Using in vivo EPID images to detect and quantify patient anatomy changes with gradient dose segmented analysis

Rapid in vivo EPID image prediction using a combination of analytically calculated attenuation and AI predicted scatter

BACKGROUND

The electronic portal imaging device (EPID) can be used in vivo, to detect on-treatment errors by evaluating radiation exiting a patient. To detect deviations from the planning intent, image predictions need to be modeled based on the patient's anatomy and plan information. To date in vivo transit images have been predicted using Monte Carlo (MC) algorithms. A deep learning approach can make predictions faster than MC and only requires patient information for training.

PURPOSE

To test the feasibility and reliability of creating a deep-learning model with patient data for predicting in vivo EPID images for IMRT treatments.

METHODS

In our approach, the in vivo EPID image was separated into contributions from primary and scattered photons. A primary photon attenuation function was determined by measuring attenuation factors for various thicknesses of solid water. The scatter component of in vivo EPID images was estimated using a convolutional neural network (CNN). The CNN input was a 3-channel image comprised of the non-transit EPID image and ray tracing projections through a pretreatment CBCT. The predicted scatter component was added to the primary attenuation component to give the full predicted in vivo EPID image. We acquired 193 IMRT fields/images from 93 patients treated on the Varian Halcyon. Model training:validation:test dataset ratios were 133:20:40 images. Additional patient plans were delivered to anthropomorphic phantoms, yielding 75 images for further validation. We assessed model accuracy by comparing model-calculated and measured in vivo images with a gamma comparison.

RESULTS

Comparing the model-calculated and measured in vivo images gives a mean gamma pass rate for the training:validation:test datasets of 95.4%:94.1%:92.9% for 3%/3 mm and 98.4%:98.4%:96.8% for 5%/3 mm. For images delivered to phantom data sets the average gamma pass rate was 96.4% (3%/3 mm criteria). In all data sets, the lower passing rates of some images were due to CBCT artifacts and patient motion that occurred between the time of CBCT and treatment.  CONCLUSIONS: The developed deep-learning-based model can generate in vivo EPID images with a mean gamma pass rate greater than 92% (3%/3 mm criteria). This approach provides an alternative to MC prediction algorithms. Image predictions can be made in 30 ms on a standard GPU. In future work, image predictions from this model can be used to detect in vivo treatment errors and on-treatment changes in patient anatomy, providing an additional layer of patient-specific quality assurance.

Preliminary application of EPID three-dimensional dose reconstruction in in vivo dose verification of breast cancer intensity-modulated radiation therapy

A preliminary study was conducted using electronic portal imaging device (EPID) based dose verification in pre-treatment and in vivo dose reconstruction modes for breast cancer intensity-modulated radiation therapy (IMRT) technique with known repositioning set-up errors. For 43 IMRT plans, the set-up errors were determined from 43 sets of EPID images and 258 sets of cone beam computed tomography images. In-house developed Edose software was used to reconstruct the dose distribution using the pre-treatment and on-treatment (in vivo) EPID acquired fluence maps. The maximum setup error was < 3.5 mm. For 43 pre-treatment cases, the γ pass rate (3 %/3 mm) is 98.49 % ± 1.15 %. The chest wall target ΔV98%P, ΔV95%P, andΔV90%P are all < 5 %, while the majority of the ipsilateral lung ΔV5Gy, ΔV20Gy, and ΔV30Gy are also < 5 %. For 258 in vivo cases, the γ pass rate is 90.98 % ± 6.53 %, with the chest wall target ΔV90%P and ipsilateral lung ΔV30Gy both < 5 %, while the other volume differences all exceed 5 %. The γ pass rate for in vivo verification is significantly lower than pre-treatment values. Although the in vivo γ verification satisfies the medical physics requirements, the reconstructed coverage of the chest wall target is far below the clinical dosimetry requirements. In vivo 3D dose reconstruction directly predicts changes in the planning target volume to aid clinicians better understand the actual dose received by patients with intra-fractional motion and anatomical changes.

In vivo EPID-based daily treatment error identification for volumetric-modulated arc therapy in head and neck cancers with a hierarchical convolutional neural network: a feasibility study

We proposed a deep learning approach to classify various error types in daily VMAT treatment of head and neck cancer patients based on EPID dosimetry, which could provide additional information to support clinical decisions for adaptive planning. 146 arcs from 42 head and neck patients were analyzed. Anatomical changes and setup errors were simulated in 17,820 EPID images of 99 arcs obtained from 30 patients using in-house software for model training, validation, and testing. Subsequently, 141 clinical EPID images from 47 arcs belonging to the remaining 12 patients were utilized for clinical testing. The hierarchical convolutional neural network (HCNN) model was trained to classify error types and magnitudes using EPID dose difference maps. Gamma analysis with 3%/2 mm (dose difference/distance to agreement) criteria was also performed. The F1 score, a combination of precision and recall, was utilized to evaluate the performance of the HCNN model and gamma analysis. The adaptive fractioned doses were calculated to verify the HCNN classification results. For error type identification, the overall F1 score of the HCNN model was 0.99 and 0.91 for primary type and subtype identification, respectively. For error magnitude identification, the overall F1 score in the simulation dataset was 0.96 and 0.70 for the HCNN model and gamma analysis, respectively; while the overall F1 score in the clinical dataset was 0.79 and 0.20 for the HCNN model and gamma analysis, respectively. The HCNN model-based EPID dosimetry can identify changes in patient transmission doses and distinguish the treatment error category, which could potentially provide information for head and neck cancer treatment adaption.

