Tumor Tracking Research Articles

Study of a Geometric Score to Identify the Increased Risk of Local Failure after SBRT for Locally Advanced Pancreatic Cancer (LAPC)

A previous analysis on an institutional cohort of LAPC patients, treated with intensive induction chemotherapy followed by SBRT, proved that local failures (LFs) predominantly occurred within the RT field. This suggests a specific efficacy of a complete covering of the macroscopic (GTV) aggressive disease with high doses (i.e., simultaneous boost [SIB] doses). Still, this is often not feasible due to the proximity of extremely radiosensitive organs at risk (OARs), which forces the use of "Simultaneous Integrated Protection" (SIP), limiting the dose to safe values in a portion of the GTV. This analysis aimed to quantitatively assess the role of GTV size and incomplete GTV covering with SIB in increasing the risk of LF. We included 51 consecutive LAPC pts treated with SBRT (Nov 2016 - Nov 2019): 30 Gy in 5 fractions to the tumor PTV, 50 Gy SIB to the region of vessel abutment/encasement, 25 Gy SIP to the overlap between tumor PTV, and the Planning OAR volumes. We used univariate/multivariable survival analysis (UVA/MVA, Cox regression, Kaplan-Meier Curves, Log-Rank test) to assess the association of LF with the GTV size and the amount of the GTV which the SIB did not cover. Specifically, we defined a set of incomplete GTV-covering levels: GTV75 (= yes if less than 75% of the GTV was included in the 50 Gy SIB isodose), GTV70, GTV50 and GTV30. Variables with p≤0.2 at UVA were included for MVA assessment. MVA models were considered relevant if their likelihood was significantly higher when compared to UVA. With a median follow-up of 17 months (range 1.4-47) 12 pts experienced LF. The GTV size (p = 0.007, risk factor for LF, Hazard Ratio [HR] 1.04 for 1 cc increase in GTV size) and GTV70 (p = 0.2, risk factor, HR = 3.73 for GTV70 = yes) were selected at UVA and included in MVA. To build a simple decision tool, we dichotomized the GTV size as below/above 25 cc (selected from Youden Index on the ROC curve): HR = 9.3 if GTV>25 cc. We used dichotomized GTV size and GTV70 to build a 3-level Geometric Score for the prediction of the risk of LF: [Low Risk (LR)] if "GTV<25cc AND GTV70 = no"; [Intermediate Risk (IR)] if "GTV>25cc OR GTV70 = yes"; [High Risk (HiR)] if "GTV>25cc AND GTV70 = yes". Pts classified at HiR had a significantly higher probability of LF: HR = 6.9 (95% CI 1.5-32.9) compared to LR, and HR = 13.2 (95% CI 3.6-48.4) when compared to IR. 10/12 LFs are in the HiR group. A large GTV size, coupled with an incomplete (<70%) covering of GTV from the SIB, highly increases the risk of LF: 62% actuarial probability in the HiR group vs 7% in the LR/IR groups. Full coverage of the GTV with SIB would be of clinical relevance for pts with large macroscopic tumors. The Geometric Score could be used to select pts that would effectively benefit from online tumor tracking (e.g., with an MRI-Linac), allowing a reduction of the SIP volume and a consequent decrease in the amount of GTV left uncovered by the SIB.

Evaluation of Patient-Specific Quality Control (QC) for Markerless Dynamic Tumor Tracking (MDTT) Deliveries

