Artificial intelligence for treatment delivery: image-guided radiotherapy.

Radiation therapy (RT) is a highly digitized field relying heavily on computational methods and, as such, has a high affinity for the automation potential afforded by modern artificial intelligence (AI). This is particularly relevant where imaging is concerned and is especially so during image-guided RT (IGRT). With the advent of online adaptive RT (ART) workflows at magnetic resonance (MR) linear accelerators (linacs) and at cone-beam computed tomography (CBCT) linacs, the need for automation is further increased. AI as applied to modern IGRT is thus one area of RT where we can expect important developments in the near future. In this review article, after outlining modern IGRT and online ART workflows, we cover the role of AI in CBCT and MRI correction for dose calculation, auto-segmentation on IGRT imaging, motion management, and response assessment based on in-room imaging.

A direct comparison of multi-energy x-ray and proton CT for imaging and relative stopping power estimation of plastic andex-vivophantoms.

Objective.Proton therapy administers a highly conformal dose to the tumour region, necessitating accurate prediction of the patient's 3D map of proton relative stopping power (RSP) compared to water. This remains challenging due to inaccuracies inherent in single-energy computed tomography (SECT) calibration. Recent advancements in spectral x-ray CT (xCT) and proton CT (pCT) have shown improved RSP estimation compared to traditional SECT methods. This study aims to provide the first comparison of the imaging and RSP estimation performance among dual-energy CT (DECT) and photon-counting CT (PCCT) scanners, and a pCT system prototype.Approach.Two phantoms were scanned with the three systems for their performance characterisation: a plastic phantom, filled with water and containing four plastic inserts and a wood insert, and a heterogeneous biological phantom, containing a formalin-stabilised bovine specimen. RSP maps were generated by converting CT numbers to RSP using a calibration based on low- and high-energy xCT images, while pCT utilised a distance-driven filtered back projection algorithm for RSP reconstruction. Spatial resolution, noise, and RSP accuracy were compared across the resulting images.Main results.All three systems exhibited similar spatial resolution of around 0.54 lp/mm for the plastic phantom. The PCCT images were less noisy than the DECT images at the same dose level. The lowest mean absolute percentage error (MAPE) of RSP,(0.28±0.07)%, was obtained with the pCT system, compared to MAPE values of(0.51±0.08)%and(0.80±0.08)%for the DECT- and PCCT-based methods, respectively. For the biological phantom, the xCT-based methods resulted in higher RSP values in most of the voxels compared to pCT.Significance.The pCT system yielded the most accurate estimation of RSP values for the plastic materials, and was thus used to benchmark the xCT calibration performance on the biological phantom. This study underlined the potential benefits and constraints of utilising such a novelex-vivophantom for inter-centre surveys in future.

MRI-based ventilation and perfusion imaging to predict radiation-induced pneumonitis in lung tumor patients at a 0.35T MR-Linac.

BACKGROUND AND PURPOSE: Radiation-induced pneumonitis (RP), diagnosed 6-12 weeks after treatment, is a complication of lung tumor radiotherapy. So far, clinical and dosimetric parameters have not been reliable in predicting RP. We propose using non-contrast enhanced magnetic resonance imaging (MRI) based functional parameters acquired over the treatment course for patient stratification for improved follow-up. MATERIALS AND METHODS: 23 lung tumor patients received MR-guided hypofractionated stereotactic body radiation therapy at a 0.35T MR-Linac. Ventilation- and perfusion-maps were generated from 2D-cine MRI-scans acquired after the first and last treatment fraction (Fx) using non-uniform Fourier decomposition. The relative differences in ventilation and perfusion between last and first Fx in three regions (planning target volume (PTV), lung volume receiving more than 20Gy (V20) excluding PTV, whole tumor-bearing lung excluding PTV) and three dosimetric parameters (mean lung dose, V20, mean dose to the gross tumor volume) were investigated. Univariate receiver operating characteristic curve - area under the curve (ROC-AUC) analysis was performed (endpoint RP grade≥1) using 5000 bootstrapping samples. Differences between RP and non-RP patients were tested for statistical significance with the non-parametric Mann-Whitney U test (α=0.05). RESULTS: 14/23 patients developed RP of grade≥1 within 3 months. The dosimetric parameters showed no significant differences between RP and non-RP patients. In contrast, the functional parameters, especially the relative ventilation difference in the PTV, achieved a p-value<0.05 and an AUC value of 0.84. CONCLUSION: MRI-based functional parameters extracted from 2D-cine MRI-scans were found to be predictive of RP development in lung tumor patients.

