Long-time clinical experience in patient setup for several particle therapy clinical indications: management of patient positioning and evaluation of setup reproducibility and stability

Abstract

Objective:

Accurate patient positioning is crucial in particle therapy due to the geometrical selectivity of particles. We report and discuss the National Center for Oncological Hadrontherapy (CNAO) experience in positioning accuracy and stability achieved with solid thermoplastic masks fixed on index base plates and assessed by daily orthogonal X-ray imaging.

Methods:

Positioning data were retrospectively collected (between 2012 and 2018) and grouped according to the treated anatomical site. 19696 fractions of 1325 patients were evaluated.

The study was designed to assess:

(i) the number of fractions in which a single correction vector was applied(SCV);

(ii) the number of fractions in which further setup verification was performed (SV);

(iii) the number of fractions in which SV lead to an additional correction within (MCV<5min) or after (MCV>5min) 5 minutes from the first setup correction;

(iv) the systematic (Σ) and random (σ) error components of the correction vectors applied.

Results:

A SCV was applied in 71.5% of fractions, otherwise SV was required. In 30.6% of fractions with SV, patient position was not further revised. In the remaining fractions, MCV<5min and MCV>5min were applied mainly in extracranial and cranial sites respectively.

Interfraction Σ was ≤ 1.7 mm/0.7° and σ was ≤ 1.2 mm/0.6° in cranial sites while in extracranial sites Σ was ≤ 5.5 mm/0.9° and σ was ≤4.4 mm/0.9°. Setup residuals were submillimetric in all sites. In cranial patients, maximum intrafractional Σ was 0.8 mm/0.4°.

Conclusion:

This report extensively quantifies inter- and intrafraction setup accuracy on an institutional basis and confirms the need of image guidance to fully benefit from the geometrical selectivity of particles.

Advances in knowledge:

The reported analysis provides a board institutional data set on the evaluation of patient immobilization and bony anatomy alignment for several particle therapy clinical indications.

Introduction

The inverse depth-dose profile and the sharp lateral penumbra make charged particles particularly suitable for high-precision dose-escalated radiotherapy.^1–5 However, the peculiar physical properties of charged particles make this treatment modality very sensitive to targeting uncertainties.^6–10 Inter- and intrafractional variation in the alignment of patient with respect to the treatment beams can result in a distortion of the planned dose distribution,^8,11–14 thus requiring a personalized assessment of treatment plan robustness against geometrical uncertainties.

Accurate patient alignment with respect to the beam is crucial in particle therapy and image-guided patient positioning is mandatory to benefit from the geometrical accuracy of particles.^15–17 However, published data on setup uncertainties during a course of radiotherapy are usually based on small patient cohorts.¹⁸ Differences in the study methodology, e.g. patient immobilization, patient preparation, treatment workflow and statistical description of the acquired data, do not readily facilitate intercomparison of reported results.¹⁸ Moreover, image-guided evaluation of geometrical setup uncertainties is strongly dependant on the technical implementation of the in-room imaging systems and their related applications. Indeed, patient setup evaluation depends on the imaging methods used (the planar projective verification images or volumetric imaging), on registration algorithms and approaches (bony anatomy rigid alignment or deformable registration), on technical features (resolution of the digitally reconstructed radiographs (DRR) used as reference images), on human factors (e.g. inter- and intraobserver variability).¹⁹ Besides, image guidance procedures vary substantially between particle therapy centers, since most centers have developed their own strategies tightly connected to a specific technical implementation and widespread commercial solutions or consensus guidelines are not yet consolidated like in photon therapy clinical practice.^15,20 For these reasons, setup uncertainties evaluation should be performed in a comprehensive way on an institutional basis.

In this study, we report and discuss the positioning accuracy and the setup stability achieved with the main use of solid thermoplastic masks and indexed base plates on a large patient cohort, who underwent particle therapy between September 2012 and November 2018 at National Center for Oncological Hadrontherapy (CNAO). This comprehensive analysis provides an overview of our institutional experience in patient positioning from the beginning of our clinical activities until today.²¹ Positioning data are reported according to different treatment sites (intracranial, head and neck, craniospinal, paraspinal, upper abdomen and thorax, lower abdomen and limb) which mainly guide the choice of the combination of immobilization devices. This report could serve as an evaluation of adequacy of setup uncertainties quantification for plan robustness assessment or safety margins definition.

Methods and materials

Patient population—data selection criteria

Positioning data of patients who underwent particle therapy between September 2012 and November 2018 at the National Center for Oncological Hadrontherapy(CNAO) were retrospectively collected. The patients gave written informed consent for radiotherapy and for the use of their anonymized data for educational and research purposes.

Positioning data of 19,696 treatment fractions of 1325 patients were retrospectively collected and analyzed.

