Dicentric chromosome assay using a deep learning-based automated system

Automated system for establishing standard radiation dose-response curves and dose estimation for the Korean population

Biological dosimetry is crucial for estimating the doses from biological samples and guiding medical interventions for accidental radiation exposure. This study aimed to derive rapid and precise dose estimates using a dicentric chromosome assay. To address the challenges of manual scoring of dicentric chromosomes, we upgraded an automatic system aimed at enhancing the precision of dicentric chromosome detection while reducing the need for human intervention. We collected blood from 30 individuals aged 20-67 years to create 30 dose-response curves aiming to investigate the differences in responses among individuals. To validate dose-estimate accuracy within a 95% confidence interval, blinded samples were categorized into three groups according to the radiation dose as follows: ≥2, ≤ 1, and 0.1 Gy. When scoring dicentric chromosomes without human review and constructing a dose-response curve, individual differences were observed. For doses ≤ 1 Gy, the standard root formula was effective; conversely, for doses ≥ 2 Gy, the regression deep neural network proved to be more ac-curate. Our developed program allowed for the rapid analysis of a large volume of dicentric chromosome images.

Collaborative activities in a biological dosimetry network for radiation emergencies in South Korea

PURPOSE

Biological dosimetry is an essential analytic method to estimate the absorbed radiation dose in the human body by measuring changes in biomolecules after radiation exposure. Joint response in a network to mass-casualty radiation incidents is one way to overcome the limitations of biological dosimetry, sharing the workload among laboratories. This study aimed to investigate the current performance, collaborative activities and technical advances of the Korea biodosimetry network (K-BioDos), and suggest the future directions toward successful joint response.

MATERIALS AND METHODS

A survey was performed to investigate the capacities of each laboratory and their expectations for the K-BioDos network. We summarized the capacities, expectations and technical advances of K-BioDos members. Based on the results, in-depth discussion was carried out to determine the future plan and activities of K-BioDos.

RESULTS

K-BioDos has grown to six laboratories since its establishment with three functional laboratories of biological dosimetry in South Korea. We constructed long-term strategy according the survey results, and performed various activities for enhanced biological dosimetry capabilities - including intercomparison exercises, education, and resource sharing. Through these active collaborations we achieved harmonization of biodosimetry protocols and technical improvement such as better image quality.

CONCLUSIONS

K-BioDos network performed various activities for joint response and constructed long-term plans, considering the expectations and feedbacks of members. K-BioDos continue to support members to establish and develop biodosimetry tools. These efforts and findings could serve as a fundamental guide for coordinated network responses in the event of large-scale radiological disaster.

Adaptive Segmentation of DAPI-stained, C-banded, Aggregated and Overlapping Chromosomes

Existing algorithms for automated segmentation of chromosomes and centromeres do not work well for condensed, C-banded and DAPI-stained chromosomes and centromeres. Overlapping and aggregation, which frequently occur in metaphase spreads, introduce additional challenges to the counting of chromosomes and centromeres in the Dicentrics Chromosome Assay (DCA). In this paper, we introduce adaptive algorithms, for segmentation of difficult metaphase spreads that include overlapping and aggregated chromosomes. In order to enhance and segment chromosomes, two optimizations are done: (1) the best algorithm among several options is automatically chosen based on predefined figures of merit, (2) the algorithm is automatically optimized with a binary search to modify its parameters to achieve predefined thresholds. These algorithms are designed to separate mildly or moderately aggregated chromosomal clusters. The clusters are segmented by skeleton junctions, reduction of the overall object thickness, and the watershed algorithm. The chromosomes are characterized by rules we establish, using minimal assumptions. Centromeres are detected by detecting bright spots on the surface of the chromosomes, and then using cluster analysis and shape and intensity profiles to identify them as centromeres. High sensitivity and specificity for chromosome and centromere detection were achieved.

Prenatal diagnosis of dicentric chromosome X mosaicism: a case report and review

A dicentric chromosome is an abnormal chromosome with two centromeres on the same chromosome. It has been reported that dicentric chromosomes are specific biomarkers of radiation exposure, but dicentric chromosomes are rarely identified in newborns with multiple congenital anomalies. At 16 weeks of gestation, a 39-year-old pregnant woman (gravida 2, para 1) was referred to the prenatal diagnosis center for genetic counseling. The fetal ultrasonography indicated multiple anomalies. Subsequently, amniocentesis was performed, and the G-banding karyotype analysis showed a rare type of mosaicism. The C-banding karyotype analysis indicated a pseudo-dicentric chromosome X [psu dic (X; 18) (p11.2; p11.2)]. A single-nucleotide polymorphism array (SNP array) revealed three pathogenic copy number variations (CNVs). After genetic counseling, the parents chose to terminate this pregnancy. This study provides new evidence for a better understanding of the diagnosis of dicentric chromosomes and emphasizes on the importance of genetic counseling.

