A deep learning model (FociRad) for automated detection of γ-H2AX foci and radiation dose estimation

Abstract

DNA double-strand breaks (DSBs) are the most lethal form of damage to cells from irradiation. γ-H2AX (phosphorylated form of H2AX histone variant) has become one of the most reliable and sensitive biomarkers of DNA DSBs. However, the γ-H2AX foci assay still has limitations in the time consumed for manual scoring and possible variability between scorers. This study proposed a novel automated foci scoring method using a deep convolutional neural network based on a You-Only-Look-Once (YOLO) algorithm to quantify γ-H2AX foci in peripheral blood samples. FociRad, a two-stage deep learning approach, consisted of mononuclear cell (MNC) and γ-H2AX foci detections. Whole blood samples were irradiated with X-rays from a 6 MV linear accelerator at 1, 2, 4 or 6 Gy. Images were captured using confocal microscopy. Then, dose–response calibration curves were established and implemented with unseen dataset. The results of the FociRad model were comparable with manual scoring. MNC detection yielded 96.6% accuracy, 96.7% sensitivity and 96.5% specificity. γ-H2AX foci detection showed very good F1 scores (> 0.9). Implementation of calibration curve in the range of 0–4 Gy gave mean absolute difference of estimated doses less than 1 Gy compared to actual doses. In addition, the evaluation times of FociRad were very short (< 0.5 min per 100 images), while the time for manual scoring increased with the number of foci. In conclusion, FociRad was the first automated foci scoring method to use a YOLO algorithm with high detection performance and fast evaluation time, which opens the door for large-scale applications in radiation triage.

Machine Learning Classification of 53BP1 Foci

“…These methods are not as time-consuming as deep learning. For example, the total computational training time for FociRad's deep learning focus detection was over 37 hours [32].…”

“…While manual counting is regarded as the standard, it is subject to large time requirements and biased results [22][23][24][25]. Automated methods have increased the efficiency of detection and have the capability of handling multi-channel and multi-dimensional images [23][24][25][26][27][28][29][30][31][32].…”

“…Several object segmentation techniques have been explored to address the uncertainty of 2D focus aggregates. Methods differentiating grouped foci have ranged from simple signal thresholding, to complex application of the watershed transform in 3D, convolutional neural networks, and other algorithms [24,25,[28][29][30][31][32]34]. Despite the success of these focus segmentation approaches, many come at a high computational cost and overlook the DNA damage complexity information found in clustered foci.…”

“…Resolving overlapping foci that form due to increasing radiation dose remains a challenge in the detection of foci imaged by 2D IF microscopy [25, 32, 33]. Clusters could represent separate foci from distinct planes in the cell or a larger focus from a single plane.…”

“…31,32) were strongly correlated (>0.80). Properties that did not exhibit linear relationships included the eccentricity of the focus FT (property 20) and focus orientation (property 0), scoring correlation values under 0.1 for several properties.The histogram comparison via correlation (property 6) demonstrated moderate negative correlation (-0.50) with a variety of properties including the focus eccentricity, histogram comparison via standard and alternative Chi-Squared distance (properties 5, 7, 10), and a strong negative correlation (-0.80) with the histogram comparison via Bhattacharrya and Hellinger distances (properties 9, 11).…”