Automated Fiber Quantification Analysis For Brain Diffusion Tensor Imaging In Patients With Systemic Lupus Erythematosus

Part Ⅰ:Application of Automated Fiber Quantification Based on Diffusion Tensor Imaging in the Microstructure of White Matter in SLE PatientsObjective:Automated fiber quantification(AFQ)analysis based on diffusion tensor imaging(DTI)was performed to investigate the white matter fiber tracts of the brain in patients with systemic lupus erythematosus(SLE).The relationship between damage to the white matter fiber tracts and cognitive dysfunction in SLE patients was investigated.Methods:Clinical and imaging data of 39 female SLE patients(SLE group)and 44 healthy volunteers matched their age,gender and education level(HC group)were prospectively collected.Automated Fiber Quantification(AFQ)analysis using diffusion tensor imaging was used to track each participant’s whole brain white matter fiber bundles,and each white matter fiber bundle was divided into 100 equal parts,and each equal fractional anisotropy(FA),mean diffusivity(MD),axial diffusivity(AD),radial diffusion(RD)values were quantitatively analyzed,the differences in the diffusion tensor coefficients of each fiber bundle were compared between the two groups,and the DTI index values of the damaged white matter bundles in the SLE patient group were correlated with the neuropsychological scale.Results:Compared with the HC group,the FA values of the left inferior frontooccipital fasciculus(96-100 segments)and the left inferior longitudinal fasciculus(69-79 segments)were reduced in the SLE group.The MD values of the left thalamic radiation tract(segments 1-11),the left corticospinal tract(segments 2124,87-100),the right corticospinal tract(segments 20-25,70-90,94-100),the right cingulum cingulate(segments 45-54),the callosum forceps minor(segments 14-19,23-31,68-91),the left inferior fronto-occipital fasciculus(segments 8-12,16-21,27-31,92-97),the right inferior fronto-occipital fasciculus(segments 1019),the right inferior longitudinal fasciculus(segments 28-34,43-46),the left arcuate fasciculus(segments 38-42)increased.The AD values of the left corticospinal tract(segments 66-100),the right corticospinal tract(segments 6890,93-100),the right inferior longitudinal fasciculus(segments 41-46),the left superior longitudinal fasciculus(segments 54-83)increased.The RD values of the left thalamic radiation(segments 1-10),the left corticospinal tract(segments 2025),the callosum forceps minor(segments 14-31,64-77),the left inferior frontooccipital fasciculus(94-100 segments)and the left inferior longitudinal fasciculus(70-79 segments)increased(both FDR corrected).In addition,correlation analysis found that the FA values ofthe left inferior fronto-occipital fasciculus segment 97-100(97:r=0.356,p=0.026;98:r=0.345,p=0.031;99:r=0.336,p=0.037;100:r=0.340,p=0.034)were positively correlated with MMSE scale scores.Conclusion:The use of AFQ analysis can more finely and accurately identify the damaged segments of white matter fiber tracts in the brain of SLE patients.In addition,changes in DTI-related indicators correlate with cognitive dysfunction and can be used as an imaging indicator for early identification of cognitive dysfunction in SLE.Part 2 Application of Machine Learning Model Based on Automatic Fiber Quantitative Analysis Based on Diffusion Tensor Imaging to the Diagnostic Value of SLE DiseaseObjective:The results of automated fiber quantification(AFQ)analysis based on magnetic resonance diffusion tensor imaging(DTI)were used to explore the application value of changes in the microstructure of white matter in patients with systemic lupus erythematosus(SLE)in the diagnosis and prediction of machine learning models.Methods:DTI images were collected from 39 female SLE patients(SLE group)and 44 healthy volunteers matched for their age,sex and education(HC group).The whole white matter fiber tracts of each subject were traced by automatic fiber quantification(AFQ)analysis,and each fiber tract was divided into 100 aliquots,and FA,MD,RD,AD values were extracted as the white matter fiber tract characteristics of the two groups of subjects,using the maximum correlation and minimum redundancy(mMRM)method.The maximum relevancy and minimum redundancy(mMRM)and least absolute shrinkage and selection operator(LASSO)logistic regression algorithms were used to filter the machine learning features with intergroup differences and construct FA,MD,AD,and RD.And two joint machine learning models,FA+MD and FA+MD+AD+RD,and the predictive classification of the SLE and HC groups by the support vector machine(SVM)five-fold cross-validation method,using the receiver The operating characyeristic curve(ROC)was used to evaluate the diagnostic efficacy.Results:A total of 116 features(18 FA features;14 MD features;15 AD features;15 RD features;26 FA+MD features;28 FA+MD+AD+RD features).Among them,FA(training group:AUC=0.96;test group:AUC=0.91),MD(training group:AUC=0.96;test group:AUC=0.93),AD(training group:AUC=0.97;test group:AUC=0.86),RD(training group:AUC=0.94;test group:AUC=0.85),FA+MD(training group:AUC=0.98;test group:AUC=0.94),FA+MD+AD+RD(training group:AUC=0.99;test group:AUC=0.96),the classification characteristics of all fiber bundle nodes had good diagnostic performance in the classification of SLE group and HC group,Among them,the MD model in the independent model had the best diagnostic efficiency,with an AUC value of 0.96,a sensitivity of 94.88%,a specificity of 92.60%,and an accuracy of 93.68%in the training group.The AUC value of the test group was 0.93,the sensitivity was 84.29%,the specificity was 91.11%,and the accuracy was 88.01%.The FA+MD+AD+RD model had the best diagnostic efficiency in the joint model,with an AUC value of 0.93,sensitivity of 99.35%,specificity of 97.71%,and accuracy of 98.49%in the test group.The AUC value of the training group was 0.96,the sensitivity was 87.14%,the specificity was 88.61%,and the accuracy was 87.87%.Conclusion:The machine learning model constructed based on the automatic fiber quantitative analysis results of diffusion tensor imaging has high application value for the classification prediction of SLE group and HC group.Among which the MD independent model and the FA+MD+AD+RD joint model have the best diagnostic efficiency and higher value of prediction classification.