IPEM topical report: guidance for the clinical implementation of online treatment monitoring solutions for IMRT/VMAT

This report provides guidance for the implementation of online treatment monitoring (OTM) solutions in radiotherapy (RT), with a focus on modulated treatments. Support is provided covering the implementation process, from identification of an OTM solution to local implementation strategy. Guidance has been developed by a RT special interest group (RTSIG) working party (WP) on behalf of the Institute of Physics and Engineering in Medicine (IPEM). Recommendations within the report are derived from the experience of the WP members (in consultation with manufacturers, vendors and user groups), existing guidance or legislation and a UK survey conducted in 2020 (Stevenset al2021). OTM is an inclusive term representing any system capable of providing a direct or inferred measurement of the delivered dose to a RT patient. Information on each type of OTM is provided but, commensurate with UK demand, guidance is largely influenced byin vivodosimetry methods utilising the electronic portal imager device (EPID). Sections are included on the choice of OTM solutions, acceptance and commissioning methods with recommendations on routine quality control, analytical methods and tolerance setting, clinical introduction and staffing/resource requirements. The guidance aims to give a practical solution to sensitivity and specificity testing. Functionality is provided for the user to introduce known errors into treatment plans for local testing. Receiver operating characteristic analysis is discussed as a tool to performance assess OTM systems. OTM solutions can help verify the correct delivery of radiotherapy treatment. Furthermore, modern systems are increasingly capable of providing clinical decision-making information which can impact the course of a patient's treatment. However, technical limitations persist. It is not within the scope of this guidance to critique each available solution, but the user is encouraged to carefully consider workflow and engage with manufacturers in resolving compatibility issues.

The application of gradient dose segmented analysis of in-vivo EPID images for patients undergoing VMAT in a resource-constrained environment

The gamma analysis metric is a commonly used metric for VMAT plan evaluation. The major drawback of this is the lack of correlation between gamma passing rates and DVH values. The novel GDSA_mean metric was developed by Steers et al. to quantify changes in the PTV mean dose (D_mean ) for VMAT patients. The aim of this work is to apply the GDSA retrospectively on head-and-neck cancer patients treated on the newly acquired Varian Halcyon, to assess changes in GDSA_mean , and to evaluate the cause of day-to-day changes in the time-plot series. In-vivo EPID transmission images of head-and-neck cancer patients treated between August 2019 and July 2020 were analyzed retrospectively. The GDSA_mean was determined for all patients treated. The changes in patient anatomy and rotational errors were quantified using the daily CBCT images and added to a time-plot with the daily change in GDSA_mean . Over 97% of the delivered treatment fractions had a GDSA_mean  < 3%. Thirteen of the patients received at least one treatment fraction where the GDSA_mean  > 3%. Most of these deviations occurred for the later fractions of radiotherapy treatment. Additionally, 92% of these patients were treated for malignancies involving the larynx and oropharynx. Notable deviations in the effective separation diameters were observed for 62% of the patients where the change in GDSA_mean  > 3%. For the other five cases with GDSA_mean  < 3%, the mean pitch, roll, and yaw rotational errors were 0.90°, 0.45°, and 0.43°, respectively. A GDSA_mean  > 3% was more likely due to a change in separation, whereas a GDSA_mean  < 3% was likely caused by rotational errors. Pitch errors were shown to be the most dominant. The GDSA_mean is easily implementable and can aid in scheduling new CT scans for patients before significant deviations in dose delivery occur.

Predicted Inferior Outcomes for Lung SBRT With Treatment Planning Systems That Fail Independent Phantom-Based Audits

PURPOSE

More than 15% of radiation therapy clinics fail external audits with anthropomorphic phantoms conducted by Imaging and Radiation Oncology Core-Houston (IROC-H) while passing other industry-standard quality assurance (QA) tests. We seek to evaluate the predicted effect of such failed plans on outcomes for patients treated with stereotactic body radiation therapy (SBRT) for lung tumors.

METHODS AND MATERIALS

We conducted a retrospective study of 55 patients treated with SBRT for lung tumors with a prescription biologically equivalent dose (BED)₁₀ ≥100 Gy using a treatment planning system (TPS) that passed IROC-H phantom audits. Sample linear accelerator beam models with introduced errors were commissioned by varying the multileaf collimator leaf-tip offset parameter (ie, dosimetric leaf gap) over the range ±1.0 mm relative to the validated model. These models mimic TPS that pass internal QA measures but fail IROC-H tests. Patient plans were recalculated on sample beam models. The predicted tumor control probability (TCP) and normal tissue complication probability (NTCP) were calculated based on published dose-response models.