LBDNet interlaboratory comparison for the dicentric chromosome assay by digitized image analysis applying weighted robust statistical methods

PURPOSE

This interlaboratory comparison was conducted to evaluate the performance of the Latin-American Biodosimetry Network (LBDNet) in analyzing digitized images for scoring dicentric chromosomes from in vitro irradiated blood samples. The exercise also assessed the use of weighted robust algorithms to compensate the uneven expertise among the participating laboratories.

METHODS

Three sets of coded images obtained through the dicentric chromosome assay from blood samples irradiated at 1.5 Gy (sample A) and 4 Gy (sample B), as well as a non-irradiated whole blood sample (sample C), were shared among LBDNet laboratories. The images were captured using the Metafer4 platform coupled with the AutoCapt module. The laboratories were requested to perform triage scoring, conventional scoring, and dose estimation. The dose estimation was carried out using either their laboratory calibration curve or a common calibration curve. A comparative statistical analysis was conducted using a weighted robust Hampel algorithm and z score to compensate for uneven expertise in dicentric analysis and dose assessment among all laboratories.

RESULTS

Out of twelve laboratories, one had unsatisfactory estimated doses at 0 Gy, and two had unsatisfactory estimated doses at 1.5 Gy when using their own calibration curve and triage scoring mode. However, all doses were satisfactory at 4 Gy. Six laboratories had estimated doses within 95% uncertainty limits at 0 Gy, seven at 1.5 Gy, and four at 4 Gy. While the mean dose for sample C was significantly biased using robust algorithms, applying weights to compensate for the laboratory's analysis expertise reduced the bias by half. The bias from delivered doses was only notable for sample C. Using the common calibration curve for dose estimation reduced the standard deviation (s*) estimated by robust methods for all three samples.

CONCLUSIONS

The results underscore the significance of performing interlaboratory comparison exercises that involve digitized and electronically transmitted images, even when analyzing non-irradiated samples. In situations where the participating laboratories possess different levels of proficiency, it may prove essential to employ weighted robust algorithms to achieve precise outcomes.

Radiation dose estimation with multiple artificial neural networks in dicentric chromosome assay

PURPOSE

The dicentric chromosome assay (DCA), often referred to as the 'gold standard' in radiation dose estimation, exhibits significant challenges as a consequence of its labor-intensive nature and dependency on expert knowledge. Existing automated technologies face limitations in accurately identifying dicentric chromosomes (DCs), resulting in decreased precision for radiation dose estimation. Furthermore, in the process of identifying DCs through automatic or semi-automatic methods, the resulting distribution could demonstrate under-dispersion or over-dispersion, which results in significant deviations from the Poisson distribution. In response to these issues, we developed an algorithm that employs deep learning to automatically identify chromosomes and perform fully automatic and accurate estimation of diverse radiation doses, adhering to a Poisson distribution.

MATERIALS AND METHODS

The dataset utilized for the dose estimation algorithm was generated from 30 healthy donors, with samples created across seven doses, ranging from 0 to 4 Gy. The procedure encompasses several steps: extracting images for dose estimation, counting chromosomes, and detecting DC and fragments. To accomplish these tasks, we utilize a diverse array of artificial neural networks (ANNs). The identification of DCs was accomplished using a detection mechanism that integrates both deep learning-based object detection and classification methods. Based on these detection results, dose-response curves were constructed. A dose estimation was carried out by combining a regression-based ANN with the Monte-Carlo method.

RESULTS

In the process of extracting images for dose analysis and identifying DCs, an under-dispersion tendency was observed. To rectify the discrepancy, classification ANN was employed to identify the results of DC detection. This approach led to satisfaction of Poisson distribution criteria by 32 out of the initial pool of 35 data points. In the subsequent stage, dose-response curves were constructed using data from 25 donors. Data provided by the remaining five donors served in performing dose estimations, which were subsequently calibrated by incorporating a regression-based ANN. Of the 23 points, 22 fell within their respective confidence intervals at p < .05 (95%), except for those associated with doses at levels below 0.5 Gy, where accurate calculation was obstructed by numerical issues. The accuracy of dose estimation has been improved for all radiation levels, with the exception of 1 Gy.