Validation of dual-energy CT-based composition analysis using fresh animal tissues and composition-optimized tissue equivalent samples.

Objective.Proton therapy allows for highly conformal dose deposition, but is sensitive to range uncertainties. Several approaches currently under development measure composition-dependent secondary radiation to monitor the delivered proton rangein-vivo. To fully utilize these methods, an estimate of the elemental composition of the patient's tissue is often needed.Approach.A published dual-energy computed tomography (DECT)-based composition-extraction algorithm was validated against reference compositions obtained with two independent methods. For this purpose, a set of phantoms containing either fresh porcine tissue or tissue-mimicking samples with known, realistic compositions were imaged with a CT scanner at two different energies. Then, the prompt gamma-ray (PG) signal during proton irradiation was measured with a PG detector prototype. The PG workflow used pre-calculated Monte Carlo simulations to obtain an optimized estimate of the sample's carbon and oxygen contents. The compositions were also assessed with chemical combustion analysis (CCA), and the stopping-power ratio (SPR) was measured with a multi-layer ionization chamber. The DECT images were used to calculate SPR-, density- and elemental composition maps, and to assign voxel-wise compositions from a selection of human tissues. For a more comprehensive set of reference compositions, the original selection was extended by 135 additional tissues, corresponding to spongiosa, high-density bones and low-density tissues.Results.The root-mean-square error for the soft tissue carbon and oxygen content was 8.5 wt% and 9.5 wt% relative to the CCA result and 2.1 wt% and 10.3 wt% relative to the PG result. The phosphorous and calcium content were predicted within 0.4 wt% and 1.1 wt% of the CCA results, respectively. The largest discrepancies were encountered in samples whose composition deviated the most from tabulated compositions or that were more inhomogeneous.Significance.Overall, DECT-based composition estimations of relevant elements were in equal or better agreement with the ground truth than the established SECT-approach and could contribute toin-vivodose verification measurements.

A joint ESTRO and AAPM guideline for development, clinical validation and reporting of artificial intelligence models in radiation therapy.

BACKGROUND AND PURPOSE: Artificial Intelligence (AI) models in radiation therapy are being developed with increasing pace. Despite this, the radiation therapy community has not widely adopted these models in clinical practice. A cohesive guideline on how to develop, report and clinically validate AI algorithms might help bridge this gap. METHODS AND MATERIALS: A Delphi process with all co-authors was followed to determine which topics should be addressed in this comprehensive guideline. Separate sections of the guideline, including Statements, were written by subgroups of the authors and discussed with the whole group at several meetings. Statements were formulated and scored as highly recommended or recommended. RESULTS: The following topics were found most relevant: Decision making, image analysis, volume segmentation, treatment planning, patient specific quality assurance of treatment delivery, adaptive treatment, outcome prediction, training, validation and testing of AI model parameters, model availability for others to verify, model quality assurance/updates and upgrades, ethics. Key references were given together with an outlook on current hurdles and possibilities to overcome these. 19 Statements were formulated. CONCLUSION: A cohesive guideline has been written which addresses main topics regarding AI in radiation therapy. It will help to guide development, as well as transparent and consistent reporting and validation of new AI tools and facilitate adoption.

Impact of bias field correction on 0.35 T pelvic MR images: evaluation on generative adversarial network-based OARs' auto-segmentation and visual grading assessment.

