|
Application Report Different Electron Detector System for TEM |
P.Bele, Technische Universität München, Department of Physics E19, James-Franck-Straße 1, 85748 Garching, Germany |
(Dieser Fachartikel wurde uns freundlicherweise von der DITABIS Digital Biomedical Imaging Systems AG zur Verfügung gestellt)
|
|
Der State of the Art für die Bestimmung von z.B. biologischen Strukturen, die Messung von elektronischen Beugungsmustern oder elektronischen Energieausgleichsspektrum mit der Transmissions-Elektronen-Mikroskopie benötigt eine exakte Erfassung des Elektronenflusses. Dieser Report handelt von den Eigenschaften verschiedener Elektronendetektor-Systemen bzgl. Ihrer charakteristischen Parameter und Evaluierung. Im Bereich der Transmissions-Elektronen-Mikroskopie (TEM) werden die folgenden Detektionssysteme bei Standard Set-Ups genutzt: |
- Phosphor Screen
- Negativfilm
- Slow Scan Charge Coupled Device (ss-CCD- Kamera)
- Imaging Plate
|
Alle genannten Detektionssysteme sind 2-dimensionale rauschende Detektorsysteme. Was wird benötigt um einen idealen Elektronendetektor zu erhalten? Die Eigenschaften eines idealen Elektronendetektors sollten sein: |
- On-line Fähigkeit
- Eine große Detektionsfläche
- Eine hohe räumliche Auflösung mit einer Pixelgröße von 10µm
- Ein hoher Umrechnungsfaktor (Zählung pro Elektron)
- Eine perfekte Modulation Transfer Function (MTF=1)
- Ein hoher dynamischer Bereich größer als 5 Größenordnungen
- Eine hohe und konstante Detection Quantum Efficiency mit dem Wert ≈ 0.9 über einen weiten Bereich der elektronischen Dosis
|
Diesen Artikel als pdf-Dokument zum Download |
|
|
1. Introduction |
The state-of the-art for the determination of i.e. biological structures, or the measurement of electron diffraction patterns or electron energy loss spectra using Transmission Electron Microscopy necessitate an accurate registration of the electron flux. This report deals with the features of different electron detector systems in terms of their characteristic parameters and evaluations. In the field of Transmission Electron Microscopy (TEM) the following detector systems are used in standard set-ups: |
- Phosphor screen;
- negative film material;
- slow scan charge coupled device (ss-CCD camera);
- Image Plate (IP).
|
All mentioned detector systems are 2-dimensional noisy detectors systems [1]. What is needed in order to have an ideal electron detector? The properties of an ideal detector should be |
- on-line capability;
- a large detection area;
- a high spatial resolution with a pixel size in the order of 10 μm;
- high conversion factor (counts per electrons);
- a perfect modulation transfer function (MTF = 1);
- a high dynamic range, lager than 5 orders of magnitude and
- a high and constant detection quantum efficiency, with a value ≈ 0.9 over a wide range of electron doses.
