https://github.com/fengtuan/DML_SS.

Dim Space Target Detection via Convolutional Neural Network in Single Optical Image

Real-time dim space target detection is a significant challenge in space situation awareness. This paper proposes a single-frame space object segmentation and detection method based on deep learning. Firstly, Channel and Space Attention U-net (CSAU-Net) is presented based on space image features. We remove unnecessary feature layers and add attention modules in the traditional encoder and decoder structure to enhance feature fusion and better use original feature layers. The proposed network structure can achieve accurate segmentation of space objects with fewer data training. At the same time, we construct a space target dataset for training, which contains targets with different signal-to-noise ratios to enhance the generalization of convolutional neural networks. After obtaining the segmentation masks, a simple connected component labeling method is applied to extract the centroid of the space target. Experiments show that our approach can achieve an ideal segmentation effect when the signal-to-noise ratio of the space target in the simulated dataset is 0.3. In addition, the proposed algorithm can realize fast segmentation and achieve an accuracy of 98.5%, which is similar to the traditional multi-frame space target detection method in real space image detection.

MRI-guided cardiac-induced target motion tracking for atrial fibrillation cardiac radioablation

Background and purposeAtrial fibrillation (AF) cardiac radioablation (CR) challenges radiotherapy tracking: multiple small targets close to organs-at-risk undergo rapid differential cardiac contraction and respiratory motion. MR-guidance offers a real-time target tracking solution. This work develops and investigates MRI-guided tracking of AF CR targets with cardiac-induced motion. Materials and methodsA direct tracking method (Trackingdirect) and two indirect tracking methods leveraging population-based surrogacy relationships with the left atria (Trackingindirect_LA) or other target (Trackingindirect_target) were developed. Tracking performance was evaluated using transverse ECG-gated breathhold MRI images from 15 healthy and 10 AF participants. Geometric and volumetric tracking errors were calculated, defined as the difference between the ground-truth and tracked target centroids and volumes respectively. Transverse, breath-hold, noncardiac-gated cine images were acquired at 4 Hz in 5 healthy and 5 AF participants to qualitatively characterize tracking performance on images more comparable to MRILinac acquisitions. ResultsThe average 3D geometric tracking errors for Trackingdirect, Trackingindirect_LA and Trackingindirect_target respectively were 1.7 ± 1.2 mm, 1.6 ± 1.1 mm and 1.9 ± 1.3 mm in healthy participants and 1.7 ± 1.3 mm, 1.5 ± 1.0 mm and 1.7 ± 1.2 mm in AF participants. For Trackingdirect, 88% of analyzed images had 3D geometric tracking errors <3 mm and the average volume tracking error was 1.7 ± 1.3 cc. For Trackingdirect on non-cardiac-gated cine images, tracked targets overlapped organsat-risk or completely missed the target area on 2.2% and 0.08% of the images respectively. ConclusionThe feasibility of non-invasive MRI-guided tracking of cardiac-induced AF CR target motion was demonstrated for the first time, showing potential for improving AF CR treatment efficacy.

Multi-target pig tracking algorithm based on joint probability data association and particle filter

In order to evaluate the health status of pigs in time, monitor accurately the disease dynamics of live pigs, and reduce the morbidity and mortality of pigs in the existing large-scale farming model, pig detection and tracking technology based on machine vision are used to monitor the behavior of pigs. However, it is challenging to efficiently detect and track pigs with noise caused by occlusion and interaction between targets. In view of the actual breeding conditions of pigs and the limitations of existing behavior monitoring technology of an individual pig, this study proposed a method that used color feature, target centroid and the minimum circumscribed rectangle length-width ratio as the features to build a multi-target tracking algorithm, which based on joint probability data association and particle filter. Experimental results show the proposed algorithm can quickly and accurately track pigs in the video, and it is able to cope with partial occlusions and recover the tracks after temporary loss. Keywords: joint probability data association, pig tracking, particle filter, centroid DOI: 10.25165/j.ijabe.20211404.6105 Citation: Sun L Q, Li Y Y. Multi-target pig tracking algorithm based on joint probability data association and particle filter. Int J Agric & Biol Eng, 2021; 14(4): 199–207.

Computed Tomography-Based Evaluation of Volume and Position Changes of the Target Region and Organs at Risk During Radiotherapy for Esophageal Cancer: A Pilot Study.