Purpose: Magnetic resonance imaging (MRI)-guided radiotherapy enables adaptive treatment plans based on daily anatomical changes and accurate organ visualization. However, the bias field artifact can compromise image quality, affecting diagnostic accuracy and quantitative analyses. This study aims to assess the impact of bias field correction on 0.35 T pelvis MRIs by evaluating clinical anatomy visualization and generative adversarial network (GAN) auto-segmentation performance. Materials and methods: 3D simulation MRIs from 60 prostate cancer patients treated on MR-Linac (0.35 T) were collected and preprocessed with the N4ITK algorithm for bias field correction. A 3D GAN architecture was trained, validated, and tested on 40, 10, and 10 patients, respectively, to auto-segment the organs at risk (OARs) rectum and bladder. The GAN was trained and evaluated either with the original or the bias-corrected MRIs. The Dice similarity coefficient (DSC) and 95th percentile Hausdorff distance (HD95th) were computed for the segmented volumes of each patient. The Wilcoxon signed-rank test assessed the statistical difference of the metrics within OARs, both with and without bias field correction. Five radiation oncologists blindly scored 22 randomly chosen patients in terms of overall image quality and visibility of boundaries (prostate, rectum, bladder, seminal vesicles) of the original and bias-corrected MRIs. Bennett's S score and Fleiss' kappa were used to assess the pairwise interrater agreement and the interrater agreement among all the observers, respectively. Results: In the test set, the GAN trained and evaluated on original and bias-corrected MRIs showed DSC/HD95th of 0.92/5.63 mm and 0.92/5.91 mm for the bladder and 0.84/10.61 mm and 0.83/9.71 mm for the rectum. No statistical differences in the distribution of the evaluation metrics were found neither for the bladder (DSC: p = 0.07; HD95th: p = 0.35) nor for the rectum (DSC: p = 0.32; HD95th: p = 0.63). From the clinical visual grading assessment, the bias-corrected MRI resulted mostly in either no change or an improvement of the image quality and visualization of the organs' boundaries compared with the original MRI. Conclusion: The bias field correction did not improve the anatomy visualization from a clinical point of view and the OARs' auto-segmentation outputs generated by the GAN.

Intrafractional motion detection for spine SBRT via X-ray imaging using ExacTrac Dynamic.

Purpose: Due to its close vicinity to critical structures, especially the spinal cord, standards for safety for spine stereotactic body radiotherapy (SBRT) should be high. This study was conducted, to evaluate intrafractional motion during spine SBRT for patients without individualized immobilization (e.g., vacuum cushions) using high accuracy patient monitoring via orthogonal X-ray imaging. Methods: Intrafractional X-ray data were collected from 29 patients receiving 79 fractions of spine SBRT. No individualized immobilization devices were used during the treatment. Intrafractional motion was monitored using the ExacTrac Dynamic (ETD) System (Brainlab AG, Munich, Germany). Deviations were detected in six degrees of freedom (6 DOF). Tolerances for repositioning were 0.7 mm for translational and 0.5° for rotational deviations. Patients were repositioned when the tolerance levels were exceeded. Results: Out of the 925 pairs of stereoscopic X-ray images examined, 138 (15 %) showed at least one deviation exceeding the predefined tolerance values. In all 6 DOF together, a total of 191 deviations out of tolerance were recorded. The frequency of deviations exceeding the tolerance levels varied among patients but occurred in all but one patient. Deviations out of tolerance could be seen in all 6 DOF. Maximum translational deviations were 2.6 mm, 2.3 mm and 2.8 mm in the lateral, longitudinal and vertical direction. Maximum rotational deviations were 1.8°, 2.6° and 1.6° for pitch, roll and yaw, respectively. Translational deviations were more frequent than rotational ones, and frequency and magnitude of deviations showed an inverse correlation. Conclusion: Intrafractional motion detection and patient repositioning during spine SBRT using X-ray imaging via the ETD System can lead to improved safety during the application of high BED in critical locations. When using intrafractional imaging with low thresholds for re-positioning individualized immobilization devices (e.g. vacuum cushions) may be omitted.

The role of online MR-guided multi-fraction stereotactic ablative radiotherapy in lung tumours.