The data set includes treatment fractions during which patient positioning was image-guided using a complete set of stereoscopic kV images. In order to streamline data processing without affecting the consistency and comparability of the data, the data set includes positioning data of patients that underwent a single radiotherapy course at the National Center for Oncological Hadrontherapy (CNAO) either in prone or supine position. Only positioning data generated from uninterrupted treatment fractions were included. The analyzed data set excludes ocular proton therapy treatment fractions that require a completely different treatment setup with the patient seated on a robotic treatment chair.²²

Detailed patient and treatment data grouped by tumour localization are shown in Table 1.

Table 1.

Patient and treatment data overview grouped by tumor localization

	Intracranial	Head and neck	Craniospinal	Paraspinal	Upper abdomen and thorax	Lower abdomen	Limb
No Patients	230	597	145	26	55	249	23
Gender – no pts (%)
Male	119 (51.7)	339 (56.8)	85 (58.6)	18 (69.2)	32 (58.2)	151 (60.6)	12 (52.2)
Female	111 (48.3)	258 (43.2)	60 (41.4)	8 (30.8)	23 (41.8)	98 (39.4)	11 (47.8)
Age [years]
Median (IQR)	54.5 (22.5)	58.1 (22.5)	54.6 (22.2)	53.0 (26.4)	66.3 (26.2)	63.5 (20.1)	54.1 (38.6)
Weight – [Kg]
Median (IQR)	73.6 (22.6)	70.6 (22.6)	78.5 (24.4)	76.2 (29.1)	63.5 (20.3)	72.4 (18.3)	71.0 (13.2)
Not Available – n° pts (%)	11 (4.8)	50 (8.4)	15 (10.3)	3 (11.5)	1 (1.8)	17 (6.8)	2 (8.7)
Positioning – n° pts (%)
Supine	225 (97.8)	593 (99.3)	129 (89.0)	2 (7.7)	29 (52.7)	56 (22.5)	19 (82.6)
Prone	5 (2.2)	4 (0.7)	16 (11.0)	24 (92.3)	26 (47.3)	193 (77.5)	4 (17.4)
Particle – n° pts (%)
Proton	189 (82.2)	140 (23.5)	49 (33.8)	4 (15.4)	6 (10.9)	19 (7.6)	2 (8.7)
Carbon	41 (17.8)	449 (75.2)	96 (66.2)	22 (84.6)	49 (89.1)	228 (91.6)	21 (91.3)
Both	0 (0.0)	8 (1.3)	0 (0.0)	0 (0.0)	0 (0.0)	2 (0.8)	0 (0.0)

IQR, interquartile range; n° pts, number of patients.

Immobilization procedure and treatment simulation

Patient immobilization is applied for a fast and reproducible setup of the patient before each treatment fraction as a way to minimize intrafraction movements during irradiation.^18,23,24

The anatomical district, the disease extension, the eventual previous radiotherapy courses together with the clinical assessment of the patient compliance guide the specific choice of the combination of the immobilization devices and the patient setup. The illustrative Figure 1 summarizes some exemplary setups for each analyzed treated site.

Figure 1.

Illustrative table summarizing some exemplary setups for each analyzed treated site. (*) bite block used to stabilize the mandible for the target in its close proximity.

All patients included in the study were positioned using a personalized solid thermoplastic mask fixed on an indexed base plate. In particular, for cranial treatments type-s head or head-and-shoulder masks with nose perforation were used together with standardized head supports of different height and curvature as polyethylene cushions or silverman^® headrests (CIVCO, Medical Solutions, Kalona, Iowa). An intraoral bite block was employed for stabilizing the mandible when clinically required. For extracranial treatments, setup points were tattooed on patient skin to provide anatomical references for daily laser alignment. All patients were immobilized on the treatment couch with a non-perforated pelvic thermoplastic mask and an indexed base plate for pelvic region. Customized Moldcare^® large cushions (40 × 60 cm² or 60 × 60 cm²) (Qfix Systems, Avondale, PA) were used together with polyethylene knee-support or foot holders. In particular, for the treatments in the upper abdomen in supine position, an angled baseplate with supports for arms and wrists was generally used.

Computed Tomography (CT) images were acquired on a SOMATOM Sensation Open CT scanner (Siemens, Germany). For the majority of cases, a slice thickness of 2 mm with a pixel spacing of 0.9766 × 0.9766 mm was applied. Scanning protocols used energy levels according to body sites, varying in the interval of 70–320 mAs at 120 kV.