ObjectiveTo analyze changes in volume and position of target regions and organs at risk (OARs) during radiotherapy for esophageal cancer patients.MethodsOverall, 16 esophageal cancer patients who underwent radiotherapy, including 10 cases of intensity-modulated radiation therapy (IMRT) and six of three-dimensional conformal radiotherapy (3D-CRT), were enrolled. The prescription doses for the planning target volumes (PTVs) were as follows: PTV1, 64 Gy/32 fractions; and PTV2, 46 Gy/23 fractions. Repeat computed tomography (CT) was performed for patients after the 5th, 10th, 15th, 20th, and 25th fractions. Delineation of the gross tumor volume (GTV) and OAR volume was determined using five repeat CTs performed by the same physician. The target and OAR volumes and centroid positions were recorded and used to analyze volume change ratio (VCR), center displacement (ΔD), and changes in the distance from the OAR centroid positions to the planned radiotherapy isocenter (distance to isocenter, DTI) during treatment.ResultsNo patient showed significant changes in target volume (TV) after the first week of radiotherapy (five fractions). However, TV gradually decreased over the following weeks, with the rate slowing after the fourth week (40 Gy). The comparison of TV from baseline to 40 Gy (20 fractions) showed that average GTVs decreased from 130.7 ± 63.1 cc to 92.1 ± 47.2 cc, with a VCR of −29.21 ± 13.96% (p<0.01), while the clinical target volume (CTV1) decreased from 276.7 ± 98.2 cc to 246.7 ± 87.2 cc, with a VCR of −10.34 ± 7.58% (p<0.01). As TVs decreased, ΔD increased and DTI decreased. After the fourth week of radiotherapy (40 Gy), centroids of GTV, CTV1, and prophylactic CTV (CTV2) showed average deviations in ΔD of 7.6 ± 4.0, 6.9 ± 3.4, and 6.0 ± 3.0 mm, respectively. The average DTI of the heart decreased by 4.53 mm (from 15.61 ± 2.96 cm to 15.16 ± 2.27 cm).ConclusionDuring radiotherapy for esophageal cancer, Targets and OARs change significantly in volume and position during the 2nd–4th weeks. Image-guidance and evaluation of dosimetric changes are recommended for these fractions of treatment to appropriate adjust treatment plans.

Tracking of Multiple Maneuvering Random Hypersurface Extended Objects Using High Resolution Sensors

With the increased resolution capability of modern sensors, an object should be considered as extended if the target extent is larger than the sensor resolution. Multiple maneuvering extended object tracking (MMEOT) uses not only measurements of the target centroid but also high-resolution sensor measurements which may resolve individual features or measurement sources. MMEOT aims to jointly estimate object number, centroid states, and extension states. However, unknown and time-varying maneuvers of multiple objects produce difficulties in terms of accurate estimation. For multiple maneuvering star-convex extended objects using random hypersurface models (RHMs) in particular, their complex maneuvering behaviors are difficult to be described accurately and handled effectively. To deal with these problems, this paper proposes an interacting multiple model Gaussian mixture probability hypothesis density (IMM-GMPHD) filter for multiple maneuvering extended object tracking. In this filter, linear maneuver models derived from RHMs are utilized to describe different turn maneuvers of star-convex extended objects accurately. Based on these, an IMM-GMPHD filtering recursive form is given by deriving new update and merging formulas of model probabilities for extended objects. Gaussian mixture components of different posterior intensities are also pruned and merged accurately. More importantly, the geometrical significance of object extension states is fully considered and exploited in this filter. This contributes to the accurate estimation of object extensions. Simulation results demonstrate the effectiveness of the proposed tracking approach—it can obtain the joint estimation of object number, kinematic states, and object extensions in complex maneuvering scenarios.

ScAdapt: virtual adversarial domain adaptation network for single cell RNA-seq data classification across platforms and species.

In single cell analyses, cell types are conventionally identified based on expressions of known marker genes, whose identifications are time-consuming and irreproducible. To solve this issue, many supervised approaches have been developed to identify cell types based on the rapid accumulation of public datasets. However, these approaches are sensitive to batch effects or biological variations since the data distributions are different in cross-platforms or species predictions. In this study, we developed scAdapt, a virtual adversarial domain adaptation network, to transfer cell labels between datasets with batch effects. scAdapt used both the labeled source and unlabeled target data to train an enhanced classifier and aligned the labeled source centroids and pseudo-labeled target centroids to generate a joint embedding. The scAdapt was demonstrated to outperform existing methods for classification in simulated, cross-platforms, cross-species, spatial transcriptomic and COVID-19 immune datasets. Further quantitative evaluations and visualizations for the aligned embeddings confirm the superiority in cell mixing and the ability to preserve discriminative cluster structure present in the original datasets.