|
|
2. Characteristic Detector Parameters |
2.1 Dynamic Range (DR) |
The dynamic range quantifies the largest signal that can be recorded before saturation of the detector relative to the smallest signal that can be distinguished from the noise. The dynamic range is defined by the following equation. |
DR = Imax / Imin |
For example, to record the entire Bragg diffraction pattern it is necessary to have a DR which is at least equal to the range of reflection intensities of the sample of interest. 4 orders of magnitude are sufficient for most macromolecules [2]. |
|
2.2 Signal-to-Noise ratio (SNR) |
This parameter is a measure of the sensitivity regarding the detector system signal output. It is equal to the ratio of the sensitivity of the detector to an ideal photon-counting detector and it is defined as the ratio between a signal and the background noise, |
SNR = Psignal / Pnoise |
or as an alternative definition as the ratio of the mean signal and the standard deviation of the noise [2]. |
SNR = Asignal / σ |
2.3 Point Spread Function (PSF) The PSF describes the response of an imaging system to a point source or point object; i.e. for the IP scanner system it is the laser source. One can say, in functional terms it is the spatial domain version of the Modulation Transfer Function (MTF). A more general term for the PSF is a system‘s impulse response. The Figure below shows a closer look to the PSF. |
|
The image of a complex object is always described as the convolution of the true object with the PSF. The degree of the spreading (blurring) of the point object is a measure for the quality of the imaging system [2]. |
|
The image of a complex object is always described as the convolution of the true object with the PSF. The degree of the spreading (blurring) of the point object is a measure for the quality of the imaging system [2]. |
|
2.4 Modulation Transfer Function (MTF) The MTF characterizes the resolution and the performance of a detector system in one parameter. It is a measure for the ability of a system to transfer contrast from the subject to the image in Fourier space. Or in other words represents the Bode plot of an imaging system (microscope or the human eye), and thus depicts the filtering characteristic of the imaging system. The MTF is defined as the normalized magnitude of the FFT of the imaging system’s point spread function (see schemes below for details). |
|
The Nyquist frequency is defined as the highest frequency that can be coded at a given sampling rate in order to be able to fully reconstruct the signal. It can be calculated for the detector system as 1 over 2 pixelnominal [2]. |
|
2.5 The Detection Quantum Efficiency (DQE) The DQE is the last and on of the most important parameter for a detector system. It is a parameter introduced to assess the varying levels of performance of imaging detectors, in order to compare their imaging capabilities by an unified approach [3, 4]. One can say the DQE quantifies the sensitivity of the detector system and is equal to the ratio of the sensitivity of the detector to an ideal photon-counting detector. At present, the DQE definition was generalized for all types of detecting devices. This includes also those using analog signals at certain stages of signal conversion. The following definition of DQE as a measure of information conversion quality is used [3, 5, 6, 7]: |
DQE = (SNRout)2/(SNRin)2 |
It is a measure of how the available signal-to-noise ratio is degraded by the imaging system and is calculated by this equation above. One can also be defined the DQE as the ratio of the (power) SNR at the detector output to the maximum possible SNR. The DQE will always in the range 0 to 1. |
|
3. Different Detection Media 3.1 Negative Film This medium can actually be seen as a photon counter, where each absorbed electron renders one grain of film developable. The practical difficulty of using film as a photon counter is in counting the grains. For this reason, it is always used as an integrator, where the determination of cumulative darkened film grains is made by measuring the optical density of the developed film, for a review see i.e. [8]. The negative film material has the following advantages:
|
- Film has excellent spatial resolution (with a typical high-end scanner device 7μm nominal pixel size);
- Large detection area;
- Inexpensive (only the film material);
- Very long exposures are possible with negligible ‘dark current’ accumulation;
- High spatial resolution.
|
Besides that there are also disadvantages, which have to be taken into account: |
- Has to be seen as an analog detection medium;
- Very low dynamic range in the order of 103 and only a linearity of 102 of the detection range between from 0.01 to 0.2 e-/μm2;
Basically, negative film is non-linear when the density of developed grains is high enough that they start shadowing one another. - Low DQE with a max. value 0.2 (depending on the used film material) [9];
- In general, the film emulsion is best for electrons in the acceleration voltage range from 40 up to 100 kV;
- Film has a high noise of about 107 ‘fog’ grains per cm2. These are nearly Poisson distributed, so the noise is the square root of 107 ≈3000. Therefore, film material is an insensitive detector and its sensitivity is 10 times smaller comparable to IP or ss-CCD cameras.
- Film material ages and the final signal detected correlates directly to the storage and the exact development procedure.
- Negative Film requires a slow, messy wet ‘off-line’ development process and a complex optical density scanner (high costs). This is in contrary to the initial low cost of the film material itself.
|
|
3.3.1 Detection Quantum Efficiency DQE (Kodak SO 163 Negative film + Zeiss SCAI negative scanner; 6.8 μm nominal pixel size) |
|
The graph above shows a distinct example for the electron dose dependency of the DQE for the combination of Kodak SO 163 negative films and Zeiss SCAI negative scanner with a nominal pixel size of 6.8 μm [10]. Here only the electron dose regime was measured, where a linear SNRout is guaranteed. In general, these properties of the negative film material allow normal imaging applications. The limit of the use of this detection medium is: |
- the low electron dose imaging;
- for electron diffraction recording;
- or the detection of parallel EELS.
