=Paper=
{{Paper
|id=Vol-2743/paper-1
|storemode=property
|title=Information System for Radiobiological Studies
|pdfUrl=https://ceur-ws.org/Vol-2743/1-6-paper-1.pdf
|volume=Vol-2743
|authors=Inna Kolesnikova,Andrey Nechaevskiy,Dmitry Podgainy,Alexey Stadnik,Alexej Streltsov, Oksana Streltsova
}}
==Information System for Radiobiological Studies==
https://ceur-ws.org/Vol-2743/1-6-paper-1.pdf
Proceedings of the Information System for the Tasks of Radiation Biology Workshop (ISRB2020)
Dubna, Russia, June 18, 2020
INFORMATION SYSTEM FOR RADIOBIOLOGICAL
STUDIES
I.A. Kolesnikova1,2, A.V. Nechaevskiy1, D.V. Podgainy1, A.V. Stadnik1,2,
A.I. Streltsov3, O.I. Streltsova1,2
1
Joint Institute for Nuclear Research, Dubna, Russia
2
Federal State-Funded Educational Institution of Higher Education of Moscow Region
"Dubna University", Dubna, Russia
3
SAP SE, Germany
E-mail: podgainy@jinr.ru
The article discusses the concept of building an information system (IS) for radiobiological
studies underway at the Laboratory of Radiation Biology of JINR. The information system
under development should have the following properties: to provide storage and access to
experimental data and methods of its processing, to contain a set of methods for experimental
results systematization and for the detection of hidden patterns that appear in the response of
biological systems to the effects of damaging factors, to ensure data presentation in a form
convenient for complex statistical analysis, to provide opportunities for research automation
based on machine and deep learning methods and neural network approaches, as well as a
comfortable environment for interaction and collaboration of different research groups. The
implementation of this system will significantly enhance the efficiency of radiobiological
studies.
Keywords: information systems, neuromorphology, machine learning, data analysis, neural
networks, behavioral responses, tracking.
Inna Kolesnikova, Andrey Nechaevskiy, Dmitry Podgainy,
Alexey Stadnik, Alexej Streltsov, Oksana Streltsova
Copyright © 2020 for this paper by its authors.
Use permitted under Creative Commons License Attribution 4.0 International (CC BY 4.0).
1
Proceedings of the Information System for the Tasks of Radiation Biology Workshop (ISRB2020)
Dubna, Russia, June 18, 2020
1. Introduction
Modern information technologies and nuclear medicine are essential components of
advances in medical technologies. Studies in this field are impossible without high-
performance computing complexes, adequate mathematical support and software. The
general rapid development of technologies and research in the field of the neural network
approach and deep learning leads to the emergence of new developments for the automation
of medical and biological studies. Technologies of machine learning and deep neural
networks for creating systems of automated diagnosis of diseases have the most practical
application [1-5].
At present, the use of proton beam therapy for tumor treatment is the most promising
direction in radiation oncology. To make up an optimal plan for proton therapy of small deep-
seated brain tumors, it is highly important to detect possible morphological and functional
changes in normal and tumor tissues of the central nervous system (CNS) that can be caused
by exposure to ionizing radiation. It is related both to the rather compact special arrangement
of many brain structures (particularly in the limbic system) and to the set of a number of
important functions that they possess.
This kind of the analysis of the damaging effect of ionizing radiation on tissues and
organs is also applicable to the assessment of risks of human space flights. Recent studies by
Russian and international experts indicate that radiation exposure of brain structures to heavy
charged particles (HCP) and high-energy protons of cosmic origin can result in cognitive
impairment. This, in turn, entails a partial or complete loss of the operator’s functions of
spacecraft crew members. A new strategy for planning further experimental work on
modeling the biological effects of space types of radiation and on assessing the risks of their
damaging effects in the condition of human interplanetary flights is the organization of
complex neuro-radiobiological studies on the effect of heavy charged particles on the
CNS [6].
