27 June 2021 to 1 July 2021
Europe/Brussels timezone

Poisson noise analysis of the Pixirad-2/PIXIE-III photon-counting detector

Not scheduled
Gather.town (Online)



Poster presentation + pitch Sensor Materials, Device Processing & Technologies Poster session 2


Mr Carlos Navarrete-León (Department of Medical Physics and Biomedical Engineering. University College London)


X-ray photon-counting detectors (XPCD) have gained substantial interest for biomedical and materials science applications due to their low-electronic noise, high-detection efficiency and energy-discrimination capabilities, compared to charge-integrating devices. The Pixirad-2/PIXIE-III is a hybrid XPCD with a 650 µm CdTe Schottky type diode with electron collection at pixel, bonded to the PIXIE-III CMOS ASIC. The readout system is organized in two blocks of a 512 x 402 matrix of 62 µm pitch square pixels, giving a global active area of 63.6 x 25 mm2. It implements two independent discriminators with a programmable threshold per pixel, allowing the acquisition of two spectral images in a single exposure [1].
Due to their capability to set a threshold above the electronics noise of each channel, XPCD are often approximated to an ideal detector in which each pixel’s output signal is only limited by the Poisson noise inherent to the quantized nature of light [2]. However, because of the charge sharing effect characteristic of small pixel sizes and thick sensor materials, XPCD data is often spatially and energetically correlated [3]. The Poisson nature of noise in XPCD has been previously studied with the Pixirad [4] and Medipix [5] technologies. This has been done by comparing theoretical and experimental standard deviation as a function of mean counts for different energies [4] or assessing the ratio between observed and expected standard deviation for different sensor materials [5].
In this work, the noise in the Pixirad-2/PIXIE-III detector is studied with a χ2 hypothesis test for the Poisson distribution. The analysis is done locally, initially on a pixel-by-pixel basis with 512x1s frames, then on a 20x20 pixels sliding window in a single frame after pixel-wise equalisation. The effect of low-energy threshold, ROI window size and number of frames in the flat field image is also considered. The experiments were conducted with a W-anode microfocus Hamamatsu source at 40 kVp and 250 µA. The detector was cooled to −20°C and the CdTe crystal was polarised with a 400 V voltage.
Figure 1 shows results of the pixel-wise hypothesis test with significance level of α=0.05. It was observed that nearly 80% of pixels follow Poisson distribution. Non-Poisson noise was found at the detector edges, the gap between the two tiles and a few cold spots in the Flat Field image. These agree with previous studies that showed higher sensitivity in sensor edges due to manufacturing imperfections, and lower sensitivity in the dark regions due to charge trapping [5]. In addition, almost an entire column of pixels on the left tile showed non-Poisson behaviour, which is not observed in the flat field images, and could be related to defects in the readout system, adding electronic noise.
The effect of flat field correction was also studied as shown in Figure 2. It was found that increasing the number of frames in the Flat Field image improves the Poisson behaviour (from 36% to 82% of the studied windows when 4 and 32 frames are used, respectively). However, when more than 64 frames are used in the Flat Field, the correction becomes less effective, especially in the gap between tiles. This is likely due to sensor instabilities over time and/or polarisation caused by the Schottky type diode. Ongoing work is focused on establishing reliable rules of thumb for optimising the system in long acquisitions. Optimal exposure and bias cycle trade-offs are being identified based on the required statistics, the intensity of the radiation reaching the detector and the required exposure time.
The methodology of the hypothesis test, along with the effect of low-energy threshold and sliding window size will be presented with further detail. This work in progress is aiming to better understand noise in the Pixirad-2/PIXIE-III for more accurate X-ray system modelling and optimization, and more robust image post-processing.

Primary author

Mr Carlos Navarrete-León (Department of Medical Physics and Biomedical Engineering. University College London)


Dr Charlotte K. Hagen (Department of Medical Physics and Biomedical Engineering. University College London) Dr Peter R. T. Munro (Department of Medical Physics and Biomedical Engineering. University College London) Prof. Alessandro Olivo (Department of Medical Physics and Biomedical Engineering. University College London) Dr Marco Endrizzi (Department of Medical Physics and Biomedical Engineering. University College London)

Presentation materials

There are no materials yet.