·Home ·Table of Contents ·Methods and Instrumentation

Image Quality comparison of Digital Radiographic Systems for NDT P. Willems, P. Soltani, B. Vaessen Agfa NDT, Septestraat 27, B-2640 Mortsel, Belgium Phone: 32-3-4442881, Fax: 32-3-4447655 Contact

1. Introduction

Various digital X-ray detectors have become available for NDT applications
Recently introduced flat panel X-ray detectors compete with storage phosphor and CCD based systems, which were introduced several years ago.
Various materials are used to convert the X-rays into visual photons or directly into charge carriers (electrons & holes). Scintillator screens consist either of coated powder phosphor or of vapor deposited needle crystals.
The multitude of systems offered is confusing for the NDT user. The following questions are often asked:
• Which technology fits best a specific application?
• How does the image quality of the different digital technologies compare with that of film?

Agfa not only supplies a vast assortment of NDT films but also offers digital solutions ranging from film digitizers over storage phosphor scanners to direct conversion flat panels.
In order to match application requirements and system specifications we need a tool to predict image quality for a proposed detector system in a given application context.

Here we report on work in progress. We will focus solely on image quality and develop a simple model that allows us to estimate the attainable image quality for a given system and for given exposure conditions. The estimated image quality of digital systems will be contrasted with that of film.

2. Storage Phosphors (Computed Radiography)

The system consists of a set of image plates usually contained in cassettes and a central scanner. The scanner is coupled to a workstation for image processing. The workflow is shown in fig. 1

Fig 1: Workflow for Storage Phosphers (CR)

For the image quality calculations we made the following assumptions:
Pixel size is 100µ and the scan spot has a FWHM of 56µ. Electronic noise was 0, fixed pattern noise was 0,1% and the swank factor (Poisson excess noise) was 0,588. An overall gain of 19 was assumed; corresponding with a screen gain of 755, an overall collection efficiency of 10% and a PMT quantum efficiency of 25%.

3. Flat Panel Detectors

The detector consists of a a-silicon pixel matrix coupled to a scintillator. Each pixel has a photodiode to capture the visual photons emitted by the phosphor. The photodiode converts the visual photons into charge carriers (electrons or holes) that are stored on the capacity of the pixel switch. Upon read-out the pixels are connected to charge amplifiers.

The scintillator layer may consist of a powder phosphor layer or of a vacuum deposited needle crystal layer. Needle crystals reduce light scatter and screens can be made thicker i.e. a higher X-ray absorption can be achieved for the same resolution.

For a direct conversion flat panel a photoconductor is used instead of a scintillator. X-rays are converted directly into charge carriers. A simple collector electrode replaces the photodiode. A bias field is applied to separate the charge carriers and to collect them. For a-selenium systems the resolution is only limited by pixel size.

For the calculation we made the following assumptions:

Direct conversion Flat Panel using a 500µ Selenium layer (Selenium)
Specifications of the Agfa DRC detector: pixel size is 139µ, the fill factor is 86%. We assume that electronic noise per pixel s_add = 2000e and that residual fixed pattern noise is 550ppm. No Poisson excess noise was assumed for radiation energies far above the K-edge of Selenium

Indirect Conversion Flat Panel using a Regular Gd2O2S powder phosphor screen. (GOS Reg)
Pixel size is 126µ, the fill factor is 57% . We assume that electronic noise and that residual fixed pattern noise are equal to II. The diode quantum efficiency was assumed 75%. The swank factor was set to 0,75.

Indirect Conversion Flat Panel using 500µ CsI vapor deposited needle crystal screen (CsI -C)
Pixel size is 143µ, the fill factor is 70%. We assume that electronic noise and that residual fixed pattern noise are equal to II. The diode quantum efficiency was assumed 65%. The swank factor was set to 0,85.

4. Image Quality

Image quality may be characterized by physical parameters such as system gain, granularity (s_D or noise power spectrum) and resolution (Modulation Transfer Function).

For many NDT applications the highest possible image quality is required. This necessitates the optimization of the exposure conditions with regard to exposed dose, radiation hardness, primary to scatter ratio and geometrical unsharpness. Finally the detector choice will determine the maximum attainable image quality.
A simple equation that relates the minimum detectable image contrast C_min to the physical parameters characterizing image quality has been proposed [1].


C_min(q₀,A) = k . (1+S/P) . 1/ SNR(q₀,0) . 1/ [q(A). A]^1/2 (eq. 1)

C_min is a function both of the exposed dose characterized here by the impacting X-ray flux q₀ and the detail aperture A (for a circular detail with radius, A=pr²). The subjective viewing conditions are compounded in the constant k, the observer's threshold criterion. The exposed dose determines the attainable signal to noise ratio at zero spatial frequency SNR(q₀,0).
Radiation hardness determines the radiation contrast and hence the signal amplitude.
The contrast is degraded both by radiation scatter (build-up factor (1+S/P) and resolution (resolution degradation factor q(A)). The resolution degradation is due to both geometrical unsharpness and to detector MTF.

