The Quantitative Study Of Lumbar Disc Herniation Based On The Machine Learning Of The 3D Reconstruction Of The Magnetic Resonance Image

Purpose:Based on magnetic resonance imaging(MRI)3D reconstruction,a series of 2D/3D indicators were established for describing the relationship between each indicator and lumbar disc herniation(LDH)staging,and exploring quantitative staging methods for LDH.Methods:Magnetic resonance images of 43 patients with LDH collected at the 971 st Naval Hospital between February 2022 and May 2022 were selected to establish a series of 2D/3D metrics to describe the changes in image and spatial structure caused by LDH,including Left Intervertebral Disc And Flavum Ligamentum Space(LIDFS),Reft Intervertebral Disc And Flavum Ligamentum Space(RIDFS),Maximum Anteroposterior Diameter Of The Dural Sac(MADDS),Sacral Table Angle(STA),Sacral Slope(SS),Lumbar Lordosis(LL),Lumbar Sacral Angle(LSA),Kyphotic Angle Of The Herniated Disc(KA),Index Of Disc Herniation(IDH),Nucleus Protrusion Rate(NPR),Ratio Between The Protruded Part And The Dural Sac(RPPDS),[Distribution Of The Protruded Disc(DPD),including x/L,y/W,z/H,∠A1/∠A2],Relative Signal Intensity(RSI).The MRI images of 43 patients were segmented,reconstructed and the values of each index were measured using the intervertebral disc visualization and quantitative analysis system we built to analyze the relationship between each index and the grade of LDH patients and to explore the quantitative staging method of LDH.Results:For quantitative typing of LDH disease classes,all 14 indicators were statistically significant,except for two indicators,SS and LSA,which were not statistically significant.KA,IDH,NPR,∠A1/∠A2,RPPDS,RSI,and x/L indicators were strongly correlated with LDH disease class;LIDFS,RIDFS,STA,LL,and y/W indicators were moderately correlated with LDH disease class;MADDS and z/H indicators were weakly correlated with LDH disease class.The regression analysis of the decision regression tree for the severity score of LDH patients showed that the three indicators of MADDS,LIDFS,and NPR contributed the most;the classification analysis of the decision classification tree for the grade of LDH patients(bulging or protruding)showed that the KA indicator had the best classification effect,and the bulging or protruding could be classified by KA;the 14 indicators with statistically significant K-means mean clustering,and the corresponding confusion matrix judgment was obtained according to the training results,and the classification accuracy was 91% for the bulging group and 100% for the protruding group.Conclusion:This project established an intervertebral disc visualization and quantitative analysis system based on 3D reconstruction of MRI images,summarized and designed 16 quantitative analysis indexes,and initially verified the validity of the system and indexes through measurement and statistical tests,and explored the feasibility of quantitative typing of LDH disease levels.IDH,NPR,∠A1/∠A2,RPPDS,y/W,x/L indicators were positively correlated with LDH patients’ disease level;RSI,LIDFS,RIDFS,STA,LL,KA indicators were negatively correlated with LDH patients’ disease level;in the regression analysis of decision regression tree for LDH patients’ severity score,MADDS,LIDFS,NPR three In the regression analysis of the severity score of LDH patients,MADDS,LIDFS,and NPR were the indicators with the greatest contribution;in the classification analysis of the classification tree of LDH patients,KA indicators were the best,and KA was able to classify the bulky or prominent;according to the clustering of each indicator parameter,the classification accuracy of the bulky group was 91%,and the classification accuracy of the prominent group was 100%.