High-precision automatic identification method for dicentric chromosome images using two-stage convolutional neural...

Area of Science:

• Cytogenetics
• Radiation Biology
• Computational Biology

Background:

• Dicentric chromosome analysis is the standard for biological dose assessment.
• Efficient and objective methods are needed for large-scale radiation incidents.
• Automated image analysis offers a promising solution.

Purpose of the Study:

• To develop an automated method for identifying dicentric chromosome images.
• To enhance the efficiency and objectivity of biological dose assessment.
• To improve the detection of challenging chromosome morphologies.

Main Methods:

• A two-stage convolutional neural network (CNN) approach was developed for image identification.
• K-means and watershed algorithms were used for segmenting adhesive chromosome masses.
• Giemsa-stained microscopic images were utilized for analysis.

Main Results:

• The two-stage CNN achieved a high identification accuracy of 99.4%, with 85.8% sensitivity and 99.6% specificity.
• The method effectively identified difficult cases, including concealed centromeres and entangled chromosomes.
• Analysis speed was increased by 20 times compared to manual detection, reducing false positives.

Conclusions:

• The proposed two-stage CNN method provides an accurate and efficient automated solution for dicentric chromosome identification.
• This approach significantly enhances biological dose assessment capabilities in radiation emergencies.
• The methodology offers a valuable reference for automated image identification tasks with low target rates.