基于深度学习的人体肋骨骨折智能检测技术

目的将人工智能中的深度学习技术应用到人体肋骨骨折识别,实现人体肋骨骨折智能检测,提高法医肋骨骨折诊断效率。方法采集3143例人体胸部X线数字影像(2602例用于训练,541例用于测试),标注肋骨骨折特征点,通过多层网络堆叠,分层、分级主动学习原始数据高度抽象的特征表述,并将此特征反馈至检测器进行骨折检测,输出骨折位置及相应置信度。结果基于深度学习的人体肋骨骨折检测准确率在90%以上。结论基于深度学习的人体肋骨骨折检测准确率较高,可用于辅助法医进行肋骨骨折识别诊断、检验鉴定等,本研究可为人体其他部位骨骼损伤智能检测提供参考。

Deep Learning into Intelligent Detection of Rib Fracture from X-Ray Imagery

Objective To apply deep learning intelligence into rib fracture detection from X-ray medical imagery to realize the artificial intelligent detection of human rib fracture so as to improve the forensic diagnostic efficiency of rib fracture. Methods 3143 human chest digital X-ray radiographs were collected, having the relevant rib fractures marked so that such labelled images were taken as the input. With 2602 radiographs to be used for training and the other 541 ones for testing, intelligent deep learning launched actively to learn those abstract featuring representations in a hierarchical way from the raw image through stacking multiple neural network layers. The derived featuring representations were further fed into a detector to have the fracture area localized. The output indicated both the image coordinates referring to the rib fracture area and the corresponding confidence. Results An accuracy was greater than 90% obtained from the deep learning intelligence to detect human rib fracture. Conclusions Deep learning intelligence is promising in X-ray medical rib fracture detection, capable of assisting forensic diagnosis for rib fracture detection and reference to intelligent detection about other bone fracture.