Performance of the Diagnostic Value of Bone Age Assessment Software Based on Deep Learning in Chinese Children

High accuracy and precision bone age assessment is very important for the diagnosis and treatment monitoring of various pediatric diseases. The commonly used bone age assessment methods include GP atlas, TW3 score and Zhonghua 05. GP method is to compare wrist X-ray films with atlas reference X-ray films. Its main disadvantages are strong subjectivity and long atlas standard interval. Different from GP method, TW3 method is to grade and score each bone, add each epiphyseal score to calculate the total score of bone maturity, and obtain the corresponding final bone age value. Although TW3 scoring method is relatively accurate, it is complex and time-consuming, and there is great variability among evaluators. In order to evaluate bone age more efficiently and accurately, a method based on computer image automatic recognition technology can help to overcome these problems.

In this study, 1000 children aged 1-18 in 5 hospitals are selected as the research objects. After taking bone age films with bone age instrument, the film reading results and evaluation time of AI Group, artificial group and standard group are recorded. One month later, the artificial group re-analyzes 1000 films with the assistance of AI system, and the evaluation time is recorded. Finally, the accuracy and time difference of artificial group, AI Group, artificial combined AI Group and standard group are compared.

The purpose of this study is to use the most advanced artificial intelligence deep learning bone age evaluation software to explore the value of bone age instrument to improve the accuracy and diagnostic efficiency of bone age evaluation by pediatricians.