Analyses of the factors influencing the accuracy of three-dimensional ultrasound in comparison with cone-beam CT in image-guided radiotherapy for prostate cancer with or without pelvic lymph node irradiation

This retrospective study evaluated the setup verification for 49 patients who received external beam radiation therapy for localized prostate cancer from June 2015 to February 2017 at Sun Yat-sen University Cancer Center. Thirty patients underwent definitive radiotherapy with a total dose of 67.5 Gy over 25 fractions, and the other 19 patients received adjuvant radiotherapy after prostatectomy with a total dose of 64–72 Gy over 32–36 fractions. Each patient was treated with a 6-MV linear accelerator (Elekta Versa HD; Elekta, Stockholm, Sweden) with a volumetric modulated arc therapy (VMAT) plan. Both 3DUS and CBCT scans were repeated before each treatment if possible after obtaining patient consent. This study was approved by the ethics committee of the Cancer Center of Sun Yat-sen University. Written informed consent was obtained from all the patients before treatment.

The workflow of Clarity was recommended by the manufacturer and was shown in Additional file 1: Figure S1. Daily quality control (QC) procedures were performed by physical technicians prior to scanning patients. All patients were asked to follow a protocol to ensure a filled bladder and an empty rectum before planning CT and each treatment session. In case of an empty bladder or flatulent rectum, patients were asked to drink water to fill the bladder or undergo an enema to empty the rectum [16]. Bladder volume and rectal volume were 409.32 ± 149.27 mL and 54.53 ± 25.50 mL during the course of the treatment, respectively. All patients were scanned in the supine position with a 3-mm slice thickness on a Big Bore CT scanner (Brilliance™ CT, Philips, The Netherlands). They were immobilized using a cushion under the knees to create a reproducible setup. The same position was kept immediately after CT acquisition to acquire the reference US image (US_ref). For TAUS scanning, the US probe was manually placed 5–10 cm supra-pubic on the abdomen with a moderate pressure for a good quality of US images. Then the US probe is swept from superior to inferior, without translatory movement of the probe, to scan the prostate and bladder from retropubic to the top of bladder. For TPUS scanning, the transperineal US probe was placed on the perineum with a moderate pressure for good image quality, which allowed complete visualization of the penile bulb, prostate, and bladder. The US images were acquired using transabdominal ultrasonography (TAUS) for 30 patients undergoing definitive radiotherapy and using transperineal ultrasonography (TPUS) for the other 19 patients receiving adjuvant radiotherapy after prostatectomy.

All simulation CT images and information for all contours, including the gross tumor volume, clinical target volume (CTV), planning target volume (PTV), and organs at risk in Digital Imaging and Communications in Medicine (DICOM) format were imported into the workstation (Monaco 5.11.01; Elekta) as a reference for CBCT scans. The radiation treatment plan along with the CT images and all contour-related information were transferred into the Clarity planning workstation to define the reference positioning volume (RPV) on the US_ref image. For patients undergoing definitive radiotherapy, the RPV was the entire prostate. For those undergoing post-prostatectomy radiotherapy, the RPV corresponded to the bladder neck since it is included in the CTV according to the European Organization for Research on Treatment of Cancer (EORTC) guideline [20].

Daily QC was performed using the Clarity quality assurance (QA) phantom by physical technicians, and all 3DUS and CBCT measurements were retrospectively revised by one senior physician (Additional file 2: Figure S2). Our results were in accordance with the previous report in which the errors are only about 1 mm radially [21].

For image guiding, all patients first underwent alignment according to skin landmarks. Patient repositioning was performed on the basis of CBCT/CT registration results. US images were collected for comparison purposes only in this study. A daily US image (US_daily) was acquired before each treatment and registered on the US_ref image by manual movement in three orthogonal directions by a trained operator. CBCT images were acquired directly after the US_daily/ US_ref registration.

The clip box for CBCT/CT registration was defined as the area which just covered the PTVs. In detail, for patients without pelvic lymph node irradiation, the clip box included the PTV of prostate ± seminal vesicle for definitive radiotherapy or tumor bed for postoperative radiotherapy; for patients with pelvic lymph node irradiation, the clip box included both the PTV of prostate ± seminal vesicle or tumor bed and the PTV of the pelvic lymphatic drainage area. We matched CBCT to the reference CT based on the whole pelvic and the prostate ± seminal vesicle or tumor bed for patients treated with and without pelvic lymph node radiotherapy, respectively. The registration was performed automatically based on grey values using Elekta software, followed by a manual adjustment by the operators on the soft-tissue target volume. No rotational setup errors were determined, and only translational shifts were considered. The times required for the acquisition and registration of US and CBCT were estimated to be 3 and 4 min, respectively.

A total of 1166 paired US and CBCT translational shifts were collected. For patient p and session s, setup errors were denoted as T_CBCT,p,s and T_US,p,s for the CBCT and US modalities, respectively. For each session, the difference between CBCT and US shifts was calculated in three orthogonal directions as follows: δ_CBCT-US,p,s = T_CBCT,p,s - T_US,p,s. Means and standard deviations (SD) of the differences were calculated for all patients. To determine whether CBCT and US imaging had the same accuracy for prostate repositioning, the 95% limits of agreement (LOA) were calculated using the Bland–Altman method for each localization and each direction as follows: LOA = b ± 1.96*SD, where b is the bias (i.e., the mean of the differences between CBCT and US modalities), and SD is the standard deviation of the differences by assuming that the differences are normally distributed, indicating whether the difference between the upper and lower limits is acceptable in clinical practice [22]. Shift agreements, which were defined as the percentage of the number of sessions for which the difference between CBCT and US modalities was below 5 mm, were calculated.