Patient positioning and verification systems

The National Center for Oncological Hadrontherapy (CNAO) is a synchrotron-based facility featuring three treatment rooms, two (lateral) with fixed-horizontal and one (central) with both horizontal and vertical beam lines all integrated with patient positioning (PPS) and image-based setup verification (PVS) systems.²⁵

The PPS is a robotic pantograph (Schaer Proton AG, Flaach, Switzerland) featuring 6 degrees of freedom motion with a positioning accuracy of ±0.3 mm in translation and ±0.1° in rotations.²⁵

In the two lateral rooms, PVS consists of a rotating stereoscopic X-ray imaging system (Schaer Proton AG, Flaach, Switzerland), mounted on the room ceiling, that allows for the simultaneous acquisition of two oblique projections at the treatment isocenter. This system is completed with a commercial software for rigid 2D–3D image registration (Verisuite^® MedCom GmbH, Darmstadt, Germany) of the acquired images to the DRRs generated from the planning CT. The detailed technology commissioning of these systems for clinical use is reported in Pella et al.²⁵

In the central room, a customized robotic imaging solution performs an out of treatment isocenter imaging and allows for the acquisition of double planar orthogonal kV-images in anteroposterior and right-left direction or volumetric come beam computed tomography (CBCT). A dedicated software platform for X-ray patient position verification (XPPV) was designed and implemented²⁶ integrating an iterative 2D–3D rigid image registration algorithm.

Quality assurance (QA) tests of the entire image-guided positioning procedure are daily performed before clinical operations, as described in detail by Pella et al. in “Quality Assurance Tests Section,”²⁵ for a submillimetric setup accuracy verification. A dedicated phantom provided for X-ray QA and geometric calibration (Brandis Medizintechnik Vertriebs GmbH, Weinheim, Germany) containing radiopaque seeds (diameter: 1.5 mm) is mechanically fixed on the treatment couch in a predetermined position indicated by the couch indexing system. The PPS with the phantom fixed on the treatment couch is automatically moved in order to bring the reference central seed of the QA phantom in correspondence to the imaging isocentre. Double X-ray images projections are acquired and the 2D distances between the centre of the reference seed in the phantom projections and the centre of each flat panel are measured. In the lateral rooms the measurements are performed at different couch rotations allowing to test couch isocentricity over 180° motion range. In the central room these measurements are performed acquiring the anteroposterior and laterolateral projections of the phantom. Each day, before clinical activity, these measurements are performed and QA tests pass if the measured 2D distances are lower than 0.8 mm.

Image guidance procedure

For each treatment fraction, the patient is positioned on the treatment couch and immobilized according to the treatment simulation (see immobilization procedure paragraph). In particular, extracranial patients are set up in a dedicated room using calibrated positioning lasers and skin tattooed reference points before being transferred into treatment room using a manual transport system compatible with the PPS. Cranial patients are positioned directly in the treatment room using the in-room calibrated positioning lasers.

Once treatment couch is in the patient-specific nominal setup-configuration, a pair of stereoscopic kV images is acquired and automatic bony alignment (after manual preliminary alignment when necessary) to the corresponding DRRs is performed. The resulting translations and rotations are compensated using the robotic couch. If the resulting position correction vector entails rotations larger than 3° or translations larger than 10 or 20 mm for cranial and extracranial treatments respectively, patients are manually repositioned on the treatment couch. This initial image-based correction vector is a measure of the interfraction setup accuracy and reproducibility mainly accounting for the efficacy of immobilization system.

For verification purposes, patient positioning procedure may envisage the acquisition of a further pair of images after the initial setup correction. In particular, this additional X-ray verification is carried out if the initial correction vector leads to large robotic couch movements or if it is clinically required during the first fractions of the treatment course in patients with critical anatomical structures in close proximity to the target. According to our institutional guidelines, this additional setup verification is generally advised if the initial correction vector includes a rotation above 2° around one axis for cranial treatment or a rotation above 2° around one axis along with a translation between 5 and 10 mm for extracranial treatments. Additional X-ray acquisition is also performed in long lasting treatment fractions to evaluate setup stability over time after several minutes from the initial setup correction. Every time new images are acquired, the rigid automatic 2D–3D image registration is repeated and a new correction vector is estimated. If this correction vector is larger than 0.3 mm in at least one direction of translation or 0.3° in at least one axis of rotation and in case the bony structures alignment between the acquired images and the DRR were evaluated clearly improved by the radiation therapists (RTTs), the robotic couch position is consequently corrected.

Over the years, patient positioning protocols have been slightly revised in order to improve the clinical workflow. In particular, until 2013 an integrated optical- and image-based procedure for patient setup was routinely implemented for cranial and extracranial irradiation.²⁷ More recently, the infrared optical tracking systems installed in the three treatment rooms are solely used to monitor intrafraction patient motion through continuous 3D detection of the position of the external markers during treatment delivery.²⁸

Data analysis

Image-based daily positioning data were grouped as a function of the treated anatomical site according to the recommendations for organ depending optimized fixation systems provided by The Union of Light Ion Therapy Centres in Europe (ULICE collaborative research group).¹⁸ The identified body regions were: intracranial, head and neck, craniospinal, paraspinal, upper abdomen and thorax and lower abdomen. Additionally, we analyzed the position reproducibility of limb-extremities, when multiple combinations of immobilization devices were used.

