基于卷积神经网络的全髋关节置换术后参数测量与人工测量方法的比较研究

A comparative study between parameters measurement method based on convolutional neural network and manual measurement method after total hip arthroplasty

目的探讨通过卷积神经网络测量全髋关节置换(total hip arthroplasty,THA)术后假体角度及双下肢长度差与人工测量的差异。方法收集2024年1至3月,在解放军总医院第四医学中心骨科实施单侧THA患者100例,其中50例使用陶瓷内衬髋关节假体、50例使用聚乙烯内衬髋关节假体。男58例、女42例,年龄(51.6±12.71)岁,均为单侧,左侧45髋、右侧55髋。利用Attention UNet网络分割假体并识别关键点,计算髋臼假体外展角、前倾角和双下肢长度差。分别使用卷积神经网络和人工方法测量假体前倾角、外展角和双下肢长度差,并进行比较。采用Bland-Altman方法评估两种测量方法的一致性。结果卷积神经网络测量100例假体的外展角、前倾角和双下肢长度差分别为40.55°±4.76°、21.5°(19.0°,27.0°)和3.6(1.7,6.8)mm,人工测量为40.58°±4.93°、21.0°(17.6°,23.2°)和3.0(2.0,7.0)mm。两种方法测量前倾角的差异有统计学意义(Z=-2.189,P=0.029),而外展角和双下肢长度差的差异无统计学意义(P>0.05)。卷积神经网络测量50例聚乙烯内衬假体的外展角和前倾角分别为40.31°±4.70°和20.0°(18.0°,22.0°),人工测量分别为40.49°±4.92°和20.0°(17.0°,21.5°),差异均无统计学意义(P>0.05);卷积神经网络测量50例陶瓷内衬假体的外展角和前倾角分别为40.79°±4.86°和26.5°(20.8°,33.0°),人工测量分别为40.68°±4.98°和20.0°(18.0°,22.0°),两种方法测量假体前倾角的差异有统计学意义(Z=-3.426,P=0.001),而外展角的差异无统计学意义(t=0.994,P=0.325)。Bland-Altman方法显示人工与卷积神经网络测量100例假体外展角的差值为0.0310°±1.091°[95%CI(-2.107°,2.169°)],聚乙烯内衬假体的差值为0.178°±1.296°[95%CI(-2.362°,2.718°)],陶瓷内衬假体的差值为-0.116°±0.825°[95%CI(-1.733°,1.501°)];100例假体前倾角的差值为-3.280°±6.920°[95%CI(-16.840°,10.280°)],聚乙烯内衬假体的差值为-0.002°±1.471°[95%CI(-2.885°,2.881°)],陶瓷内衬假体的差值为-6.560°±8.523°[95%CI(-23.270°,10.150°)];双下肢长度差的差值为(-0.101±0.947)mm[95%CI(-1.956,1.755)mm]。结论卷积神经网络测量THA术后髋臼假体外展角、聚乙烯内衬假体前倾角、双下肢长度差与人工测量值无差异,但测量陶瓷内衬假体前倾角与人工测量方法有差异。Bland-Altman分析表明两种测量方法在测量外展角、聚乙烯内衬假体前倾角和双下肢长度差时具有较好的一致性。

ObjectiveTo investigate the differences between convolutional neural network(CNN)measurements and manual measurements of prosthetic angles and leg length discrepancies after total hip arthroplasty(THA).MethodsData were collected from 100 patients who underwent unilateral THA at the Department of Orthopaedics,Fourth Medical Center of PLA General Hospital,between January and March 2024.Fifty patients received a ceramic-lined hip prosthesis,and 50 received a polyethylene-lined hip prosthesis.The cohort included 58 males and 42 females,with an average age of 51.6±12.71 years.The left hip was affected in 45 cases,and the right hip in 55 cases.The Attention UNet network was used to segment the prostheses and identify key points,from which the acetabular prosthesis abduction angle,anteversion angle,and leg length discrepancy were calculated.The CNN measurements of anteversion angle,abduction angle,and leg length discrepancy were compared with manual measurements.The consistency between the two methods was evaluated using the Bland-Altman method.ResultsCNN measurements of the abduction angle,anteversion angle,and leg length discrepancy for the 100 prostheses were 40.55°±4.76°,21.5°(19.0°,27.0°),and 3.6(1.7,6.8)mm,respectively,while manual measurements were 40.58°±4.93°,21.0°(17.6°,23.2°),and 3.0(2.0,7.0)mm,respectively.There was a statistically significant difference in anteversion angle between the two methods(Z=-2.189,P=0.029),but no significant differences in abduction angle or leg length discrepancy(P>0.05).For the 50 cases with polyethylene-lined prostheses,CNN-measured abduction and anteversion angles were 40.31°±4.70°and 20.0°(18.0°,22.0°),respectively,while manual measurements were 40.49°±4.92°and 20.0°(17.0°,21.5°);there were no statistically significant differences(P>0.05).For the 50 ceramic-lined prostheses,CNN-measured abduction and anteversion angles were 40.79°±4.86°and 26.5°(20.8°,33.0°),respectively,while manual measurements were 40.68°±4.98°and 20.0°(18.0°,22.0°),with a significant difference in anteversion angle(Z=-3.426,P=0.001)but not in abduction angle(t=0.994,P=0.325).Bland-Altman analysis showed a difference of 0.031°±1.091°for the abduction angle of the 100 prostheses between the two methods,0.178°±1.296°for the polyethylene-lined prostheses,and-0.116°±0.825°for the ceramic-lined prostheses.For the anteversion angle,the difference was-3.280°±6.920°for the 100 prostheses,-0.002°±1.471°for the polyethylene-lined prostheses,and-6.560°±8.523°for the ceramic-lined prostheses.The difference in leg length discrepancy was-0.101±0.947 mm.ConclusionCNN measurements of acetabular prosthesis abduction angle,polyethylene-lined prosthesis anteversion angle,and leg length discrepancy after THA showed no significant differences from manual measurements.However,CNN and manual measurements of ceramic-lined prosthesis anteversion angle differed.Bland-Altman analysis indicates that the two methods have good consistency when measuring the abduction angle,the lower limb discrepancy,and the anteversion angle of the polyethylene liner prosthesis.