All the risks described above make investigations aimed at studying the pathogenesis
in different body tissues after exposure to ionizing radiation extremely relevant. The
development and study of new methods of pharmacological protection against radiation
damage, the testing of radiation-resistant materials and the elaboration of prevention methods
can be the solution to the existing problems. The process of organizing this kind of studies is
complex and rather laborious, for example, it takes 7-10 days from the moment of collecting
tissues for research to receiving a report from a pathologist. At least two groups of animals,
10 rodents each, participate in experiments within scientific studies, i.e. at least 20 drugs need
to be analyzed. With such a method of data processing, the human factor plays a key role
(individual characteristics, pathomorphologist’s competence).
In the light of the above, the automation of the entire process of working with data
from the moment of carrying out an experiment to the detection and visualization of the
obtained patterns and models of the ongoing processes is of particular relevance.
2. Structure and methods for the implementation of the information system
One of the reasons for the laboriousness of automation is the complexity in the
analysis of heterogeneous experimental data, which may include: morphological data (images
of slices of different biological tissues), behavioral data (video data of experimental animals)
and others obtained by different research groups. A complete understanding of the process of
exposure and a qualitative picture of the consequences of exposure to ionizing radiation on
biosystems require the systematization and simultaneous processing of a significant amount
of data related to different aspects of the manifestation of exposure.
2
Proceedings of the Information System for the Tasks of Radiation Biology Workshop (ISRB2020)
Dubna, Russia, June 18, 2020
A schematic diagram of the radiobiological research workflow and the corresponding
data flows is shown in the figure below.
Fig. 1. Radiobiological research workflow schema
The heterogeneity of experimental data defines the creation of a developed subsystem
for acquiring, storing and systematizing experimental data, which is capable of working with
data with the same efficiency for both conducted and current experiments when studying the
effects of ionizing radiation and other factors on biological objects.
A complete picture and a model of the ongoing processes require the development of
a qualitative set of algorithms for experimental data processing based on machine and deep
learning methods for the tasks of pathomorphology and behavioral analysis in the study of the
effects of ionizing radiation and other factors.
Close interaction of different research groups and the requirement to ensure
information security when accessing data and research results entail the application of a
maximum of modern IT solutions, including web technologies, reliable modern means of
authentication and hierarchical access management, as well as components for convenient
operation and visualization of data analysis results.
The first component of the information system, ensuring work with data, should
correspond in its structure to the logic of conducted radiobiological experiments. It dictates
both the hierarchical structure of data storage and the corresponding organization of an
interface for interacting with the user of the information system. Schematically, such an
organization can be illustrated as follows:
Experiment is the major element inside of which all the contents of the experiment
will be stored. It has a standard set of attributes: name, creator’s name, creation date,
description; and a special attribute such as exposure (for example, gamma radiation).
Group is an auxiliary element related to a specific experiment, inside of which data on
“objects” of research (experimental animals) will be stored. In addition to the standard
attributes, it has special ones: organ under study, dye, microscope magnification, drug.
“Object” of research (experimental animal) is an auxiliary element related to a
specific group; metadata of the “objects” of the experiment (photos and video files) will be
stored inside.
The client-side (frontend) is the graphical interface that the user sees on the page. It is
responsible for the appearance of the components (stylization) and their position on the page
(layout). With the help of the frontend, the user can interact with the server-side (backend)
and the database using HTTP requests to the API. From the user’s side it happens unnoticed
and feels like, for example, pressing a button, entering data into a form, clicking on a page
3
Proceedings of the Information System for the Tasks of Radiation Biology Workshop (ISRB2020)
Dubna, Russia, June 18, 2020
element. The following technologies are used as means for implementing the frontend of the
service: HTML5/SCSS, JavaScript, WebPack, React.js, Redux.
The component of the information system related to the algorithmic part can be
divided into two large subgroups:
1. Subgroup of computer vision methods associated with morphological data
processing (images of slices of different biological tissues);
2. The second subgroup of methods comprises methods for analyzing video sequences
in the study of behavioral patterns of experimental animals (video data of experiments on
animal behavior).
Modern tasks of computer vision are conventionally divided into:
Classification – classification of an image by the type of object;
Semantic segmentation – identification of all pixels of objects of a certain type or
background in the image. When objects of the same class overlap, their pixels are not
separated from each other;
Object detection – detection of all objects of the specified classes and definition of the
size and position of the rectangular image area containing them;
Instance segmentation – identification of pixels belonging to each object of each class
separately.
