Patient Population
Following an experimental prospective design,
80 paediatric patients (age 6 - 17 y) consecutively referred for antero-posterior full spine x-ray for diagnosis or follow-up of scoliosis,
were randomly assigned to 2 groups of 40 patients,
without excluding complex or post-surgical cases.
Patients with additional lateral projections were included for radiation exposure analysis.
Image acquisition
All were imaged with the same x-ray digital flat panel detector (Digital Diagnost,
Philips Medical Systems).
All exams were done in posterior-anterior projection,
without lateral collimation,
using digital stitching as required by the spine length,
using two radiographic presets.
Group A followed manufacturer's preset and group B with an optimized dose preset (decided after phantom testing) described below:
Acquisition Parameters:
• Group A: As the standard manufacturer preset,
automatic exposure,
80 kVp,
sensitivity 600,
image density 0.
For the lateral projection,
kVp was set to 84.
• Group B: As the local optimized preset, automatic exposure,
90 kVp,
Sensitivity 800,
Image density -2.
For the lateral projection,
kVp was set to 95.
Data acquistion
Technical,
image,
and dose data were collected from the modality terminal along the examination,
and completed by DICOM header information processing.
Image reading
Five observers (2 senior radiologists,
2 junior radiologists,
and 1 senior orthopedists) performed three repeated independent readings,
along different days,
for a total of 15 readings,
of Cobb's angle and Risser grading,
blinded to any other information.
A collaborator helped loading the images and anotating the results of each reading.
All readings were performed in the same standard radiology workstation.
Statistical analysis
The Shapiro-Wilk was used to test for normality.
Mann-Whitney U test (two sided test) has been used to compare the measured values. Measurement of intra and inter observer agreement was analyzed using universal coefficient Berry-Mielke R,
a Cohen's Kappa generalization to an interval and ordinal measurement scales.
It is able to handles multiple qualifiers and multiple observation dimensions per observer,
and is appropriate for measuring reliability.
P-value of less than 0.05 was considered to indicate statistically significant difference.
Statsdirect software has been used por statistical analysis [12]