Analysis Of Influencing Factors And Artificial Intelligence-assisted Research For The Radiographic Diagnosis Of DDH In Children

BackgroundDevelopmental dysplasia of the hip(DDH)is a common skeletal deformity in children and the prominent cause of hip osteoarthritis and lower limb disability.The therapeutic difficulty and effect of DDH are closely related to an early and proper diagnosis.After the appearance of ossification nuclei in young children over 4-6 months,anteroposterior(AP)pelvic radiography is the preferred screening examination for diagnosing DDH.The radiographic diagnosis of pediatric DDH,based on measurements of hip parameters,is significantly influenced by the quality of pelvic radiographs.In addition,the radiographic diagnosis of pediatric DDH is not merely measuring hip parameters and determining the degree of hip dislocation but also requires hip parameter analysis based on clinical thought to achieve an expert-like diagnosis.PartⅠQuantification of femoral position on radiographic acetabular coverage in childrenObjectivesBased on the current status of diagnosing pediatric DDH,the objectives of this study were(1)to evaluate the effect of femoral malposition in different planes on radiographic acetabular coverage,(2)to identify the tolerance of the radiographic acetabular coverage to femoral malpositioning,and(3)to predict the actual acetabular coverage from non-standard radiographs.MethodsTwenty-one children(34 hips)with normative acetabular coverage were screened in this retrospective study.The Mimics-based local-rotation fluoroscopy simulation method(MLFS)was used to digitally reconstruct non-standard pelvic radiographs with femoral malposition by tilting,inclining,and rotating the femur in 4°increments within the range of femoral motion.The acetabular coverage indices,i.e.,acetabular-head index(AHI)and center-edge angle(CEA),were measured for each pelvic radiograph by three independent observers.Statistical analysis was conducted to determine the variance of the AHI and LCEA with the femoral malposition.Moreover,a linear regression analysis was conducted to determine the relationship between the radiographic acetabular coverage and femoral positioning.ResultsThere were no significant differences in the AHI with the femoral rotation(range:0°-16°)compared with the standard femoral position.The AHI increased with the femoral abduction but decreased with other motions.We found a linear relationship between the AHI and femoral positioning.The linear regression analysis demonstrated approximately 0.65%(R~2=0.861,P<0.001)of increase in the AHI with every degree of femoral inclination and0.14%(R~2=0.373,P<0.001),0.21%(R~2=0.622,P<0.001),0.14%(R~2=0.586,P<0.001),or 0.10%(R~2=0.227,P<0.001)of decrease in the AHI with every degree of internal rotation,external rotation,flexion,or extension,respectively.Compared to the standard femoral position,the CEA with the femoral tilt(range:-20°-4°),inclination(range:0°-4°),and rotation(range:-8°-40°)did not change significantly(P>0.05).There was a steady increase in the CEA with the femoral abduction,whereas it decreased with other motions.We found a linear correlation between the CEA and femoral positioning.The linear regression equations showed the CEA increased by about 0.20°(R~2=0.733,P<0.001)for each 1°increase in femoral inclination and decreased by about 0.01°(R~2=0.003,P=0.300),0.07°(R~2=0.191,P<0.001),0.06°(R~2=0.167,P<0.001),or0.07°(R~2=0.164,P<0.001)for each 1°increase in internal rotation,external rotation,flexion,or extension,respectively.ConclusionsThis study confirms that femoral malposition in different planes,especially femoral inclination,affects radiographic acetabular coverage in children.Therefore,each pelvic radiograph should assess potential femoral malposition before diagnosing hip disorders.The linear regression equations between femoral position and radiographic acetabular coverage will assist surgeons in predicting the acetabular coverage on non-standard pelvic radiographs.PartⅡConstruction and application of an artificial intelligence-based model for quality control of pediatric pelvic radiographsObjectivesBased on the neural network framework and developmental characteristics of obturator morphology in children,our aims were(1)to propose a universal algorithm to measure obturator diameters,and(2)to propose an efficient method for assessing the quality of pediatric pelvic radiographs and validate its clinical feasibility by comparing its measurements of evaluation parameters and final judgments with clinical experts using a randomized testing set.MethodsWe retrospectively collected 3,247 pediatric AP pelvic radiographs(F:M=1,989:1,258;0.5-11.7 