To report on the patient-specific QC for MDTT for liver and lung tumors treated using the dome-of-liver/diaphragm as a surrogate for the target position and compare to the QC of conventional fiducial-based tracking. Vero4DRT is a linear accelerator that has a gimbal-mounted waveguide and collimation system allowing the radiation beam to perform dynamic tumor tracking based on a patient's respiratory motion. A correlation model is built between an external IR marker placed on the chest and an internal structure, which may be implanted fiducial surrogates or in the case of MDTT, an anatomic landmark. During treatment an orthogonal kV image pair is taken every second. The auto detected internal structure position versus the predicted position is recorded in treatment log files. Five IMRT tracking plans were delivered to two commercial motion phantoms with in-house additions for patient-specific tracking QC. For point dose measurements, a 0.6cc farmer chamber was placed inside the dynamic tracking phantom which contains fiducials for fiducial-based tracking delivery and an anatomic landmark for MDTT. For 2D dose distribution measurements, a radiographic film insert for the Quasar Respiratory Motion Phantom was constructed. The phantom contains both fiducials and has a liver-dome shape at one end to facilitate MDTT. Both motion platforms were interfaced with software that enabled patient-specific respiratory motion traces acquired during a 4DCT scan. Each plan was delivered three times on each phantom: 1) in static mode, 2) during fiducial-based tracking, and 3) during MDTT. Chamber measurements taken during static and tracked delivery were compared to the average chamber dose from the treatment planning software (TPS). Film distributions measured during tracked deliveries were compared to the static film distribution using gamma analysis with FILMQA PRO software. Finally, dynamic tracking treatment statistics, extracted from log files, were compared between fiducial and markerless tracking deliveries. All chamber measurements resulted in dose differences <2.5% compared with the TPS. Dose differences between fiducial-based and MDTT were 0.26% on average (range 0.03-0.62%). For film dose map analysis, gamma pass rates were >95% for all plans for all tracking methods. The average gamma pass rate for 3%/3mm was 99.4% (fiducial-based) vs 98.4% (markerless). For 2%/2mm, the average gamma pass rate was 96.4% (fiducial-based tracking) vs 95.6% (MDTT). Dynamic treatment statistics from logs files reported an average 3D absolute deviation from predicted position of 0.52mm (±0.24) for fiducial tracking and 1.19mm (±0.50) for MDTT. Both fiducial-based and MDTT plans meet passing criteria for patient-specific QC. 3D absolute deviation of detected versus predicted position are larger for MDTT compared to fiducial tracking, however no significant impact on dose delivery was measured.

Assessment of interfraction dose variation in pancreas SBRT using daily simulation MR images

Pancreatic Cancer is associated with poor treatment outcomes compared to other cancers. High local control rates have been achieved by using hypofractionated stereotactic body radiotherapy (SBRT) to treat pancreatic cancer. Challenges in delivering SBRT include close proximity of several organs at risk (OARs) and target volume inter and intra fraction positional variations. Magnetic resonance image (MRI) guided radiotherapy has shown potential for online adaptive radiotherapy for pancreatic cancer, with superior soft tissue contrast compared to CT. The aim of this study was to investigate the variability of target and OAR volumes for different treatment approaches for pancreatic cancer, and to assess the suitability of utilizing a treatment-day MRI for treatment planning purposes. Ten healthy volunteers were scanned on a Siemens Skyra 3 T MRI scanner over two sessions (approximately 3 h apart), per day over 5 days to simulate an SBRT daily simulation scan for treatment planning. A pretreatment scan was also done to simulate patient setup and treatment. A 4D MRI scan was taken at each session for internal target volume (ITV) generation and assessment. For each volunteer a treatment plan was generated in the Raystation treatment planning system (TPS) following departmental protocols on the day one, first session dataset (D1S1), with bulk density overrides applied to enable dose calculation. This treatment plan was propagated through other imaging sessions, and the dose calculated. An additional treatment plan was generated on each first session of each day (S1) to simulate a daily replan process, with this plan propagated to the second session of the day. These accumulated mock treatment doses were assessed against the original treatment plan through DVH comparison of the PTV and OAR volumes. The generated ITV showed large variations when compared to both the first session ITV and daily ITV, with an average magnitude of 22.44% ± 13.28% and 25.83% ± 37.48% respectively. The PTV D95 was reduced by approximately 23.3% for both plan comparisons considered. Surrounding OARs had large variations in dose, with the small bowel V30 increasing by 128.87% when compared to the D1S1 plan, and 43.11% when compared to each daily S1 plan. Daily online adaptive radiotherapy is required for accurate dose delivery for pancreas cancer in the absence of additional motion management and tumour tracking techniques.

Exploring beamline momentum acceptance for tracking respiratory variability in lung cancer proton therapy: a simulation study