|
For example, for diffraction experiments strong reflexes, or reflexes on a high background noise, one will receive a systematic change of the peak intensities. This is due to the very limited linear dynamic range and will always leads to an error in the reflex intensity. |
|
3.2 CCD Camera The development of charge coupled device (CCD) cameras for the consumer market has been dramatic over the last several years. While there have been tremendous advances in the technology for producing many types of CCD chips, the requirements for effective performance in microscopy have so far made it impractical to take advantage of the mass produced chips. Still, ss-CCDs (slow scan charge coupled device) are now in routine use for a number of TEM applications, particularly where large series of images are needed such as in tomography and wavefront reconstruction from de-focus series, and where the benefits of rapid, digital data recording outweigh any of the disadvantages in CCD performance compared to other data recording media.
All CCD cameras used with electron microscopes are built-up on the so-called quantum optical chain (QOC). The detector is made up of electro-optical elements, which typically consists of a chain of elements which convert the electrons to more readily manipulated quanta, and a possible gain element to increase the number of quanta, in order to better compete with noise in the final image sensing element. For example, the energy converter is generally a thin phosphorous screen. Its luminescence signal is then amplified by an image amplifier and viewed by a CCD image sensor. For a review see for example Gruner et al. [11]. The Advantages of this detector are: |
- Direct on-line capability (direct access to the digitized image);
- high sensitivity;
- low readout noise (only if ss-CCD camera);
- moderate to high conversion factor between 4 to 9 counts per electron (depending on the scintillator material used);
- for the example of the GATAN ss-CCD camera the conversion factor is 7.2 counts per electrons [10];
- moderate to high linear dynamic range of 103 to 105 orders of magnitude with up to max. 14 bit; (depending on the scintillator material used);
- DQE in the range of about 0.4 – 0.8 (depending on the scintillator material) [12, 13].
|
But there are also some disadvantages:
|
- Limited detection area (1kx1k or max. 2kx2k standard, 4kx4k cameras available, but very expensive and not standard yet);
- Nominal pixel size of 14 – 24 μm for commercial CCD cameras;
- The electron-to-photon conversion generally utilizes only a fraction of the electron energy. Some fraction of the resultant photons is lost before reaching the CCD;
- Broader point spread function (PSF), which is dependent on the thickness of the phosphorous layer and the coupling of the scintillator by fiber optics to the CCD chip, due to possible cross-talk between neighbouring pixel;
- Several inherent noise components:
- High dark current noise especially in the low electron dose regime;
- Always some readout noise introduced.
|
|
3.2.1 Point Spread Function PSF (1kx1k GATAN ss-CCD; 24 μm nominal pixel size) In the picture below the calculated PSF for a standard GATAN ss-CCD camera is shown.
|
|
This graph refers to old laboratory results [10] and here the evaluation was made by using the tilted-slit method, developed by Fujita et al. [14] instead of the now common edge method developed by Weikenmeier et al. [15]. Also the PSF shows a circular symmetry. |
|
|
3.2.2 Modulation Transfer Function MTF (1kx1k GATAN ss-CCD; 24 μm nominal pixel size) Also the MTF of the standard ss-CCD camera was evaluated using the above calculated PSF [10]. |
|
As one can clearly see, the MTF drops down drastically and the value at ½ Nyquist frequency is only 0.3. This rapid decrease of the MTF is due to the photon scattering and a long-range blurring caused by the re-entering of electrons into the scintillator. Also, one major point to note, is that due to aliasing phenomena the MTF for ss-CCD cameras are often been over-estimated. This could be a reason for the discrepancy or mismatch between experimental results and simulations for high resolution TEM [16]. But one has to keep in mind, the PSF and the MTF varies for the different types of ss-CCD cameras, depending on the scintillator material and thickness of the scintillator layer. |
|
3.2.3. Detection Quantum Efficiency DQE (1kx1k GATAN ss-CCD; 24 μm nominal pixel size) Here the electron dependent DQE for the 1kx1k GATAN ss-CCD camera is presented. The overall performance of the ss-CCD camera hampers from the decay of the DQE value in the extreme low to medium electron dose regime. |
|
The maximal DQE value for any large area ss-CCD camera can be calculated to a value of ≤ 0.8, which is limited by absorption, from the scintillator material noise factor and the detector gain. In the lower electron dose regime the DQE is massively limited by the electronic read out noise (background level or dark current and its noise) of the detector system [10]. |
|
3.3 Imaging Plates (IP) 3.3.1 Introduction The Imaging Plate is a flexible electron detector, where an active layer of very small photostimulable crystals locally store high energetic radiation. |
|
The IP consists of a layered structure, starting with a metal support layer. After a flexible polyester base the main layer is applied. The highly dispersed storage crystals are embedded in a blue coloured resin. The last layer consists of a protective polymer layer (see scheme above).
