AUTHOR=Yang Chen , Dai Wei , Qin Bin , He Xiangqian , Zhao Wenlong 

TITLE=A real-time automated bone age assessment system based on the RUS-CHN method

JOURNAL=Frontiers in Endocrinology

VOLUME=Volume 14 - 2023

YEAR=2023

URL=https://www.frontiersin.org/journals/endocrinology/articles/10.3389/fendo.2023.1073219

DOI=10.3389/fendo.2023.1073219

ISSN=1664-2392

ABSTRACT=Bone age is the age of skeletal development and is a direct indicator of physical growth and development in children. In this study, a lightweight model based on the RUS_CHN method was used to obtain key skeletal grades and locations in one step, and bone age was calculated by regressing the grades using Lightgbm, which was able to output bone age in real time.
Critical bone position and grade were determined using three lightweight target detection models using the RUS_CHN approach, and then the age of the bones was predicted using a Lightgbm regression model. Intersection over Union (IOU) was used to evaluate the precision of critical bone placement, while Mean Absolute Error (MAE) and Root Mean Square Error were used to evaluate the discrepancy between predicted and true bone age (RMSE). In order to test the CPU speed, the model was finally transformed into an Open Neural Network Exchange (ONNX) model.
The three lightweight models achieved good results with an average Intersection over Union above 0.9 in all key bones. The best results for the bone age prediction component MAE were 0.37 years, and the best results for RMSE were 0.49 years. Using the CPU 5600X for prediction, the critical bone level and position prediction time was 36 ms. The bone age prediction time was 2 ms.
We developed an RUS_CHN BAA system based on lightweight target detection that can output results instantly in settings with limited computing capacity, such a CPU, while keeping good accuracy and robustness to adjust to circumstances without manual segmentation.