The tasks of semantic and instance segmentation are in demand for the tasks of the
first subgroup of methods.
Image processing is based on a set of classical computer vision algorithms (image
filtering, statistical analysis methods) and techniques for using deep convolutional neural
network architectures, among which the following should be considered:
1. Mask R-CNN: convolutional neural network architecture for instance segmentation
in images. It extends the Faster R-CNN architecture by adding another branch that predicts
the position of the mask covering the found object, thus solving the task of instance
segmentation. The mask represents a simple rectangular matrix, in which ones in the cells
indicate that the corresponding pixel belongs to the object of the specified class, and zeros
respectively mask the pixels that do not belong to the object. In the architecture of this
network one conventionally distinguishes a separate convolutional neural network for
calculating image features, the so-called backbone, and a head, i.e. the union of parts
responsible for predicting the position and size of the object, classifying the object and
defining its mask.
The Mask R-CNN shows high results in instance segmentation and object
detection [7].
2. U-Net: convolutional neural network architecture designed for image segmentation,
which is originally developed for biomedical imaging. The network architecture is a sequence
of convolution + pooling layers, which first reduce the spatial resolution of the image and
then increase it, having previously combined it with image data and passed through other
convolutional layers, which turns the neural network into a kind of complex filter that carries
out instance segmentation. The U-Net convolutional network architecture performs very well
in machine learning competitions. In this case, it can be used not only for segmentation, but
also for object detection in images [8].
3. Xception: compact deep neural network. It extends the Inception architecture, the
idea of which is to eliminate the choice of the size of the convolution kernel by taking several
options and using them, while combining the results. This increases the number of operations
that need to be performed to calculate the activations of one layer, therefore, before each
convolutional block, a convolution with a 1х1 kernel is done, which reduces the dimension of
the signal fed to the input to convolutions with large kernel sizes.
The architecture has proven itself well in its versatility and performance [9].
4
Proceedings of the Information System for the Tasks of Radiation Biology Workshop (ISRB2020)
Dubna, Russia, June 18, 2020
In the second subgroup of techniques, the study of video data entails additional
methods of work using classical computer vision methods, such as the multivariate Gaussian
distribution for assessing the static background of a scene, the analysis of the optical flow to
evaluate and predict movement in frames, algorithms for interframe tracking of an
experimental animal to assess the trajectory of its movement and its significant parameters.
Some methods can be taken from the open computer vision library OpenCV [10].
The development of these algorithms is planned to be performed on the “ML/DL
Ecosystem”, which provides wide opportunities both for the development of mathematical
models and algorithms and for carrying out resource-intensive calculations and data analysis
on the basis of the HybriLIT platform [11]. The created ecosystem has two components: the
first component is designed for the development of models and algorithms based on
JupyterHub, i.e. a multi-user platform for working with Jupyter Notebook (known as IPython
with the ability to work in a web browser); the second component is intended for performing
resource-intensive, massively parallel calculations, for example, for training neural networks
using NVIDIA graphics accelerators.
The ecosystem enables the development of services based on ML\DL algorithms, the
debugging process of the corresponding software and provides visualization tools for the
results of experimental data analysis, as well as allows implementing different modern
approaches for data analysis, image and video processing. The services developed on the
basis of the ecosystem allow users to access the computing resources of the “Govorun”
supercomputer for conducting massively parallel calculations [12].
The third component of the information system is used for the tasks of implementing
the information system based on modern IT solutions, including web technologies, modern
solutions for inference and components for the visualization of data analysis.
The choice of the client-server architecture for the tasks being solved is determined by
the enumerated features of the project. Using the web service, users will be able to interact
with the database (to add data, delete or edit), to analyze images and video materials on high-
performance computing resources and obtain the results of processing in a convenient form.
