Purpose: To develop and evaluate a novel deep learning framework that leverages prior fraction information to improve organ-at-risk segmentation accuracy in fractionated adaptive radiotherapy.

Methods: Data from 74 pancreatic cancer patients treated with 5-fraction magnetic resonance guided adaptive radiotherapy were analyzed. Images were pre-processed with resampling (1.5×1.5×3.0mm voxel spacing), density normalization, and cropping (128×128×64 dimensions). The proposed dual-path architecture incorporates both current fraction imaging (infer path) and prior fraction imaging with segmentation data (support path), with feature fusion blocks integrating multi-scale spatial and temporal features. Performance was evaluated using both 3D UNet and SwinUNETR backbones, comparing baseline implementations (without prior fraction data) against our sequential framework. Dice Similarity Coefficient (DSC), 95th percentile Hausdorff Distance (HD95), and Average Symmetric Surface Distance (ASSD) were assessed with Mann-Whitney U-test for statistical significance.

Results: The sequential approach yielded consistent improvements across all evaluated organs (colon, duodenum, small bowel, stomach) with both network backbones. With 3D UNet, significant increases in DSC were observed for colon (0.851 to 0.871, p=0.0525) and small bowel (0.778 to 0.820, p=0.0399), while HD95 for duodenum significantly decreased (32.46mm to 26.37mm, p=0.0087). Using SwinUNETR, significant improvements were seen in DSC for duodenum (0.690 to 0.730, p=0.0129) and HD95 for stomach (26.22mm to 14.82mm, p=0.0350).

Conclusions: The proposed framework effectively leverages temporal information across treatment fractions to enhance segmentation accuracy in adaptive radiotherapy, providing consistent improvements with both convolutional and transformer-based architectures, suggesting potential for clinical implementation and improved efficiency in ART workflows.

Purpose
Effective radiotherapy treatment planning and dose delivery depends on high-quality CT imaging. This work evaluates the dosimetric accuracy of the HyperSight CBCT system (Varian Medical Systems) to determine its reliability for treatment planning.

 

Methods
Images of 30 H&N cancer patients were acquired on a FBCT (GE Healthcare) and HyperSight CBCT during CT simulation, and on a conventional TrueBeam CBCT (Varian Medical Systems) on the first treatment day. The HyperSight images were reconstructed using the iCBCT metal artifact reduction (HS-MAR) and iCBCT Acuros (HS-Acuros) algorithms. For each patient, the CBCTs were rigidly registered to the FBCT, and the clinically-accepted reference plans were forward-calculated on all images. For an initial patient subset, dose-volume histogram (DVH) metrics were evaluated for the target and surrounding organs, and a 3D gamma evaluation (3%/3mm) was performed using FBCT as the reference.

 

Results
The maximum deviations of all PTV DVH-metrics (D2%,D95%,D99%,Dmean,Dmax,V100%,V105%) were within 14.4%, 13.2% and 62.9% of plans calculated on FBCT for the HS-Acuros, HS-MAR and conventional TrueBeam images, respectively. For the oral cavity, the D2%,Dmean,V70%,V80%,V90% and V95% metric deviations were within 3.2% for all images, withHS-MAR images having the highest variability. This may result from reduced dental-metal artifacts in 22/30 HS-MAR images compared to FBCT, not necessarily implying reduced dosimetric accuracy. The median(±SD) gamma pass rates were 98.8%±1.3% for HS-Acuros, 98.9%±1.0% for HS-MAR, and 96.8%±2.5% for TrueBeam.

 

Conclusions
HyperSight demonstrated greater consistency with FBCT in PTV DVH-metrics and dose distributions compared to TrueBeam, highlighting a significant advancement towards CBCT-based direct-dose calculation for adaptive radiotherapy.

Purpose: To investigate the effects of respiratory motion during fast (~6 s) and slow (~60 s) cone beam computed tomography (CBCT) acquisitions, with a focus on implications for online and offline adaptive radiotherapy (ART).

Methods: The fast and slow CBCT modes are compared with 4D fan beam CT, considering average (“AVE”) and maximum (“MIP”) intensity projections. Data are acquired using a respiratory motion phantom representing a human thorax with a lung tumor. A range of sup-inf motion amplitudes (3-11 mm) and periods (3-5 s) are considered.

Results: Fast mode CBCT motion artefacts are more severe for larger amplitudes and longer periods. Motion artefacts are minimal in slow mode. Target volumes contoured using an HU thresholding approach on slow mode CBCTs are smaller (by up to 7%) than those on the AVE. On the MIP, volumes are even larger, with differences up to 29%. Fast mode motion artefacts were judged to be too severe for contouring. Gamma pass rates (1%, 1 mm, 10% dose threshold) for dose calculated on fast and slow mode CBCTs compared to the AVE are ≥ 99%. Dose differences (fast mode minus AVE) are generally larger for larger amplitudes and longer periods. Dose differences between slow mode and AVE are similar across all motion patterns considered.

