The present study aims to evaluate whether a LFT is adequate as a clinical radiographic for initial evaluation and/or the subsequent follow-up monitoring of AIS by evaluating image quality, reliability, and agreement and measuring radiation dose.
Materials and methods Study design This study was divided into two parts: an experimental part and a clinical part. The aim of the experimental part was to evaluate the physical characteristics of image quality and observer-based evaluation of the LFT and ORT using a pediatric trunk phantom and to measure the radiation exposure. Image quality was also evaluated for radiographs of two "3D-printed" models of scoliotic spines (3DPSs), where the agreement and reliability for LFT and ORT were determined. The aim of the clinical part was to evaluate the inter-rater reliability of the actual clinical examinations for LFT radiographs of AIS from our outpatient clinic. Ideally, "agreement" should be determined on the clinical radiograph using both the LFT and ORT techniques. However, this would require "simultaneous" double examinations, thus exposing the participants to excess radiation. This was deemed ethically unviable; instead, we performed double examinations on the 3DPS. Reliability and agreement were evaluated as proposed by Langensiepen et al.2 This study was evaluated and approved by the local Research Ethics Committee (Journal number: 17025334).
Imaging system setups of the LFT and the ORT The examinations were performed in either the posteroanterior or anteroposterior projection for the ORT and LFT, respectively. The patient was standing facing the image intensifier/detector or radiation tube with extended hips and knees and with their feet 10 cm apart. Two lead aprons were placed in the interscapular (mammary) and sacral (genitals) regions. The distances to the radiation tube (source to detector) were 100 cm for LFT and 230 cm for ORT. For the LFT, two images of the thoracic and lumbar spine were necessary to cover the thoracolumbar spine. For ORT, only one exposure was needed. The LFT was performed on a DelftDI D2RS system with fluoroscopic exposure of 4 pulses per second/frame per second, and the ORT was performed using a digital Carestream DRX-Evolution system with automatic exposure control. The following acquisition parameters were assessed: tube potential (85 kVp, density set at low for LFT and 71 kVp, density set at 0 for ORT), both with grid, using manual exposures (LFT) and automated mA selection exposure control (ORT) and no additional filtration for LFT and additional filtration (1 mm Al + 0.1 mm Cu) for ORT. The pixel size of 0.139 mm for ORT gives a maximum resolution of 3.6 lp/mm. The pixel size of 0.159 mm for LFT gives a maximum resolution of 3.1 lp/mm.
Evaluators In the experimental part, observer-based assessments were performed once by 3 evaluators. Evaluator 1 was an experienced (22-year practice) pediatric orthopedic surgeon, evaluator 2 was an experienced (16-year practice) pediatric orthopedic surgeon, and evaluator 3 was a reporting radiographer with 10 years of practice. Evaluator 1 conducted all analyses for the primary physical characteristics. For the 3DPS, measurements of CA and classification according to Nash and Moe (NM) were conducted three times as single sessions at least 1 week apart. In the clinical part, the clinical radiographs were analyzed in a single session by evaluators 1 and 2. Evaluators were blinded to patient identity and clinical information. All image evaluations were performed on a PACS system (Impax 6.4.0, Agfa ® HealthCare, Mortsel, Belgium) on a 3-megapixel viewing station by the 3 evaluators separately.
Part 1. Evaluation of image quality of the LFT and ORT imaging systems The pediatric trunk phantom The Pediatric Whole Body Phantom "PBU-70" was examined radiographically for image quality. We chose only the phantom torso of a 4-year-old child with a height of 105 cm with life-size, full-body anthropomorphic measurements with embedded soft tissue substitutes of a synthetic skeleton, lungs, liver, mediastinum, and kidneys (phantom).16 We performed 25 radiographic examinations with both techniques, where the phantom was placed and repositioned for every recording.