On the evaluation of mobile target trajectory between four-dimensional computer tomography and four-dimensional cone-beam computer tomography.

PurposeFor mobile lung tumors, four‐dimensional computer tomography (4D CT) is often used for simulation and treatment planning. Localization accuracy remains a challenge in lung stereotactic body radiation therapy (SBRT) treatments. An attractive image guidance method to increase localization accuracy is 4D cone‐beam CT (CBCT) as it allows for visualization of tumor motion with reduced motion artifacts. However, acquisition and reconstruction of 4D CBCT differ from that of 4D CT. This study evaluates the discrepancies between the reconstructed motion of 4D CBCT and 4D CT imaging over a wide range of sine target motion parameters and patient waveforms.MethodsA thorax motion phantom was used to examine 24 sine motions with varying amplitudes and cycle times and seven patient waveforms. Each programmed motion was imaged using 4D CT and 4D CBCT. The images were processed to auto segment the target. For sine motion, the target centroid at each phase was fitted to a sinusoidal curve to evaluate equivalence in amplitude between the two imaging modalities. The patient waveform motion was evaluated based on the average 4D data sets.ResultsThe mean difference and root‐mean‐square‐error between the two modalities for sine motion were −0.35 ± 0.22 and 0.60 mm, respectively, with 4D CBCT slightly overestimating amplitude compared with 4D CT. The two imaging methods were determined to be significantly equivalent within ±1 mm based on two one‐sided t tests (p < 0.001). For patient‐specific motion, the mean difference was 1.5 ± 2.1 (0.8 ± 0.6 without outlier), 0.4 ± 0.3, and 0.8 ± 0.6 mm for superior/inferior (SI), anterior/posterior (AP), and left/right (LR), respectively.ConclusionIn cases where 4D CT is used to image mobile tumors, 4D CBCT is an attractive localization method due to its assessment of motion with respect to 4D CT, particularly for lung SBRT treatments where accuracy is paramount.

Study on Rotation and Scaling Invariance of Retina-Like Imaging

Inspired by human vision, retina-like imaging has the feature of rotation and scaling invariance, which is beneficial for target tracking and recognition. However, there is a limitation that the center of field of view (FOV) should match with the target centroid. Currently, there is no effective method to quantitatively estimate the effects from this mismatch. To solve the issue, a novel evaluation method is proposed by applying the log-polar transform and Fourier transform. Then reconstruct the target image by exchanging magnitude and phase information between the target image and its rotated or scaled counterpart. Using two current methods of digital image processing, scale-invariant feature transform (SIFT) and biaxial projection similarity analysis (BPS), as evaluation criteria, the proposed method performs efficiently for all eccentric angles. Simulation results show that for a specific target, the invariance of rotation and scaling works effectively for a value above 0.69.

Heterogeneous Graph Attention Network for Unsupervised Multiple-Target Domain Adaptation.

Domain adaptation, which transfers the knowledge from label-rich source domain to unlabeled target domains, is a challenging task in machine learning. The prior domain adaptation methods focus on pairwise adaptation assumption with a single source and a single target domain, while little work concerns the scenario of one source domain and multiple target domains. Applying pairwise adaptation methods to this setting may be suboptimal, as they fail to consider the semantic association among multiple target domains. In this work we propose a deep semantic information propagation approach in the novel context of multiple unlabeled target domains and one labeled source domain. Our model aims to learn a unified subspace common for all domains with a heterogeneous graph attention network, where the transductive ability of the graph attention network can conduct semantic propagation of the related samples among multiple domains. In particular, the attention mechanism is applied to optimize the relationships of multiple domain samples for better semantic transfer. Then, the pseudo labels of the target domains predicted by the graph attention network are utilized to learn domain-invariant representations by aligning labeled source centroid and pseudo-labeled target centroid. We test our approach on four challenging public datasets, and it outperforms several popular domain adaptation methods.

A motion prediction confidence estimation framework for prediction-based radiotherapy gating.

