Research Design
In an anthropomorphic phantom study,
three phantoms were exposed,
each conforming to a standard patient size at a given patient age (one-year-old,
five-years-old and ten-years-old).
These phantoms simulate human tissue in both attenuation properties and in structure (Fig. 1).
Phantom use removed the need to irradiate live patients,
resolving several ethical issues and allowing unlimited exposure repetition [6].
This also removed the possibility of case-variability regarding patient size [7].
Antero-posterior (AP) pelvis radiographs represent one of the most common paediatric projections [8,9],
and are the largest contributor to patient dose in paediatric projection radiography [10].
Dose and image quality indicators were recorded with each AP pelvis exposure.
An inherent Dose Area Product (DAP) meter was used to record phantom doses,
giving a digital dose read-out with each exposure.
Image quality analysis involved the measurement of a Signal Difference to Noise Ratio (SDNR),
which is an entirely objective approach,
eliminating bias [12].
This was on the basis that digital radiography is noise-limited.
To reflect clinical practice,
some images were also appraised by two independent observers [12].
As anatomic detail was limited within the phantoms,
chicken specimen radiographs simulating an infant hip and thigh region were acquired,
with perspex sheets adding simulated tissue depth and thus attenuation (See Fig. 2).
Phantom Exposures
Each individual phantom was positioned for an AP pelvis.
Optimal centring and collimation were applied,
as per published good practice [13].
For each phantom,
the first exposure maintained manual exposure factors to generate a reference image and reference dose and image quality values,
against which all other images would be compared.
Further exposures utilised the AEC,
with the Dose Constant Control (DCC) (commonly referred to as the "density" setting) set to '0',
before being reduced in equal increments of '1',
until the lower limit of '-5' was reached.
All other parameters remained fixed throughout for each phantom.
All exposures were repeated three times and the mean DAP value was taken to reduce random error arising from signal fluctuation [14].
Chicken Specimen Exposures
A chicken leg and thigh was placed on sheets of perspex,
and the perspex depth was varied until the DAP delivered to the chicken specimen equated to that received by each phantom with identical exposure conditions.
At this point,
the chicken specimen simulated the phantoms as best possible.
Manual and AEC exposures were then obtained as per the phantom exposures.
Analysis of Phantom Images by SDNR
SDNR calculation involved measurement of noise levels within unprocessed phantom radiographs.
This was achieved through placement of an elliptical region of interest (ROI) onto an area of unattenuated beam within each phantom image, representing an area of absolute signal [15].
Mean values and standard deviations within this region were recorded.
A second ROI was placed onto an area of homogenous,
attenuated beam near the femoral head of each phantom.
Again,
mean values and standard deviations were recorded.
All ROIs were of equal size,
shape and site on each image (See Fig. 3).
The SDNR of each image was then established using the formula shown in Fig. 4.
Analysis of Specimen Images by Visual Grading Analysis (VGA)
VGA applied evaluation criteria (Table 1),
assessing image features similar to those used in clinical image appraisal [1].
Two observers independently appraised the 25 specimen images.
Each observer was trained in the visual assessment task.
Experimental images were randomised and presented on high resolution monitors,
with the manually exposed image displayed side-by-side for comparison.
A maximum possible score of 3.0 denoted similar or superior image quality to that of the reference image.
Intra- and inter-observer reliabilities were subsequently tested,
showing strong correlation (for intra-observer correlation,
Spearman's rho = 1.0 and 0.73 for observer 1 and 2 respectively; for inter-observer correlation,
Spearmans rho = 0.76) and thus high reliability.
The average VGA scores of the two observers were therefore recorded for each image.
Overall Analysis
DAP,
SDNR and VGA scores from each image were then compared with those of the relevant reference image,
allowing comparison between experimental research and current clinical practice,
wherein manual exposures are currently favoured [11,3].