Western University Western University
Scholarship@Western Scholarship@Western
Biochemistry Publications Biochemistry Department
10-6-2020
Estimating partial body ionizing radiation exposure by automated Estimating partial body ionizing radiation exposure by automated
cytogenetic biodosimetry cytogenetic biodosimetry
Ben Shirley
Cytognomix
, bshirley@cytognomix.com
Peter Rogan
The University of Western Ontario
, progan@uwo.ca
Follow this and additional works at: https://ir.lib.uwo.ca/biochempub
Part of the Analytical, Diagnostic and Therapeutic Techniques and Equipment Commons,
Biochemistry Commons, Bioinformatics Commons, and the Genetics Commons
Citation of this paper: Citation of this paper:
Ben C. Shirley, Joan H. M. Knoll, Jayne Moquet, Elizabeth Ainsbury, Ngoc-Duy Pham, Farrah Norton, Ruth
C. Wilkins & Peter K. Rogan (2020) Estimating partial-body ionizing radiation exposure by automated
cytogenetic biodosimetry, International Journal of Radiation Biology, 96:11, 1492-1503, DOI: 10.1080/
09553002.2020.1820611
Full Terms & Conditions of access and use can be found at
https://www.tandfonline.com/action/journalInformation?journalCode=irab20
International Journal of Radiation Biology
ISSN: (Print) (Online) Journal homepage: https://www.tandfonline.com/loi/irab20
Estimating partial-body ionizing radiation
exposure by automated cytogenetic biodosimetry
Ben C. Shirley , Joan H. M. Knoll , Jayne Moquet , Elizabeth Ainsbury , Ngoc-
Duy Pham , Farrah Norton , Ruth C. Wilkins & Peter K. Rogan
To cite this article: Ben C. Shirley , Joan H. M. Knoll , Jayne Moquet , Elizabeth Ainsbury ,
Ngoc-Duy Pham , Farrah Norton , Ruth C. Wilkins & Peter K. Rogan (2020) Estimating partial-
body ionizing radiation exposure by automated cytogenetic biodosimetry, International Journal of
Radiation Biology, 96:11, 1492-1503, DOI: 10.1080/09553002.2020.1820611
To link to this article: https://doi.org/10.1080/09553002.2020.1820611
View supplementary material
Published online: 06 Oct 2020.
Submit your article to this journal
Article views: 86
View related articles
View Crossmark data
ORIGINAL ARTICLE
Estimating partial-body ionizing radiation exposure by automated cytogenetic
biodosimetry
Ben C. Shirley
a
, Joan H. M. Knoll
a,b
, Jayne Moquet
c
, Elizabeth Ainsbury
c
, Ngoc-Duy Pham
d
, Farrah Norton
e
,
Ruth C. Wilkins
f
, and Peter K. Rogan
a,g
a
CytoGnomix Inc., London, Canada;
b
Department of Pathology and Laboratory Medicine, University of Western Ontario, London, Canada;
c
Public Health England, Oxford, Great Britain;
d
Dalat Nuclear Research Institute, Dalat, Vietnam;
e
Canadian Nuclear Laboratories, Chalk River,
Canada;
f
Health Canada, Ottawa, Canada;
g
Departments of Biochemistry and Oncology, University of Western Ontario, London, Canada
ABSTRACT
Purpose: Inhomogeneous exposures to ionizing radiation can be detected and quantified with
the dicentric chromosome assay (DCA) of metaphase cells. Complete automation of interpretation
of the DCA for whole-body irradiation has significantly improved throughput without compromis-
ing accuracy, however, low levels of residual false positive dicentric chromosomes (DCs) have con-
founded its application for partial-body exposure determination.
Materials and methods: We describe a method of estimating and correcting for false positive
DCs in digitally processed images of metaphase cells. Nearly all DCs detected in unirradiated cali-
bration samples are introduced by digital image processing. DC frequencies of irradiated calibra-
tion samples and those exposed to unknown radiation levels are corrected subtracting this false
positive fraction from each. In partial-body exposures, the fraction of cells exposed, and radiation
dose can be quantified after applying this modification of the contaminated Poisson method.
Results: Dose estimates of three partially irradiated samples diverged 0.2–2.5 Gy from physical
doses and irradiated cell fractions deviated by 2.3%–15.8% from the known levels. Synthetic par-
tial-body samples comprised of unirradiated and 3 Gy samples from 4 laboratories were correctly
discriminated as inhomogeneous by multiple criteria. Root mean squared errors of these dose esti-
mates ranged from 0.52 to 1.14 Gy
2
and from 8.1 to 33.3%
2
for the fraction of cells irradiated.
Conclusions: Automated DCA can differentiate whole- from partial-body radiation exposures and
provides timely quantification of estimated whole-body equivalent dose.
