Automated patient positioning in CT using a 3D camera for body contour detection: accuracy in pediatric patients

The study was conducted in accordance with the Declaration of Helsinki and international standards of Good Clinical Practice. The medical ethics committee of our hospital waived the need for informed consent. All consecutive pediatric (< 18 years old) patients that underwent a CT examination of the head, thorax, and/or abdomen during routine clinical care in our hospital during a 5-month period from September 2018 to February 2019 were retrospectively included. All scans were performed on a dual-source CT scanner (DSCT) (SOMATOM Drive (software version Syngo CT VA62A), Siemens Healthineers) that was also equipped with a commercially available 3D camera for body contour detection (Siemens Healthineers).

Positioning and scanning of the pediatric patients were done by a team of dedicated CT radiographers as per normal clinical routine. Patients were positioned by the radiographer with the aid of laser beams within the gantry of the CT scanner. A horizontal laser line was projected on the lateral side of the patient and the table height was adjusted by the radiographer so that the laser line was assumed to align the center of the body region to be examined with the scanner isocenter. In our hospital, radiographers are trained to pay special attention to the position of the patient. Scanning was performed after this manual positioning done by radiographers.

3D camera images of children were collected and retrospectively used for the evaluation of the accuracy in patient positioning. The 3D camera is part of the CT system (SOMATOM Drive, Siemens Healthineers) and is attached to the ceiling and in front of the CT scanner, facing down onto the patient table. To start the 3D camera patient positioning process, the radiographer triggers a planning image with the camera when the patient is lying down on the scanner table in the target pose for the CT examination. The camera obtains two images: a color image and a depth image. Each pixel in the depth image describes the distance from the camera to the closest object surface. First, the algorithm detects the pose of the patient and body regions using the depth measurements and the known table position and shape [12]. After selection of the scan range, the ideal table height for the individual patient and the scheduled examination is proposed by the 3D camera such that the isocenter of the selected body region and the scanner isocenter align. Therefore, a virtual patient Avatar is fitted to the camera data in order to cope with areas that cannot be seen by the camera, e.g., through clothing or blankets or to handle positioning aids. The Avatar is a statistical shape model which in the fitting process assumes the pose and body proportions of the patient found in the depth data. The isocenter curve of the Avatar is finally averaged across all slices of the body region selected on the localizer radiograph. If Avatar fitting is not possible, then the isocenter curve is automatically obtained as the geometric center between the camera depth data and the central part of the scanner table. This is the same fallback as described before [11]. For adult patients, the camera images are processed by an algorithm [12], as described in detail before [10, 11]. However, the algorithm installed on the scanner that was used in the (adult) reference study was not optimized for pediatric patients. Therefore, prior to the start of the inclusion for this study, an algorithm training was performed on a separate large dataset of pediatric patients (n = 267) to improve the landmark detection and to adjust the Avatar shape model to the different body proportions found in pediatric patients. With the adaptations made, the new version of the algorithm was expected to work for adult and pediatric patients, but was not yet installed at the scanner during the inclusion. Therefore, the algorithm was applied retrospectively without the need of additional data and no user input was required afterwards. By doing so, offline system performance is equal to the real-world situation. When the algorithm was not able to fit the Avatar, the regular and automatic fallback was applied for patients positioned with and without a fixation aid, such as a baby cradle (Fig. 1) or a vacuum cushion (Fig. 2). The algorithm is currently not commercially available.

CT image reconstructions with a slice thickness of 3.0 mm and a reconstruction increment of 3.0 mm were used. The reconstructed field-of-view included the entire skin surface. The skin surface was extracted from the CT data in each axial slice and used to calculate the middle of the patient in the anterior-posterior direction. These values were averaged over all slices along the z-axis, providing the patient isocenter, to determine the ideal table height as described in detail before [11]. Accuracy in patient positioning is demonstrated as the difference between this ideal table height and the table height proposed by the camera algorithm or the radiographer. The accuracy is expressed as a single and absolute value in millimeter. The distribution of patients with a table height deviation lower or higher than the ideal table height is expressed as positive or negative numbers, respectively.

Cases with obvious patient movement or repositioning between the body contour detection by the 3D camera and the CT scan or with large items blocking the camera sight were excluded.

