Patient-specific quality assurance strategies for synthetic computed tomography in magnetic resonance-only radiotherapy of the abdomen

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.

Automated hallucination detection for synthetic CT images used in MR-only radiotherapy workflows

Objective. Artificial intelligence (AI)-generated synthetic CT (sCT) images have become commercially available to provide electron densities and reference anatomies in MR-only radiotherapy workflows. However, hallucinations (false regions of bone or air) introduced in AI-generated sCT images may affect the accuracy of dose calculation and patient setup verification. We developed a tool to detect bone hallucinations and/or inaccuracies in AI-generated pelvic sCT images used in MR-only workflows.Approach. A deep learning auto segmentation (DLAS) model was trained to auto-segment bone on MR images. The model was implemented with a 3D SegResNet network architecture using the MONAI framework with a training dataset of 86 Dixon MR image sets paired with their corresponding ground truth contours derived from planning CT images deformed to the MR images. The model performance was then assessed on an independent testing dataset (n= 10).Main results. The DLAS model-based hallucination screener identified hallucinations in bone structures using daily MR images and accurately flagged these regions on sCT images. The sensitivity of the screener is adjustable based on the distance of discrepancies between bone regions derived from sCT to bone contours generated by the DLAS. The average specificity of the DLAS model was 0.78, 0.93 and 0.98 for distance parameters of 0.8, 1.0 and 1.2 cm, respectively. The screener identified false high-density hallucination regions in the abdomen of AI-generated sCT images for all testing patients, highlighting potential issues with the training data used for the AI sCT model.Significance. A hallucination screener for AI-generated pelvic sCT images was developed and implemented for routine clinical use. The screener serves as an important quality assurance tool for MR-only radiotherapy workflows. By identifying potential AI-generated errors, the hallucination screener may improve the safety and accuracy of sCT images used for dose calculation and image guidance.

Brain MR-only workflow in clinical practice: A comparison among generators for quality assurance and patient positioning

BACKGROUND AND PURPOSE

Routine quality control procedures are still required for sCT based on artificial intelligence (AI) to verify the performance of the generators. The aim of this study was to evaluate three generators based on AI or bulk density (BD) assignment for the patient-specific quality assurance (PSQA) of another AI-based generator in clinical routine. A patient positioning study based on 2D/2D kV-image comparing the performances of four sCT generators was also performed.

MATERIALS AND METHODS

On the four generators available commercially at our institution, one was chosen as the clinical one, and the three others were used for PSQA. Several dose metrics were calculated like the mean error, dose-volume histogram metrics, and 1%/1 mm gamma analysis. A comparison against CT was considered as a reference. Translations and rotations found during patient positioning based on sCT were compared to those based on CT.

RESULTS

Some of the metrics calculated against CT revealed patients outside the tolerances chosen (1% for point metrics; 90% for gamma pass rate). None of the generators was able to identify these outliers for all metrics studied. Performing a PSQA with other sCT generators introduced several false positives and false negatives. None of the generators was able to clearly identify, for all metrics studied, a true sCT failure caused by a metal implant. The smallest positioning deviations were found for the BD assignment sCT, the largest for the only AI generator not based on a T1 Dixon MR sequence.

CONCLUSIONS

PSQA of a sCT generator with another sCT generator should be performed with great care. Patient positioning is an important aspect to consider when evaluating a sCT generator. The results of this study should help medical physicists willing to set up a MR-only workflow for the brain based on a 2D/2D kV-image patient positioning.

Results of 2023 survey on the use of synthetic computed tomography for magnetic resonance Imaging-only radiotherapy: Current status and future steps

Background and purpose

The emergence of synthetic CT (sCT) in MR-guided radiotherapy (MRgRT) represents a significant advancement, supporting MR-only workflows and online treatment adaptation. However, the lack of consensus guidelines has led to varied practices. This study reports results from a 2023 ESTRO survey aimed at defining current practices in sCT development and use.

Materials and methods

An survey was distributed to ESTRO members, including 98 questions across four sections on sCT algorithm generation and usage. By June 2023, 100 centers participated. The survey revealed diverse clinical experiences and roles, with primary sCT use in the pelvis (60%), brain (15%), abdomen (11%), thorax (8%), and head-and-neck (6%). sCT was mostly used for conventional fractionation treatments (68%), photon SBRT (40%), and palliative cases (28%), with limited use in proton therapy (4%).

Results

Conditional GANs and GANs were the most used neural network architectures, operating mainly on 1.5 T and 3 T MRI images. Less than half used paired images for training, and only 20% performed image selection. Key MR image quality parameters included magnetic field homogeneity and spatial integrity. Half of the respondents lacked a dedicated sCT-QA program, and many did not apply sanitychecks before calculation. Selection strategies included age, weight, and metal artifacts. A strong consensus (95%) emerged for vendor neutral guidelines.

Conclusion

The survey highlights the need for expert-based, vendor-neutral guidelines to standardize sCT tools, metrics, and clinical protocols, ensuring effective sCT use in MR-guided radiotherapy.

A Deep Learning Approach for the Fast Generation of Synthetic Computed Tomography from Low-Dose Cone Beam Computed Tomography Images on a Linear Accelerator Equipped with Artificial Intelligence

Artificial Intelligence (AI) is revolutionising many aspects of radiotherapy (RT), opening scenarios that were unimaginable just a few years ago. The aim of this study is to propose a Deep Leaning (DL) approach able to quickly generate synthetic Computed Tomography (CT) images from low-dose Cone Beam CT (CBCT) acquired on a modern linear accelerator integrating AI. Methods: A total of 53 patients treated in the pelvic region were enrolled and split into training (30), validation (9), and testing (14). A Generative Adversarial Network (GAN) was trained for 200 epochs. The image accuracy was evaluated by calculating the mean and mean absolute error (ME and ME) between sCT and CT. RT treatment plans were calculated on CT and sCT images, and dose accuracy was evaluated considering Dose Volume Histogram (DVH) and gamma analysis. Results: A total of 4507 images were selected for training. The MAE and ME values in the test set were 36 ± 6 HU and 7 ± 6 HU, respectively. Mean gamma passing rates for 1%/1 mm, 2%/2 mm, and 3%/3 mm tolerance criteria were respectively 93.5 ± 3.4%, 98.0 ± 1.3%, and 99.2 ± 0.7%, with no difference between curative and palliative cases. All the DVH parameters analysed were within 1 Gy of the difference between sCT and CT. Conclusion: This study demonstrated that sCT generation using the DL approach is feasible on low-dose CBCT images. The proposed approach can represent a valid tool to speed up the online adaptive procedure and remove CT simulation from the RT workflow.