Objective. Investigating the aspects of proton beam delivery to track organ motion with pencil beam scanning therapy. Considering current systems as a reference, specify requirements for next-generation units aiming at real-time image-guided treatments. Approach. Proton treatments for six non-small cell lung cancer (NSCLC) patients were simulated using repeated 4DCTs to model respiratory motion variability. Energy corrections required for this treatment site were evaluated for different approaches to tumour tracking, focusing on the potential for energy adjustment within beamline momentum acceptance (dp/p). A respiration-synchronised tracking, taking into account realistic machine delivery limits, was compared to ideal tracking scenarios, in which unconstrained energy corrections are possible. Rescanning and the use of multiple fields to mitigate residual interplay effects and dose degradation have also been investigated. Main results. Energy correction requirements increased with motion amplitudes, for all patients and tracking scenarios. Higher dose degradation was found for larger motion amplitudes, rescanning has beneficial effects and helped to improve dosimetry metrics for the investigated limited dp/p of 1.2% (realistic) and 2.4%. The median differences between ideal and respiratory-synchronised tracking show minimal discrepancies, 1% and 5% respectively for dose coverage (CTV V95) and homogeneity (D5-D95). Multiple-field planning improves D5-D95 up to 50% in the most extreme cases while it does not show a significant effect on V95. Significance. This work shows the potential of implementing tumour tracking in current proton therapy units and outlines design requirements for future developments. Energy regulation within momentum acceptance was investigated to tracking tumour motion with respiratory-synchronisation, achieving results in line with the performance of ideal tracking scenarios. ±5% Δp/p would allow to compensate for all range offsets in our NSCLC patient cohort, including breathing variability. However, the realistic momentum of 1.2% dp/p representative of existing medical units limitations, has been shown to preserve plan quality.

Predicting successful clinical candidates for fiducial-free lung tumor tracking with a deep learning binary classification model.

The CyberKnife system is a robotic radiosurgery platform that allows the delivery of lung SBRT treatments using fiducial-free soft-tissue tracking. However, not all lung cancer patients are eligible for lung tumor tracking. Tumor size, density, and location impact the ability to successfully detect and track a lung lesion in 2D orthogonal X-ray images. The standard workflow to identify successful candidates for lung tumor tracking is called Lung Optimized Treatment (LOT) simulation, and involves multiple steps from CT acquisition to the execution of the simulation plan on CyberKnife. The aim of the study is to develop a deep learning classification model to predict which patients can be successfully treated with lung tumor tracking, thus circumventing the LOT simulationprocess. Target tracking is achieved by matching orthogonal X-ray images with a library of digital radiographs reconstructed from the simulation CT scan (DRRs). We developed a deep learning model to create a binary classification of lung lesions as being trackable or untrackable based on tumor template DRR extracted from the CyberKnife system, and tested five different network architectures. The study included a total of 271 images (230 trackable, 41 untrackable) from 129 patients with one or multiple lung lesions. Eighty percent of the images were used for training, 10% for validation, and the remaining 10% for testing. For all five convolutional neural networks, the binary classification accuracy reached 100% after training, both in the validation and the test set, without any falseclassifications. A deep learning model can distinguish features of trackable and untrackable lesions in DRR images, and can predict successful candidates for fiducial-free lung tumortracking.

Simultaneous object detection and segmentation for patient-specific markerless lung tumor tracking in simulated radiographs with deep learning.

Real-time tumor tracking is one motion management method to address motion-induced uncertainty. To date, fiducial markers are often required to reliably track lung tumors with X-ray imaging, which carries risks of complications and leads to prolonged treatment time. A markerless tracking approach is thus desirable. Deep learning-based approaches have shown promise for markerless tracking, but systematic evaluation and procedures to investigate applicability in individual cases are missing. Moreover, few efforts have been made to provide bounding box prediction and mask segmentation simultaneously, which could allow either rigid or deformable multi-leaf collimatortracking. The purpose of this study was to implement a deep learning-based markerless lung tumor tracking model exploiting patient-specific training which outputs both a bounding box and a mask segmentation simultaneously. We also aimed to compare the two kinds of predictions and to implement a specific procedure to understand the feasibility of markerless tracking on individualcases. We first trained a Retina U-Net baseline model on digitally reconstructed radiographs (DRRs) generated from a public dataset containing 875 CT scans and corresponding lung nodule annotations. Afterwards, we used an independent cohort of 97 lung patients to develop a patient-specific refinement procedure. In order to determine the optimal hyperparameters for automatic patient-specific training, we selected 13 patients for validation where the baseline model predicted a bounding box on planning CT (PCT)-DRR with intersection over union (IoU) with the ground-truth higher than 0.7. The final test set contained the remaining 84 patients with varying PCT-DRR IoU. For each testing patient, the baseline model was refined on the PCT-DRR to generate a patient-specific model, which was then tested on a separate 10-phase 4DCT-DRR to mimic the intrafraction motion during treatment. A template matching algorithm served as benchmark model. The testing results were evaluated by four metrics: the center of mass (COM) error and the Dice similarity coefficient (DSC) for segmentation masks, and the center of box (COB) error and the DSC for bounding box detections. Performance was compared to the benchmark model including statistical testing forsignificance. A PCT-DRR IoU value of 0.2 was shown to be the threshold dividing inconsistent (68%) and consistent (100%) success (defined as mean bounding box DSC > 0.6) of PS models on 4DCT-DRRs. Thirty-seven out of the eighty-four testing cases had a PCT-DRR IoU above 0.2. For these 37 cases, the mean COM error was 2.6 mm, the mean segmentation DSC was 0.78, the mean COB error was 2.7 mm, and the mean box DSC was 0.83. Including the validation cases, the model was applicable to 50 out of 97 patients when using the PCT-DRR IoU threshold of 0.2. The inference time per frame was 170 ms. The model outperformed the benchmark model on all metrics, and the comparison was significant (p < 0.001) over the 37 PCT-DRR IoU > 0.2 cases, but not over the undifferentiated 84 testingcases. The implemented patient-specific refinement approach based on a pre-trained baseline model was shown to be applicable to markerless tumor tracking in simulated radiographs for lungcases.