The electron irradiation excites the crystals in their luminescence centers to a semi-stable state, in form of trapped F-centers. The image information, formed by this excitation is stable for many hours and decays within days (see scheme below). |
|
By an illumination with red laser light, the crystals are excited again and stimulated to release the stored information as a blue luminescence signal. As this is a physical protective polymer flexible polyester metal support layer Reflected laser light Luminescence process it is fully reversible without degradation, so the Imaging Plate can be reused many times. In the following some information about the unique scanner design of the DITABIS Micron is mentioned. The system is a drum scanner and was tested with a nominal pixel size of 25 μm. The mirror collecting system was modified, using a mirror with a small numerical aperture. This helps to reduce scattering of the luminescence signal in the collector system. Due to the 2 channel detection system it is easy to detect the luminescence and the reflected laser light simultaneously. Also a special laser diode was built in to reduce the laser spot to 5 μm. This avoids the cross-talk of the signal of 2 adjacent pixels, due to the fact that the ration of nominal pixel size to laser spot is large enough. Also the mechanical stability was improved [17, 18]. By now the nominal pixel size of a commercial available scanner is down to 15 μm, but due to the available imaging plates and their related grain sizes its minimal nominal pixel size cannot not be smaller than ~12 μm, even if the technical possibility is possible. Also this detector has its advantages, which are: |
- Imaging Plates have a detection area comparable to negative film material;
- Low background noise (factor 10 smaller than ss-CCD cameras);
- They show a high spatial frequency response, determined by the modulation transfer function (MTF);
- Very high recordable dynamic range up to 6 orders of magnitude
- 6 orders of magnitude depending on the electron dose for the 16-bit output signal, if combining the high and low signal channels;
- IPs are suitable for various kinds of applications ranging from cryo-TEM low-dose on biological samples (weak phase objects) to diffraction with high electron doses;
- High conversion factor of 8.80 counts per electron.
|
Nevertheless, also disadvantages are to be mentioned: |
- Off-line detector system;
- For high electron doses (>0.5 e-/μm2) the SNRout is dominated by a linear noise term
- laser power fluctuations
- fluctuations of the scanner velocity;
- IP inhomogeneities
- IP topology,
- internal grain structure of the active layer.
- The signal-to-noise ratio follows the theoretical Poisson noise distribution only in the low dose regime;
- In the higher electron dose regime there is still the problem with the low detection quantum efficiency. The electron detection is hampered by a low DQE for an electron dose higher than about 1e-/μm2,
- originated from surface and active layer inhomogeneities of the IP;
- Characteristic noise of the IPs dominates the detection signal for high electron doses
- resulting in a rapid decrease of the detection quantum efficiency (DQE);
- the DQE has the desired high and constant value only in the low electron dose region (< 0.5 e-/μm2).
|
|
3.3.2 Simultaneously Detected Luminescence and Reflected Laser Light Signal Here the unique scanner feature of the simultaneously detection of the luminescence and the reflected laser light signal comes into play (DITABIS Micron). Now it is possible to detect not only the luminescence signal but also a reflected light signal of the original laser light at the same time. An important feature, which has to be mentioned, is that these 2 signals are pixel aligned. One can think about the information this additional signal carries. Of course the reflected laser light signal encodes all the laser fluctuations, but there is more to consider, if one take a look to the figure below. |
|
First let assume the laser light is reflected off of the IP surface. One will obtain a luminescence signal only from the grains of the active layer. Since the two signals originate in different areas one would not expect a positive correlation between the 2 signals. On the other hand, it is possible to get a laser light reflection off of the grain of the IP's active layer. In this case a correlation between the 2 signals is expected.