Background: The aim of this prospective observational study was to evaluate the dosimetry benefits, changes in pulmonary function, and clinical outcome of online adaptive MR-guided SBRT. Methods: From 11/2020-07/2022, 45 consecutive patients with 59 lesions underwent multi-fraction SBRT (3-8 fractions) at our institution. Patients were eligible if they had biopsy-proven NSCLC or lung cancer/metastases diagnosed via clinical imaging. Endpoints were local control (LC) and overall survival (OS). We evaluated PTV/GTV dose coverage, organs at risk exposure, and changes in pulmonary function (PF). Acute toxicity was classified per the National Cancer Institute-Common Terminology Criteria for Adverse Events version 5.0. Results: The median PTV was 14.4 cm3 (range: 3.4 - 96.5 cm3). In total 195/215 (91%) plans were reoptimised. In the reoptimised vs. predicted plans, PTV coverage by the prescribed dose increased in 94.6% of all fractions with a median increase in PTV VPD of 5.6% (range: -1.8 - 44.6%, p < 0.001), increasing the number of fractions with PTV VPD ≥ 95% from 33% to 98%. The PTV D95% and D98% (BED10) increased in 93% and 95% of all fractions with a median increase of 7.7% (p < 0.001) and 10.6% (p < 0.001). The PTV D95% (BED10) increased by a mean of 9.6 Gy (SD: 10.3 Gy, p < 0.001). At a median follow-up of 21.4 months (95% CI: 12.3-27.0 months), 1- and 2-year LC rates were 94.8% (95% CI: 87.6 - 100.0%) and 91.1% (95% CI: 81.3 - 100%); 1- and 2-year OS rates were 85.6% (95% CI: 75.0 - 96.3%) and 67.1 % (95% CI: 50.3 - 83.8%). One grade ≥ 3 toxicity and no significant reduction in short-term PF parameters were recorded. Conclusions: Online adaptive MR-guided SBRT is an effective, safe and generally well tolerated treatment option for lung tumours achieving encouraging local control rates with significantly improved target volume coverage.

Impact of daily plan adaptation on accumulated doses in ultra-hypofractionated magnetic resonance-guided radiation therapy of prostate cancer.

Background and purpose: Ultra-hypofractionated online adaptive magnetic resonance-guided radiotherapy (MRgRT) is promising for prostate cancer. However, the impact of online adaptation on target coverage and organ-at-risk (OAR) sparing at the level of accumulated dose has not yet been reported. Using deformable image registration (DIR)-based accumulation, we compared the delivered adapted dose with the simulated non-adapted dose. Materials and methods: Twenty-three prostate cancer patients treated at two clinics with 0.35 T magnetic resonance-guided linear accelerator (MR-linac) following the same treatment protocol (5 × 7.5 Gy with urethral sparing and daily adaptation) were included. The fraction MR images were deformably registered to the planning MR image. Both non-adapted and adapted fraction doses were accumulated with the corresponding vector fields. Two DIR approaches were implemented. PTV* (planning target volume minus urethra+2mm) D95%, CTV* (clinical target volume minus urethra) D98%, and OARs (urethra+2mm, bladder, and rectum) D0.2cc, were evaluated. Statistical significance was inferred from a two-tailed Wilcoxon signed-rank test (p < 0.05). Results: Normalized to the baseline, the accumulated PTV* D95% increased significantly by 2.7 % ([1.5, 4.3]%) through adaptation, and the CTV* D98% by 1.2 % ([0.1, 1.7]%). For the OARs after adaptation, accumulated bladder D0.2cc decreased by 0.4 % ([-1.2, 0.4]%), urethra+2mmD0.2cc by 0.8 % ([-1.6, -0.1]%), while rectum D0.2cc increased by 2.6 % ([1.2, 4.9]%). For all patients, rectum D0.2cc was still below the clinical constraint. Results of both DIR approaches differed on average by less than 0.2 %. Conclusions: Online adaptation in MRgRT improved target coverage and OARs sparing at the level of accumulated dose.

Repeatability quantification of brain diffusion-weighted imaging for future clinical implementation at a low-field MR-linac.