278 Evaluation of Patient-Specific Quality Control (QC) for Markerless Dynamic Tumour Tracking (MDTT) Deliveries

278 Evaluation of Patient-Specific Quality Control (QC) for Markerless Dynamic Tumour Tracking (MDTT) Deliveries

Advanced and emerging radiation therapy approaches for intrahepatic cholangiocarcinoma

Radiotherapy (RT) is an integral component of the multidisciplinary care for intrahepatic cholangiocarcinoma (iCCA). Over the past decades, RT techniques have been developed with the aim of enhancing tumor control and minimizing toxicity. The most recent technological advancements include proton beam therapy (PBT) and magnetic resonance-guided radiotherapy (MRgRT). PBT is notable for its unique physical characteristics that allow for greater sparing of surrounding normal organs, especially the liver, from low to moderate doses of radiation. MRgRT provides advantages in other aspects, including superior tumor visualization before treatment, on-board treatment plan adaptation, and tumor tracking during treatment. These features allow for precise dose delivery and safe dose escalation, especially for patients with tumors close to luminal GI structures. In this review article, the rationale, clinical outcomes, clinical applications, challenges, and future directions of PBT and MRgRT are discussed. Additionally, the potential combination of novel therapeutics with RT in iCCA is explored.

CT-CBCT deformable registration using weakly-supervised artifact-suppression transfer learning network

Objective. Computed tomography-cone-beam computed tomography (CT-CBCT) deformable registration has great potential in adaptive radiotherapy. It plays an important role in tumor tracking, secondary planning, accurate irradiation, and the protection of at-risk organs. Neural networks have been improving CT-CBCT deformable registration, and almost all registration algorithms based on neural networks rely on the gray values of both CT and CBCT. The gray value is a key factor in the loss function, parameter training, and final efficacy of the registration. Unfortunately, the scattering artifacts in CBCT affect the gray values of different pixels inconsistently. Therefore, the direct registration of the original CT-CBCT introduces artifact superposition loss. Approach. In this study, a histogram analysis method for the gray values was used. Based on an analysis of the gray value distribution characteristics of different regions in CT and CBCT, the degree of superposition of the artifact in the region of disinterest was found to be much higher than that in the region of interest. Moreover, the former was the main reason for artifact superposition loss. Consequently, a new weakly supervised two-stage transfer-learning network based on artifact suppression was proposed. The first stage was a pre-training network designed to suppress artifacts contained in the region of disinterest. The second stage was a convolutional neural network that registered the suppressed CBCT and CT. Main Results. Through a comparative test of the thoracic CT-CBCT deformable registration, whose data were collected from the Elekta XVI system, the rationality and accuracy after artifact suppression were confirmed to be significantly improved compared with the other algorithms without artifact suppression. Significance. This study proposed and verified a new deformable registration method with multi-stage neural networks, which can effectively suppress artifacts and further improve registration by incorporating a pre-training technique and an attention mechanism.