If the same areas of the reflected laser light and the luminescence signal are cross-correlated a positive correlation peak can be observed. This is depicted in the figure above on the right side [19]. Therefore, on can assume that laser light reflection on the surface of the IP takes place, but the positive correlation peak clearly indicates that some reflection also occurs on the grain of the active layer. This means that the additional signal carries information about the IP surface and the internal grain structure. It is important to note, that this information is specific for each individual IP. One can say it is the IP’s fingerprint.
This additional information in the reflected laser light signals enables a possibility to eliminate the IP characteristic noise, laser noise and other fluctuations, and in order to obtain an increased DQE.
This is made possible by a new data correction to eliminate the noise terms and is described in more detail in [17, 18]. The high spatial frequency or pixel-to-pixel noise is corrected by the so-called ‘reflected light correction’ eliminating the laser fluctuation and the IP characteristic noise.
In addition, the local misalignment, due to local geometric distortions, can also be corrected. Such distortions between gain reference and the actual data image occur for IPs as a result of thermal expansions and mechanical instabilities of the scanner. This is well known for other high-end scanner systems such as high resolution negative scanners. An average geometric distortion is found on the order of 0.2-0.3% [18]. In comparison, the corresponding value for negative film scanners is on the order of 0.5-0.6% and 0% for the ss-CCD camera due to their fixed installation [20]. After the correction of the local misalignment every single pixel of the luminescence signal is ‘re-distorted’ and results in the required new local alignment and to the extremely good DQE. |
|
3.3.3 Dynamic range (DITABIS Micron 25μm nominal pixel size) The IP has a large detection area and a high dynamic range of about 6 orders of magnitude for the 16bit signal [17] which can be reached by combining the high and low signal channel. |
|
The conversion factor is 8.80 counts per electron, which is comparable to good standard ss-CCD cameras. The high dynamic range allows a simultaneous quantitative recording of weak and strong signals in one single exposure [17, 18]. |
|
3.3.4 Spatial Resolution PSF (DITABIS Micron; 25μm nominal pixel size) In the following the spatial resolution and MTF of the DITABIS Micron scanner is described in more detail. For the evaluation the edge profile method, developed by Weikenmeier et al. [15], is used. The 4 experimental curves of the edge profiles in the following graph clearly show that there is no dependence on the applied electron dose or the laser power. Previous measurements, concerning the orientation of the original edge, showed that the scanner resolution is isotropic. |
|
This leads to the Point Spread Function of the laser signal, which was fitted with a modified 21 parameter function for a combination of 10 Gaussians [17]. In the work of Weikenmeier et al. [15], where the standard procedure is described, only 4 Gaussians are used which is still the common procedure. |
|
The new fitting function enables a more precise shape determination of the laser beam [17]. |
|
3.3.5 Modulation Transfer Form MTF (DITABIS Micron; 25μm nominal pixel size) The following graph the corresponding MTF is depicted. This graph shows by how much the contrast is reduced for different spatial frequencies. As mentioned before, an ideal detector system would show a constant MTF of 1. |
|
The resulting MTF of the IP scanner system also drops down at Nyquist frequency to a value of 0.28, but this is still a good value if compared to ss-CCD camera. In order to show a direct comparison, the MTF of a standard GATAN ss-CCD camera (1kx1k) is also depicted in this graph [2]. For all spatial frequencies, even down to the Nyquist frequency the IPs show a superior MTF. |
|
