| Objective:To study the differences in clinical outcomes and cervical sagittal parameters after two-segment Mobi-C artificial cervical disc replacement and prestige-LP artificial cervical disc combined with ROI-C fusion;To construct an automatic measurement model of the Cobb angle of the cervical spine based on deep learning.Methods:1.Part Ⅰ:Clinical studyThis study is based on the project of the National Natural Science Foundation of China,"Basic biological research on the prevention and treatment of degenerative osteoarthrosis based on the theory of tendon injury" and the Chinese Postdoctoral Science Foundation,"Re-evaluation study of Qi-musk pill for the treatment of cervical spondylosis of neurogenic type.The clinical trial protocol for the re-evaluation study of Astragalus Musk Pill for cervical spondylosis of nerve root type.The disc group(33 cases)and the hybrid group(32 cases)were studied on the basis of the subject,and the observation period was 1 year or more.The disc group was treated with a two-segment Mobi-C artificial cervical disc replacement program,and the hybrid group was treated with a Prestige-LP artificial cervical disc combined with an ROI-C fusion device.The characteristics and trends of the efficacy indexes of the two groups at each treatment time point(1 week,1 month,3 months,6 months and 1 year after surgery)were observed and compared.The above data were statistically analyzed using SPSS 25.0 software.2.Part Ⅱ:Deep learningA total of 7000 cases of lateral cervical spine X-ray films from January 2021 to March 2023 in our hospital were selected as the study subjects.6300 X-ray films were randomly selected as the training set and validation set,and the remaining 700 X-ray films were used as the test set.Four key points of the C2-7Cobb angle measurement parameters of the cervical spine at the central margin were identified,and orthopedic surgeon 1 annotated the key points of the cervical vertebrae in the training and validation sets,which were used to train a deep learning model for key point detection.Orthopedic surgeons 1 and 2 annotated the test set separately,and the average value of each parameter obtained by the two surgeons was used as the reference standard.Intra-group correlation coefficient(ICC),Pearson correlation coefficient(r),root mean square error(RMSE),mean absolute error(MAE),and scatter plot were used to compare the deep learning moodels with the reference standard.Results:1.Part Ⅰ:Clinical studies(1)VAS,JOA,DNI,and ADL scoresIntra-group comparison showed that(1)VAS and NDI scores at 1 week,3 months,6 months and 1 year postoperatively were significantly lower in the disc group and the hybrid group compared with those before surgery,with statistically significant differences(P<0.01);and VAS and NDI scores at 3 months postoperatively were significantly lower than those at 1 week postoperatively in both groups(P<0.01),(2)The JOA scores at 1 week,3 months,6 months and 1 year postoperatively were significantly higher in the disc group and the hybrid group than before surgery(P<0.01),and the JOA scores at 3 months postoperatively were significantly higher in both groups than at 1 week postoperatively(P<0.01);the JOA scores at 1 year postoperatively were significantly higher in the disc group than at 3 months postoperatively(P=0.046).(3)Postoperative ADL scores were significantly lower in the disc group than preoperatively(P<0.01).Comparison between groups showed that there was no significant difference between groups in VAS,JOA and DNI scores of patients in both preoperative and postoperative time periods(P>0.05).(2)Changes in sagittal parametersComparison within groups showed that(1)C0-2Cobb was significantly lower than preoperative(P<0.01),FSU was significantly higher than preoperative(P<0.05),both C7S and DH were significantly higher than preoperative(P<0.001),and C7S was significantly lower than preoperative 1 year after surgery(P<0.01)in the disc group;no significant differences were seen in the values of each sagittal parameter at 1 year after surgery compared with 1 week after surgery(P>0.05).(2)C2-7Cobb,C2-6Cobb,T1S,FSU,C7S,and DH were significantly higher in the hybridization group at 1 week postoperatively than preoperatively(P<0.05),and C2-7Cobb,C2-6Cobb,FSU,C7S,and DH were higher at 1 year postoperatively than preoperatively(P<0.05);no significant difference was observed in each parameter at 1 year postoperatively compared with that at 1 week postoperatively(P>0.05).Comparison between groups showed that C2-7Cobb,C2-6Cobb,T1S and FSU at 1 week postoperatively were significantly lower in the intercalated group than in the hybrid group(P>0.05),and T1S,C7S and FSU at 1 year postoperatively were significantly lower in the intercalated group than in the hybrid group(P>0.05).