Enhancing chromosomal analysis efficiency through deep learning-based artificial intelligence graphic analysis

Abstract

The objective of this study is to evaluate the efficacy and diagnostic utility of an advanced chromosomal analysis approach. A total of 2663 amniotic fluid samples were chosen for chromosomal karyotype profiling between January 2022 and June 2023. Two sets of tests were carried out: experiment 1 involved randomly selecting 1168 examples to test the accuracy of machine learning-based chromosomal karyotypes. The aim was to determine the method’s general applicability when cases were naturally dispersed. Experiment 2 concentrated on randomly selecting the most common examples of chromosomal number anomalies and cases with structural defects that did not affect the visual assessment of chromosome categories. The goal was to investigate the diagnostic efficacy of the artificial intelligence (AI) analysis system in detecting these flaws. The results of experiment 1 demonstrated the resilience of the intelligent analysis system in cases with significant differences in chromosomal karyotypes, resulting from manual shooting and film-making. Experiment 2 results showed that the intelligent analysis system surpassed the standard chromosomal image analysis program in terms of automated analysis accuracy, for both normal and defect cases. Furthermore, the intelligent analysis system demonstrated detection and analysis speeds that were 3–15 times faster. The average speed of regular case analysis increased by a factor of 4–6, cases with quantitative defects increased by a factor of 3–5, and cases with structural defects increased by a factor of 5–7. Implementing a chromosome intelligence analysis system in clinical practice could improve the efficiency of chromosome identification and analysis, allow for more widespread chromosomal examination, and reduce the likelihood of congenital defects.