When exposure conditions are optimized the remaining critical parameter for detail detection is [q(A).A)] which reduces to the detector MTF. This is why for critical applications, detector resolution should be regarded as the most important decision factor.

5. Resolution

For film the detector MTF reduces to the inherent unsharpness (U_f). It is well known that the inherent unsharpness increases with radiation energy. The effect of increasing radiation energy on inherent unsharpness is shown in fig. 2

Fig 2: MTF from inherent Unsharpness


For the systems of sections 2-3, we will compare resolution of the digital systems with film. For the indirect flat panels and for CR we have to account both for sampling and screen MTF. The intrinsic resolution of the selenium layer used in the Agfa DRC flat panel is extremely high [2] hence only sampling (pixel size) determines the MTF.

Fig 3a: MTF Film vs. Digital systems 100kVp, Object 10mm, FFD=1000mm ,Focal spot Film=2mm, Digital=1mm

Fig 3b: MTF Film vs. Digital systems 200kVp, Object 20mm, FFD=1000mm, Focal spot Film=2mm, Digital=1mm

We have depicted the following situations in fig. 3. The resulting MTF is the product of geometrical unsharpness and detector MTF:

1. 100kVp radiation, FFD= 1000mm, object of 10mm in contact with detector. The focal spot size was 2mm for film and 1mm for the digital detectors. Made possible because of the higher speed of digital detector systems.
2. 200kVp radiation, object of 20mm in contact with detector. FFD and focal spot as A.
3. Ir192 radiation, object of 35mm in contact with detector. FFD 750mm, focal spot 2mm for both detector systems.

We may conclude that for 100kVp and a 10mm object, film is superior to the digital systems. The Selenium Direct Conversion Flat Panel comes closest. For 200kVp Selenium even improves on film thanks to the smaller focal spot.

Fig 3c: MTF Film vs. Digital systems Ir192 , Object 35mm , FFD=750mm , Focal spot =2mm

Fig 4: Mean free path of secondary x-ray photons


The indirect flat panels and CR have similar MTF curves. For all situations, even for Ir192 and thick objects, they do not catch up with film and Selenium.

For practical situations the results of the indirect systems and CR will be even worse, since we have postulated that the optical scatter is the dominant source for unsharpness and neglected the X-ray contribution. For medical radiation energies this is a valid assumption but for NDT where gamma sources and linear accelerators are used, the X-ray contribution can not be neglected. The inherent unsharpness for the digital detectors will scale with the mean free path L of the secondary X-ray photons shown in fig. 4

Selenium has a much shorter mean free path than the other digital systems. From this we may expect that CR and the indirect flat panels will get even worse compared to the selenium flat panel at high radiation energies. The selenium resolution may be expected to exceed that of film for linear accelerator energies.

6. Detective Quantum efficiency

For medical X-ray applications it is customary to combine gain, MTF and noise power spectrum into a single parameter: the detective quantum efficiency or DQE .

DQE = (SNRout / SNR in)2

(eq. 2a)


This parameter depends both on the exposed dose (q₀) and on the spatial frequency (w).
Experimentally the DQE values can be derived from the X-ray absorption a, the system gain G, the modulation transfer function MTF(w) and the normalized noise power spectrum NNPS(q₀,w).

DQE(q0,w) = (a. G . q0. MTF)2 [q0 . NNPS(q0,w)]

(eq. 2b)


We have applied a simple model that yields the DQE in function of:
• System parameters such as thickness of the phosphor layer, conversion efficiency of the phosphor, collection and/or quantum efficiency of the photodiode, CCD or PMT etc.
• Exposure conditions such as radiation type, radiation quality (primary to scatter ratio) exposed dose, geometrical unsharpness etc.

The model is based on an analysis of noise transfer. The imaging process is represented as a number of cascading imaging steps. For each signal and noise transfer characteristics are derived. The use of such an analysis has become quite popular for medical X-ray systems. It is described by several authors [3-4]. It has been verified experimentally for medical radiation spectra (Tungsten source 40-140kVp).

We will estimate the DQE values for radiation with a 3,5mm Cu HVL spectrum (220kVp - 8mmCu). These values are then compared with film DQE derived from measurements of film granularity s_D and gradient g.

For film the following formula is used:


(eq.3)


Herein is:
• q₀ : dose to reach density 2 above fog
• G = (Log₁₀ e) . g (Film gain)
• g : gradient or slope of characteristic curve at density 2 above fog
• MTF(w) : Film Modulation Transfer Function
• s_D : granularity at density 2 above fog
• A : microdensitometer scanning aperture
We have assumed that the noise power spectrum of film is only weakly dependent on spatial frequency for the small interval [0,4] lp/mm. For small densities and light exposures this is a valid assumption. For X-ray exposures, quantum mottle is the main noise source at low spatial frequencies. The NPS then decreases as spatial frequency increases, hence the depicted film curve actually underestimates film DQE at higher spatial frequencies.
The DQE at zero spatial frequency may be used as a figure of merit to benchmark the various systems. We have calculated DQE(q₀,0) for a selection of digital systems. The key features of these systems are summarized in table 1.