SDPS-09 MIXED RESPONSE OF BRAIN METASTASES TREATED WITH OSIMERTINIB PREDICTS INFERIOR SURVIVAL OUTCOMES: AN ARTIFICIAL INTELLIGENCE-BASED LESION-LEVEL ASSESSMENT FOR CRANIAL CONTROL OF EGFRMUTANT NONSMALL CELL LUNG CANCER

Abstract Osimertinib has been found to be effective for intracranial control in patients with metastatic EGFR-mutant non-small cell lung cancer (NSCLC). However, the lesion-level response of brain metastases to osimertinib has not been thoroughly evaluated. This retrospective study included 73 patients with EGFR-mutant NSCLC and brain metastases who were treated with osimertinib at a single institution from 2016 to 2021. Patients with the leptomeningeal disease were excluded. The study used an FDA-approved brain tumor management artificial intelligence (AI) platform, VBrain, to identify, track, and measure brain metastases on the baseline and follow-up MRI brain scans. Mixed response (MiR) was defined as the occurrence of progressive or new lesions along with synchronous responsive shrinking intracranial lesions at the first follow-up scan. K-means grouping was used to partition patients into two clusters according to their response heterogeneity. With a median follow-up of 23.8 months, a high MiR score (higher than 103%, n=25) was associated with worse cranial-progression-free survival (3.2 vs. 17.9 months, p&amp;lt;0.0001) and significantly inferior overall survival (12.4 vs. 30.1 months, HR=2.35, p=0.016). This poses a similar negative impact on the survivals of 18 patients with pure RANO-BM progression (HR=2.52, p=0.025). Eight patients had the highest MiR score (MiRmax), in whom synchronous new lesions and completely responsive tumors were observed. In a multivariate Cox proportional-hazards regression involving performance status, extracranial failure, tumor volume, lesion number, and RANO-BM classification as variables, MiRmax remained an independent prognosticator for inferior survival (HR=5.20, 95%CI; 2.21–12.23, p=0.002). Our study demonstrates that MiR in brain metastases from EGFR-mutant NSCLC treated with osimertinib is associated with inferior survival outcomes and a higher risk of local progression. Lesion-level response assessment using AI may provide important prognostic information and aid in treatment decision-making for these patients.

Three-Dimensional Volumetric Magnetic Resonance Imaging Analysis of Syringomyelia Evolution After Posterior Fossa Decompression for Chiari Malformation Type1

This study aims to introduce a method of three-dimensional (3D) volume magnetic resonance imaging (MRI) measurements to evaluate the evolution of syringomyelia (SM) in adults with Chiari malformation type 1 (CM-1) after posterior fossa decompression (PFD), to provide reproducible and accurate measurements. Radiologic and clinical data were obtained for 28 patients. MRI 3D syrinx volumes were evaluated by 2 observers using Philips Healthcare Tumor Tracking tool. The mean SM preoperative volume was 9.1 ± 15.3 cm3 versus 4.6 ± 8.7 cm3 postoperatively, corresponding to a volume variation reduced by -50% ± 33.4%. A total of 21 patients (75%) showed >25% reduction in SM volume postoperatively and 25 patients (89.2%) showed good to excellent postoperative clinical outcomes. Very high interobserver and intraobserver agreement was noted for 3D volumetric analysis with Cohen weighted κ coefficients (0.899-0.991). A negative correlation was found between the number of involved metameres and the postoperative SM volume reduction (Pearson correlation coefficient= 0.378; P= 0.047). Patients with better clinical outcomes had fewer syrinx septae (F= 8.830; P= 0.001). Similarly, better syrinx volume reduction was associated with fewer septae (F= 1.554; P= 0.233). A positive relationship was observed between clinical outcomes and postoperative syrinx volume reduction (F= 1.554; P= 0.233), with data similar to the reported literature. This study shows that 3D volume measurement of the syrinx cavity on MRI is a reproducible and accurate method to assess the postoperative evolution of SM in CM-1 after PFD. Most patients (75%) showed a reduction of SM volume, with a mean volume decrease of 50%. The reliability is yet to be validated in further larger studies.

Population-based asymmetric margins for moving targets in real-time tumor tracking.

