Artificial intelligence-based bone age assessment using deep learning of characteristic regions in digital hand radiograph

• DOI: 10.3760/cma.j.issn.1005-1201.2019.10.020
• Publication Date: 2019-10-10
• Authors: Ying Wen; Xuhua Ren; Xiujun Yang; Lihong Li; Jun Lan; Tingting Li; Qian Wang; Lili Shi
• Keywords: Data set

Objective To detect the feasibility and efficiency of bone age(BA) artificial intelligence(AI) estimation based on deep learning features from traditional regions interest(ROI) in hand digital radiographs(DR).

Methods BA dataset left DR with 11 858 subjects aged 0 to 18 years Children′s Hospital Shanghai were split training(80.0%) validation (20.0%) set this study. An improved regression convolutional neural networks extreme gradient boosting decision tree method utilized for BA analysis ROIs images. Another data 1 229 also hospital was adopted test. Mean average precision(mAP) mean absolute error(MAE) used assess model accuracy detection prediction, respectively.

Results The mAP 0.91,and MAE all male female 0.461 0.431 respectively test sets. The difference less than year accounted 90.07% between assessment peadiatric radiologists, an rate 96.67%.The over 9.03% (with underestimation 6.43% overestimation 2.60%), which corresponding age being training or sesamoid nearby adductor pollicis fusion epiphysis appeared set.

Conclusion An AI ROIs′ images is initially achieved automatically predict rapidly effectively, yet it still needs further optimization.

Key words: Age determination by skeleton; Radiography; Diagnosis, computer-assisted; Artificial intelligence