신속한 방사선 Biodosimetry로 자동 Dicentric 염색체 식별 (ADCI) 및 복용량 추정

The overall goal of this software-based protocol is to accurately and rapidly estimate total body biological radiation dose from dicentric chromosomes found in sets of automatically filtered metaphase cell images using machine learning-based image processing techniques. This method can help answer key questions regarding the response to a mass radiation exposure event. Timely examination of a large number of samples will be necessary to determine radiation dose.

The main advantage of this software is automation for image filtering, calibration curve generation, dicentric chromosome detection, and the radiation dose estimation. These are traditionally time-consuming and labor-intensive tasks. Though this method can provide insights into radiation cytogenic biodosimetry, it could be applied to select the best chromosomal images for diagnostic analyses of inherited and acquired genetic diseases.

We first had the idea for this method when we began developing chromosome image analysis methods for automatically scoring fluorescence inside two hybridization images. Colleagues had indicated that the current capacity for testing multiple exposed individuals was also very limited. Open the software and create a new sample from the menu.

Then select the folder containing the metaphase images to analyze. At the prompt, supply a unique ID to identify the sample. Only alphanumeric, underscores, and dashes may be used.

Then provide a description of the sample if desired. Repeat this process to add additional sample image sets. A minimum of three calibration sample sets is required for the analysis, but use seven or more for good results.

They should span zero to five grays of radiation exposure. Once all the samples are loaded, highlight them in the samples list and add them to the process queue. Then process all the samples to identify various properties within each image.

Processing progress is displayed. Once the processing has finished, click the green check mark and save the processed samples as sample files. Before proceeding, it is a good idea to view the the processed images in the Metaphase Image Viewer.

Highlight samples from the list and open the viewer. Then navigate through the images. From within the image viewer, filters can be applied.

Several image selection models are bundled with the software. Filtering will screen the properties identified during processing. If desired, a customized model can be built from scratch and saved for repeated use.

If the Curve Calibration Wizard described in the next section is used, the model is applied during that step. Use the Curve Calibration Wizard to analyze process sample files. For a good curve fit, include at least seven calibration samples from zero to five gray.

In the first page of the Wizard macro, place a check mark besides each desired calibration sample. For each calibration sample, specify the radiation dose in the adjacent text field. In the next page, select a preset image selection model.

In the next page, select an SVM Sigma value from the dropdown box. A value of 1.4 or 1.5 is recommended for dose estimates greater than one gray. Otherwise, use a value of 1.0.

On the final page, review all the selected options. If they appear correct, finish the selection process. Now a pre-populated create a curve dialog will appear.

Specify a unique identity for the curve in the text box and then press Validate data to ensure the content of the the response dose list is formatted correctly. Verify that all of the fields in the response dose list are highlighted in green, which indicates valid data. Then send the data for analysis using the OK button and follow the prompts to save the results.

To perform the dose estimate analysis, start the Wizard macro found under the Dose Estimation menu. First, select a previously created calibration curve from the dropdown box. Its properties will appear below.

On the next page, toggle the test samples of unknown exposure to include them in the dose estimation. On the next page, review the description and properties of the image selection model applied during the calibration curve generation. Every sample should be analyzed using the same model.

When estimating the dose of a sample exposed to an unknown amount of radiation, we strongly recommend that the same image selection model be applied to this data that corresponded to the model used in generating the dose calibration curve. On the next page, make certain that the same SVM Sigma value used during calibration curve generation populates the fields. Make corrections if needed.

On the final page, review all the selected options. If they appear correct, finish the selection process and press OK in the pre-populated dose calculator dialog to perform dose estimation. The dose estimation results are displayed in the console in tabular format for each test sample in grays.

Accompanying the tabular data will be plots of the data shown in the plot area. After either entering a sample, preparing your calibration curve, or estimating a dose, it is advisable to generate a sample, a curve, or a dose report to create a complete record of your analysis. Now generate a Dose Estimation Report.

When prompted, provide a name for the report or use the browse option to modify an existing report folder. Audit capabilities, log file integrity issues, and statistical options are reviewed in the text protocol. The software was tested using cytogenetic image samples exposed to ionizing radiation that were obtained from the biodosimetry laboratories at Health Canada and at Canadian Nuclear Laboratories.

These samples were generated and analyzed manually by experts at both labs. The software used automated image selection filters on the samples to eliminate a wide variety of unusable images. In the current example, the best 250 images were automatically selected for each sample.

C was excluded because it contained too many overlapping chromosomes, and D has excessive numbers of separated sister chromatids. The software may incorrectly flag some objects as dicentric chromosomes, like object two. It inflates dicentric counts in samples.

False positive filters exclude these objects from calculations of dicentric frequency. After filtering, the occurrences of dicentric chromosomes in a population of cells from an irradiated sample follow a Poisson distribution, as compared to the un-filtered data set. Chi-square goodness of fit statistics showed the results from filtered data are much more reliable.

Ultimately, the automated measurements were within 0.3 gray of the manually scored doses, with a few exceptions. One sample that had a poor goodness of fit to Poisson distribution was not accurate. This was a lower quality sample.

It contained too many images with incorrectly classified dicentric chromosomes resulting from sister chromatid separation. But the results of the automated analysis still fell within the IAEA specified triage limits. After watching this video, you should have a good understanding of how to use the software to automatically select optimal images using image selection models, create calibration curves, perform automated dicentric chromosome detection in radiation dose estimation, and create reports.

Once mastered, this technique can estimate dose of a sample containing 500 metaphase images in 10 to 20 minutes from a existing set of metaphase cell images. It's important to remember to make use of the automated functionality the software provides. Manual selection of images is possible.

However, doing so reduces efficiency and may not be as reproducible as the automated procedures. This technique will enable researchers in the field of biodosimetry to explore new questions in radiation research involving population studies that could not practically be carried out in the past.