The two 3D-printed AIS models Radiographs of the two 3DPSs with small and severe lumbar scoliosis were recorded using both techniques. We performed 1 set of radiographs with both techniques. The three evaluators performed 3 measurements of CA and NM at least 1 week apart. Inter- and intra-rater reliability for the three evaluators were assessed using analysis of interclass correlation as well as the mean absolute difference (MAD), standard error of measurement (SEM), and Bland-Altman plots for the mean differences with additional analysis for systematic differences. We defined accuracy from direct measurement with a protractor for medical purposes. This was seen as a surrogate measure of overall accuracy. The phantom and 3DPSs are shown in Figure 1, and the radiographs are shown in Figure 2.
Figure 1. The pediatric whole-body phantom "PBU-70" and one of the two 3D-printed scoliosis models.
Figure 2. Radiographs using the LFT (darker images) and the ORT (brighter images) of the 3D-printed scoliosis models (top four left) and the phantom (top two right). Radiographs (ORT) of scoliosis with regions of interest (below left) and clinical LFT radiographs (below right).
Primary physical characteristics of image quality evaluation: noise, contrast, and SNR We examined the imaging characteristics of the LFT and ORT radiographs of the phantom and the two 3DPSs. Twenty-five consecutive radiographs of the phantom using both techniques were evaluated. The objective primary physical characteristics were the contrast, random noise, signal-to-noise ratio, and contrast-to-noise ratio. The contrast was defined as the relative signal difference between the two predefined locations of the bony vertebral spine and the surrounding adjacent tissue of the spine, namely, a square region of interest of the vertebral spine, where the upper and lower pedicles and endplates were included, along with a similar region of interest of the adjacent soft tissue, where no bony soft tissue was included (see Figure 2). Random noise was defined as the fluctuations of the signal over the image when uniformly exposed, as expressed by the standard deviation.14,17 SNR was defined as the ratio of the signal (defined as the signal of the object) divided by the standard deviation (of the background).17 CNR was defined as the ratio of the signal (defined as the difference of the signal of the object and the background) divided by the standard deviation (of the background). Figure 2 shows these regions of interest and radiographs of the phantom, the 3DPSs, and the clinical radiographs using the LFT.
Observer-based methods of image quality evaluation using the visibility of anatomical structures For ethical reasons, we only performed observer-based evaluations on the images from the phantom and the 3DPSs since we did not want to expose patients to double examinations of the ORT and LFT. We utilized two observer-based methods for comparison of image quality, namely, scores of the image criteria (ICS) and visual grading analysis (VGAS).14,18 We compared a reference LFT image to all 25 ORT images as well as an ORT image to all 25 LFS images. The reference images were selected randomly.14 Using the ICS, the absolute level of image quality of specific structures was determined by evaluating a specific structure compared to the same structure in the reference image by the observer; thus, the observer's decision threshold was constant. The task of the observer was to decide whether the specific structure was superiorly visually reproduced1 or inferiorly visually reproduced when compared to the reference (0) (the criterion). We calculated the ICS for the lumbar (L3) and thoracic (Th6) vertebrae as ICS=∑Ii=1∑Cc=1 ∑Oo=1 Fi,c,o IxCxO where Fi,c,o = fulfillment of criterion c for image i and observer o. Fi,c,o = 1 if criterion c is fulfilled; otherwise, Fi,c,o = 0, I = number of images, C = number of criteria, and O = number of observers. The number of images assessed for both the LFT and ORT was 25 (I = 25), the number of evaluators was 3 (O = 3), and the numbers of criteria were 6 for the thoracic spine and 7 for the lumbar spine (C = 6 for the thoracic spine and C = 7 for the lumbar spine) in accordance with the chosen CEC criteria.