CONCLUSIONS

This study successfully demonstrates a high-precision dose estimation method across a general range up to 4 Gy through fully automated detection of DCs, adhering strictly to Poisson distribution. Incorporating multiple ANNs confirms the ability to perform fully automated radiation dose estimation. This approach is particularly advantageous in scenarios such as large-scale radiological incidents, improving operational efficiency and speeding up procedures while maintaining consistency in assessments. Moreover, it reduces potential human error and enhances the reliability of results.

Clonal hematopoiesis of a novel dic(18;20) clone following allogeneic hematopoietic stem cell transplantation

A 55-year-old man in first complete remission of acute myeloid leukemia with a normal karyotype underwent allogeneic hematopoietic stem cell transplantation from a human-leukocyte-antigen-matched sibling. Bone marrow examination on day 28 confirmed complete remission, but G-banding analysis revealed a novel chromosomal abnormality, including dic(18;20)(p11.2;q11.2). The patient developed moderate chronic graft-versus-host disease on day 174, and the abnormal clones identified by dic(18;20) significantly increased after that point. Chimerism testing repeatedly confirmed complete donor type. Although next-generation sequencing showed no clonal hematopoiesis-related gene mutations, copy number analysis of the donor and the recipient revealed copy number deletion of 18p, 18q, and 20q. The patient has maintained remission for more than 2 years to date without developing a hematologic neoplasm or cytopenia. The distinctive clonal hematopoiesis with a dicentric chromosome seemed to have undergone the breakage-fusion-bridge cycle, which could cause the complex events of deletion, amplification, and inversion. These copy number alterations might have increased the number of clones with growth advantage, and the highly inflammatory environment in the recipient due to graft-versus-host disease might have contributed to the clonal selection.

Deep Neural Network-Based Automatic Dicentric Chromosome Detection Using a Model Pretrained on Common Objects

Dicentric chromosome assay (DCA) is one of the cytogenetic dosimetry methods where the absorbed dose is estimated by counting the number of dicentric chromosomes, which is a major radiation-induced change in DNA. However, DCA is a time-consuming task and requires technical expertise. In this study, a neural network was applied for automating the DCA. We used YOLOv5, a one-stage detection algorithm, to mitigate these limitations by automating the estimation of the number of dicentric chromosomes in chromosome metaphase images. YOLOv5 was pretrained on common object datasets. For training, 887 augmented chromosome images were used. We evaluated the model using validation and test datasets with 380 and 300 images, respectively. With pretrained parameters, the trained model detected chromosomes in the images with a maximum F1 score of 0.94 and a mean average precision (mAP) of 0.961. Conversely, when the model was randomly initialized, the training performance decreased, with a maximum F1 score and mAP of 0.82 and 0.873%, respectively. These results confirm that the model could effectively detect dicentric chromosomes in an image. Consequently, automatic DCA is expected to be conducted based on deep learning for object detection, requiring a relatively small amount of chromosome data for training using the pretrained network.

Biomarkers for Biodosimetry and Their Role in Predicting Radiation Injury

Radiation-related normal tissue injury sustained during cancer radiotherapy or in a radiological or mass casualty nuclear incident is a major health concern. Reducing the risk and mitigating consequences of radiation injury could have a broad impact on cancer patients and citizens. Efforts to discover biomarkers that can determine radiation dose, predict tissue damage, and aid medical triage are underway. Exposure to ionizing radiation causes changes in gene, protein, and metabolite expression that needs to be understood to provide a holistic picture for treating acute and chronic radiation-induced toxicities. We present evidence that both RNA (mRNA, microRNA, long noncoding RNA) and metabolomic assays may provide useful biomarkers of radiation injury. RNA markers may provide information on early pathway alterations after radiation injury that can predict damage and implicate downstream targets for mitigation. In contrast, metabolomics is impacted by changes in epigenetics, genetics, and proteomics and can be considered a downstream marker that incorporates all these changes to provide an assessment of what is currently happening within an organ. We highlight research from the past 10 years to understand how biomarkers may be used to improve personalized medicine in cancer therapy and medical decision-making in mass casualty scenarios.