BACKGROUND: Longitudinal assessments of apparent diffusion coefficients (ADCs) derived from diffusion-weighted imaging (DWI) during intracranial radiotherapy at magnetic resonance imaging-guided linear accelerators (MR-linacs) could enable early response assessment by tracking tumor diffusivity changes. However, DWI pulse sequences are currently unavailable in clinical practice at low-field MR-linacs. Quantifying the in vivo repeatability of ADC measurements is a crucial step towards clinical implementation of DWI sequences but has not yet been reported on for low-field MR-linacs. This study assessed ADC measurement repeatability in a phantom and in vivo at a 0.35 T MR-linac. METHODS: Eleven volunteers and a diffusion phantom were imaged on a 0.35 T MR-linac. Two echo-planar imaging DWI sequence variants, emphasizing high spatial resolution ("highRes") and signal-to-noise ratio ("highSNR"), were investigated. A test-retest study with an intermediate outside-scanner-break was performed to assess repeatability in the phantom and volunteers' brains. Mean ADCs within phantom vials, cerebrospinal fluid (CSF), and four brain tissue regions were compared to literature values. Absolute relative differences of mean ADCs in pre- and post-break scans were calculated for the diffusion phantom, and repeatability coefficients (RC) and relative RC (relRC) with 95% confidence intervals were determined for each region-of-interest (ROI) in volunteers. RESULTS: Both DWI sequence variants demonstrated high repeatability, with absolute relative deviations below 1% for water, dimethyl sulfoxide, and polyethylene glycol in the diffusion phantom. RelRCs were 7% [5%, 12%] (CSF; highRes), 12% [9%, 22%] (CSF; highSNR), 9% [8%, 12%] (brain tissue ROIs; highRes), and 6% [5%, 7%] (brain tissue ROIs; highSNR), respectively. ADCs measured with the highSNR variant were consistent with literature values for volunteers, while smaller mean values were measured for the diffusion phantom. Conversely, the highRes variant underestimated ADCs compared to literature values, indicating systematic deviations. CONCLUSIONS: High repeatability of ADC measurements in a diffusion phantom and volunteers' brains were measured at a low-field MR-linac. The highSNR variant outperformed the highRes variant in accuracy and repeatability, at the expense of an approximately doubled voxel volume. The observed high in vivo repeatability confirms the potential utility of DWI at low-field MR-linacs for early treatment response assessment.

Auto-segmentation of pelvic organs at risk on 0.35T MRI using 2D and 3D Generative Adversarial Network models.

PURPOSE: Manual recontouring of targets and Organs At Risk (OARs) is a time-consuming and operator-dependent task. We explored the potential of Generative Adversarial Networks (GAN) to auto-segment the rectum, bladder and femoral heads on 0.35T MRIs to accelerate the online MRI-guided-Radiotherapy (MRIgRT) workflow. METHODS: 3D planning MRIs from 60 prostate cancer patients treated with 0.35T MR-Linac were collected. A 3D GAN architecture and its equivalent 2D version were trained, validated and tested on 40, 10 and 10 patients respectively. The volumetric Dice Similarity Coefficient (DSC) and 95th percentile Hausdorff Distance (HD95th) were computed against expert drawn ground-truth delineations. The networks were also validated on an independent external dataset of 16 patients. RESULTS: In the internal test set, the 3D and 2D GANs showed DSC/HD95th of 0.83/9.72 mm and 0.81/10.65 mm for the rectum, 0.92/5.91 mm and 0.85/15.72 mm for the bladder, and 0.94/3.62 mm and 0.90/9.49 mm for the femoral heads. In the external test set, the performance was 0.74/31.13 mm and 0.72/25.07 mm for the rectum, 0.92/9.46 mm and 0.88/11.28 mm for the bladder, and 0.89/7.00 mm and 0.88/10.06 mm for the femoral heads. The 3D and 2D GANs required on average 1.44 s and 6.59 s respectively to generate the OARs' volumetric segmentation for a single patient. CONCLUSIONS: The proposed 3D GAN auto-segments pelvic OARs with high accuracy on 0.35T, in both the internal and the external test sets, outperforming its 2D equivalent in both segmentation robustness and volume generation time.

Comparison of deep learning networks for fully automated head and neck tumor delineation on multi-centric PET/CT images.