ARTICLE HISTORY
Received 3 July 2020
Revised 17 August 2020
Accepted 1 September 2020
KEYWORDS
Ionizing radiation;
biodosimetry; chromosomal
aberrations; inhomogeneous
exposure; soft-
ware automation
Introduction
Accurate biological doses received by individuals exposed to
ionizing radiation must be determined in order to effectively
diagnose and treat victims. The dicentric chromosome assay
(DCA) is the gold standard biological dose assessment
method and is endorsed by the International Atomic Energy
Agency (IAEA), the World Health Organization, and the
Pan American Health Organization. Dicentric chromosome
(DC) aberrations are biomarkers of radiation exposure and
the IAEA recommends a sufficient count of either images
examined or DCs encountered for accurate assessment of
biological dose. Low linear energy transfer (LET) generates
chromosome breaks that can be mis-repaired as DCs, which
exhibit a Poisson distribution in cells. However, if radiation
exposure is inhomogeneous (partial body), the portion of
exposed cells expected to conform to a Poisson distribution
of DCs must be determined prior to estimating absorbed
dose (International Organization for Standardisation (ISO)
19238 2004; International Organization for Standardisation
(ISO) 21243 2008; International Atomic Energy
Agency 2011).
Traditionally, interpreting the DCA is a painstaking pro-
cess which requires significant training to perform.
Following extensive laboratory processing (Oestreicher et al.
2017), the operator examines metaphase images, excludes
those of poor quality, documents DCs in each image, then
determines the overall frequency of DCs. The frequency of
DCs per cell is related to absorbed radiation dose (in Gray
[Gy]). The DCA has been shown to be accurate for the
0–5 Gy range of exposures by fitting DC frequencies of
known dose to a linear quadratic calibration curve. The
absorbed dose of samples of unknown exposure is inferred
from the calibration curve based on DC frequency. For
accurate dose assessment, the detection of at least 100 DCs
at higher doses is recommended. However, at low-dose or
partial-body exposures in which DCs are much less frequent,
scoring of several thousand images is necessary for accurate
dose estimation (International Atomic Energy Agency 2011)
(though scoring of fewer cells is recommended as a first
CONTACT Peter K. Rogan progan@uwo.ca Department of Biochemistry, Schulich School of Medicine and Dentistry, University of Western Ontario, London
N6A 2C1, Canada
Supplemental data for this article can be accessed here.
Copyright ß 2020 Taylor & Francis Group LLC.
INTERNATIONAL JOURNAL OF RADIATION BIOLOGY
2020, VOL. 96, NO. 11, 1492–1503
https://doi.org/10.1080/09553002.2020.1820611
step to handle large numbers of samples for rapid ‘triage’ in
emergency response (Oestreicher et al. 2017 )).
Automated approaches have been sought after to improve
the throughput of the DCA, especially for large-scale testing
(Maznyk et al. 2012). Semi-automated detection of DCs still
requires manual image selection and verification of candi-
date DCs (Schunck et al. 2004). The Automated Dicentric
Chromosome Identifier and Dose Estimator (ADCI) soft-
ware completely automates DC detection and estimates bio-
logical radiation dose (Rogan et al. 2016). Suboptimal
metaphase images are removed (Liu et al. 2017), chromo-
somes within remaining images are classified, which are
then further discriminated as either normal or DC. ADCI
generates calibration curves and estimates exposure levels of
samples of uncertain doses. ADCI can process a sample of
500 metaphase images and estimate dose in 3–5 min using
a multicore desktop computer system (Intel i7-6700HQ,
16 GB RAM) equipped with a graphics processing unit
(GPU; Nvidia
V
R
GTX 960 M or RTX 2070) (Li et al. 2019).
This benchmark estimate is equivalent to 1.7 images per
second, or 6000 images per hour.
Image selection models which eliminate and/or rank
images are a prerequisite for accurate automated dose esti-
mation. The models are optimized to filter out suboptimal
chromosome morphology and control for preparation differ-
ences that are often variable between laboratories.
Application of these models can significantly reduce mis-
classification of DCs and increase the accuracy of DC fre-
quencies (Shirley et al. 2017).
Nevertheless, residual false positive (FP) DCs, that is,
monocentric chromosomes incorrectly classified as DCs,
produce inflated dose estimates, especially in samples
exposed to low levels of radiation. A previously published
FP method removes FP DCs flagged by ADCI by applying
filters designed to detect morphological subclasses of FPs
(Liu et al. 2017). These chromosomes are reclassified as nor-
mal, monocentric chromosomes and can be visualized in
ADCI in the built-in metaphase image viewer. While 55% of
FPs on average are eliminated using this method, some FPs
remain after filtering. The impact of the residual FPs is min-
imal when both calibration and test samples are processed
using the same algorithm, resulting in the equivalent levels
of FP misclassification in all images, regardless of source.
This effectively mitigates their effect on dose estimation (Li
et al. 2019). Dose estimation accuracy is therefore
unaffected, and results fulfill IAEA criteria for triage
biodosimetry.
