Quantifying the performance of a hybrid pixel detector with GaAs:Cr sensor for transmission electron microscopy

Deep Learning Approach for High-accuracy Electron Counting of Monolithic Active Pixel Sensor-type Direct Electron Detectors at Increased Electron Dose

Electron counting can be performed algorithmically for monolithic active pixel sensor direct electron detectors to eliminate readout noise and Landau noise arising from the variability in the amount of deposited energy for each electron. Errors in existing counting algorithms include mistakenly counting a multielectron strike as a single electron event, and inaccurately locating the incident position of the electron due to lateral spread of deposited energy and dark noise. Here, we report a supervised deep learning (DL) approach based on Faster region-based convolutional neural network (R-CNN) to recognize single electron events at varying electron doses and voltages. The DL approach shows high accuracy according to the near-ideal modulation transfer function (MTF) and detector quantum efficiency for sparse images. It predicts, on average, 0.47 pixel deviation from the incident positions for 200 kV electrons versus 0.59 pixel using the conventional counting method. The DL approach also shows better robustness against coincidence loss as the electron dose increases, maintaining the MTF at half Nyquist frequency above 0.83 as the electron density increases to 0.06 e-/pixel. Thus, the DL model extends the advantages of counting analysis to higher dose rates than conventional methods.

TERSE/PROLIX (TRPX) - a new algorithm for fast and lossless compression and decompression of diffraction and cryo-EM data

High-throughput data collection in crystallography poses significant challenges in handling massive amounts of data. Here, TERSE/PROLIX (or TRPX for short) is presented, a novel lossless compression algorithm specifically designed for diffraction data. The algorithm is compared with established lossless compression algorithms implemented in gzip, bzip2, CBF (crystallographic binary file), Zstandard(zstd), LZ4 and HDF5 with gzip, LZF and bitshuffle+LZ4 filters, in terms of compression efficiency and speed, using continuous-rotation electron diffraction data of an inorganic compound and raw cryo-EM data. The results show that TRPX significantly outperforms all these algorithms in terms of speed and compression rate. It was 60 times faster than bzip2 (which achieved a similar compression rate), and more than 3 times faster than LZ4, which was the runner-up in terms of speed, but had a much worse compression rate. TRPX files are byte-order independent and upon compilation the algorithm occupies very little memory. It can therefore be readily implemented in hardware. By providing a tailored solution for diffraction and raw cryo-EM data, TRPX facilitates more efficient data analysis and interpretation while mitigating storage and transmission concerns. The C++20 compression/decompression code, custom TIFF library and an ImageJ/Fiji Java plugin for reading TRPX files are open-sourced on GitHub under the permissive MIT license.

Characterizing the resolution and throughput of the Apollo direct electron detector

Advances in electron detection have been essential to the success of high-resolution cryo-EM structure determination. A new generation of direct electron detector called the Apollo, has been developed by Direct Electron. The Apollo uses a novel event-based MAPS detector custom designed for ultra-fast electron counting. We have evaluated this new camera, finding that it delivers high detective quantum efficiency (DQE) and low coincidence loss, enabling high-quality electron counting data acquisition at up to nearly 80 input electrons per pixel per second. We further characterized the performance of Apollo for single particle cryo-EM on real biological samples. Using mouse apoferritin, Apollo yielded better than 1.9 Å resolution reconstructions at all three tested dose rates from a half-day data collection session each. With longer collection time and improved specimen preparation, mouse apoferritin was reconstructed to 1.66 Å resolution. Applied to a more challenging small protein aldolase, we obtained a 2.24 Å resolution reconstruction. The high quality of the map indicates that the Apollo has sufficiently high DQE to reconstruct smaller proteins and complexes with high-fidelity. Our results demonstrate that the Apollo camera performs well across a broad range of dose rates and is capable of capturing high quality data that produce high-resolution reconstructions for large and small single particle samples.

Event driven 4D STEM acquisition with a Timepix3 detector: Microsecond dwell time and faster scans for high precision and low dose applications

Four dimensional scanning transmission electron microscopy (4D STEM) records the scattering of electrons in a material in great detail. The benefits offered by 4D STEM are substantial, with the wealth of data it provides facilitating for instance high precision, high electron dose efficiency phase imaging via centre of mass or ptychography based analysis. However the requirement for a 2D image of the scattering to be recorded at each probe position has long placed a severe bottleneck on the speed at which 4D STEM can be performed. Recent advances in camera technology have greatly reduced this bottleneck, with the detection efficiency of direct electron detectors being especially well suited to the technique. However even the fastest frame driven pixelated detectors still significantly limit the scan speed which can be used in 4D STEM, making the resulting data susceptible to drift and hampering its use for low dose beam sensitive applications. Here we report the development of the use of an event driven Timepix3 direct electron camera that allows us to overcome this bottleneck and achieve 4D STEM dwell times down to 100 ns; orders of magnitude faster than what has been possible with frame based readout. We characterize the detector for different acceleration voltages and show that the method is especially well suited for low dose imaging and promises rich datasets without compromising dwell time when compared to conventional STEM imaging.