The analyzed treatment fractions were classified accounting for: (i) fractions in which a single initial correction vector was applied and no additional X-ray images were acquired (single correction vector, SCV); (ii) fractions in which further image-based setup verification was performed (setup verification, SV); (iii) fractions in which further SV triggered an additional setup correction (multiple correction vector, MCV) within 5 min (MCV <5 min) or after 5 min (MCV >5 min) from the first setup correction. We imposed the 5 min time threshold to distinguish those cases in which the additional correction vector was likely applied to refine the initial setup before beam delivery, from other cases in which the setup verification was performed between different treatment fields. We considered reasonable setting the time threshold to 5 min since the setup refinement evaluation of the verification images generally requires less time than the evaluation of the first setup images as long as the manual initialization of automatic registration is not necessary for verification images.

Treatment fraction classification was performed automatically by means of a custom-written Matlab script (Mathworks, Natick/MA) which navigates through clinical treatment data and retraces the image-guidance steps performed during each treatment fraction using treatment log files (Figure 2).

Figure 2.

Schematic representation of the analyzed treatment workflow. MCV, multiple correctionvector; SCV, single correction vector; SV, setup verification.

We determined the overall range of initial setup correction vectors (first CV) as a measure of interfraction setup accuracy mainly accounting for the efficacy of the immobilization systems. We reported the range of MCV <5 min as a measure of residual setup errors and the range of MCV >5 min as a measure of setup stability over time.

The distributions of the correction vector were expressed by median and interquartile ranges (IQR). For the sake of completeness and in order to ensure comparability with other published data, we also used the definitions of van Herk for the evaluation of inter- and intrafraction setup error.²⁹ The population systematic error component (Σ) was calculated as the standard deviation of the mean of the inter- and intrafractional error for each patient, whereas the root mean square of the standard deviations represented the population random error component (σ).

Translational setup correction vectors were expressed along the patient-specific right-left direction (R-L), inferosuperior direction (I-S) and posteroanterior direction (P-A). Negative translations were in right, inferior and posterior direction. Rotations around the above-mentioned axes constructed pitch, roll and yaw, respectively.

Differences between prone and supine positioning in upper and lower abdominal patients were compared using one-way analysis of variance. Finally, we investigated the influence of time on patient positioning stability: the time interval between the initial setup correction and eventual subsequent correction was correlated with the correction parameters by computing Spearman’s correlation coefficient.

Results

Detailed description of the classification of treatment fractions is shown in Table 2. In the most fractions (71.5%), a SCV was applied to couch position and no additional X-ray images were acquired for setup verification purposes. Otherwise, in the remaining 28.5% of the analyzed fractions, SV after the first correction was performed. Setup verification frequently occurred for craniospinal (45.1%), intracranial (39.6%), and head and neck (26.0%) treatments.

Table 2.

Classification of fractions and correction vectors according to the treated anatomical site.

	Intracranial	Head and neck	Craniospinal	Paraspinal	Upper abdomen and thorax	Lower abdomen	Limb	All sites
No fractions	4706	8369	2200	365	572	3162	322	1,9696
Fractions per patient—mean (min-max)	20.4 (1–36)	14.0 (1–36)	15.1 (1–40)	14.0 (3–34)	10.30 (1–27)	12.6 (1–34)	14.0 (2–28)	14.8 (1–40)
Classification —no fr (%)
SCV	2843 (60.4)	6193 (74.0)	1208 (54.9)	305 (83.6)	497 (86.9)	2755 (87.1)	290 (90.1)	1,4091 (71.5)
SV	1863 (39.6)	2176 (26.0)	992 (45.1)	60 (16.4)	75 (13.1)	407 (12.9)	32 (9.9)	5605 (28.5)
MCV < 5 min—no fr (% of SV)	334 (17.9) a	372 (17.1)a	189 (19.1)a	34 (56.7)	36 (48.0)	210 (51.6)	15 (46.9)	1190 (21.2)a
MCV > 5 min—no fr (% of SV)	1082 (58.1)a	1224 (56.3)a	682 (68.8)a	0 (0)	0 (0.0)	0 (0.0)	0 (0)	2988 (53.3)a
No MCV—no fr (% of SV)	573 (30.8)	664 (30.5)	200 (20.2)	26 (43.3)	39 (52.0)	197 (48.4)	17 (53.1)	1716 (30.6)
Treatment room—no fr (%)
Lateral rooms	4465 (94.9)	7853 (93.7)	1746 (79.4)	0 (0.0)	2 (0.3)	33 (1.0)	31 (9.6)	1,4130 (71.7)
Central room	241 (5.1)	516 (6.3)	454 (20.6)	365 (100.0)	570 (99.7)	3129 (99.0)	291 (90.4)	5565 (28.3)

MCV, multiple correction vector; SCV, single correction vector; SV, setup verification;n^o, number of fractions.

Number of treatment fractions delivered in lateral and central rooms are reported separately according to the different image guidance hardware available.

Number of fractions in which a SCV was applied and number of fractions in which a SV was required are reported. When a setup verification was required, we report the number of fractions in which an additional CV was applied (MCV < 5 min and MCV > 5 min) and the number of fractions in which no further couch position adjustment was performed.

