Development and multi-institutional validation of a convolutional neural network to detect vertebral body mis-alignments in 2D x-ray setup images.

BACKGROUND: Misalignment to the incorrect vertebral body remains a rare but serious patient safety risk in image-guided radiotherapy (IGRT). PURPOSE: Our group has proposed that an automated image-review algorithm be inserted into the IGRT process as an interlock to detect off-by-one vertebral body errors. This study presents the development and multi-institutional validation of a convolutional neural network (CNN)-based approach for such an algorithm using patient image data from a planar stereoscopic x-ray IGRT system. METHODS: X-rays and digitally reconstructed radiographs (DRRs) were collected from 429 spine radiotherapy patients (1592 treatment fractions) treated at six institutions using a stereoscopic x-ray image guidance system. Clinically-applied, physician approved, alignments were used for true-negative, "no-error" cases. "Off-by-one vertebral body" errors were simulated by translating DRRs along the spinal column using a semi-automated method. A leave-one-institution-out approach was used to estimate model accuracy on data from unseen institutions as follows: All of the images from five of the institutions were used to train a CNN model from scratch using a fixed network architecture and hyper-parameters. The size of this training set ranged from 5700 to 9372 images, depending on exactly which five institutions were contributing data. The training set was randomized and split using a 75/25 split into the final training/ validation sets. X-ray/ DRR image pairs and the associated binary labels of "no-error" or "shift" were used as the model input. Model accuracy was evaluated using images from the sixth institution, which were left out of the training phase entirely. This test set ranged from 180 to 3852 images, again depending on which institution had been left out of the training phase. The trained model was used to classify the images from the test set as either "no-error" or "shifted", and the model predictions were compared to the ground truth labels to assess the model accuracy. This process was repeated until each institution's images had been used as the testing dataset. RESULTS: When the six models were used to classify unseen image pairs from the institution left out during training, the resulting receiver operating characteristic area under the curve values ranged from 0.976 to 0.998. With the specificity fixed at 99%, the corresponding sensitivities ranged from 61.9% to 99.2% (mean: 77.6%). With the specificity fixed at 95%, sensitivities ranged from 85.5% to 99.8% (mean: 92.9%). CONCLUSION: This study demonstrated the CNN-based vertebral body misalignment model is robust when applied to previously unseen test data from an outside institution, indicating that this proposed additional safeguard against misalignment is feasible.

Modeling of kyphoplasty cement for accurate dose calculations.

We have determined the optimal method for modeling kyphoplasty cement to enable accurate dose calculations in the Eclipse treatment planning system (TPS). The cement studied (Medtronic Kyphon HV-R®) consists of 30% Barium, 68% polymethylmethacrylate (PMMA), and 2% benzoyl peroxide, formulated to be radiopaque with kV imaging systems. Neither Barium nor PMMA have a high physical density, resulting in different interaction characteristics for megavoltage treatment beams compared to kV imaging systems. This can lead to significant calculation errors if density mapping is performed using a standard CT number to density curve. To properly characterize the cement for dose calculation, we 3D printed a hemi-cylindrical container to fit adjacent to a micro-chamber insert for an anthropomorphic phantom, and filled the container with Kyphon cement. We CT scanned the combination, modeled the cement with multiple material assignments in the TPS, designed plans with different field sizes and beam geometry for five photon modes, and measured the doses for all plans. All photon energies show significant error in calculated dose when the cement is modeled based on the CT number. Of the material assignments we evaluated, polytetrafluoroethylene (PTFE) showed the best overall agreement with measurement. Calculated and measured doses agree within 3.5% for a 340-degree arc technique (which averages transmission and scatter effects) with the Acuros XB algorithm and PTFE as the assigned material. To confirm that PTFE is a reasonable substitute for kyphoplasty cement, we performed measurements in a slab phantom using rectangular inserts of cement and PTFE, showing average agreement of all photon modes within 2%. Based on these findings, we conclude that the PTFE material assignment provides acceptable dose calculation accuracy for the AAA and Acuros XB photon algorithms in the Eclipse TPS. We recommend that the cement be delineated as a structure and assigned the PTFE material for accurate dose calculation.

Development of clinically relevant QA procedures for the BrainLab ExacTrac imaging system.