Motion prediction can compensate for latency in image-guided radiotherapy and has been an active area of research. However, motion predictions are subject to error and variations. We have developed and evaluated a novel motion prediction confidence estimation framework to improve the efficacy and robustness of prediction-based radiotherapy gating decision-making. The specific scenario of adaptive gating in magnetic resonance imaging (MRI)-guided radiotherapy is studied as an example, but the method generalizes to other modalities and motion management setups. The proposed prediction confidence estimator is based on a generic training/testing paradigm and consists of a weighted combination of three components: the prediction model's goodness of fit, variation in the prediction using a leave-one-out process and the velocity of the tracked target. Roughly, these terms quantify respectively the consistency between prediction and the training data, the robustness of model inference, and the stability due to target speed. The weight parameters and the action level in triggering beam-off decision are optimized. The method is assessed and validated in 8 healthy volunteer and 13 patient studies using a 0.35T MRI-guided radiotherapy system predicting 0.25-0.33s ahead. The effect of the action level on the predicted gating decision accuracy, beam-on positive predictive value (PPV) and median distance between the predicted and ground-truth target centroids were evaluated. Statistical significance was evaluated using a paired t-test. The tradeoff between these performance metrics and gating duty cycle was assessed. Use of the confidence estimator threshold increased gating accuracy by up to 2.42%, increased PPV by up to 3.00%, and reduced the median centroid distance up to 0.28mm. The confidence estimator threshold on average increased gating accuracy to 96.5% (P=2.08×10-4 ), increased PPV to 96.7% (P=1.46×10-5 ), reduced the median centroid distance to 0.54mm (P=1.71×10-5 ) at the cost of reducing the gating duty cycle by 14.3% to 48.5%. Hyperparameter tuning revealed that contrary to intuition, the velocity term offered only minimal performance improvement in some cases but also introduced potential stability issues. The combination of goodness of fit and leave-one-out prediction variation provided the most effective confidence estimator, yielding universally better performance in gating decisions. Confidence estimation utilizing prediction model fitness criterion and validation principles can complement prediction methods to guide MRI-guided radiotherapy gating. Results from both volunteer and patient studies showed improved gating quality.

An image regression motion prediction technique for MRI-guided radiotherapy evaluated in single-plane cine imaging.

To develop and evaluate a novel motion prediction method for magnetic resonance image (MRI)-guided radiotherapy applications. This method, which we deem "image regression," predicts future tissue motion based on a weighted combination of previously observed motion states. Motion predictions are derived from a sliding window of recent motion states which are defined by a temporal sequence of images. A key advantage of this method compared to other motion prediction methods is that its computational complexity scales weekly with the number of spatial points predicted. Applications of gating latency reduction and improvement in deformable registration-based target tracking are demonstrated. The image regression (IR) motion prediction method was developed and evaluated using 26.9h of real-time imaging acquired from eight healthy volunteers and 13 patients using a 0.35T MRI-guided radiotherapy system. Motion predictions were performed 0.25-0.33s into the future using a weighted sum of previously observed motion states with image similarity-derived weights. The set of previously observed motion states were continuously updated to incorporate the changes in breathing patterns. The accuracy of the predicted radiotherapy gating decision, beam-on positive predictive value (PPV), and predicted vs ground-truth target centroid position errors are reported. The IR technique was compared against no prediction, linear extrapolation, and an established autoregressive linear prediction algorithm. The usage of IR to initialize the deformable registration and enhance the target tracking was demonstrated in the healthy volunteer studies. Deformable registration with IR initialization was compared to the initialization performed by current clinical software: no initialization, previous image registration initialization and linear motion extrapolation initialization. The average IR-predicted radiation beam gating decision accuracy was 95.8%, with a PPV of 95.7%, and median and 95th percentile centroid position errors of 0.63 and 2.08mm, respectively. Compared to the autoregressive linear prediction method, gating accuracy was 1.15% greater, PPV was 1.61% greater, and median and 95th percentile centroid distances were 0.21 and 0.23mm smaller. The IR-initialized registration on average converged within 0.50mm of the ground-truth position in fewer than 10 iterations whereas the next best initialization method required more than 25 iterations. Image regression motion prediction has the potential to reduce the gating latencies and improve the speed and accuracy of deformable registration-based target tracking in MRI-guided radiotherapy.

A photogrammetry-based experimental modal analysis method by tracking visible laser spots

A photogrammetry-based experimental modal analysis (EMA) method is developed, where a single-point laser shines a laser beam to a surface of a test object and an exterior feature is added to the test object in the form of a high-contrast laser spot. A three-dimensional kinematic model is developed to derive the kinematic relation between the laser spot and its associated image target on an image plane of a camera. A point-tracking technique is proposed to determine two-dimensional pixel coordinates of the optical intensity centroid of the image target on a digital image. The derived kinematic relation and EMA method were experimentally investigated. In the experimental investigation, displacements of multiple points at a cantilever beam were measured by a digital camera. Natural frequencies and mode shapes of the cantilever beam were estimated using the EMA method, and they compared well with those from a finite element model of the cantilever beam.