A method to predict patient‐specific table coordinates for quality assurance in external beam radiation therapy

Abstract Purpose While external beam radiotherapy treatment planning determines nearly every mechanical and dosimetric parameter of the linear accelerator (LINAC), the table coordinates in all three dimensions are generally unknown until initial patient setup at the LINAC. Knowing these parameters in advance could help verify the direction of patient shifts and prevent wrong‐site errors. This study aims to determine the feasibility and accuracy of table coordinate prediction for indexed immobilization devices. Methods A total of 303 table coordinates were predicted for patients on Varian and Elekta linear accelerators with immobilization devices including Orfit mask with baseplate, wingboard, breastboard and BodyFix. Predictions were made for all three spatial dimensions except for Body Fix setups due to the lack of a radiographically apparent indexing‐related landmark. Coordinates were predicted by measuring baseline table coordinates in all dimensions at specified landmark positions. Results Predictions were accurate within 2 cm for 86% of coordinates (71% within 1 cm). Table coordinates were predicted most accurately for head and neck patients with a base plate and the most difficult prediction was in the lateral direction for breastboard patients. Conclusions With proper indexing, table coordinates can be predicted with reasonable accuracy. The data suggest an action of level of 2 cm with certain exceptions for specific immobilization devices and directions.

only the table angle is set prior to treatment. Exact table coordinates are not necessarily required given the role of advanced image guidance as confirmation of accurate 3D anatomical localization. Prior knowledge of these table coordinates, however, may provide an early alert of an error if a mismatch is observed with observed table coordinates. Positioning errors could therefore be detected prior to exposure to imaging ionizing radiation. [5][6][7] Mismatches could be resolved by a staff member before proceeding with imaging. Prior table coordinate knowledge also provides an independent isocenter verification not subject to errors in the image-guidance process, prevents wrong direction patient shifts, and allows the medical physicist to review these coordinates during plan review as well.  table relative shift of +4.0 cm may be interpreted as the absolute coordinate itself. 8 This study states that the issue could be remedied by "having the coordinates checked by the physicists or therapist staff pretreatment". A 2009 report summarizing errors in the state of Pennsylvania showed that 32% of the reported events were from wrong location, wrong side, or wrong setup, and called for additional safety checks on location. 9 Movements of the table from reference marks is an error-prone step in the patient setup workflow, making up 10.8% of reportable radiation incidents analyzed in PHE Report No.

10 Common error pathways from Radiation Oncology Incident
Learning System ® (RO-ILS) showed that 74 of the 396 events were caused by either wrong shift instructions or wrong shift performed at treatment. 11 Wrong site errors tend to be detected by port films, but this only applies on fractions where port films are taken. Studies have also examined the constriction of values for table coordinates for prevention of errors, but these data rely on baseline values only known after the first treatment fraction. 12 The prediction of table coordinates may be a practical, valuable additional safety check that can be used without exposing the patient to additional ionization radiation. Moreover, this prediction completes the set of machine parameters set in the treatment fields prior to patient treatment, serving as an engineering control preventing treatment without higher approval, just as other machine geometrical parameters are treated (gantry angle, collimator angle, etc).