PURPOSE: The aim of this study was to develop Quality Assurance procedures for the BrainLab ExacTrac (ET) imaging system following the TG 142 recommendations for planar kV imaging systems. MATERIALS AND METHODS: A custom-designed 3D printed holder was used to position the Standard Imaging QCkV-1 phantom at isocenter, facing the ET X ray tubes. The linac's light field (collimator at 45°) was used to position the phantom holder. The ET images were exported to ARIA where geometric distortion was checked. The DICOM images were analyzed in the PIPSpro software. The following parameters were recorded (technique 80 kV/2mAs): spatial resolution (Modulated Transfer Function (MTF) F50/F40/F30), contrast-to-noise ratio (CNR), and noise. A baseline was generated for future image analysis. Beam quality and exposure were measured using the Unfors R/F detector. Using a rod holder, the detector was placed at isocenter, facing each ET X-ray tube. The measurements were performed for all preset protocols ranging from cranial low (80 kV/6.3 mAs) to abdomen high (145 kV/25 mAs). The total exposure was converted to dose. RESULTS AND DISCUSSION: The image quality parameters were close for the two tubes. A common baseline was therefore generated. The average baseline values (both tubes, both images/tube) were 1.06/1.18/1.30, 1.32, and 67.3 for the MTF F50/F40/F30, noise, and CNR respectively. The procedure described here was used for another 24 sets. Using a positioning template and 3D printed phantom holder, experimental reproducibility has been acceptably high. The measured phantom dimensions were within 1 mm from the nominal values. The measured kV values were within 2% of the nominal values. The exposure values for the two tubes were comparable. The range of total measured dose was 0.099 mGy (cranial low) to 1.353 mGy (abdomen high). CONCLUSIONS: A reliable process has been implemented for QA of the ET imaging system by characterizing the system's performance at isocenter, consistent with clinical conventions.

Varian HDR surface applicators - commissioning and clinical implementation.

The purpose of this study was to validate the dosimetric performance of Varian surface applicators with the source vertically positioned and develop procedures for clinical implementation. The Varian surface applicators with the source vertically positioned provide a wide range of apertures making them clinically advantageous, though the steep dose gradient in the region of 3-4 mm prescription depth presents multiple challenges. The following commissioning tests were performed: 1) verification of functional integrity and physical dimensions; and 2) dosimetric measurements to validate data provided by Varian as well as data obtained using the Acuros algorithm for heterogeneity corrected dose calculation. A solid water (SW) phantom was scanned and the Acuros algorithm was used to compute the dose at 5 mm depth and at surface for all applicators. Two sets of reference dose measurements were performed, with the source positioned at (i) -10 mm and (ii) -15 mm from the center of the first nominal dwell position. Measurements were taken at 5 mm depth in a SW phantom and in air at the applicator surface. The results were then compared to the vendor's data and to the Acuros calculated dose. Relative dose measurements using Gafchromic films were taken at a depth of 4 mm in SW. Percent depth ionization (PDI) measurements using ion chamber were performed in SW. The profiles generated from film measurements and the PDI plots were compared with those computed using the Acuros algorithm and vendor's data, when available. Preliminary leakage tests were performed using optically stimulated luminescence dosimeters (OSLDs) and the results were compared with Acuros predictions. All applicators were found to be functional with physical dimensions within 1 mm of specifications. For scenario (ii) measurements taken in SW at 5 mm depth and in air at the surface of each applicator were within 10% and 4% agreement with vendor's data, respectively. Compared with Acuros predictions, these measurements were within 6% and 5%, respectively. Measurements taken for scenario (i) showed reduced agreement with both the vendor's data as well as the Acuros calculations, especially when using the 10 mm applicator. The full widths of the measured dose profiles were within 2 mm of those predicted by Acuros at the 90% dose level. The PDI plots and measured leakage dose were in good agreement with vendor's data and Acuros predictions. Based on the dosimetric results, a quality assurance program and procedures for clinical implementation were developed. Treatment planning will be performed using scenario (ii). The 10mm applicator will not be released for clinical use. A prescription depth of 4mm is recommended, to ensure full coverage at 3 mm and a minimum dose of 90% of prescribed dose at 4 mm depth.

Commissioning and quality assurance for the treatment delivery components of the AccuBoost system.