Both geometric and dosimetric components are commonly considered when determining the margin for planning target volume (PTV). As dose distribution is shaped by controlling beam aperture in peripheral dose prescription and dose-escalated simultaneously integrated boost techniques, adjusting the margin by incorporating the variable dosimetric component into the PTV margin is inappropriate; therefore, geometric components should be accurately estimated for margin calculations. We introduced an asymmetric margin-calculation theory using the guide to the expression of uncertainty in measurement (GUM) and intra-fractional motion. The margins in fiducial marker-based real-time tumor tracking (RTTT) for lung, liver, and pancreatic cancers were calculated and were then evaluated using Monte Carlo (MC) simulations. A total of 74705, 73235, and 164968 sets of intra- and inter-fractional positional data were analyzed for 48 lung, 48 liver, and 25 pancreatic cancer patients, respectively, in RTTT clinical trials. The 2.5th and 97.5th percentiles of the positional error were considered representative values of each fraction of the disease site. The population-based statistics of the probability distributions of these representative positional errors (PD-RPEs) were calculated in six directions. A margin covering 95% of the population was calculated using the proposed formula. The content rate in which the clinical target volume (CTV) was included in the PTV was calculated through MC simulations using the PD-RPEs. The margins required for RTTT were at most 6.2, 4.6, and 3.9mm for lung, liver, and pancreatic cancer, respectively. MC simulations revealed that the median content rates using the proposed margins satisfied 95% for lung and liver cancers and 93% for pancreatic cancer, closer to the expected rates than the margins according to van Herk's formula. Our proposed formula based on the GUM and motion probability distributions (MPD) accurately calculated the practical margin size for fiducial marker-based RTTT. This was verified through MC simulations.

Comparison of dynamic tumor tracking error measurement methods for robotic radiosurgery.

Dynamic tumor motion tracking is used in robotic radiosurgery for targets subject to respiratory motion, such as lung and liver cancers. Different methods of measuring tracking error have been reported, but the differences among these methods have not been studied, and the optimal method is unknown. The purpose of this study was to assess and compare tracking errors encountered with individual patients using different evaluation methods for method optimization. We compared the beam's eye view (BEV), machine learning (ML), log (addition error: AE), and log (root sum square: RSS) methods. Log (AE) and log (RSS) were calculated from log files. These tracking errors were compared, and the optimal evaluation method was ascertained. A t-test was performed to evaluate statistically significant differences. Here, the significance level was set at 5%. The mean values of BEV, log (AE), log (RSS), and ML were 2.87, 3.91, 2.91, and 3.74mm, respectively. The log (AE) and ML were higher than BEV (p<0.001), and log (RSS) was equivalent to the BEV, suggesting that the log (RSS) calculated with the log file method can substitute for the BEV calculated with the BEV method. As RSS error calculation is simpler than BEV calculation, using it may improve clinical practice throughput. This study clarified differences among three tracking error evaluation methods for dynamic tumor tracking radiotherapy using a robotic radiosurgery system. The log (RSS) calculated by the log file method was found to be the best alternative to BEV method, as it can calculate tracking errors more easily than the BEV method.

Enhancing the target visibility with synthetic target specific digitally reconstructed radiograph for intrafraction motion monitoring: A proof-of-concept study.

Intrafraction motion monitoring in External Beam Radiation Therapy (EBRT) is usually accomplished by establishing a correlation between the tumor and the surrogates such as an external infrared reflector, implanted fiducial markers, or patient skin surface. These techniques either have unstable surrogate-tumor correlation or are invasive. Markerless real-time onboard imaging is a noninvasive alternative that directly images the target motion. However, the low target visibility due to overlapping tissues along the X-ray projection path makes tumor tracking challenging. To enhance the target visibility in projection images, a patient-specific model was trained to synthesize the Target Specific Digitally Reconstructed Radiograph (TS-DRR). Patient-specific models were built using a conditional Generative Adversarial Network (cGAN) to map the onboard projection images to TS-DRR. The standard Pix2Pix network was adopted as our cGAN model. We synthesized the TS-DRR based on the onboard projection images using phantom and patient studies for spine tumors and lung tumors. Using previously acquired CT images, we generated DRR and its corresponding TS-DRR to train the network. For data augmentation, random translations were applied to the CT volume when generating the training images. For the spine, separate models were trained for an anthropomorphic phantom and a patient treated with paraspinal stereotactic body radiation therapy (SBRT). For lung, separate models were trained for a phantom with a spherical tumor insert and a patient treated with free-breathing SBRT. The models were tested using Intrafraction Review Images (IMR) for the spine and CBCT projection images for the lung. The performance of the models was validated using phantom studies with known couch shifts for the spine and known tumor deformation for the lung. Both the patient and phantom studies showed that the proposed method can effectively enhance the target visibility of the projection images by mapping them into synthetic TS-DRR (sTS-DRR). For the spine phantom with known shifts of 1 mm, 2 mm, 3 mm, and 4 mm, the absolute mean errors for tumor tracking were 0.11 ± 0.05 mm in the x direction and 0.25 ± 0.08 mm in the y direction. For the lung phantom with known tumor motion of 1.8 mm, 5.8 mm, and 9 mm superiorly, the absolute mean errors for the registration between the sTS-DRR and ground truth are 0.1 ± 0.3 mm in both the x and y directions. Compared to the projection images, the sTS-DRR has increased the image correlation with the ground truth by around 83% and increased the structural similarity index measure with the ground truth by around 75% for the lung phantom. The sTS-DRR can greatly enhance the target visibility in the onboard projection images for both the spine and lung tumors. The proposed method could be used to improve the markerless tumor tracking accuracy for EBRT.