Conclusions: Dosimetric perturbations resulting from motion artefacts are not severe for the amplitudes and periods considered herein. However, motion artefacts (especially in fast mode) have implications for registration during image-guided patient positioning, target contouring, and treatment plan optimization for ART.

Purpose
To assess the dosimetric impact of CT guided online adaptive radiotherapy (CTgART) on target coverage and bowel dose.

Methods
Eight patients with locally advanced cervical cancer were treated with CTgART on the Varian Ethos platform. Patients received 180cGy for 25 fractions to the uterus/parametrium (CTVp) and elective pelvic and para-aortic (PA) nodal regions (CTVn) with a 9-field IMRT plan. Gross nodes in the pelvis (GTVn_pel) and PA (GTVn_pa) regions received a boost to 220 and 230cGy per fraction respectively. The daily adaptive process included AI-assisted contouring, scheduled treatment plan assessment, and generation of an adapted plan optimized to the daily anatomy.

Results
Interfraction anatomical changes reduced the target volume receiving 100% of the prescribed dose (V100) in the scheduled plan, and adapting the treatment daily improved V100 for CTVp and CTVn from 96.7% in the scheduled plan to 100% in the adapted plan. Gross nodal coverage improved significantly, with GTVn_pel V100 increasing from 83.4 to 99.3%, indicating variable interfraction positioning of these nodes. GTVn_pa positioning was more stable, as the average V100 increased from 95.0 to 98.6% in the adapted plan. Daily adaptation reduced all bowel dose metrics, including V4000cGy, V3000cGy, and a large reduction in bowel hotspot (D0.04cc) from 245.2 to 223.6cGy for patients with PA nodes.

Conclusions
Daily online CTgART for locally advanced cervix cancer improves target coverage while reducing bowel dose. This is beneficial for mobile targets such as the uterus, extended length targets that include PA nodes, and nodes in close proximity to small bowel.

Purpose: To develop an efficient offline adaptive workflow using Ethos 2.0 and enhanced HyperSight imaging for head-and-neck cancer patients. These proposed methods utilize a template-based planning workflow, with direct dose calculation on mid-treatment cone beam computed tomography (CBCTs).

Methods: Ethos 2.0 template-generated plans were created for seventeen head-and-neck cancer patients for comparison with manual plans created in Eclipse. The template was optimized for multi-dose level head-and-neck cases to produce clinical-quality volumetric modulated arc therapy (VMAT) plans. To test the viability of direct HyperSight CBCT dose calculation, measured CBCT Hounsfield Units (HU) were compared to CT simulation HUs. Dose calculated on a CT simulation and CBCT were compared for both phantom and patient deliveries using 3D global gamma analyses and a pass rate criterion of 90%.

Results: The template-generated plans were comparable to the manually created plans. Mean parotid gland dose was lower in the Ethos-templated plans (18.16 Gy vs. 23.56 Gy). However, 105% prescription dose coverage was higher in the templated plans at intermediate dose levels (75.2% vs. 61.8%). Differences in HU between the CT and CBCT were minimal, with noticeable differences only in the densest (bone) materials. For dose comparisons on the phantom and patient plans, the gamma pass rates were 99.5% (1%/1mm) and above 94.6% (1%/1mm), respectively.

Conclusions: A robust Ethos planning template was created for head-and-neck cancers. Templated plans used in combination with direct dose calculation on CBCTs allows for rapid, automated offline adaptive workflows and eliminates the need for resource-intensive repeat CTs.

Purpose: To analyze intrafraction motion of prostate SBRT patients treated with fiducial-free online adaptive radiation therapy, OART, on Ethos with HyperSight. The goal is to compare adaptive IGRT procedures to our experience using auto-beam hold on TrueBeam with implanted gold fiducials.

Methods: Nineteen prostate patients received 36.25 Gy/5 fx (PTV margin = 4 mm) adaptive radiation therapy. Image guidance via CBCTs was performed twice during treatment. Time between imaging and applied shifts were analyzed for all fractions. Data was compared to previous analysis of patients with fiducials treated with the same fractionation on a TrueBeam STx linac.

 

Results: 192 CBCTs over 95 fractions of prostate OART were analyzed (excluding verification CBCTs). Average time between intrafraction CBCTs was (4.3 ± 0.7) min. This includes time for IGRT and beam-on. Shifts greater than 2 mm, or otherwise considered necessary, were applied. 9.4% (18/192) of intrafraction CBCTs resulted in shifts to patient position. For triggered imaging treatments, time following a CBCT before beam-hold was initiated was (6.1 ± 2.3) min. 98.8% (79/80) of fractions had ≤ 2 interruptions that resulted in a shift of ≥ 2 mm.

 

Conclusions: Fiducial-based treatments informed imaging protocols for online adaptive prostate SBRT. Imaging was previously triggered based on fiducial motion, while CBCTs were taken at specified time points for adaptive treatments. Transient motion between CBCTs is not captured here, however the low occurrence of intrafraction shifts indicates that persistent changes in position are accurately captured with current procedures.