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

Abstract

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.

Keywords: chromosome metaphases image; cytogenetic dosimetry; deep learning; dicentric chromosome assay; object detection; transfer learning; you only look once.

Figure 1

Comprehensive procedure for training the object detection network with chromosome images and individual dicentric chromosome patches. The object detection model is initialized with the pretrained weight, regardless of the chromosome or the metaphase image.

Figure 2

Network architecture of YOLOv5. The network consists of a backbone (CSPDarknet), neck (PANet), and head (YOLO layer).

Figure 3

Plots of training losses (location loss, objectness loss, and classes loss) versus the epochs for the pretrained and randomized initial weights.

Figure 4

Plots of evaluation metrics for the model pretrained on the MS-COCO dataset. (a) F1 score vs. confidence score and (b) precision–recall curve for validation dataset. (c) F1 score vs. confidence score and (d) precision–recall curve for test dataset. “N” and “D” denote normal and dicentric chromosome, respectively.

Figure 5

Plots of evaluation metrics for the randomly initialized model. (a) F1 score vs. confidence score and (b) precision–recall curve for the validation dataset. (c) F1 score vs. confidence score and (d) precision–recall curve for the test dataset. “N” and “D” denote normal and dicentric chromosome, respectively.

Figure 6

Visualization examples of the input (first row) and corresponding detection results of the input (second row). The red bounding boxes contain the dicentric chromosomes and the green ones the normal monocentric chromosomes.