The inter-operator variability (IOV) of the registration process for CBCT and 3DUS modalities was evaluated in each direction. A total of 40 images were selected for analysis. Twenty images of two patients undergoing lymph node irradiation and 20 images of two patients who did not undergo lymph node irradiation were retrospectively registered by two well-trained operators. For each session s of the patient p, the standard deviation σ_p,s was calculated over the assessments made by the two operators. The IOV was calculated using the following formula: IOV = RMS_p(RMS_s(σ_p,s)), with RMS_p indicating the root mean square over all patients and RMS_s indicating the root mean square over all sessions for the same patients.

Forty-nine prostate cancer patients who underwent 1340 treatment sessions were included in this study. CBCT examinations were successfully performed in a total of 1331 sessions (99.33%) and 3DUS examinations were successfully performed in 1175 sessions (87.69%). Insufficient bladder filling was the most frequent obstacle for successful 3DUS. A total of 1166 paired US and CBCT translational shifts were calculated. All the following analyses are based on data from these 1166 paired shifts, and the accuracy evaluation of 3DUS was presented on the basis of the values provided by the soft-tissue match in CBCT.

Setup errors in target localization between CBCT and the simulation CT were − 0.77 ± 3.98 mm, 0.05 ± 3.43 mm, and − 0.14 ± 3.33 mm in the SI, LR, and AP directions, respectively. For US imaging, setup errors were − 0.48 ± 6.09 mm, 0.23 ± 4.17 mm, and 0.40 ± 5.19 mm in the SI, LR, and AP directions, respectively (Fig. 1a). The average discrepancies between 3DUS and CBCT were − 0.28 ± 5.28 mm, − 0.16 ± 3.48 mm, and − 0.47 ± 4.31 mm in the SI, LR, and AP directions, respectively (Fig. 1b, Additional file 3: Table S1).

In evaluations using the Student’s two-sided one-sample t-test, the discrepancies between 3DUS and CBCT were not significantly different from zero in the SI (P = 0.066) or LR (P = 0.124) directions but showed significant differences in the AP direction (P < 0.001). Shift agreements for a range of ±5 mm were determined, and 67.6, 87.3, and 77.7% of the discrepancies were less than 5 mm longitudinally, laterally, and vertically, respectively. The upper and lower LOA values were 10.07 mm/− 10.63 mm, 6.66 mm/− 6.98 mm, and 7.98 mm/− 8.92 mm in the SI, LR, and AP directions, respectively (Additional file 3: Table S1).

Patients were divided into two groups based on whether they received pelvic lymph node irradiation (group A) or not (group B). As shown in Fig. 2, the best concordance between the 3DUS and CBCT registrations was found in the LR direction, with 84.9 and 89.2% of the discrepancies being less than 5 mm in groups A and B, respectively. Larger absolute discrepancies between 3DUS and CBCT registrations were found in the SI and AP directions. The mean discrepancies between 3DUS and CBCT were significantly different in all the directions in group A (SI, P < 0.001; LR, P < 0.001; AP, P < 0.001) and in none of the directions in group B (SI, P = 0.143; LR, P = 0.153; AP, P = 0.227; Table 1).

We investigated whether the relative accuracy of 3DUS with reference to CBCT is influenced by the approach used for acquiring US images. Patients were divided into TAUS and TPUS groups for further analysis. The best concordance between 3DUS and CBCT registrations was again noted in the LR direction, with 83.4 and 92.1% of the discrepancies being less than 5 mm in the TAUS and TPUS groups, respectively (Fig. 3, Table 2). In the other directions, all shift agreements were less than 80% and even under 70% in the SI direction (Fig. 3, Table 2). The discrepancies not significantly different in all directions in the TPUS group (SI, P = 0.899; LR, P = 0.151; AP, P = 0.103), but were significantly different in all directions in the TAUS group (SI, P = 0.019; LR, P = 0.005; AP, P < 0.001; Table 2).

In order to identify the potential factors that influence the accuracy of 3DUS with reference to CBCT, the logistic regression analysis was performed. For the entire cohort, patients aged ≤ 70 years showed a significantly higher rate of shift agreement in all three directions than that shown by patients aged > 70 years (OR = 1.37, P = 0.012 for the SI direction; OR = 1.46, P = 0.036 for the LR direction; and OR = 1.70, P < 0.001 for the AP direction). However, the rectum volume showed no significant influence on the discrepancies between 3DUS and CBCT in any direction. Treatment without pelvic lymph node irradiation, bladder volume < 300 mL, and TPUS were significant influencing factors for the discrepancies between 3DUS and CBCT in the SI (OR = 1.30, P = 0.036) and LR directions (OR = 1.47, P = 0.03), in the LR (P = 0.004) and AP directions (P = 0.031), and in the LR direction (OR = 0.43, P < 0.001), respectively (Table 3).

The IOV for both modalities is shown in Additional file 4: Table S2. The IOV values in the SI, LR, and AP directions were 1.4 mm, 1.3 mm, and 1.7 mm, respectively, for CBCT, and the corresponding values for US were 1.7 mm, 2.1 mm, and 2.6 mm.