^aIn 126, 84, 79 treatment fractions for intracranial, head and neck and craniospinal treatments respectively, both MCV < 5 min and MCV > 5 min were applied.

In 30.6% of fractions with SV, the additional image acquisition did not lead to a revision of patient position. In those cases, the resulting correction vector was lower than the system accuracy or bony structures registration did not clearly improve patient alignment. Most of the SV that did not lead to the update of couch position were recorded for extracranial treatments.

In the other cases, a second correction vector was applied to the treatment couch position. In particular, the eventual couch position adjustment occurred mostly within 5 min from the first setup correction for paraspinal (56.7% of fractions with SV), lower (51.6% of fractions with SV), upper (48.0% of fractions with SV), abdominal and limb (46.9% of fractions with SV) fractions. Conversely, for intracranial (58.1% of fractions with SV), head and neck (56.3% of fractions with SV) and craniospinal (68.8% of fractions with SV) fractions, the couch position revision occurred mainly after 5 min from the application of the first correction vector.

It is worth mentioning that extracranial treatments are generally delivered in central room and they do not show any MCV >5 min.

The distribution of first CV is reported in Figure 3. Related medians (IQR) and Σ (σ) are shown in Table 3. The median translational values in R-L and in P-A direction for the cranial sites were lower than 2 mm, while for lower abdomen median I-S translation is greater than 5 mm. The medians of rotations reached a maximum of 0.2°. The variability of interfraction and interpatient positioning is greater in extracranial rather than in cranial regions showing translational and rotational IQR values greater than 8 mm and 1° respectively. For cranial sites, interfraction Σ was lower than 1.7 mm and 0.7°, while σ was lower than 1.2 mm and 0.6°. On the contrary, for extracranial treatment sites, Σ was lower than 5.5 mm and 0.9°, while σ was lower than 4.4 mm and 0.9°. Statistically significant differences in positioning reproducibility for prone and supine treatments delivered in upper and lower abdomen are depicted in Figure 4.

Figure 3.

Distribution of setup correction vectors applied over the course of treatment fractions grouped by treated anatomical site. Boxplots report the distribution of the initial setup correction parameters (first CV), the distribution of the additional correction vectors applied within 5 min (MCV <5 min) and after 5 min (MCV >5 min) from the first setup correction. On each boxplot, the central mark indicates the median, and the bottom and top edges of the box indicate the 25th and 75th percentiles, respectively. The whiskers extend to the most extreme data points not considered outliers. Box whiskers contains approximately 99.3% of data. R-L, right left direction; I-S, inferosuperior direction; MCV, multiple correctionvector; P-A, posteroanterior direction.

Table 3.

Initial setup correction vector evaluation.

	Initial correction vector (1st CV)
	Median (IQR) [mm, °]						∑ (σ) [mm, °]
Anatomical sites	R-L	I-S	P-A	Pitch	Roll	Yaw	R-L	I-S	P-A	Pitch	Roll	Yaw
Intracranial	1,5 (1.9)	−0,1 (2.5)	−1,7 (2.2)	0,1 (0.8)	−0,1 (1.0)	0,2 (0.9)	1,2 (0.9)	1,6 (1.2)	1,5 (0.6)	0,6 (0.6)	0,6 (0.6)	0,6 (0.6)
Head and neck	1,3 (2.0)	0,3 (2.4)	−1,8 (2.4)	−0,1 (0.9)	−0,1 (1.1)	0,1 (0.8)	1,3 (1.0)	1,5 (1.1)	1,7 (0.7)	0,6 (0.5)	0,7 (0.6)	0,6 (0.5)
Craniospinal	1,0 (2.3)	−0,1 (2.2)	−1,9 (2.3)	0,0 (0.8)	−0,1 (1.0)	0,1 (1.1)	1,4 (1.0)	1,5 (1.2)	1,6 (0.8)	0,6 (0.5)	0,6 (0.6)	0,7 (0.6)
Paraspinal	−0,2 (4.7)	−5,5 (8.7)	−3,3 (5.7)	−0,2 (1.1)	0,2 (0.8)	0,0 (1.2)	2,5 (2.8)	5,5 (4.4)	3,2 (1.8)	0,6 (0.7)	0,5 (0.7)	0,9 (0.7)
Upper abdomen and thorax	0,5 (4.7)	−1,7 (7.4)	0,0 (6.0)	−0,1 (1.1)	0,0 (0.9)	0,0 (1.4)	3,0 (2.8)	4,7 (3.6)	3,6 (1.9)	0,7 (0.7)	0,5 (0.8)	0,7 (0.8)
Lower abdomen	−0,4 (4.2)	−5,4 (7.1)	−3,3 (5.9)	−0,2 (1.0)	0,1 (0.7)	0,0 (1.2)	2,6 (2.7)	4,3 (3.7)	4,1 (2.1)	0,7 (0.7)	0,4 (0.7)	0,7 (0.7)
Limb	0,4 (5.2)	0,0 (4.8)	2,6 (4.8)	0,1 (0.9)	0,1 (0.8)	0,0 (1.1)	2,9 (3.0)	2,8 (3.3)	2,9 (2.5)	0,5 (0.7)	0,3 (0.7)	0,6 (0.9)

CV, correction vector; IQR, interquartile range.

