THE PROJECT OF ANOTHER LOW-COST METAPHASE FINDER (SECOND REPORT-APPLICATION OF ARTIFICIAL INTELLIGENCE).
Journal:
Radiation protection dosimetry
PMID:
30806467
Abstract
Biological dosimetry is used to estimate individual absorbed radiation dose by quantifying an appropriate biological marker. The most popular gold-standard marker is the appearance of dicentric chromosomes in metaphase. The metaphase finder is a tool for biological dosimetry that finds metaphase cells on glass slides. The author and a software company have designed a new system and are now preparing to produce the system commercially. The metaphase finder consists of an automated microscope, a camera, and a computer. To enhance the accuracy of the system, an artificial intelligence (AI) with deep learning was tested. A total of 1709 images of the metaphase finder detected as 'metaphases' were read into a nine-layer artificial neural network to detect true metaphases. A total of 456 images were used for training, and the rest of the images were used for validation. The accuracy of AI was 0.89 for metaphases and 0.90 for non-metaphases.