人工智能在胸部骨折CT诊断中的应用

Application of artificial intelligence in CT diagnosis of thoracic fracture

目的探讨人工智能骨疾病诊断系统(AI)在胸部骨折中的诊断效能及应用价值。方法回顾性分析深圳大学总医院因创伤急诊就诊行胸部CT并经3~6周复查胸部CT证实的726例胸部骨折,计算AI和两名放射科医师以及放射科医师在AI辅助下对胸部骨折诊断的召回率、精确率和F1分数。结果AI检测肋骨骨折的召回率和F1分数分别为0.91和0.92,均高于放射科医师1(0.77与0.85)和放射科医师2(0.84与0.90),AI的精确率(0.92)低于放射科医师1(0.95)和放射科医师2(0.96)。放射科医师1和2在AI辅助下肋骨骨折检测的召回率、精确率和F1分数分别为0.94、0.95、0.94和0.97、0.98、0.97。AI检测胸部其他骨折的召回率和F1分数(0.90与0.90)高于放射科医师1(0.62与0.74)和放射科医师2(0.73与0.81),放射科医师1和2在AI辅助下,胸部其他骨折检测的召回率、精确率和F1分数分别为0.94、0.95、0.94和0.97、0.97、0.97。结论AI能高效高敏感地检测急诊创伤患者胸部CT中的胸部骨折,有望优化急诊创伤患者的诊疗流程。

Objective To investigate the diagnostic performance and application value of an artificial intelligence(AI)bone disease diagnosis system in the diagnosis of thoracic fractures.Methods A retrospective analysis was performed on 726 cases of thoracic fractures confirmed by chest CT re-examination 3-6 weeks after trauma emergency admission at Shenzhen University General Hospital.The recall rate,precision rate,and Fl score of AI,two radiologists,and the radiologists assisted by AI in diagnosing thoracic fractures were calculated.Results The recall rate and Fl score of AI in detecting rib fractures were 0.91 and 0.92,respectively,both higher than those of Radiologist 1(0.77,0.85)and Radiologist 2(0.84,0.90).The precision rate of AI(0.92)was lower than that of Radiologist 1(0.95)and Radiologist 2(0.96).With AI assistance,the recall rate,precision rate,and F1 score of Radiologist 1 and Radiologist 2 in detecting rib fractures were 0.94,0.95,0.94 and 0.97,0.98,0.97,respectively.For detecting other thoracic fractures,the recall rate and F1 score of AI(0.90,0.90)were higher than those of Radiologist 1(0.62,0.74)and Radiologist 2(0.73,0.81).With AI assistance,the recall rate,precision rate,and F1 score of Radiologist 1 and Radiologist 2 in detecting other thoracic fractures were 0.94,0.95,0.94 and 0.97,0.97,0.97,respectively.Conclusion AI can efficiently and sensitively detect thoracic fractures in chest CT scans of emergency trauma patients,potentially optimizing the diagnosis and treatment process for emergency trauma patients.