Deep learning-based markerless lung tumor tracking in stereotactic radiotherapy using Siamese networks.

Radiotherapy (RT) is involved in about 50% of all cancer patients, making it a very important treatment modality. The most common type of RT is external beam RT, which consists of delivering the radiation to the tumor from outside the body. One novel treatment delivery method is volumetric modulated arc therapy (VMAT), where the gantry continuously rotates around the patient during the radiation delivery. Accurate tumor position monitoring during stereotactic body radiotherapy (SBRT) for lung tumors can help to ensure that the tumor is only irradiated when it is inside the planning target volume. This can maximize tumor control and reduce uncertainty margins, lowering organ-at-risk dose. Conventional tracking methods are prone to errors, or have a low tracking rate, especially for small tumors that are in close vicinity to bonystructures. We investigated patient-specific deep Siamese networks for real-time tumor tracking, during VMAT. Due to lack of ground truth tumor locations in the kilovoltage (kV) images, each patient-specific model was trained on synthetic data (DRRs), generated from the 4D planning CT scans, and evaluated on clinical data (x-rays). Since there are no annotated datasets with kV images, we evaluated the model on a 3D printed anthropomorphic phantom but also on six patients by computing the correlation coefficient with the breathing-related vertical displacement of the surface-mounted marker (RPM). For each patient/phantom, we used 80% of DRRs for training and 20% forvalidation. The proposed Siamese model outperformed the conventional benchmark template matching-based method (RTR): (1) when evaluating both methods on the 3D phantom, the Siamese model obtained a 0.57-0.79-mm mean absolute distance to the ground truth tumor locations, compared to 1.04-1.56 mm obtained by RTR; (2) on patient data, the Siamese-determined longitudinal tumor position had a correlation coefficient of 0.71-0.98 with the RPM, compared to 0.07-0.85 for RTR; (3) the Siamese model had a 100% tracking rate, compared to 62%-82% forRTR. Based on these results, we argue that Siamese-based real-time 2D markerless tumor tracking during radiation delivery is possible. Further investigation and development of 3D tracking iswarranted.

Intra-Patient Lung CT Registration through Large Deformation Decomposition and Attention-Guided Refinement.

Deformable lung CT image registration is an essential task for computer-assisted interventions and other clinical applications, especially when organ motion is involved. While deep-learning-based image registration methods have recently achieved promising results by inferring deformation fields in an end-to-end manner, large and irregular deformations caused by organ motion still pose a significant challenge. In this paper, we present a method for registering lung CT images that is tailored to the specific patient being imaged. To address the challenge of large deformations between the source and target images, we break the deformation down into multiple continuous intermediate fields. These fields are then combined to create a spatio-temporal motion field. We further refine this field using a self-attention layer that aggregates information along motion trajectories. By leveraging temporal information from a respiratory cycle, our proposed methods can generate intermediate images that facilitate image-guided tumor tracking. We evaluated our approach extensively on a public dataset, and our numerical and visual results demonstrate the effectiveness of the proposed method.

PO-1905 Towards tumor tracking by using deep learning to correlate lung contour deformation feature and skin

PO-1905 Towards tumor tracking by using deep learning to correlate lung contour deformation feature and skin

PO-1889 Feasibility of markerless lung tumor tracking on radiographs with Retina U-Net

PO-1889 Feasibility of markerless lung tumor tracking on radiographs with Retina U-Net

Effect of scattered megavoltage x-rays on markerless tumor tracking using dual energy kilovoltage imaging.