OBJECTIVES: Deep learning-based auto-segmentation of head and neck cancer (HNC) tumors is expected to have better reproducibility than manual delineation. Positron emission tomography (PET) and computed tomography (CT) are commonly used in tumor segmentation. However, current methods still face challenges in handling whole-body scans where a manual selection of a bounding box may be required. Moreover, different institutions might still apply different guidelines for tumor delineation. This study aimed at exploring the auto-localization and segmentation of HNC tumors from entire PET/CT scans and investigating the transferability of trained baseline models to external real world cohorts. METHODS: We employed 2D Retina Unet to find HNC tumors from whole-body PET/CT and utilized a regular Unet to segment the union of the tumor and involved lymph nodes. In comparison, 2D/3D Retina Unets were also implemented to localize and segment the same target in an end-to-end manner. The segmentation performance was evaluated via Dice similarity coefficient (DSC) and Hausdorff distance 95th percentile (HD95). Delineated PET/CT scans from the HECKTOR challenge were used to train the baseline models by 5-fold cross-validation. Another 271 delineated PET/CTs from three different institutions (MAASTRO, CRO, BERLIN) were used for external testing. Finally, facility-specific transfer learning was applied to investigate the improvement of segmentation performance against baseline models. RESULTS: Encouraging localization results were observed, achieving a maximum omnidirectional tumor center difference lower than 6.8 cm for external testing. The three baseline models yielded similar averaged cross-validation (CV) results with a DSC in a range of 0.71-0.75, while the averaged CV HD95 was 8.6, 10.7 and 9.8 mm for the regular Unet, 2D and 3D Retina Unets, respectively. More than a 10% drop in DSC and a 40% increase in HD95 were observed if the baseline models were tested on the three external cohorts directly. After the facility-specific training, an improvement in external testing was observed for all models. The regular Unet had the best DSC (0.70) for the MAASTRO cohort, and the best HD95 (7.8 and 7.9 mm) in the MAASTRO and CRO cohorts. The 2D Retina Unet had the best DSC (0.76 and 0.67) for the CRO and BERLIN cohorts, and the best HD95 (12.4 mm) for the BERLIN cohort. CONCLUSION: The regular Unet outperformed the other two baseline models in CV and most external testing cohorts. Facility-specific transfer learning can potentially improve HNC segmentation performance for individual institutions, where the 2D Retina Unets could achieve comparable or even better results than the regular Unet.

Reduction of cone-beam CT artifacts in a robotic CBCT device using saddle trajectories with integrated infrared tracking.

BACKGROUND: Cone beam computed tomography (CBCT) is widely used in many medical fields. However, conventional CBCT circular scans suffer from cone beam (CB) artifacts that limit the quality and reliability of the reconstructed images due to incomplete data. PURPOSE: Saddle trajectories in theory might be able to improve the CBCT image quality by providing a larger region with complete data. Therefore, we investigated the feasibility and performance of saddle trajectory CBCT scans and compared them to circular trajectory scans. METHODS: We performed circular and saddle trajectory scans using a novel robotic CBCT scanner (Mobile ImagingRing (IRm); medPhoton, Salzburg, Austria). For the saddle trajectory, the gantry executed yaw motion up to ± 10 ∘ $\pm 10^{\circ }$ using motorized wheels driving on the floor. An infrared (IR) tracking device with reflective markers was used for online geometric calibration correction (mainly floor unevenness). All images were reconstructed using penalized least-squares minimization with the conjugate gradient algorithm from RTK with 0.5 × 0.5 × 0.5 mm 3 $0.5 \times 0.5\times 0.5 \text{ mm}^3$ voxel size. A disk phantom and an Alderson phantom were scanned to assess the image quality. Results were correlated with the local incompleteness value represented by tan ( ψ ) $\tan (\psi)$ , which was calculated at each voxel as a function of the source trajectory and the voxel's 3D coordinates. We assessed the magnitude of CB artifacts using the full width half maximum (FWHM) of each disk profile in the axial center of the reconstructed images. Spatial resolution was also quantified by the modulation transfer function at 10% (MTF10). RESULTS: When using the saddle trajectory, the region without CB artifacts was increased from 43 to 190 mm in the SI direction compared to the circular trajectory. This region coincided with low values for tan ( ψ ) $\tan (\psi)$ . When tan ( ψ ) $\tan (\psi)$ was larger than 0.02, we found there was a linear relationship between the FWHM and tan ( ψ ) $\tan (\psi)$ . For the saddle, IR tracking allowed the increase of MTF10 from 0.37 to 0.98 lp/mm. CONCLUSIONS: We achieved saddle trajectory CBCT scans with a novel CBCT system combined with IR tracking. The results show that the saddle trajectory provides a larger region with reliable reconstruction compared to the circular trajectory. The proposed method can be used to evaluate other non-circular trajectories.

Real-time motion management in MRI-guided radiotherapy: Current status and AI-enabled prospects.

