人工智能在肋骨骨折诊断中应用价值  

Value of artificial intelligence to the diagnosis of rib fractures

作　　者：白岩 蔡显圣[2] 张传臣[3] 魏里[1] BAI Yan;CAI Xiansheng;ZHANG Chuanchen;WEI Li(CT Room,Liaocheng People's Hospital,Liaocheng,Shandong 252000,China;Equipment Department,Liaocheng People's Hospital,Liaocheng,Shandong 252000,China;MR Room,Liaocheng People's Hospital,Liaocheng,Shandong 252000,China)

机构地区：[1]聊城市人民医院CT室,山东聊城252000 [2]聊城市人民医院医疗设备处,山东聊城252000 [3]聊城市人民医院MR室,山东聊城252000

出　　处：《中华实用诊断与治疗杂志》2023年第10期1020-1024,共5页Journal of Chinese Practical Diagnosis and Therapy

基　　金：国家自然科学基金(61976110)。

摘　　要：目的比较人工智能软件与住院医师在CT影像中诊断肋骨骨折的效能,探讨人工智能在肋骨骨折诊断中的应用价值。方法2021年7—12月聊城市人民医院诊治肋骨骨折患者350例,均行胸部CT检查,分别采用人工智能软件和人工阅片(2名影像科住院医师独立完成)分析CT影像,记录阅片时间、骨折数量、骨折类型及骨折部位。以2名影像科副主任医师判定结果为金标准,比较人工智能与住院医师诊断肋骨骨折及肋骨骨折数、骨折类型、骨折部位的灵敏度。结果人工智能阅片时间[(45.62±27.54)s]均短于住院医师1[(110.57±31.45)s]、住院医师2[(127.36±21.65)s](t=48.685,P<0.001;t=52.684,P<0.001)。人工智能诊断肋骨骨折的灵敏度(92.6%)均高于住院医师1(78.9%)、住院医师2(77.4%)(χ^(2)=26.880,P<0.001;χ^(2)=31.473,P<0.001)。人工智能诊断肋骨骨折数的灵敏度(96.8%)均高于住院医师1(83.8%)、住院医师2(82.3%)(χ^(2)=89.610,P<0.001;χ^(2)=104.407,P<0.001)。在诊断肋骨骨折类型方面,人工智能诊断轻微骨折、骨皮质扭曲的灵敏度(95.7%、94.2%)均高于住院医师1(82.9%、62.1%)、住院医师2(80.9%、59.6%)(P<0.05),诊断错位骨折的灵敏度(99.2%)与住院医师1(99.5%)、住院医师2(99.0%)比较差异均无统计学意义(P>0.05)。在诊断肋骨骨折部位方面,人工智能诊断前肋、腋肋骨折的灵敏度(96.6%、96.2%)均高于住院医师1(75.7%、85.6%)、住院医师2(73.8%、84.6%)(P<0.05),诊断后肋骨折的灵敏度(94.2%)与住院医师1(93.5%)、住院医师2(91.9%)比较差异均无统计学意义(P>0.05)。结论与人工诊断相比,人工智能可提高诊断肋骨骨折的效率,诊断前肋、腋肋轻微骨折及骨皮质扭曲的灵敏度较高。Objective To compare the efficiencies of artificial intelligence(AI)software versus radiology residents on the diagnosis of rib fractures,and to investigate the value of AI to the diagnosis of rib fractures.Methods From July to December 2021,350 patients with rib fractures were diagnosed and treated in Liaocheng People's Hospital.All patients underwent chest CT scan,and the CT images were analyzed by AI software and by 2 radiology residents,respectively.The time of reading,and the number,type and location of fractures were recorded.Taking the judgment results of two deputy chief physicians as the gold standard,the sensitivities of AI and residents in diagnosing rib fractures,number of rib fractures,fracture type and fracture location were compared.Results The reading time of AI[(45.62±27.54)s]was shorter than that of resident 1[(110.57±31.45)s]and resident 2[(127.36±21.65)s](t=48.685,P<0.001;t=52.684,P<0.001).The sensitivity of AI in diagnosing rib fracture(92.6%)was higher than that of resident 1(78.9%)and resident 2(77.4%)(χ^(2)=26.880,P<0.001;χ^(2)=31.473,P<0.001).The sensitivity of AI in diagnosing the number of rib fractures(96.8%)was higher than that of resident 1(83.8%)and resident 2(82.3%)(χ^(2)=89.610,P<0.001;χ^(2)=104.407,P<0.001).The sensitivities of AI in diagnosing minor fractures and osteocortical torsion(95.7%,94.2%)were higher than those of resident 1(82.9%,62.1%)and resident 2(80.9%,59.6%)(P<0.05),and the sensitivity of AI in diagnosing malunion fractures(99.2%)was not significantly different from that of resident 1(99.5%)and resident 2(99.0%)(P>0.05).The sensitivities of AI in diagnosing anterior and axillary rib fractures(96.6%,96.2%)were higher than those of resident 1(75.7%,85.6%)and resident 2(73.8%,84.6%)(P<0.05),and the sensitivity of AI in diagnosing posterior rib fractures(94.2%)was not significantly different from that of resident 1(93.5%)and resident 2(91.9%)(P>0.05).Conclusion Compared with m anual diagnosis,AI can improve the diagnosing efficiency on rib fractures,and the sens

关 键 词：肋骨骨折 人工智能 灵敏度 轻微骨折 骨皮质扭曲 

分 类 号：TP18[自动化与计算机技术—控制理论与控制工程] R683[自动化与计算机技术—控制科学与工程]