To determine the effect of megavoltage (MV) scatter on the accuracy of markerless tumor tracking (MTT) for lung tumors using dual energy (DE) imaging and to consider a post-processing technique to mitigate the effects of MV scatter on DE-MTT. A Varian TrueBeam linac was used to acquire a series of interleaved 60/120kVp images of a motion phantom with simulated tumors (10 and 15 mm diameter). Two sets of consecutive high/low energy projections were acquired, with and without MV beam delivery. The MV field sizes (FS) ranged from 2×2cm2 -6×6cm2 in steps of 1×1cm2 . Weighted logarithmic subtraction was performed on sequential images to produce soft-tissue images for kV only (DEkV ) and kV with MV beam on (DEkV+MV ). Wavelet and fast Fourier transformation filtering (wavelet-FFT) was used to remove stripe noise introduced by MV scatter in the DE images ( ). A template-based matching algorithm was then used to track the target on DEkV, DEkV+MV , and images. Tracking accuracy was evaluated using the tracking success rate (TSR) and mean absolute error (MAE). For the 10 and 15 mm targets, the TSR for DEkV images was 98.7% and 100%, and MAE was 0.53 and 0.42mm, respectively. For the 10mm target, the TSR, including the effects of MV scatter, ranged from 86.5% (2×2cm2 ) to 69.4% (6×6cm2 ), while the MAE ranged from 2.05mm to 4.04mm. The application of wavelet-FFT algorithm to remove stripe noise ( ) resulted in TSR values of 96.9% (2×2cm2 ) to 93.4% (6×6cm2 ) and subsequent MAE values were 0.89mm to 1.37mm. Similar trends were observed for the 15mm target. MV scatter significantly impacts the tracking accuracy of lung tumors using DE images. Wavelet-FFT filtering can improve the accuracy of DE-MTT during treatment.

Accuracy of abdominal organ motion estimation in radiotherapy using the right hemidiaphragm top as a surrogate during prolonged breath-holds quantified with MRI.

Respiratory motion presents a challenge in radiotherapy of thoracic and upper abdominal tumors. Techniques to account for respiratory motion include tracking. Using magnetic resonance imaging (MRI) guided radiotherapy systems, tumors can be tracked continuously. Using conventional linear accelerators, tracking of lung tumors is possible by determining tumor motion on kilo voltage (kV) imaging. But tracking of abdominal tumors with kV imaging is hampered by limited contrast. Therefore, surrogates for the tumor are used. One of the possible surrogates is the diaphragm. However, there is no universal method for establishing the error when using a surrogate and there are particular challenges in establishing such errors during free breathing (FB). Prolonged breath-holding might address these challenges. The aim of this study was to quantify the error when using the right hemidiaphragm top (RHT) as surrogate for abdominal organ motion during prolonged breath-holds (PBH) for possible application in radiation treatments. Fifteen healthy volunteers were trained to perform PBHs in two subsequent magnetic resonance imaging (MRI) sessions (PBH-MRI1 and PBH-MRI2). From each MRI acquisition, we selected seven images (dynamics) to determine organ displacement during PBH by using deformable image registration (DIR). On the first dynamic, the RHT, right and left hemidiaphragm, liver, spleen and right and left kidney were segmented. We used the deformation vector fields (DVF), generated by DIR, to determine the displacement of each organ between two dynamics in inferior-superior (IS), anterior-posterior (AP), left-right (LR) direction and we calculated the 3D vector magnitude (|d|). The displacements of the RHT, both hemidiaphragms and the abdominal organs were compared using a linear fit to determine the correlation (R2 of the fit) and the displacement ratio (DR, slope of the fit) between displacements of the RHT and each organ. We quantified the median difference between the DRs of PBH-MRI1 and PBH-MRI2 for each organ. Additionally, we estimated organ displacement in the second PBH by applying the DR from the first PBH to the displacement of the RHT measured during the second PBH. We compared the estimated organ displacement to the measured organ displacement during the second PBH. The difference between the two values was defined as the estimation error of using the RHT as a surrogate and assuming a constant DR over MRI sessions. The linear relationships were confirmed by the high R2 values of the linear fit between the displacements of the RHT and the abdominal organs (R2 > 0.96) in the IS and AP direction and |d|, and high to moderate correlations in the LR direction (0.93 > R2 > 0.64). The median DR difference between PBH-MRI1 and PBH-MRI2 varied between 0.13 to 0.31 for all organs. The median estimation error of the RHT as a surrogate varied between 0.4 and 0.8 mm/min for all organs. The RHT could serve as an accurate surrogate for abdominal organ motion during radiation treatments, e.g. in tracking, provided the error of the RHT as motion surrogate is taken into account in the margins. The study was registered in the Netherlands Trial Register (NL7603). This article is protected by copyright. All rights reserved.