Medians and IQRs and population systematic (∑) and random errors (σ) as a measure of interfraction setup accuracy and reproducibility for each treatment site analyzed.

Figure 4.

Distribution of the initial setup correction vectors (first CV) for prone and supine treatments grouped by treated anatomical site. Significant difference according to ANOVA test (p < 0.005) are marked with an asterisk (*). CV, correction vector; ANOVA, analysis of variance.

The distribution of MCV <5 min is shown in Figure 3 and corresponding medians (IQR) and Σ (σ) are shown in Table 4. For each treatment site, less than half of the analyzed patients had MCV <5 min in some fractions of their treatment course. Particularly, residual setup errors are related to a numerous patient cohort for intracranial (113 patients/334 fractions), head and neck (157 patients/372 fractions) and lower abdomen (113 patients/210 fractions) treatments. Residual setup errors are submillimetric for both cranial and extracranial patients. In particular, for cranial treatments maximum median translation is 0.2 mm in R-L direction while for extracranial treatment it is 0.4 mm in I-S direction for lower abdominal treatments. For each treated site, the distribution of the additional correction vector had lower median values and less variability than the distribution of the initial correction vector. After the application of the daily initial correction vector, Σ was reduced to a maximum of 0.8 mm and 0.7° for cranial sites and to a maximum of 1.4 mm and 0.7° for extracranial sites. Analogously, σ are reduced to less than 0.5 mm and 0.3° for cranial sites and less to 1.1 mm and 0.5° for extracranial sites. All of these MCV <5 min were corrected by updating treatment couch position.

Table 4.

Correction vectors applied within 5 min from the first setup correction (MCV <5 min)

	Multiple correction vector (MCV < 5 min)
			Median (IQR) [mm, °]						∑ (σ) [mm, °]
Anatomical sites	Npz (%)	Nfr(%%)	R-L	I-S	P-A	Pitch	Roll	Yaw	R-L	I-S	P-A	Pitch	Roll	Yaw
Intracranial	113 (49.1)	334 (7.1)	0,0 (0.7)	0,0 (0.6)	0,0 (0.4)	0,0 (0.5)	0,0 (0.5)	0,1 (0.7)	0,5 (0.6)	0,4 (0.5)	0,5 (0.4)	0,3 (0.3)	0,4 (0.4)	0,5 (0.4)
Head and neck	157 (26.3)	372 (4.4)	0,1 (0.8)	−0,1 (0.9)	0,0 (0.7)	0,1 (0.6)	0,0 (0.6)	0,1 (0.6)	0,6 (0.5)	0,6 (0.6)	0,6 (0.5)	0,6 (0.5)	0,6 (0.5)	0,5 (0.4)
Craniospinal	72 (49.7)	189 (8.6)	0,2 (0.9)	−0,1 (0.9)	−0,1 (0.7)	0,0 (0.6)	0,1 (0.6)	0,1 (0.8)	0,7 (0.6)	0,9 (0.6)	0,8 (0.7)	0,6 (0.4)	0,8 (0.6)	0,5 (0.4)
Paraspinal	16 (61.5)	34 (9.3)	0,3 (1.3)	0,1 (0.8)	0,0 (0.6)	0,1 (0.5)	0,2 (0.6)	0,0 (0.6)	1,1 (0.9)	0,4 (0.4)	0,5 (0.8)	0,2 (0.2)	0,3 (0.3)	0,3 (0.2)
Upper abdomen and thorax	18 (32.7)	36 (6.3)	0,1 (1.3)	0,3 (1.1)	0,1 (0.8)	0,0 (0.4)	0,1 (0.7)	0,0 (0.5)	0,9 (1.0)	0,4 (0.7)	0,6 (0.9)	0,4 (0.4)	0,3 (0.4)	0,4 (0.5)
Lower abdomen	113 (45.4)	210 (6.6)	0,2 (1.2)	0,4 (1.1)	0,0 (0.8)	−0,1 (0.4)	0,2 (0.4)	0,0 (0.4)	1,0 (0.9)	0,9 (0.8)	0,7 (0.8)	0,4 (0.3)	0,5 (0.4)	0,3 (0.3)
Limb	8 (34.8)	15 (4.7)	−0,5 (1.1)	−0,3 (1.9)	0,1 (1.4)	0,0 (1.2)	0,1 (0.7)	0,0 (0.4)	1,0 (1.1)	0,4 (1.4)	0,6 (0.9)	0,4 (0.6)	0,2 (0.4)	0,7 (0.6)

IQR, interquartile range; MCV, multiple correction vector.