The objective for this work was to develop a commissioning methodology for the treatment delivery components of the AccuBoost system, as well as to establish a routine quality assurance program and appropriate guidance for clinical use based on the commissioning results. Various tests were developed: 1) assessment of the accuracy of the displayed separation value; 2) validation of the dwell positions within each applicator; 3) assessment of the accuracy and precision of the applicator localization system; 4) assessment of the combined dose profile of two opposed applicators to confirm that they are coaxial; 5) measurement of the absolute dose delivered with each applicator to confirm acceptable agreement with dose based on Monte Carlo modeling; 6) measurements of the skin-to-center dose ratio using optically stimulated luminescence dosimeters; and 7) assessment of the mammopad cushion's effect on the center dose. We found that the difference between the measured and the actual paddle separation is < 0.1 cm for the separation range of 3 cm to 7.5 cm. Radiochromic film measurements demonstrated that the number of dwell positions inside the applicators agree with the values from the vendor, for each applicator type and size. The shift needed for a good applicator-grid alignment was within 0.2 cm. The dry-run test using film demonstrated that the shift of the dosimetric center is within 0.15 cm. Dose measurements in water converted to polystyrene agreed within 5.0% with the Monte Carlo data in polystyrene for the same applicator type, size, and depth. A solid water-to-water (phantom) factor was obtained for each applicator, and all future annual quality assurance tests will be performed in solid water using an average value of 1.07 for the solid water-to-water factor. The skin-to-center dose ratio measurements support the Monte Carlo-based values within 5.0% agreement. For the treatment separation range of 4 cm to 8cm, the change in center dose would be < 1.0% for all applicators when using a compressed pad of 0.2 cm to 0.3 cm. The tests performed ensured that all treatment components of the AccuBoost system are functional and that a treatment plan can be delivered with acceptable accuracy. Based on the commissioning results, a quality assurance manual and guidance documents for clinical use were developed.

Treatment planning methodology for the Miami Multichannel Applicator following the American Brachytherapy Society recently published guidelines: the Lahey Clinic experience.

The objective of this study was to develop a standardized procedure from simulation to treatment delivery for the multichannel Miami applicator, in order to increase planning consistency and reduce errors. A plan is generated prior to the 1st treatment using the CT images acquired with the applicator in place, and used for all 3 fractions. To confirm the application placement before each treatment fraction, an AP image is acquired and compared with the AP baseline image taken at simulation. A preplanning table is generated using the EBRT doses and is used to compute the maximum allowable D2cc for bladder, rectum, and sigmoid, and the mean allowable dose for the upper vaginal wall per HDR brachytherapy fraction. These data are used to establish the criteria for treatment planning dose optimization. A step-by-step treatment planning approach was developed to ensure appropriate coverage for the tumor (D90 > 100% prescribed dose of 700 cGy/fraction) and the uninvolved vaginal surface (dose for the entire treatment length > 600 cGy/fraction), while keeping the organs at risk below the tolerance doses. The equivalent dose 2 Gy (EQD2) tolerances for the critical structures are based on the American Brachytherapy Society (ABS) recently published guidelines. An independent second check is performed before the 1st treatment using an in-house Excel spreadsheet. This methodology was successfully applied for our first few cases. For these patients: the cumulative tumor dose was 74-79 EQD2 Gy10 (ABS recommended range 70-85); tumor D90 was >100% of prescribed dose (range 101%-105%); cumulative D2cc for bladder, rectum, and sigmoid were lower than the tolerances of 90, 75, and 75 EQD2 Gy3, respectively; cumulative upper vaginal wall mean dose was below the tolerance of 120 EQD2 Gy3; the second check agreement was within 5%. By using a standardized procedure the planning consistency was increased and all dosimetric criteria were met.

Retrospective review of Contura HDR breast cases to improve our standardized procedure.

To retrospectively review our first 20 Contura high dose rate breast cases to improve and refine our standardized procedure and checklists. We prepared in advance checklists for all steps, developed an in-house Excel spreadsheet for second checking the plan, and generated a procedure for efficient contouring and a set of optimization constraints to meet the dose volume histogram criteria. Templates were created in our treatment planning system for structures, isodose levels, optimization constraints, and plan report. This study reviews our first 20 high dose rate Contura breast treatment plans. We followed our standardized procedure for contouring, planning, and second checking. The established dose volume histogram criteria were successfully met for all plans. For the cases studied here, the balloon-skin and balloon-ribs distances ranged between 5 and 43 mm and 1 and 33 mm, respectively; air_seroma volume/PTV_Eval volume≤5.5% (allowed≤10%); asymmetry<1.2mm (goal≤2 mm); PTV_Eval V90%≥97.6%; PTV_Eval V95%≥94.9%; skin max dose≤98%Rx; ribs max dose≤137%Rx; V150%≤29.8 cc; V200%≤7.8 cc; the total dwell time range was 225.4 to 401.9 seconds; and the second check agreement was within 3%. Based on this analysis, more appropriate ranges for the total dwell time and balloon diameter tolerance were found. Three major problems were encountered: balloon migration toward the skin for small balloon-to-skin distances, lumen obstruction, and length change for the flexible balloon. Solutions were found for these issues and our standardized procedure and checklists were updated accordingly. Based on our review of these cases, the use of checklists resulted in consistent results, indicating good coverage for the target without sacrificing the critical structures. This review helped us to refine our standardized procedure and update our checklists.