Heterogeneous, partial-body exposure is prevalent in
cases of accidental radiation exposure (Prasanna et al. 2010).
Partially irradiated samples deviate from the expected
Poisson distribution, as the unirradiated portion of cells
inflates the percentage of cells lacking DCs. This deviation
must be considered to avoid underestimating exposures. The
impact of FP DCs on dose estimates of partially irradiated
samples was not predictable and affected the accuracy of
some estimates, especially at low-dose exposures. We
describe a framework for automated estimation of partially
irradiated samples using ADCI, which effectively corrects
DC counts of FPs resulting from image segmentation and
machine learning-based misclassification (Figure 1).
Methods
Sample preparation and image capture
Samples were irradiated by biodosimetry laboratories at
Health Canada (HC), Canadian Nuclear Laboratories (CNL),
Public Health England (PHE), and Dalat Nuclear Research
Institute (DNRI) using established protocols (International
Atomic Energy Agency 2011; Oestreicher et al. 2017; Pham
et al. 2019). HC irradiated samples using 250 kVp X-rays
(X-RAD-320 (Precision X-ray, North Branford, CT)) at a
dose rate of 0.8 Gy/min, CNL used a
137
Cs GammaCell40
(Atomic Energy of Canada Ltd., Ottawa, ON) at a dose rate
of 4.5 rad/s, DNRI used 200 kVp X-rays (Radioflex-
200EGM (Rigaku, Japan)) at a dose rate of 0.497 Gy/min.
Samples obtained from PHE were irradiated ex vivo in a
water phantom at 37
Cto
60
Co gamma rays, with a dose
rate of 0.27 Gy/min, at the University of Ghent irradiation
facility. Dosimetry was performed with a NE2571 Farmer
ionization chamber (Thermo Electron, UK) calibrated in
terms of air kerma using the IAEA TRS-277 code of prac-
tice. To simulate partial-body irradiations, irradiated blood
samples were mixed with sham-irradiated control blood
from the same donor in a ratio of 1:1 and sent to PHE, at
room temperature, for processing using standard techniques
(International Atomic Energy Agency 2011).
All laboratories captured images of metaphase cells utiliz-
ing a Metafer slide scanning platform (Metasystems, Newton,
MA). HC scanned slides on a Zeiss AxioImager.Z2 micro-
scope connected through a CoolCube 1 CCD camera using
Metafer4 v3.10.7 software. CNL scanned slides on a Zeiss
AxioImager.Z2 microscope equipped with a CoolCube 1
CCD camera using Metafer4 v3.11.8 software. PHE scanned
slides on a Zeiss AxioImager.M1 microscope and CoolCube 1
CCD camera using Metafer4 v3.9.10 software. PHE manually
selected images that appeared to contain approximately 46
chromosomes of good morphology that were reasonably well
spread from the low magnification (10) scan image gallery.
DNRI scanned slides on an AxioImager.Z2 microscope with
CCD camera using Metafer4 v3.10 software. DNRI further
selected images based on the following criteria: metaphase
cells at first mitotic division post-irradiation, with 46 chromo-
somes that are non-overlapping, well spread with chromatids
separated. HC, DNRI, and PHE utilized MSearch to eliminate
images lacking metaphase cells. CNL used MSearch to
capture all unsorted images automatically without applying
any selection criteria; ADCI was used to eliminate those
which did not contain metaphase cells. Images were exported
as TIFF files.
Sample transfer and image processing
Calibration samples of known dose ranging from 0 to 5Gy
(0–4.5 Gy for PHE, 0–4Gy for HC) were obtained from each
laboratory. All test samples were derived from whole-body
INTERNATIONAL JOURNAL OF RADIATION BIOLOGY 1493
exposures, except for PHE, which provided four whole-body
and three partial-body irradiated samples. Except for HC
and CNL (Li et al. 2019), transfer of metaphase image data
was performed via secured internet connection using
Synology Cloud Station software to a centralized Network
Attached Storage device at the University of Western
Ontario. Results for each laboratory were separated, and
images were grouped by sample dose. Transfers took
12–24 h on average, depending on image count and internet
connection speed. To assess transfer success, file counts
were matched to the expected number of images (refer to
Supplemental Table for image counts from each laboratory),
and random images were opened to assess potential
data corruption.
ADCI software was used to examine metaphases using
image processing, image segmentation, and machine learn-
ing methods (Li et al. 2016, 2019; Rogan et al. 2016; Liu
et al. 2017; Shirley et al. 2017). This process removes
Figure 1. Flowchart of major steps taken to synthesize partially irradiated samples, perform dose estimation, and predict fraction of cells irradiated. Rounded
shapes denote start and end points, rectangular shapes represent operations which must be performed, slanted parallelograms represent datasets. The flowchart
presents the steps necessary to analyze samples originating from a single laboratory, all steps were repeated for each laboratory.
1494 B. C. SHIRLEY ET AL.