Medians and IQRs and population systematic (∑) and random errors (σ) as a measure of residual setup error

The distribution of MCV > 5 min is described in Figure 3 and corresponding medians (IQR) and Σ(σ) are shown in Table 5. About half of patients with intracranial, head and neck and craniospinal lesions had MCV > 5 min in some fractions of their treatment course.

Table 5.

Correction vectors applied after 5 min from the first setup correction (MCV >5 min)

	Multiple correction vector (MCV > 5 min)
Median (IQR) [mm, °]						∑ (σ) [mm, °]
Anatomical sites	Npz(%)	Nfr(%)	R-L	I-S	P-A	Pitch	Roll	Yaw	R-L	I-S	P-A	Pitch	Roll	Yaw
Intracranial	143 (62.2)	1082 (23.0)	0,1 (0.7)	0,5 (1.2)	0,1 (0.4)	0,1 (0.5)	0,1 (0.6)	0,2 (0.7)	0,5 (0.5)	0,6 (0.5)	0,2 (0.3)	0,3 (0.3)	0,3 (0.3)	0,4 (0.4)
Head and neck	290 (48.6)	1224 (14.6)	0,1 (0.9)	0,5 (1.2)	0,0 (0.4)	0,2 (0.5)	0,0 (0.6)	0,1 (0.6)	0,6 (0.5)	0,8 (0.6)	0,4 (0.4)	0,4 (0.3)	0,4 (0.3)	0,4 (0.4)
Craniospinal	91 (62.8)	682 (31.0)	0,2 (1.0)	0,9 (0.8)	0,1 (0.4)	0,2 (0.6)	0,1 (0.6)	0,4 (0.8)	0,6 (0.6)	0,7 (0.6)	0,4 (0.4)	0,4 (0.3)	0,4 (0.4)	0,4 (0.3)

IQR, interquartile range; MCV, multiple correction vector.

Medians and IQRs and population systematic (Σ) and random errors(σ) as a measure of setup stability during irradiation.

Setup stability for the analyzed treatment sites is reported with translational and rotational median values lower than 0.9 mm and 0.4°. Σ reached a maximum of 0.8 mm in I-S direction in head and neck patients and 0.4°. All of these MCV > 5 min were applied to update treatment couch position.

Median time (IQR) between the first CV and MCV <5 min was 2.5 (0.9 min), 2.9 (1.0) min, 3.2 (1.0) min, 3.6 (0.9) min, 3.3 1.2) min, 3.8 (1.1) min, for intracranial, head and neck, paraspinal and craniospinal, upper abdomen and thorax, lower abdomen and limb, respectively. Additionally, median (IQR) time between first CV and MCV >5 min was 11.8 (5.0) min, 13.6 (6.2) min, 13.8 (5.3) min, for intracranial, head and neck, paraspinal and craniospinal respectively. For each treatment site analyzed, additional translational and rotational displacements were moderate to week correlated with time (Spearman’s correlation coefficient ranging from −0.2 to 0.4) (Supplementary Figure 1).

Discussion

This report extensively quantifies inter- and intrafraction setup accuracy achievable with the combination of multiple immobilization devices and daily image guidance based on 2D–3D registration on a large patient cohort, who underwent particle therapy between September 2012 and November 2018 at the National Center for Oncological Hadrontherapy (CNAO). This analysis provides an overview of our institutional experience in patient positioning from the beginning of our clinical activities until today, according to different treatment sites.

We analyzed positioning data retrospectively collected over 6-year time span. The strategy adopted to classify treatment fractions and correction vectors was based on a retrospective and systematic revision of the log files generated during clinical operations by the imaging and the patient positioning systems. This is an alternative approach to the extrapolation and evaluation of positioning data saved in the oncology information system. The adopted method is capable of detecting the modifications of the clinical procedures, positioning guidelines and data saving methods occurred over the years that forced us to define reasonable a-posteriori criteria to make data comparable and consistent and to streamline the data processing. Nevertheless, we put forward the idea that the number of patients and fractions included in this analysis is sufficient to provide robust results.

When interpreting the reported results, one has to consider that due to a strong increase in patient number over the last 2 years, the sum of the analyzed fractions in the recent period is about twice the fractions performed in the previous years, thus potentially being biased by the currently applied clinical setup procedures which underwent modifications and refinement since the start of the clinical activities at the National Center for Oncological Hadrontherapy (CNAO).

The reported setup uncertainties inherently derive from different sources along the whole treatment path from planning to dose delivery specific to our institution. Nevertheless, we believe that a fruitful benchmark can be performed with respect to other particle therapy centres and that the reported analysis could serve as a guidance for improving patient positioning procedures and estimate treatment plans robustness on a solid quantitative basis.

Some considerations on the specificity of the institutional image-guidance equipment are mandatory. The National Center for Oncological Hadrontherapy (CNAO) in-room imaging systems for daily patient position verification are different between the lateral and central rooms. Despite these differences, we opted to group the analyzed data set without considering the treatment room as a classification factor, since there is a predominant distinction of anatomical regions treated in the lateral rooms (typically cranial) and in central room (typically extra cranial) (Table 1).