Detector	Material	mg/cm2	a	pixel (mu)	fill factor	w (MTF20%)	Nyquist	MTF Nyq.
CsI -A	CsI	50	4,7%	100	75%	3,50	5,0	9%
CsI -B	CsI	160	14,0%	200	90%	2,25	2,5	16%
CsI-C	CsI	200	17,0%	143	70%	2,25	3,5	8%
GOS Fine	Gd2O2S	34	3,8%	126	57%	(5,80)	4,0	40%
GOS Reg	Gd2O2S	70	7,6%	126	57%	2,75	4,0	10%
GOS Fast	Gd2O2S	133	14,0%	126	57%	2,00	4,0	6%
DRC	a - Se	220	4,7%	139	86%	(6,50)	3,6	68%
CR	BaFBr	75	5,6%	100	100%	2,50	5,0	4%
Table 1: Digital Detectors

The other parameters are as specified in sections 2-3.

Film	dose D=2	g D=2	sD D=2
D2	29,0	5,4	0,015
D3	14,0	4,5	0,016
D4	8,7	4,4	0,019
D5	4,6	4,4	0,026
D7	3,2	4,4	0,031
D8	2,5	4,0	0,035
	mGy
Table 2: Film Data

Fig 5: DQEO 220kVp-8mm Cu


In fig. 5 we compare the digital systems with current Agfa NDT films. The values we used to calculate film DQE₀ are shown in table 2. For film the DQE refers to the dose of table 2. For 220kVp + 8mm Cu this corresponds to an X-ray flux q₀ = 1,86E+06 at D7 dose. The digital detectors were exposed with a dose of 10% of D7.

We see that all films exhibit superior DQE₀ values compared to the digital systems. The best results are obtained for the midrange films D4, D5 and D7 , D8 suffers from a granularity that has increased more than speed increase, for D2 and D3 speed is reduced more than the granularity reduction.
For the indirect TFT systems it is obvious that when the scintillator screen is optimized in order to obtain a better resolution (CsI -A , GOS Fine) the DQE₀ value is less compared to that of the selenium system. Now let us compare resolution i.e. the spatial frequency where MTF = 20% (shown in table 1). Although a significant improvement has been realized for CsI-A and GOS Fine compared to the thick converter screens, resolution is still far inferior compared to that of Selenium.
The thickest CsI screen (CsI - C) has the highest X-ray absorption and hence its DQE0 value is nearly 3x that of the selenium panel. Now referring to eq. 1, we may quantify the benefit of this high DEQ₀ value. Suppose we are looking to a weak detail with large dimensions where resolution degradation is insignificant. To reach the same minimum detectable contrast detectors need equal SNR values. The signal to noise value is related to the DQE by:

SNR(q0,w) = [q0 . DQE(q0,w) ]1/2

(eq. 4)

Hence for a detector X with an inferior DQE₀ value the exposed dose has to be increased by a factor that equals the square of the ratio DQE₀(X) / DQE₀(CsI-C). For the selenium panel this results in an increase of dose of 176%. For smaller details the resolution degradation becomes important and the situation is reversed e.g. already for a 500µ; detail both panels need the same dose.
Finally in fig. 6 we show the variation of DQE with spatial frequency. Structurix D7 is compared with the systems of sections 2-3.

Fig 6: DQE(w) 200kVp - 8mm Cu


We see that film is superior. For higher spatial frequencies the difference between film and the digital detectors increases drastically. Beyond 3lp/mm the differences between the different flat panels fade out. CR is no match for the flat panels.

7. Conclusion

The resolution should be the most important criterion for detector choice for critical applications. For NDT radiation hardness (e.g. 220kVp - 8 mmCu) film DQE is superior compared to all available CR and flat panel detector systems currently available. Current CR has a low DQE compared to the flat panel systems. For large details the selenium Agfa DRC flat panel needs ± 75% higher exposure to give the same minimum detectable contrast as a thick CsI flat panel (CsI -C). However, this situation reverses for details smaller than 500µ;

References

1. A. Workman and A.R. Cowen in Phys. Med. Biol. 38 (1993) pp. 1789
2. W. Zhao et. al. , Med. Phys. 24 (12) , December 1997, pp. 1834, see fig. 10
3. W. Zhao and J. Rowlands, Med. Phys. 24(12), December 1997, pp. 1819
4. J. Siewerdsen, Ph. D. Thesis, Michigan 1998

© AIPnD , created by NDT.net

|Home|    |Top|