(3)Establishment of prediction modelUnivariate analysis showed that gender,C2-7Cobb at 1 week postoperatively,C2-6Cobb at 1 year postoperatively,NDI score(1 week,3 months,6 months,and 1 year postoperatively),VAS score(1 week,3 months,6 months,and 1 year postoperatively),and JOA score(1 week,3 months,6 months,and 1 year postoperatively)were associated with healing with statistically significant differences(P<0.05).The multifactorial analysis was modeled using the stepwise and all-inclusive methods,with model 1 using the all-inclusive method and model 2 using the stepwise method to screen variables.A total of eight variables were entered into the regression equation in model 1,of which VAS score(OR=14.127,P=0.02)was an independent predictor of healing,with a partial regression coefficient of 2.648.One factor was entered into the regression equation in model 2,of which VAS score(OR=7.666,P=0.000)was an independent predictor of healing,with a partial regression coefficient of 2.037.Multi-factor regression analysis showed that VAS score at 1 week postoperatively was an independent predictor of healing(OR=7.666,p<0.001)The area under the ROC curve(AUC)of Model 1 was 0.955,p<0.0001,and the cutoff value(Cutoff)of this prediction model was 0.62,with a sensitivity of 78.3%and a specificity of 88.9%when the Yordon index was 0.79;the area under the ROC curve(AUC)of Model 2 was 0.871,p<0.0001.When the Jorden index was 0.596,the cutoff value(Cutoff)of this prediction model was 0.39,the sensitivity was 65.2%,and the specificity was 94.4%.The AUC area of the above 2 prediction models was>0.7 and P<0.05,indicating that the prediction models were well differentiated.The results of the Hosmer-Lemeshaw(H-L)goodness-of-fit test showed that the P values of both Model 1 and Model 2 were greater than 0.05,indicating that there was no significant difference between the predicted and observed values,so these 2 prediction models were considered to have good calibration ability.(4)Disease-free survival analysis(1)Using time to healing as the median survival time,the median survival time(time to healing)estimates for patients in the disc and hybrid groups were 12 and 24 months,respectively,and the survival curves of patients after receiving the two procedures were the same,and the difference was not statistically significant.(2)The median survival time(whether recurrence or symptom remission)was estimated to be 12 and 24 months for patients in the disc and hybrid groups,respectively,after the two procedures,and the survival curves of patients were the same,and the difference was not statistically significant.2.Part Ⅱ:Deep Learning1.With the gradual increase in the amount of training data,the model demonstrates satisfactory predictive ability in the same validation sample.2.Comparing the cobb angle and mean values measured by the three orthopedic surgeons with the predicted values,it was concluded that there was no significant difference in the measured values among the three surgeons,and the predicted value 3 was not significantly different compared to physician 1,physician 2,physician 3 and the mean value(p>0.05).Correlation analysis indicated that physician 1 was significantly positively correlated with physician 2(r=0.993,P<0.001);physician 1 was significantly positively correlated with physician 3(r=0.993,P<0.001);and physician 2 was significantly positively correlated with physician 3(r=0.992,P<0.001).There was a high degree of agreement among the three physicians’ measurements3.The predicted results for the 2000 sample were more discrete from the mean of physician measurements,and the predicted results for the 7000 sample were in high agreement with the mean of physician measurements.4.Model 1(2000 sample)had an ICC of 0.587,Spearman correlation coefficient of 0.514,mean difference of 0.23,mean absolute error of 2.80,and root mean square error of 8.12;Model 2(7000 sample)had an ICC of 0.764,Spearman correlation coefficient of 0.595,mean difference of 0.1,mean absolute error of 2.26 and root mean square error of 5.99,model 2 has better predictive ability.Conclusion:Both the two-segment Mobi-C artificial cervical disc replacement and the prestige-LP artificial cervical disc combined with the ROI-C fusion significantly improved clinical symptoms and were able to restore sagittal balance of the cervical spine.The newly developed deep learning algorithm-based model can automatically identify key points on cervical lateral radiographs and generate cervical C2-7Cobb angles with good agreement with manual measurements. |
PDF Full Text Request