As a whole, the observed clinically relevant initial setup errors demonstrate that daily image guidance is essential for correct patient localization and that treatment couches capable of accurate motion featuring 6 degrees of freedom are fundamental for setup adjustments.

A revision of published data on inter- and intrafraction setup errors according to different treatment sites and immobilization devices is presented in a ULICE report^18,30 whereas, the present work comprehensively reports a single institutional experience in patient positioning.

The reported interfraction accuracy and reproducibility of fractions in the cranial anatomical site are comparable to other setup accuracy/reproducibility studies based on in-room kV planar images and using similar immobilization devices, even though smaller patient cohorts was considered.^18,31–34 A trend toward posterior direction was observed. The magnitude of this displacement could be explained by taking into account the differences between the simulation and treatment couches. This issue was pointed out also by Amelio et al³¹ and Shafai-Erfani et al³⁴ when analyzing setup errors in cranial treatments. The use of thermoplastic masks in pelvic treatments is less common than in cranial treatments.¹⁸ At the best of our knowledge, studies regarding interfraction accuracy of thermoplastic masks for pelvic immobilization report quite comparable or smaller initial setup errors.^18,35–37 However, a trend toward inferior direction was observed in prone paraspinal, upper and lower abdominal patients. Possible causes of such predominant setup errors could be the loosening of mold with the increasing number of fractions, or the patient relaxation prior the planning CT acquisition. In fact, during the personalization of the immobilization devices the patient remains lying on the CT couch for a long time and he/she may drift before the CT scan acquisition. Increasing the time between the patient positioning and setup verification could reduce these discrepancies. Modifications to the patient positioning procedure for this category of patients treated in prone position are being considered in order to mitigate such biased initial setup errors. It is noteworthy that, especially for extracranial treatment (e.g. prostate, uterus or rectum irradiations), an additional source of geometric uncertainty is the target movement with respect bony anatomy used as a surrogate for patient alignment.³⁸ However, proper intestinal preparation with strict dietary/voiding regime (allowing reproducible filling condition of bladder and rectum) at planning CT and before each treatment fraction can significantly reduce the interfraction target movement.^39,40 To overcome this issue, we are recently performing CBCT prior treatment fraction for the daily evaluation of soft tissues volumes and positions reproducibility. In addition, multiple CT images have always been acquired at different fractions of the treatment course for dose distribution evaluation or replanning in case of significant changes of soft tissue anatomy. Proper immobilization of patients treated for limb-extremities tumours can be challenging and existing guidelines on this topic are not univocal.⁴¹ Our results suggest that the use of multiple combinations of immobilization devices for limb-extremities allowed for relatively small setup errors with respect to previous studies.^42–45

As previously commented, the first correction vector intrinsically contains residuals uncertainties not completely related to the reproducibility of the immobilization system but derived from the total institutional treatment preparation procedures: treatment planning images are acquired outside the treatment room and the treatment isocentre position is defined as relative to the CT couch which do not include weight compensation systems as the treatment couch. These differences in the planning and treatment couches are completely eliminated when the first correction vector is applied to the PPS. Submillimetre and subdegree accuracy of the PPS and PVS has been assessed during the treatment rooms commissioning²⁵ and are daily determined during pre-clinical QA measurements and periodically assessed with the use of independent control systems (e.g. measurements with laser tracker systems), and possibly in-loop controls.²⁵

Setup stability evaluation has been focused on intracranial, head and neck and craniospinal patients showing submillimeter Σ (σ) (Table 5). These results are in line with other published data of patients immobilized with thermoplastic masks and positioned with planar kV planar images grouped in the ULICE report.¹⁸ It should be recalled that the evaluation of setup stability is clinically required in poorly compliant or painful patients with potentially drifting setups during long-lasting treatments or with safety anatomical structures in close proximity to the target. Therefore, the presented analysis focused more on potentially critical and unstable cases and could be interpreted as a stability worst-case scenario evaluation. Extracranial sites are mainly treated in the central room where setup stability is continuously monitored using infrared optical tracking systems. Abdominal compression using thermoplastic masks, together with multifield planning approach and multifractionation, have also been assessed as additional strategies to mitigate the effect of patient motion.^46,47

For each treatment site, translational and rotational displacements have a moderate or week correlation with time. Amelio et al³¹ found the same results by analyzing setup stability over time in patients with head and neck, brain and skull base tumours (Supplementary Figure 1).

A critical point when investigating the dosimetric effects of setup errors is selecting the range of errors to simulate.⁷ Robustness evaluation should be performed on a case-specific basis or on massive plan recomputation, with real patient data and a comprehensive, fine-grained set of realistic setup errors. This report is intended to provide a contribution in this direction, as well as to pave the way for studies investigating the potential correlation of patient set-up uncertainties with tumor control and treatment toxicity.