Recognition And Classification Of Scoliosis Based On Deep Learning Technology And The Studies On Selection Of Lowest Instrumented Vertebra In Adolescent Idiopathic Scoliosis

Chapter 2 Recognition and classification system of patients with scoliosis based on deep learning techniques——Bilateral CNN for image-features extractionObjective We developed a system that can automatically classify common types of scoliosis and subtypes of scoliosis secondary to neurofibromatosis type 1(NF1-S)using deep learning algorithms(DLAs)and improve the accuracy and effectiveness of classification.Methods A total of 100 adolescent idiopathic scoliosis(AIS),100 congenital scoliosis(CS),211 NF1-S patients,and 114 normal controls were included in this study for primary classification experiments using whole spine AP and lateral X-ray images.Among them,comprehensive experiments in classification of NF1 subtypes(131 dystrophic and 80 nondystrophic NF1-S)were performed.For identification of scoliosis,especially for image features of NF1-S with dystrophic curves,we devised a novel network(i.e.,Bilateral CNN)utilizing a bilinear-like operation to discover the similar interest features between whole spine AP and lateral X-ray images.The performance of Bilateral CNN was compared with spine surgeons,conventional DLAs(i.e.,VGG-16,Res Net-50,BCNN,Two-path BCNN),recently proposed DLAs(i.e.,Shuffle Net,Mobile Net,Efficient Net).Results In primary classification,our proposed Bilateral CNN with an accuracy of 87.92% also outperformed the other seven DLAs with varied accuracies between 52.58% to 83.35% with five-fold cross-validation.In classification of NF1-S subtypes,our proposed method with 80.36% accuracy also outperformed the other seven DLAs ranging from 61.90% to 76.19% with five-fold cross-validation.It also outperformed the spine surgeons(with an average accuracy of 77.5% for the senior surgeons and 65.0% for the junior surgeons).Our method is highly generalizable due to the innovative algorithm and data augmentation.Furthermore,the heatmaps extracted by Bilateral CNN showed curve pattern and morphology of ribs and vertebrae contributing most to the classification results.Conclusion The proposed Bilateral CNN can automatically capture representative features for classifying types of scoliosis and subtypes of NF1-S utilizing AP and lateral X-ray images,leading to a relatively good performance.Therefore,the proposed method can assist surgeons with the auxiliary diagnosis.Chapter 3(Section 1)The rotation of preoperative-presumed lowest instrumented vertebra: Is it a risk factor for distal adding-on in Lenke 1A/2A curve treated with selective thoracic fusion?Objective To investigate whether the rotation of preoperative-presumed lowest instrumented vertebra(LIV)is a risk factor for adding-on(AO)in adolescent idiopathic scoliosis(AIS)treated with selective posterior thoracic fusion(s PTF).Methods A total of 196 AIS patients of Lenke type 1A or 2A with minimum 2-year follow-up after s PTF with all pedicle screw instrumentation were included.Radiographical parameters were measured as follows: preoperative rotation angle of presumed LIV and LIV + 1,LIV + 1/LIV rotation difference,postoperative rotation angle of LIV and LIV derotation angle on CT scans.Patients were classified into AO group and non-AO group during the follow-up.The parameters were compared between the two groups to investigate risk factors for AO.Results Among 196 patients,40(20.4%)patients developed with AO at the final follow-up.Compared with non-AO group,patients with AO had significantly larger preoperative rotation angle of presumed LIV(8.8° ± 3.4° vs.3.4° ± 2.9°,P < 0.001)and LIV + 1(5.9° ± 4.0° vs.3.6° ± 2.9°,P = 0.004),LIV + 1/LIV rotation difference(-2.6° ± 3.7° vs.0.6° ± 3.2°,P < 0.001)and postoperative rotation angle of LIV(7.2° ± 4.3° vs.3.0° ± 2.9°,P < 0.001).The last substantially touched vertebrae(LSTV)was selected as LIV in 148 patients,among which the above-described parameters were found to be remarkably larger in AO group than non-AO group as well.Multivariate analysis presented Risser grade and preoperative rotation angle of presumed LIV as independent predictors of AO.Conclusion AIS patients with low Risser grade and large preoperative rotation angle of presumed LIV are more likely to develop with AO after s PTF.Moreover,for the patients