| METHODS
Linear accelerators included in this study include the Elekta Versa HD, Varian 23EX, and Novalis Tx. Virtual simulation and treatment planning is performed in Philips Pinnacle 3 version 9.16 and the record-and-verify system in use is MOSAIQ ® . Table coordinates consist of parameters in the three spatial directions, referred to in this study as lateral (x), vertical (y), and longitudinal (z). The overall methodology is to identify baseline table coordinates (T x,0, T y,0, T z,0 ) corresponding to a coordinate (x 0, y 0, z 0 ) in the tomographic slice of a landmark in an immobilization device and adjust them by planned patient isocenter patient shifts from this landmark point (x′, y′, z′).
The longitudinal direction is the most difficult to predict, as it requires consistent indexing of immobilization devices. 13 If this condition is met, the table longitudinal coordinate when a landmark slice in each immobilization device is set to isocenter is required. To predict the longitudinal table coordinate at the patient's isocenter, only the difference in slice position between this landmark and patient isocenter in the CT scan is needed (z′). This approach must be specific for each immobilization device and for each linear accelerator with a unique coordinate system.
The vertical dimension is more simply predicted by measuring the vertical distance between the planned isocenter and the treatment table in the CT-simulation scan (y′). This process is facilitated by the inclusion of the treatment table in the treatment planning system. This has the additional advantage of verifying the correct vertical placement of the treatment table in the treatment planning process. It is required that the vertical table calibration is such that the table is at its nominal position, T y,0 , when vertically set to isocenter.
Finally, the accuracy of the lateral coordinate depends on how laterally centered reference points are made at CT-simulation and subsequently centered at treatment. If the patient is laterally centered, then an assumption can be made that the table is set at its nominal lateral position, T x,0 , when the patient is aligned to his or her reference marks, and the final lateral coordinate is predicted by the lateral displacement of the isocenter coordinate from the reference coordinate (x′). The lateral table coordinate prediction becomes most difficult for patients reference marks are not placed at midline.
Fortunately, most patients at our institution have reference marks placed in this manner, even those with a clear sidedness (e.g., right or left breast). However, this lateral coordinate prediction system is not applicable at sites such as extremities, where the patient is centered on the table, but both the reference and isocenter coordinates are laterally off-center in an extremity.
The prediction of the table coordinates in three dimensions is dictated by the immobilization device. Primarily, four devices are used at our institution: thermoplastic masks on a base plate for brain or head and neck treatments, a wingboard for chest and upper abdomen treatments, a breastboard for breast treatments, and finally a BodyFIX ® for most other treatments. All except BodyFIX ® have readily apparent landmarks radiographically apparent in the CT-scans.
The head and neck base plate (Orfit Industries, Belgium) has three pairs of holes, the lowest of which was identified as the landmark ( Fig. 1). The bottom edge of the wingboard (Civco Radiotherapy, Orange City, IA, USA) was used as a landmark as was the bottom edge of the arm support structure for the breastboard (Qfix, Avondale, PA, USA) ( Fig. 1). For the Bodyfix (Elekta, Stockholm, Sweden), there is no landmark to assist in the longitudinal prediction, so this parameter is not a candidate for patient-specific prediction. It is possible to perform population-based predictions for certain sites such as pelvis, but this is beyond the scope of this study.  Figure 2 outlines the geometry of the prediction method. Note that when using the 6 degree-of-freedom

2.A | Clinical verification
Current on-treatment patient data was used to verify the prediction model. While the vast majority of table coordinates can be predicted, some coordinates cannot be predicted. These include the longitudinal coordinates for BodyFIX ® patients and coordinates for patients treated with electrons. Patients to be treated with electron therapy were excluded due to the direct clinical verification of the light field on the skin being the most important consideration in our clinic, which overshadows the role of table coordinates for quality assurance. As an indication of the general applicability of this system, the total number of table coordinates required on a typical day was estimated to be 180 (60 patients with three dimensions of couch coordinates) of which all but 28 could be predicted (84%). Of those, 22 were expected BodyFIX ® treatments (primarily pelvis, lumbar spine, or extremities) for which the longitudinal position is not predictable.
Finally two patients were expected electron treatments (a total of six parameters not predictable).
Of the 303 couch coordinates from 100 patients studied, 84 were from breastboard setups, 66 from wingboard setups, 111 from mask and base plate setups, and 42 from BodyFIX ® setups. These data include patients receiving 2D, 3D-conformal, step and shoot IMRT, and VMAT. Data collected included the site, the indexing/immobilization device, the field ID and field names, isocenter name, the For each immobilization device, predictions were made separately for the Varian vs. Elekta vault (with or without use of HexaPOD TM ).
The differences between predicted and actual table coordinates were then analyzed in a spreadsheet.

| CONCLUSION S
Prediction of couch coordinates is achievable with minimal additional steps required in the simulation process. The vertical dimension was predicted accurately regardless of site, immobilization device, or treatment machine. The longitudinal dimension required consistent use of table indexing, a procedure commonly practiced by radiation therapy clinics. Laterally, the predictions were accurate, but had larger uncertainty in cases where sidedness dictated a laterally off-center patient position. Action levels at or around 2 cm might be appropriate for the practice described in this study, and the use of this prediction tool can be used to improve patient safety by detecting setup errors before any sort of ionization radiation is used for verification imaging or treatment.

CONFLI CTS OF INTEREST
The authors have no other relevant conflicts of interest to disclose.