To implement the web service, a technology stack based on the Node.js platform and the
React.js framework and the MySQL database management system to store data have been
chosen. Node.js is a cross-platform backend JavaScript runtime environment that executes
backend JavaScript scenarios. Node.js provides the ability to implement one’s own web
application server and write a REST API, i.e. a special interface through which data is
exchanged between the client and the server via the HTTP protocol, and interaction with the
database and the file system of the server itself is performed. This allows one to organize the
acquisition and storage of experimental data and store the entire amount of downloaded
digital data directly in the server file system. React is a JavaScript library for user interface
development. Using HTTP requests to the REST API, the client receives data from the
server, and the data is processed and used to build an interface. React enables the
implementation of a convenient web interface and, with its help, the provision of user access
to the REST API.
Conclusion
At present, the team of authors is actively developing the information system in
accordance with the methods and approaches considered in the previous sections. The reader
can find a detailed description of what has already been done in each of the indicated
directions in the articles of the present collection.
The designed and implemented IS will make it possible to perform a comprehensive
analysis of heterogeneous experimental data, including from different research groups, and to
5
Proceedings of the Information System for the Tasks of Radiation Biology Workshop (ISRB2020)
Dubna, Russia, June 18, 2020
automate most of the work within data analysis and results presentation, which, as a result,
will accelerate the acquisition of qualitatively new results.
In conclusion, it is noteworthy that the information system under development can be
used not only for medical and research purposes, but also for education. The IS will enable
further testing of specialists on histology of nervous tissue to confirm their qualification,
moreover, the IS can also be used as a training system for training new personnel in the field
of radiobiology.
References
[1] R.A. Tomakova, S.A. Philist, S.A. Gorbatenko, N.A. Shvetsova, Analysis of histological
images using morphological operators synthesized on the basis of the Fourier transform and
neural network modeling. // Automatic analysis and image recognition, no.3 (9), 2010, pp.
54-60, in Russian. https://cyberleninka.ru/article/n/analiz-gistologicheskih-izobrazheniy-
posredstvom-morfologicheskih-operatorov-sintezirovannyh-na-osnovepreobrazovaniya-furie-
i/viewer
[2] M.M. Lukashevich, V.V. Starovoitov, Technique for counting the number of cell nuclei in
medical histological images. // System analysis and applied informatics, 2, 2016, pp. 37-42,
in Russian. https://cyberleninka.ru/article/n/metodika-podscheta-chisla-yader-kletok-na-
meditsinskih-gistologicheskih-izobrazheniyah/viewer
[3] Eulenberg, P., Köhler, N., Blasi, T. et al. Reconstructing cell cycle and disease
progression using deep learning. Nat Commun 8, 463 (2017). https://doi.org/10.1038/s41467-
017-00623-3
[4] AlphaFold: Using AI for scientific discovery. https://deepmind.com/blog/alphafold/
[5] Masayasu Toratani, Masamitsu Konno et al. A Convolutional Neural Network Uses
Microscopic Images to Differentiate between Mouse and Human Cell Lines and Their
Radioresistant Clones. DOI: 10.1158/0008-5472.CAN-18-0653 Published December 2018
[6] O. Taranina, RAS Council on Space. New concept of risk. // Joint Institute for Nuclear
Research weekly, № 51-52, 25.12.2017. http://jinrmag.jinr.ru/2017/51/kr51.htm
[7] Kaiming He, Georgia Gkioxari, Piotr Dollár, Ross Girshick, Mask R-CNN.
https://arxiv.org/abs/1703.06870
[8] Olaf Ronneberger, Philipp Fischer, Thomas Brox, U-Net: Convolutional Networks for
Biomedical Image Segmentation. https://arxiv.org/abs/1505.04597
[9] François Chollet, Xception: Deep Learning with Depthwise Separable Convolutions.
https://arxiv.org/abs/1610.02357
[10] Open Source Computer Vision Library. Electronic resource: https://opencv.org/
[11] HybriLIT heterogeneous computing platform. Electronic resource:
http://hlit.jinr.ru/ecosystem-for-ml_dl_bigdataanalysis-tasks
[12] Gh. Adam et al., IT‑ecosystem of the HybriLIT heterogeneous platform for
high‑performance computing and training of IT‑specialists // CEUR Workshop Proceedings,
Selected Papers of the 8th International Conference «Distributed Computing and Grid-
technologies in Science and Education» (GRID 2018), Dubna, Russia, September 10‑14,
2018, http://ceur-ws.org/Vol-2267/638-644-paper-122.pdf
6