with LSTV selected as LIV,preoperative rotation of presumed LIV might be still a risk factor associated with the occurrence of AO.Chapter 3(Section 2)Selecting the Last Substantially Touching Vertebra as Lowest Instrumented Vertebra in Lenke type 2A-R and 2A-L CurvesObjective The aim of this study was to determine whether the last substantially touching vertebra(LSTV)can be selected as the optimal lowest instrumented vertebra(LIV)for Lenke 2A adolescent idiopathic scoliosis(AIS)with different lumbar modifiers(2A-R and 2A-L)and to investigate its relationship with the distal adding-on(AO).Methods A total of 101 female patients were included with a minimum of 2-year follow-up after selective posterior surgery.Patients were classified based on the direction of L4 tilt: 2A-L and 2A-R.Patients with LSTV-1,LSTV,or LSTV+1 selected as LIV were assigned to three groups.Factors associated with AO were analyzed through comparison among the three groups.Results The level of LSTV was more distal in the 2A-R group than that in the 2A-L group(P=0.011).Distal AO was observed in 24 patients(23.8%).In the 2A-R curves,26.1% patients were found to have AO.The incidence of AO was significantly higher in LSTV-1 than LSTV or LSTV+1 group.Logistic regression analysis showed the distance between LIV and LSTV(LIV-LSTV <0)was the independent factor associated with AO(OR=8.7,95% CI=3.1-45.5,P=0.011).In the 2A-L curves,21.8% patients were found to have AO.The incidence of AO was significantly lower in LSTV+1 than LSTV-1 or LSTV group.Similarly,logistic regression showed the distance between LIV and LSTV(LIV-LSTV?0)had significant association with AO(OR=11.9,95% CI=2.5-53.2,P=0.009).Conclusion The distance between LIV and LSTV was a significant factor associated with AO for both 2A-R and 2A-L patients.The rule of selecting LIV should be different between 2A-R and 2A-L curves.We recommend to extend the fusion level to LSTV in 2A-R curve and to LSTV+1 in 2A-L curve to avoid distal AO.Chapter 3(Section 3)Posterior spinal fusion for Lenke modifier C adolescent idiopathic scoliosis: can lowest instrumented vertebra stop at the apical vertebra of the lumbar curve?Objective This study aimed to investigate the radiographic risk factors related to the occurrence of distal adding-on(AO)in posteriorly treated Lenke modifier C adolescent idiopathic scoliosis(AIS)patients with the apical vertebra of the lumbar curve(L-AV)selected as the lowest instrumented vertebra(LIV).Methods A total of 73 consecutive Lenke modifier C AIS patients were analyzed with a minimum of 2-year follow-up after posterior spinal fusion surgery with L-AV selected as LIV.Patients were grouped according to the occurrence of AO.Radiographical parameters were measured as follows: Cobb angle,curve flexibility and AV translation of the thoracic curve and lumbar curve,L-AV rotation and tilt,coronal balance,Harrington stable zone on anteroposterior(AP)film and convex side-bending(SB)film,L-AV derotation and L-AV/AV+1 disc opening or closing on convex SB film,etc.The SRS-22 score was used to evaluate clinical outcomes.Radiographic and clinical parameters were compared between the two groups.Results There were 23 patients in AO group and 50 patients in non-AO group.Preoperatively,the AO group had proximal L-AV,lower flexibility of the thoracic curve,coronal imbalance shifted to the convex side of the lumbar curve,lower Harrington stable zone on AP film and convex SB film,and less L-AV/AV+1 disc opening on convex SB film compared to non-AO group.The logistic regression revealed that the flexibility of the thoracic curve,coronal balance,Harrington stable zone on convex SB film,and L-AV/AV+1 disc opening or closing on convex SB film were significant predictors of AO.Specifically,the flexibility of the thoracic curve >40.0%,coronal balance< 19.6mm,and Harrington stable zone on convex SB film >77.8% might be optimal thresholds for selecting L-AV as LIV.At the final follow-up,AO group had larger lumbar curves and a lower correction rate.No difference was found in the SRS-22 between the two groups.Conclusion For Lenke modifier C AIS patients,LIV might be considered to stop at LAV if there were large flexibility of the thoracic curves,good coronal balance,large Harrington stable zone on convex SB film and L-AV/AV+1 disc opening on convex SB film,to reduce the occurrence of AO.