Smart Engines and RAS: A software package for reconstruction of x-ray tomographic pictures

Main article: Computer tomography

2020: The announcement of a software package for reconstruction of x-ray tomographic pictures

On October 8, 2020 it became known that the collaboration the Russian scientists from developer company of systems recognitions Smart Engines and institutes Russian Academy of Sciences develops a domestic software package for reconstruction x-ray of tomographic pictures in real time. Approbation is carried out on civil processors of production of the Russian company MCST, including domestic processor Elbrus"-8CB". The received results show that processors of domestic development have sufficient performance for the solution of tomographic tasks in real time. Creation of completely domestic hardware and software system is intended to solve a problem of substitution of foreign products when developing Russian industrial and medical tomographs.

Rice: example of a sinogramma of a thorax (at the left); result of KT — the section of the 3D image (on the right).

As it was explained, computer tomographic roentgenography – for October, 2020 widely used method of noninvasive diagnostics based on image generation of a form and inner pattern of an object using x-ray emission. Objects of live and inanimate nature, for medical diagnostics or carrying out scientific research of materials are exposed to tomographic researches. At x-ray KT a final set of images forms from a set of the roentgenograms removed under different corners. After collecting of a full range of projections, their processing – reconstruction which purpose – recovery of an inner morphological pattern of an object begins.

In medical researches of a gentra (the mobile device containing the system of detectors and x-ray radiators) rotates around the motionless patient. Spatial resolution in such tomographs reaches 0.2-0.5 mm. Results of KT remain in the DICOM format – the medical industry standard developed for creation, storage, transfer of digital medical images and related documents of the inspected patient. For scientific research of in vitro in vitro, other experimental scheme is applied – the source and the detector are not mobile, and a set of roentgenograms receive, rotating a sample.

The Russian developers of research company Smart Engines, in collaboration with scientists from institutes of the Russian Academy of Sciences (IPPI RAS and FITs IU RAS) and engineers of the producer of Elbrus processors of JSC MCST tested the domestic software intended for reconstruction of the x-ray images received during the tomographic researches.

The algorithm of fast hafovsky return projection (HFBP) which is steadily overtaking on performance – direct Fourier reconstruction algorithm is the cornerstone of development (Direct Fourier Reconstruction – DFR). Distinctive feature of the provided software – support of the Russian operating system Elbrus along with extended by MS Windows, macOS and also the different Linux distribution kits, and an opportunity for work on processor architecture Elbrus, x86, x86_64.

Being completely domestic development, as a part of a hardware and software system on the Elbrus platform the Smart Tomo Engine technology is suitable for medical and industrial scanners of all generations, for nano-tomographs of submicronic permission and also for the synchrotron centers.

Demonstration reconstruction was carried out on the synthetic data imitating pictures of dental tomographs and the real images received when shooting a may-bug on the FNITs KF RAS microtomograph. Measurements of speed of accomplishment of the implemented reconstruction algorithms at 5 workstations are carried out: Elbrus-401, Elbrus-804, Elbrus-801CB (development and production of JSC MCST), AMD Ryzen 7 2700 and AMD Ryzen Threadripper 3970X.

Table. Measurements of operating time of the program at reconstruction of a may-bug, sec.

The received results showed that processors of domestic development have sufficient performance for the solution of tomographic tasks in real time and capable to compete with processors of foreign production. the 4-processor Elbrus-804 server spent for reconstruction of 511 layers of synthetic these 19 seconds, i.e. each layer was recovered in 0.037 seconds, and layer-by-layer frequency was 26.8 layers per second (26.8 ips).

Elbrus-4C is the processor in large quantities supplied to the market. Has 4 cores with a frequency of 750 … 800 MGTS, 3 channels of memory DDR3-1600. Elbrus - 8C - 8 of cores, frequency 1.2 … 1.3 GHz, 4 channels of memory DDR3-1600, and at the same time – in each core are 1.5 times more than actuators (ALU) for calculations from a floating comma. Elbrus-8CB is further updating: 8 cores with a frequency of 1.5 GHz, memory of DDR4-2400 and are even twice more than ALU. Elbrus-8CB works with not aligned data better, and in it a lot of other small software updatings to comparison about Elbrus-8C. Konstantin Trushkin, the deputy CEO for marketing of JSC MCST told

Gentri of the 16-slice cardiological tomograph in a second makes slightly less than two revolutions, registering about 30 sinogramm. We recover 26.8 layers per second, i.e. we carry out reconstruction almost in real time. Thus, reconstruction using the Russian platform meets requirements for speed imposed in cardiology where the key fiducial parameter is the period of reduction of heart which averages one second. Reconstruction in real time is required also for scanning protocol implementation — controlled reconstruction. Use of this protocol reduces beam loading because collecting of roentgenograms is interrupted at once as soon as their set is sufficient for recovery. Vladimir Arlazarov, PhD in Technological Sciences, the CEO of Smart Engines told