Preliminary Study On Three-Dimensional Printed Model Application Of Lumbosacral Plexus Based On Rapid Magnetic Resonance Imaging

BackgroundThree-dimensional printed models of pelvis are widely used in guiding surgical plan design,physician-patient communication and clinical teaching due to its personalization and precision.However,the main existing problem is that the data source relies on CT,which cannot meet the requirements of accurate 3D printing of various structures such as bone and lumbosacral plexus.Three-dimensional Magnetic Resonance Neurography(MRN)can display the anatomy and course of nerves clearly,which provides the necessary data source for accurate 3D printing of the multi-level structure of pelvis.How to achieve rapid and accurate segmentation of lumbosacral plexus based on MRN is the primary problem must be solved.For this reason,the research will be:1.To solve the limitation of long MRN scanning time by using Compressed Sensing(CS)technology;2.To solve the dilemma of time-consuming and subj ective artificial segmentation and reconstruction of the lumbosacral plexus by using Deep Learning(DL);3.On the basis of the studies in the first two parts,accurate modeling and 3D printing of the lumbosacral plexus were realized,and the clinical application value of 3D printing model of the lumbosacral plexus nerve was preliminarily discussed.So far,relatively few studies on accurate segmentation modeling and 3D printing model of lumbosacral plexus based on MRN have been reported.Purposes1.To evaluate the feasibility and performance of combine MRN imaging with CS for LSP imaging.2.A fully auto-segmentation deep learning model of lumbosacral plexus based on rapid MRN image(U-Net model)was constructed.3.To briefly discuss the application value of lumbosacral plexus 3D printed model based on rapid accurate 3D modeling.Methods1.A total of 40 healthy volunteers were recruited to undergo conventional 3D Fast Field Echo(FFE)based on principle pf selective excitation technique(ProSet)and CSFFE sequence(include:FFE-CS2 and FFE-CS3).Subjective image quality was evaluated using a four-grade scoring system by medical radiologists.Signal to noise ratio(SNR),nerve-to-muscle signal intensity ratio(SIR),and morphological measurement were analyzed for objective evaluation.Above-mentioned parameters were compared among FFE,FFE-CS2 and FFE-CS3.2.The raw FFE images of 49 patients were retrospectively collected to train the U-Net automatic segmentation model.The positive predictive value(PPV),sensitivity(SEN),dice similarity coefficient(DSC),Jaccard similarity coefficient(JACC)and 95%Hausdorff distance(95%HD)of automatic segmentation were calculated using the manually segment images as the gold standard.At the same time,the segmentation images were subjective evaluated to verify the accuracy of automatic segmentation.3.The MRN images of LSP were semi-automatically segmented,and 3D printed models were used for clinical diagnosis and medical education.To observe the difference between the experimental group(3D printing combined with PowerPoint and PACS)and the control group(PowerPoint and PACS teaching)in the theoretical exam scores,and to observe the attitude of the experimental group to the model by questionnaire during medical teaching.Results1.Compared with conventional FFE,the scanning time of FFE-CS2 and FFE-CS3 sequence was reduced by 93 and 94 seconds,respectively.The conventional FFE sequence showed the highest bilateral L5 and S1 DRGs SNR than both two FFE-CS sequences with different accelerating factors(all,P<0.05).SIR and morphological measurements showed no significant difference among these three sequences.and there was no significant difference among the three groups in subjective images quality score(P=0.257).2.All data were randomly divided into training set and test set at 39:10.The PPV,SEN,JACC,DSC and 95%HD of the U-Net model for LSP auto-segmentation was 0.864±0.017,0.848±0.041,0.855±0.017,0.747±0.027,44.884±28.067,respectively.3.Rapid MR imaging combined with automatic segmentation technology can achieve rapid and accurate 3D modeling and 3D printing of lumbosacral plexus.The lumbosacral plexus model can clearly display and specify the anatomical features of the nerve,combined with CT-based 3D printing model of lumbosacral vertebrae,is helpful for clinical diagnosis,treatment and clinical teaching,especially in clinical teaching.The score difference of the experimental group(5.1 ± 2.4)was higher than that of the control group(2.9±2.2),the difference was statistically significant(P<0.05).The questionnaire survey showed that the students in the experimental group thought the 3D printed model was realistic and could clearly display the anatomical and spatial information,which greatly stimulated their interest in medical imaging education.Conclusions1.Compressed sensing is feasible in LSP MR imaging.Compared with conventional FFE imaging,FFE-CS sequences reduce the SNR to a certain extent,but provides similar nerve-to-muscle SIR and morphological measurement results,and can shorten the scanning time.2.A fully automatic segmentation U-Net model can accurately realize the automatic delineation of LSP,which could shorten the segmentation time obviously in clinical practice and improve the working efficiency in 3D segmentation of LSP.3.Rapid MR imaging combined with automatic segmentation technology can achieve rapid and accurate 3D modeling and 3D printing of lumbosacral plexus.LSP model combined with CT-based 3D printing model of lumbosacral vertebrae can help guide clinical diagnosis and treatment and improve clinical teaching effect.