RESULTS

A leaf-tip offset value of -1.0 mm decreased the fraction of plans receiving a planning treatment volume of BED₁₀ ≥100 Gy from 95% to 27%. This translated to a significant decrease in 2-year TCP of 4.8% (95% CI: 2.0%-5.5%) with a decrease in TCP up to 21%. Conversely, a leaf-tip offset of +1.0 mm resulted in 36% of patients exceeding previously met organs at risk (OAR) constraints, including 2 instances of spinal cord and brachial plexus overdoses and a small increase in chest wall NTCP of 0.7%, (95% CI: 0.5%-0.8%).

CONCLUSIONS

Simulated treatment plans with modest MLC leaf offsets result in lung SBRT plans that significantly underdose tumor or exceed OAR constraints. These dosimetric endpoints translate to significant detriments in TCP. These simulated plans mimic planning systems that pass internal QA measures but fail independent phantom-based tests, underscoring the need for enhanced quality assurance including external audits of TPS.

Using eclipse scripting to fully automate in-vivo image analysis to improve treatment quality and safety

PURPOSE

An automated, in-vivo system to detect patient anatomy changes and machine output was developed using novel analysis of in-vivo electronic portal imaging device (EPID) images for every fraction of treatment on a Varian Halcyon. In-vivo approach identifies errors that go undetected by routine quality assurance (QA) to compliment daily machine performance check (MPC), with minimal physicist workload.

METHODS

Images for all fractions treated on a Halcyon were automatically downloaded and analyzed at the end of treatment day. For image analysis, compared to first fraction, the mean difference of high-dose region of interest is calculated. This metric has shown to predict changes in planning treatment volume (PTV) mean dose. Flags are raised for: (Type-A) treatment fraction whose mean difference exceeds 10%, to protect against large errors, and (Type-B) patients with three consecutive fractions with mean exceeding ±3%, to protect against systematic trends. If a threshold is exceeded, a physicist is e-mailed, a report for flagged patients, for investigation. To track machine output changes, for all patients treated on a day, the average and standard deviations are uploaded to a QA portal, along with the reviewed MPC, ensuring comprehensive QA for the Halcyon. To guide clinical implementation, a retrospective study from November 2017 till December 2020 was conducted, which grouped errors by treatment site. This framework has been used prospectively since January 2021.

RESULTS

From retrospective data of 1633 patients (35 759 fractions), no Type-A errors were found and only 45 patients (2.76%) had Type-B errors. These Type-B deviations were due to head-and-neck weight loss. For 6 months of prospective use (345 patients), 13 patients (3.7%) had Type-B errors and no Type-A errors.

CONCLUSIONS

This automated system protects against errors that can occur in vivo to provide a more comprehensive QA. This fully automated framework can be implemented in other centers with a Halcyon, requiring a desktop computer and analysis scripts.

Assessment of predictive indicators of acute gastrointestinal toxicity using in vivo transmission images

PURPOSE

For pelvic and abdominal treatments, excess dose to the bowel can result in acute toxicities. Current estimates of bowel toxicity are based on pre-treatment dose-volume histogram data. However, the actual dose the bowel receives depends on interfraction variations, such as patient anatomy changes. We propose a method to model bowel toxicities, incorporating in vivo patient information using transit electronic portal imaging device (EPID) images.

METHODS AND MATERIALS

For 63 patients treated to the lower thorax, abdomen, or pelvis on the Varian Halcyon, weekly chart review was performed to obtain incidences of grade 2 or higher toxicity, RTOG scale. Twenty patients presented with acute gastrointestinal (GI) toxicity. All patients were treated with conventional fractionation. For each treatment plan, the absolute volume dose-volume histogram of the bowel was exported and analyzed. Additionally, for each fraction of treatment, in vivo EPID images were collected and used to estimate the change in radiation transmission during the course of treatment. A logistic model was used to test correlations between acute GI toxicity and bowel dosimetric parameters as well as metrics obtained from in vivo image measurements. After performing the fit to the in vivo EPID data, the bootstrap resampling method was used to create confidence intervals. In vivo EPID image metrics from an additional 42 patients treated to the lower thorax, abdomen, or pelvis were used to validate the logistic model fit.

RESULTS

The incidence of toxicity versus the volume of 40 Gy to the bowel space was fitted with a logistic function, which was superior to an average model (p < 0.0001) and agrees with previously published models. For the initial in vivo EPID data, the incidence of toxicity versus the sum of in vivo transmission measurements showed marginal significance after 15 fractions (p = 0.10) of treatment and a significance of p = 0.038 is seen at the 20th fraction, when compared to an average model. For the validation data set, the logistic model of the in vivo transmission measurement after 20 fractions was superior to the average model (p = 0.043), with the model falling within the 68% confidence interval of the fit of the initial data set.

CONCLUSIONS

Dose-volume constraints to reduce the incidence of acute GI toxicity have been validated. The presented novel EPID transmission-based metric can be used to identify GI toxicity as patients progress through treatment.