years)from January 2014 to June 2022.They were randomly divided into the training set(2,560 images),verifying set(325 images),and testing set(362 images),using the random number tables in the ratio of 8:1:1.The AI model was constructed from the training set using the convolutional neural network and optimized by the verifying set.Finally,the final judgments of the AI model were tested by the randomized testing set using the receiver operator characteristic(ROC)curves.Meanwhile,the objective evaluation parameters,i.e.,pelvic tilt index(PTI)and pelvic rotation index(PRI),were further assessed using Bland-Altman plots and t-tests.ResultsIn the testing set of 362 pelvic radiographs,the diagnostic accuracy,sensitivity,specificity,and area under the ROC curve(AUC)of the AI model were 99.4%(360/362),98.6%(138/140),100.0%(222/140),and 0.993,respectively.Compared with clinicians,the95%limits of agreement(Bland-Altman analysis)for pelvic tilt index(PTI)and pelvic rotation index(PRI),as determined by the AI model,were-0.052-0.072(bias 0.010,P=0.633)and-0.088-0.055(bias-0.016,P=0.333),respectively.ConclusionsThe proposed method is the first attempt to apply AI algorithms in the quality assessment of pelvic radiographs with great consistency and efficiency.This method can be used for real-time validation of the quality of pelvic radiographs in current Picture archiving and communications systems(PACS),which will remarkably improve the accuracy of interpreting pelvic radiographs and diagnosing DDH.PartⅢClinical thought-based software for diagnosing pediatric DDH on anteroposterior pelvic radiographsObjectivesGiven the practical problems in diagnosing pediatric DDH,this study aimed to(1)investigate the effectiveness of several hip parameters and radiographic classification methods and summarize a reasonable radiographic diagnostic strategy for pediatric DDH,and(2)develop an auxiliary diagnostic tool based on the diagnostic strategy and validate its clinical feasibility.MethodsIn this study,we retrospectively collected 543 standard pelvic radiographs(Female:Male=408:135,mean age 2.0 years)from January 2017 to December 2021.Based on the radiographic diagnostic strategy for pediatric DDH,our auxiliary diagnostic software for DDH was developed with a three-step design concept of manual labeling of key landmarks,computer algorithm correction,and matching pediatric DDH imaging diagnostic strategy with the patient’s general information.Two independent clinicians,including a junior and an intermediate orthopedist,measured hip parameters,the overall diagnosis,and specific classification to compare the diagnoses made by the software and conventional manual method.Taking the manual measurements by the expert panel as the standard,the software’s accuracy was evaluated using ROC curves and confusion matrix,Bland-Altman plots,and t-tests.Results1.Diagnostic performanceConventional manual method:The AUC,accuracy,sensitivity,and specificity for the junior orthopedist were 0.894,90.79%(986/1,086),85.12%(309/363),and 93.64%(677/723),respectively,while they were 0.957,95.67%(1,039/1,086),95.87%(348/363)and 95.57%(691/723)for the intermediate orthopedist.DDH diagnostic software:The AUC,accuracy,sensitivity,and specificity for the junior orthopedist were 0.988,99.08%(1,076/1,086),98.07%(356/363)and 99.59%(720/723),respectively,while they were 0.994,99.72%(1,083/1,086),100.00%(363/363),and 99.59%(720/723)for the intermediate orthopedist.De Long’s test showed that the software was more effective than the conventional method in diagnosing DDH(P<0.001).2.Hip parametersConventional manual method:Compared with the expert panel,the Bland-Altman 95%Lo A for acetabular index(Ax)were-3.26°-6.15°(bias 1.45°,P=0.522;junior orthopedist)and-3.29°-3.74°(bias 0.23°,P=0.509;intermediate orthopedist).As for CEA,the 95%Lo A were-7.12°-11.60°(bias 2.24°,P=0.079;junior orthopedist)and-7.65°-6.64°(bias-0.50°,P=0.690;intermediate orthopedist).DDH diagnostic software:The Bland-Altman 95%Lo A for Ax were-2.09°-2.91°(bias0.41°,P=0.216;junior orthopedist)and-1.98°-2.72°(bias 0.37°,P=0.264;intermediate orthopedist).As for CEA,the 95%Lo A were-3.68°-5.28°(bias-0.80°,P=0.522;junior orthopedist)and-2.94°-4.59°(bias-0.83°,P=0.509;intermediate orthopedist).ConclusionsOur clinical thought-based software can provide expert-like analysis of pelvic radiographs and obtain the radiographic diagnosis of pediatric DDH with great consistency and efficiency.Meanwhile,the initial success of the auxiliary tool in this research based on the radiographic diagnostic strategy for pediatric DDH lays the groundwork for developing more efficient and intelligent diagnostic tools for DDH.