3.3.6 Luminescence Signal-to-Noise Ratio SNRout (DITABIS Micron; 25μm nominal pixel size) Here the SNRout of the scanner output signal of homogeneously exposed IPs is depicted [18]. The different curves show the dramatic enhancement, when using the different now possible data correction steps. As mentioned before, in the low dose region the SNRout follows the theoretical Poisson distribution. For higher electron doses the linear noise term becomes the main distribution. This means that the square root dependency is no longer valid. |
|
After applying the different data correction steps a dramatic increase of the SNRout is obtained. We see that the interval which can be described as an ideally Poisson-distributed signal is extended to higher electron doses. |
|
3.3.7 Detection Quantum Efficiency DQE (DITABIS Micron; 25μm nominal pixel size) This graph shows the corresponding results of the DQE. The reflected light correction increases the DQE by about 35%. The conventional gain normalization improves the DQE but the effect is much smaller than expected. |
|
But after the geometric distortion corrected gain normalization one will found the desired increase of the DQE [18]. That the DQE still drops a little bit at higher electron doses could be related to the gain reference image, which was taken at an electron dose of 50 e-/μm2. Maybe this electron dose was still not high enough to provide the perfect gain reference image. |
|
4. Conclusion As a final conclusion of this report a comparison of the DQEs for different electron detector systems (ss-CCD camera and 2 different IP scanner systems) is presented [17]. The data for the negative film is neglected here, due to its small electron dose regime (here see again 3.1). As mentioned before, for an ideal electron detector we would expect a constant value of 1. One can clear that the ss-CCD camera would be the best choice only in the high electron dose regime after conventional gain normalization. In the region lower than 0.5 e-/μm2 the DQE is low as a result of the detector gain which is limited by the background level or dark current and its noise. |
|
In the case of the IP scanner system the background noise is about 10 times smaller compared to the ss-CCD camera. The comparison of our uncorrected data with the ones for the FUJI IP scanner system also illustrates the necessity to apply all the previously discussed data corrections to obtain a good overall performance. Now, with the advanced data processing the tested IP readout device shows the best performance of all systems. |
|
5. Literature [1] G. McMullan, S. Chen, R. Henderson, and A.R. Faruqi, Ultramicroscopy, 109, 9, 1126, 2009. [2] P. Bele, extracted from Teaching Seminar University of Yamanashi, 2010. [3] A. Rose, J. Soc. Motion. Pict. Eng., 47, 273, 1946. [4] R.C. Jones, Adv. in Electronics and Electron Phys. 11, 87, 1959. [5] M. M. Butslov, B. M. Stepanov, and S. D. Fanchenko, Optoelectronic Intensifiers and Their Use in Scientific Research [in Russian], Moscow, 1978. [6] I. A. Cunningham, In: The Expanding Role of Medical Physics in Diagnostic Imaging, Madison, Wisconsin, USA, pp. 231, 1997. [7] H. Gfirtner, Quality Assurance and Patient Radiation Protection in Diagnostic Radiology, Berlin, 1996. [8] E. Zeitler, Ultramicroscopy, 46,405, 1992. [9] C. Burmester, Diploma Thesis, Ruprecht-Karls Universität, Heidelberg, 1992. [10] P. Bele, private lab communication; Data presented at the - 12th Euopean Congress on Electron Microscopy (EUREM12), Brno, Czech Republic, 2000 (oral contribution). [11] S. M. Gruner , M. W. Tate, E.F. Eikenberry, Rev. Sci. Instr., 73, 2815, 2002. [12] O.L. Krivanek, Proc. 10th Eur. Congr. on Electron Microscopy, Granada, Spain, Vol. 1, 83, 1992. [13] S. Kujawa and D. Krahl, Ultramicroscopy, 46, 395, 1992. [14] H. Fujita, D. Tsai, T. Itoh, K. Doi, J. Morishita, K. Ueda, A. Ohtzuka, IEEE Trans. Med. Imag. II, 1, 34, 1992. [15] A. L. Weickenmeier, W. Nichter, J. Mayer, Optik, 99, 147 , 1995. [16] C.B. Boothroyd, J. Micros., 190, 99, 1998. [17] P. Bele, R. Ochs, I. Angert, R.R. Schroder, Microscopy Research&Technique. 49 (3), 281, 2000. [18] P. Bele, Microscopy and Analysis, 23, 5, 5, 2009. [19] P. Bele, World Journal of Engineering, accepted for publication, 2011. [20] J. Hesse, H. Hebert, P.J.B. Koeck, Microscopy Research and Technique, 49, 3, 292, 2000. |
|
|
Quelle: DITABIS Digital Biomedical Imaging Systems AG (04/2011) |
|
|
|
|
|
|
|
|
|
|
|
|
|
|