We also graded the appearance using the VGA method by grading the visibility of specific anatomical structures between images of the two radiographic techniques according to Månsson.18 The visibility was graded according to a five-level scale: clearly inferior visibility of a specific structure in the image compared to the same structure in the reference image (-2), slightly inferior to (-1), equal to (0), slightly better than (+1), or clearly better than (+2). We calculated the VGAS for the lumbar (L3) and thoracic (Th6) vertebrae as VGAS=∑Ii=1∑Ss=1 ∑Oo=1 Gi,s,o IxSxO where Gi,s,o = grading (-2, -1, 0, +1, or +2) for image i, structure s, and observer o; I = number of images; S = number of structures; and O = number of observers. The number of images assessed for both the LFT and ORT was 25 (I = 25), the number of evaluators was 3 (O = 3), and the numbers of criteria were 6 for the thoracic spine and 7 for the lumbar spine (C = 6 for thoracic spine and C = 7 for lumbar spine) in accordance with the chosen CEC criteria. The specific evaluated (vertebral) structures were in accordance with the European guidelines on Quality Criteria for Diagnostic Radiographic Images in 1990 and 1996.19-21 The CEC lumbar spine criteria, 1990, were chosen for the thoracic evaluation of the images since they focused on osseous vertebral spine structures as well as adjacent soft tissues. For the lumbar evaluation, the CEC lumbar spine criteria, 1996, were chosen since they also included an evaluation of the sacroiliac joints.
Radiation dose The recording systems of LFT and ORT measured the dose area product (DAP) by an integrated DAP meter, and these were stored with the images. Monte Carlo calculations using X-ray dosimetry software (PCXMC, version 2.0; Stuk, Helsinki, Finland) were utilized to determine the effective radiation doses. This was based on the recorded DAP and the principal patient size of the phantom as well as technical and geometric exposure parameters.
Part 2: Evaluation of AIS radiographs using the LFT for clinical reliability Clinical routine radiographs using LFT Low-dose fluoroscopic technique was our regular routine clinical radiographic method for AIS for 3 years. One hundred thirty-six adolescent patients with AIS were included as participants. The sex ratio F:M was approximately 2:1. The average age was 13.4 years (range 6-17). We retrieved 342 LFT images of 136 patients with AIS. Two independent evaluators performed measurements of six radiographic parameters once and separately, where they were blinded to the clinical data and previous evaluations. The parameters were the CA, the level of the upper and lower vertebrae used for determining CA, the NM, and the Metha angles on the left and right sides at the apex vertebrae. Table 1 shows the radiographic characteristics of the participants. Figure 2 illustrates the standard clinical LFT radiographs.
Table 1. Distribution of participants according to the classification of King and Moe. Type 1: an "S"-shaped deformity, in which both curves are structural and cross the CSVL (midline), with the lumbar curve being larger than the thoracic curve. Type 2: an "S"-shaped deformity, in which both curves are structural and cross the CSVL, with the thoracic curve being larger than or equal to the lumbar curve. Type 3: major thoracic curve in which only the thoracic curve is structural and crosses the CSVL. Type 4: long "C"-shaped thoracic curve in which the fifth lumbar vertebra is centered over the sacrum and the fourth lumbar vertebra is tilted into the thoracic curve. Type 5: double thoracic curve.
Dextro convex Sinistro convex Type 1 7 29 Type 2 10 0 Type 3 24 5 Type 4 22 28 Type 5 2 1 Unclassifiable 8 Statistical analyses Reliability was assessed using the interclass correlation coefficient (ICC). In part 1 of the two 3DPS models, inter- and intra-reliability for CA and NM were assessed using a 2-way mixed model for consistency for the 3 evaluators for their 3 separate measurements. Agreement was defined as the MAD, SEM, and Bland-Altman plots for the mean differences with additional analyses of one-way t-test and logistic regression for significant and systematic differences, respectively. In part 2 of the clinical evaluation, the single measurements of CA and the other 5 radiographic parameters of the two evaluators were assessed using a 2-way mixed model of ICC for absolute agreement for inter-rater reliability.
Inter- and intra-rater reliability assessed by the ICC was considered with the following limits of agreement: poor; 0.0-0.20 slight; 0.21-0.40 fair; 0.41-0.60 moderate; 0.61-0.80 substantial; and 0.81-1 almost perfect. Independent sample t-tests or Mann-Whitney nonparametric tests were conducted for differences in SNR and CNR, differences in image quality, and radiation dose. This depended on if normal distribution was present. This was tested by Kolmogorov-Smirnov and Shapiro-Wilk tests and evaluation of QQ plots. We considered a p-value of \<0.05 as a significant result. ICC and other statistical evaluations were performed using IBM SPSS Statistics, version 22 (IBM©, Chicago, Il, USA).