MRI-guided radiotherapy (MRIgRT) is a highly complex treatment modality, allowing adaptation to anatomical changes occurring from one treatment day to the other (inter-fractional), but also to motion occurring during a treatment fraction (intra-fractional). In this vision paper, we describe the different steps of intra-fractional motion management during MRIgRT, from imaging to beam adaptation, and the solutions currently available both clinically and at a research level. Furthermore, considering the latest developments in the literature, a workflow is foreseen in which motion-induced over- and/or under-dosage is compensated in 3D, with minimal impact to the radiotherapy treatment time. Considering the time constraints of real-time adaptation, a particular focus is put on artificial intelligence (AI) solutions as a fast and accurate alternative to conventional algorithms.

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

BACKGROUND: 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 collimator tracking. PURPOSE: 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 individual cases. METHODS: 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 for significance. RESULTS: 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 testing cases. CONCLUSIONS: 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 lung cases.

Intra-frame motion deterioration effects and deep-learning-based compensation in MR-guided radiotherapy.

BACKGROUND: Current commercially available hybrid magnetic resonance linear accelerators (MR-Linac) use 2D+t cine MR imaging to provide intra-fractional motion monitoring. However, given the limited temporal resolution of cine MR imaging, target intra-frame motion deterioration effects, resulting in effective time latency and motion artifacts in the image domain, can be appreciable, especially in the case of fast breathing. PURPOSE: The aim of this work is to investigate intra-frame motion deterioration effects in MR-guided radiotherapy (MRgRT) by simulating the motion-corrupted image acquisition, and to explore the feasibility of deep-learning-based compensation approaches, relying on the intra-frame motion information which is spatially and temporally encoded in the raw data (k-space). METHODS: An intra-frame motion model was defined to simulate motion-corrupted MR images, with 4D anthropomorphic digital phantoms being exploited to provide ground truth 2D+t cine MR sequences. A total number of 10 digital phantoms were generated for lung cancer patients, with randomly selected eight patients for training or validation and the remaining two for testing. The simulation code served as the data generator, and a dedicated motion pattern perturbation scheme was proposed to build the intra-frame motion database, where three degrees of freedom were designed to guarantee the diversity of intra-frame motion trajectories, enabling a thorough exploration in the domain of the potential anatomical structure positions. U-Nets with three types of loss functions: L1 or L2 loss defined in image or Fourier domain, referred to as NNImgLoss-L1 , NNFloss-L1 and NNL2-Loss were trained to extract information from the motion-corrupted image and used to estimate the ground truth final-position image, corresponding to the end of the acquisition. Images before and after compensation were evaluated in terms of (i) image mean-squared error (MSE) and mean absolute error (MAE), and (ii) accuracy of gross tumor volume (GTV) contouring, based on optical-flow image registration. RESULTS: Image degradation caused by intra-frame motion was observed: for a linearly and fully acquired Cartesian readout k-space trajectory, intra-frame motion resulted in an imaging latency of approximately 50% of the acquisition time; in comparison, the motion artifacts exhibited only a negligible contribution to the overall geometric errors. All three compensation models led to a decrease in image MSE/MAE and GTV position offset compared to the motion-corrupted image. In the investigated testing dataset for GTV contouring, the average dice similarity coefficients (DSC) improved from 88% to 96%, and the 95th percentile Hausdorff distance (HD95 ) dropped from 4.8 mm to 2.1 mm. Different models showed slight performance variations across different intra-frame motion amplitude categories: NNImgLoss-L1 excelled for small/medium amplitudes, whereas NNFloss-L1 demonstrated higher DSC median values at larger amplitudes. The saliency maps of the motion-corrupted image highlighted the major contribution of the later acquired k-space data, as well as the edges of the moving anatomical structures at their final positions, during the model inference stage. CONCLUSIONS: Our results demonstrate the deep-learning-based approaches have the potential to compensate for intra-frame motion by utilizing the later acquired data to drive the convergence of the earlier acquired k-space components.

Prior knowledge based deep learning auto-segmentation in magnetic resonance imaging-guided radiotherapy of prostate cancer.