Significant differences in patient positioning between the radiographers and the 3D camera were analyzed by means of normality and a nonparametric test. The absolute table height deviation is a continuous paired variable reported as median (interquartile range (IQR)), calculated with Microsoft Excel (Microsoft Office Professional Plus 2016). Data distribution was tested with the Shapiro-Wilk test. Wilcoxon signed-rank test was used for comparison of the absolute table height deviation for the different body regions within and between the camera and radiographer group. A two-sided p value of < 0.05 was considered statistically significant. Statistical analyses were performed using SPSS (version 25, IBM Corp). Continuous measures of absolute table height deviation (mm) were calculated and evaluated with Microsoft Excel (Microsoft Office Professional Plus 2016). A post hoc power analysis was performed with G*Power (version 3.1.9.6) for the patients positioned with a fixation aid to determine the effect size, given the power to be achieved (80%) and the sample size available [13, 14]. The purpose of this test is to estimate the smallest possible difference in patient positioning accuracy between the 3D camera and the radiographer that can be detected in this study.

After exclusion of sixteen scans from the analysis due to patient repositioning after body contour detection by the 3D camera or because of large objects blocking the camera sight, one hundred and ninety-one scans were available for analysis. Of which, 149 pediatric patients were without and 42 patients were with a baby cradle or vacuum cushion. Within the group without fixation aids, the median age (IQR) was 11 years (6) and ranged between 3 months and 17 years old. For patients positioned in the baby cradle or the vacuum cushion, the median age (IQR) was 0.8 years (1.4) and ranged between 1 day and 6 years.

Within the group without fixation aids, median (IQR) absolute table height deviation was 10.2 (16.8) for abdomen, 16.4 (16.6) for head, 4.1 (5.1) for thorax-abdomen, and 9.7 (9.7) mm for thorax CT scans positioned by radiographers (Fig. 3; Table 1). A total of 41 (28%) patients were positioned higher than the scanner isocenter. The majority of patients were positioned lower than the scanner isocenter (Table 1).

For the 42 patients positioned in the baby cradle or the vacuum cushion, the median (IQR) absolute table height deviation was 8.7 (1.1) for abdomen, 9.1 (12.9) for head, 8.0 (3.1) for thorax-abdomen, and 15.3 (15.8) mm for thorax CT scans (Table 2). A total of 12 (29%) patients were positioned higher than the scanner isocenter. Within this group, the majority of patients were also positioned lower than the scanner isocenter (Table 2).

Within the group without fixation aids, median (IQR) absolute table height deviation in millimeter was 3.1 (4.7) for abdomen, 3.9 (6.3) for head, 2.2 (4.3) for thorax-abdomen, and 4.8 (6.7) for thorax CT scans (Fig. 3; Table 1). A small majority of the patients were positioned lower than the scanner isocenter (Table 1).

Within the patient group positioned in the baby cradle or vacuum cushion, the median (IQR) absolute table height deviation was 10.8 (8.3) for abdomen, 10.2 (15.3) for head, 17.4 (16.0) for thorax-abdomen, and 15.2 (15.0) mm for thorax CT scans. Within this group, the majority of patients were also positioned lower than the scanner isocenter (Table 2). An Avatar could be used in three out of nine cases when a patient was positioned in a baby cradle and the fallback had to be applied for all cases positioned in a vacuum cushion.

For patients positioned without the baby cradle or vacuum cushion, the median absolute table height deviations were higher for all four body parts when positioned by the radiographer compared with table height suggestion by the 3D camera (Table 1). For each of the four body parts, the largest deviation from the ideal table height was also higher for patients positioned by the radiographer. The largest deviation was 80.5 mm for an abdominal scan with the pediatric patient positioned lower than the isocenter by the radiographer. For the 3D camera, the largest deviation was 30.1 mm. Patient positioning accuracy for the 3D camera system and the radiographers differed significantly for all body parts (p < 0.005) with the exception of thorax-abdomen scans (p = 0.064). Figure 4 shows a case presentation with small deviations from the ideal table height of 0.13 mm and − 0.82 mm for the 3D camera and radiographer, respectively.

For patients positioned with the baby cradle and vacuum cushion, the deviation from the ideal table height by the 3D camera was up to 46.4 mm for a thoracic scan (Table 2). For the radiographer, the largest deviation was 42.4 mm below the isocenter. Difference between patient positioning accuracy of the 3D camera system and the radiographers was not significant (Table 2). For both groups, the mean and standard deviation (SD) of the table height deviation for all body parts combined (n = 42) was calculated for the post hoc power analysis. The mean (SD) was 17.1 mm (15.6) and 12.2 mm (9.4) for the 3D camera and radiographers group, respectively, with a correlation coefficient between groups of 0.2. The effect size or Cohen’s d corresponding with a power of 80% was 0.45. This is equivalent to a difference of 8.2 mm between camera and radiographers.

Positioning of young patients with lots of cabling, vital monitoring devices, or blankets was challenging for accurate positioning by both the 3D camera and the radiographers. Such a case presentation is presented in Fig. 5 with an ideal table height deviation of − 10.3 mm and 23.8 mm for the 3D camera and radiographer, respectively.