This work covers most of the innovations.
It embraces 8 direct radiography (DR) systems of 7 different models and 5 different manufacturers; 2 computed radiography (CR) systems are reported for comparison (list in Fig.
2).
System A1 and A2 are twin systems which underwent an x-ray tube replacement in 2010; Systems A3 and A4 are identical apart from the detector technology (Fig. 3 and 4).
Data of System D1 are preliminary.
The technological innovations introduced are related to 3 different aspects of a mammography unit,
as explained in Fig.
1:
- X-ray production [3,
4,
5,
6],
- X-ray acquisition technique,
- X-ray detection technology [7,
8,
9].
The work pursued two goals,
the physical characterisation of the different systems and the comparison of their image quality with respect to dose.
A.
Physical Characterisation: Performed according to IEC 62220-1-2 with either beam quality RQA-M2 (Mo/30 mm Mo + 2 mm Al,
28 kV) or target/filter combination W/50 mm Rh in equal conditions and neither antiscatter grid nor compression paddle in place.
Evaluation of
- modulation transfer function (MTF) with edge device placed directly on detector cover; entrance air kerma corresponding to levels in “normal” use (ca.
90 μGy);
- normalized noise power spectrum (NNPS) – mean value of 3 subsequent images;
- detective quantum efficiency (DQE).
Evaluation has been carried out on linearized dicom images “for processing” with the open source software ImageJ and the plugin “qa-distri/DQE Panel v7” extended for DEQ calculation.
Air Kerma values were determined with a regularly calibrated ion chamber (Radcal 9010 + 10X6-6M,
Radcal Corporation,
USA).
Uncertainty: <2% for MTF,
<4% for NNPS, 8% for DQE (precision <5%).
B.
Phantom image quality evaluation: The process was carried out in two steps,
applying two different methods for image quality evaluation [10,
11]:
- optimization of exposure technique: Using contrast to noise ratio (CNR) as working parameter,
for each phantom thicknesses (2 cm – 7 cm) those high voltage values and target/filter combinations were determined,
which exhibited the best CNR at equal average glandular dose (AGD).
CNR measurement followed the European Guidelines.
The test object consisted of several polimethylmethacrylate (PMMA) layers covering the whole detector area with a small contrast object (foil of 0.2 mm Al) positioned above 2 cm PMMA,
as indicated in Fig.
5.
The adopted formula was
Fig.: Formula 1
where MPV = mean pixel value
SD = standard deviation.
- image quality assessment: The method of contrast detail (CD) evaluation was used with the phantom CDMAM ver.
3.4 and the commercial software CDMAM Analyser (both Artinis,
The Netherlands) for automated image analysis (see Fig. 6 and Fig.
7).
The CDMAM phantom consists of several PMMA layers and a detail tablet,
which in this study was always positioned above 2 cm of PMMA.
Additional PMMA layers needed to achieve the desired total phantom thickness were put on top of the detail tablet.
In order to reduce uncertainty,
for each exposure technique a set of 8 raw images was considered (detection rate of 75%) [12].
In DR systems the phantom block was slightly moved between successive images.
Results are expressed in terms of a group contrast detail curve and,
additionally,
of an overall quality index,
the Image Quality Figure Inverse (IQFinv),
defined as
Fig.: Formual 2
which increases when detail detection rises.
This figure has prevously been shown to represent an objective and absolute measure of the image quality [10],
suitable for comparison of different equipments.
The linear relation between CNR and IQFinv was verified in all systems (see Fig. 8 for an example).
Under the hypothesis that image noise is dominated by quantum noise,
image quality improves applying a higher dose according to CNR2 prop.
dose.
That means CNR2/dose is constant for each kV/phantom setting.
In analogy we defined the dose independent figure of merit IQFinv2/AGD to compare systems performances in various conditions.
Fig.: Formula 3
Uncertainty of the CNR value was estimated as standard deviation of the evaluation of 10 repeated exposures.
AGD was calculated according to Dance.
Uncertainty: CNR < 3% in DR,
< 5% in CR; IQFinv < 4%,
AGD 10% (precision 3%),
mean IQFinv2/AGD 5 % - 7%.