Background and purpose: Automation is desirable for organ segmentation in radiotherapy. This study compared deep learning methods for auto-segmentation of organs-at-risk (OARs) and clinical target volume (CTV) in prostate cancer patients undergoing fractionated magnetic resonance (MR)-guided adaptive radiation therapy. Models predicting dense displacement fields (DDFMs) between planning and fraction images were compared to patient-specific (PSM) and baseline (BM) segmentation models. Materials and methods: A dataset of 92 patients with planning and fraction MR images (MRIs) from two institutions were used. DDFMs were trained to predict dense displacement fields (DDFs) between the planning and fraction images, which were subsequently used to propagate the planning contours of the bladder, rectum, and CTV to the daily MRI. The training was performed either with true planning-fraction image pairs or with planning images and their counterparts deformed by known DDFs. The BMs were trained on 53 planning images, while to generate PSMs, the BMs were fine-tuned using the planning image of a given single patient. The evaluation included Dice similarity coefficient (DSC), the average (HDavg) and the 95th percentile (HD95) Hausdorff distance (HD). Results: The DDFMs with DSCs for bladder/rectum of 0.76/0.76 performed worse than PSMs (0.91/0.90) and BMs (0.89/0.88). The same trend was observed for HDs. For CTV, DDFM and PSM performed similarly yielding DSCs of 0.87 and 0.84, respectively. Conclusions: DDFMs were found suitable for CTV delineation after rigid alignment. However, for OARs they were outperformed by PSMs, as they predicted only limited deformations even in the presence of substantial anatomical changes.

The role of artificial intelligence in radiotherapy clinical practice.

This review article visits the current state of artificial intelligence (AI) in radiotherapy clinical practice. We will discuss how AI has a place in the modern radiotherapy workflow at the level of automatic segmentation and planning, two applications which have seen real-work implementation. A special emphasis will be placed on the role AI can play in online adaptive radiotherapy, such as performed at MR-linacs, where online plan adaptation is a procedure which could benefit from automation to reduce on-couch time for patients. Pseudo-CT generation and AI for motion tracking will be introduced in the scope of online adaptive radiotherapy as well. We further discuss the use of AI for decision-making and response assessment, for example for personalized prescription and treatment selection, risk stratification for outcomes and toxicities, and AI for quantitative imaging and response assessment. Finally, the challenges of generalizability and ethical aspects will be covered. With this, we provide a comprehensive overview of the current and future applications of AI in radiotherapy.

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

BACKGROUND: 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. PURPOSE: 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. METHODS: Experiments were performed on a prototype 1.0 T 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 4 Hz 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. RESULTS: 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%. CONCLUSIONS: 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 with MLC-tracking.

Continuous time-resolved estimated synthetic 4D-CTs for dose reconstruction of lung tumor treatments at a 0.35 T MR-linac.

Objective.To experimentally validate a method to create continuous time-resolved estimated synthetic 4D-computed tomography datasets (tresCTs) based on orthogonal cine MRI data for lung cancer treatments at a magnetic resonance imaging (MRI) guided linear accelerator (MR-linac).Approach.A breathing porcine lung phantom was scanned at a CT scanner and 0.35 T MR-linac. Orthogonal cine MRI series (sagittal/coronal orientation) at 7.3 Hz, intersecting tumor-mimicking gelatin nodules, were deformably registered to mid-exhale 3D-CT and 3D-MRI datasets. The time-resolved deformation vector fields were extrapolated to 3D and applied to a reference synthetic 3D-CT image (sCTref), while accounting for breathing phase-dependent lung density variations, to create 82 s long tresCTs at 3.65 Hz. Ten tresCTs were created for ten tracked nodules with different motion patterns in two lungs. For each dataset, a treatment plan was created on the mid-exhale phase of a measured ground truth (GT) respiratory-correlated 4D-CT dataset with the tracked nodule as gross tumor volume (GTV). Each plan was recalculated on the GT 4D-CT, randomly sampled tresCT, and static sCTrefimages. Dose distributions for corresponding breathing phases were compared in gamma (2%/2 mm) and dose-volume histogram (DVH) parameter analyses.Main results.The mean gamma pass rate between all tresCT and GT 4D-CT dose distributions was 98.6%. The mean absolute relative deviations of the tresCT with respect to GT DVH parameters were 1.9%, 1.0%, and 1.4% for the GTVD98%,D50%, andD2%, respectively, 1.0% for the remaining nodulesD50%, and 1.5% for the lungV20Gy. The gamma pass rate for the tresCTs was significantly larger (p< 0.01), and the GTVD50%deviations with respect to the GT were significantly smaller (p< 0.01) than for the sCTref.Significance.The results suggest that tresCTs could be valuable for time-resolved reconstruction and intrafractional accumulation of the dose to the GTV for lung cancer patients treated at MR-linacs in the future.