Quality assurance methodology for Varian RapidArc treatment plans.

With the commercial introduction of the Varian RapidArc, a new modality for treatment planning and delivery, the need has arisen for consistent and efficient techniques for performing patient-specific quality assurance (QA) tests. In this paper we present our methodology for a RapidArc treatment plan QA procedure. For our measurements we used a 2D diode array (MapCHECK) embedded at 5 cm water equivalent depth in MapPHAN 5 phantom and an Exradin A16 ion chamber placed in six different positions in a cylindrical homogeneous phantom (QUASAR). We also checked the MUs for the RapidArc plans by using independent software (RadCalc). The agreement between Eclipse calculations and MapCHECK/MapPHAN5 measurements was evaluated using both absolute distance-to-agreement (DTA) and gamma index with 10% dose threshold (TH), 3% dose difference (DD), and 3 mm DTA. The average agreement was 94.4% for the DTA approach and 96.3% for the gamma index approach. In high-dose areas, the discrepancy between calculations and ion chamber measurements using the QUASAR phantom was within 4.5% for prostate cases. For the RadCalc calculations, we used the average SSD along the arc; however, for some patients the agreement for the MUs obtained with RadCalc versus Eclipse was inadequate (discrepancy > 5%). In these cases, the plan was divided into partial arc plans so that RadCalc could perform a better estimation of the MUs. The discrepancy was further reduced to within ~4% using this approach. Regardless of the variation in prescribed dose and location of the treated areas, we obtained very good results for all patients studied in this paper.

Dosimetric consequences of manual pullback procedure for coronary artery radiotherapy with 90Sr/90Y beta-source.

PURPOSE: This work presents a quantitative dosimetric analysis of the Novoste (90)Sr/(90)Y beta-source cardiovascular brachytherapy treatments using a manual pullback technique for patients with in-stent restenosis. METHODS AND MATERIALS: Based on our previous measurements, a model was developed to estimate the dose in the middle of the junction region for tandem irradiation expressed as fraction of prescription dose (FPD) and dosimetric overlap length (DOL) receiving more/less than a threshold dose. The overlap/gap size was measured using the digital cine images recorded during treatment and then FPD and DOL were quantified. RESULTS: Statistical analysis of 55 patients showed that the overlap size and the FPD at 2 mm radial distance were in range of 0 to 23 mm and 13-200% of prescription dose (Rx), respectively. Four gaps out of 76 pullback cases were found, but their size was at most 5 mm. CONCLUSION: Use of a 5 mm overlap avoided underdosed regions in the vast majority of the cases. These results are the first step towards an analysis of the clinical outcome of these patients.

GAF film dosimetry of a tandem positioned beta-emitting intravascular brachytherapy source train.

Coronary artery brachytherapy may require treatment of lesions longer than a single source length. A treatment option is tandem positioning of the single source. This study presents relative dosimetric measurements of a cardiovascular brachytherapy source and the dosimetric characteristics in the junction region of tandem treatments. Measurements were carried out using a Novoste Beta Cath 90Sr/90Y 40 mm beta source in a plastic water phantom. Radiochromic MD-55-2 film, calibrated using both 6 MV photon and 6 MeV electron beams from a linear accelerator, was used as the dosimeter. Dose distributions around a single source and in the junction region of tandem irradiation were measured. Measurements of the near field dose as close as 1.2 mm from the source are presented. Significant over- or underdoses in the junction region of tandem irradiation were quantified. At a radial distance of 2 mm from the longitudinal axis of the source, the dose value in the middle of the junction region, normalized to the dose at 2 mm midline single source, was about 182% for a 2-seed overlap and 16% for a 2-seed gap, respectively. Dose distributions in the junction region as a function of source overlap and radial distance have fairly high gradients and exhibit characteristic patterns. The fraction of prescription dose was found to have a sigmoidal dependence on overlap size, for radial distances ranging between 1.2 and 3 mm. The parameters of these sigmoids, quantified as functions of radial distance, could be used to provide quick and reasonable over/underdose estimates, given any potential overlap or gap in the junction area, with an uncertainty within 10%.