Prediction of low back fasciitis by machine learning method based on radiomics

Prediction of low back fasciitis by machine learning method based on radiomics | Research Square window.SnipcartSettings = { analytics: { enabled: false } }; (function() { var accessVector = localStorage.getItem('access_vector') || ''; window.dataLayer = window.dataLayer || []; if (accessVector) { window.dataLayer.push({ user: { profile: { profileInfo: { snid: accessVector } } } }); } })(); (function(w,d,s,l,i){w[l]=w[l]||[];w[l].push({'gtm.start':new Date().getTime(),event:'gtm.js'});var f=d.getElementsByTagName(s)[0],j=d.createElement(s),dl=l!='dataLayer'?'&l='+l:'';j.async=true;j.src='https://www.googletagmanager.com/gtm.js?id='+i+dl;f.parentNode.insertBefore(j,f);})(window,document,'script','dataLayer','GTM-K279D39R'); Browse Preprints In Review Journals COVID-19 Preprints AJE Video Bytes Research Tools Research Promotion AJE Professional Editing AJE Rubriq About Preprint Platform In Review Editorial Policies Our Team Advisory Board Help Center Sign In Submit a Preprint Cite Share Download PDF Article Prediction of low back fasciitis by machine learning method based on radiomics mingxin song, ling Jiang, hui Yang, cheng Jia, Jian Qin, Qiang Xiao This is a preprint; it has not been peer reviewed by a journal. https://doi.org/ 10.21203/rs.3.rs-4239014/v1 This work is licensed under a CC BY 4.0 License Status: Posted Version 1 posted You are reading this latest preprint version Abstract Purpose: The aim of this study was to analyze lumbar fascial changes in patients with lower back pain using lumbar spine soft tissue-based MR imaging radiomics. Methods: We retrospectively analyzed the lumbar MRI of 380 patients with low back pain. Patients were randomly assigned to either the training (n = 187) or validation (n = 193) cohorts. Seven different machine learning algorithms were used to establish classification prediction models, and their prediction efficiency were evaluated with the best performance was selected as the final classification prediction model. Multivariate logistic regression analysis was used to create radiomics model and combined nomogram model and their predictive performance were evaluated using receiver operating characteristic (ROC) curves. Results: Among the seven machine learning models, the Lasso model exhibited the highest diagnostic efficiency with an AUC of 0.835. The radiomics nomogram integrated clinical and radiomics signature features, demonstrating strong performance in both the training and validation sets with AUC values of 0.97 and 0.96, respectively. AUC and DCA indicated that the radiomics nomogram model effectively diagnosed lumbar fasciitis. Conclusion: We developed an nomogram model that integrated clinical and radiomics features to help clinicians identify and predict low back fasciitis through soft tissue magnetic resonance. Health sciences/Diseases/Trauma Physical sciences/Physics/Information theory and computation Fascia Radiomics Low Back Pain Nomogram MRI Figures Figure 1 Figure 2 Figure 3 Figure 4 Figure 5 Figure 6 Figure 7 Figure 8 Introduction Low back pain is a very common problem that has a huge impact on individuals, families and businesses around the world. Globally, disability due to low back pain has increased by 54% in 25 years. [1, 2] Low back pain can be caused by any one of a number of anatomical structures. Bones, discs, joints, ligaments, muscles, etc. are mostly considered to be one of the common causes of low back pain. [3] Fascia, however, is often overlooked. Recent studies have shown that the fascia is a potential source of lower back pain, and that inflammation and micro-damage to the fascia may contribute to lower back pain. [4] Therefore, it is necessary to accurately diagnose the low back pain caused by fascia and to prevent and treat it early. It has been proved that conventional MRI（magnetic resonance imaging) can better observe the anatomical structure and change of fascia. [5-7] Compared with CT, MR is more distinguishable in the anatomy of fascial structures and adjacent structures, and more reliable in the identification of subtle morphological differences. However, MRI diagnosis is often subjective based on the experience of the radiologist. This makes it difficult to make early and detailed observations of fascia. To overcome the influence of examiner's experience and subjectivity on diagnostic results, it is a challenge for imaging doctors to find a more objective, stable and efficient diagnostic method. Radiomics extracts quantitative features from medical images with high throughput, and makes diagnosis and prediction through quantitative analysis. It is objective, non-invasive and exploitable. [8] We have achieved good diagnostic efficacy in previous studies by establishing an Radiomics nomogram model. [9] In this study, we continued to expand the number of cases and add clinical factors. And we used seven different machine learning models, lasso、XGBlinear(Extreme gradient boosting)、SVMradial(support vector machine)、SVMlinear、RF(Random forest)、NNet(Neural Network)、KNN(K Nearest Neighbors) model,respectively. The final radiomics model with the best performance were selected.basing on the two-dimensional MRI imagesThe radiomics features were extracted and the optimal model was selected to establish the diagnostic model of lumbar fasciitis, which would reduce the missed diagnosis rate and provide a basis for the selection of treatmen. Materials And Methods Patients The institutional review board approved this retrospective study, and informed consent was waived. The primary cohort for this study was retrieved from the Institutional Picture Archiving and Communication System (PACS) database for patients who underwent lumbar MRI due to low back pain at our institution in September 2020 and September 2023. Inclusion Criteria for the Current Study Were as Follows (1) Patients with low back pain symptoms confirmed by clinical and MRI diagnosis of shallow lumbar fasciitis, defined as abnormal. The definition standard of superficial fasciitis of the back was as follows： The clinical manifestation was lumbar tenderness with local soft tissue stiffness. Some can touch the granular induration in the painful part. MRI showed striated or lamellar abnormal signals in the superficial lumbar fascia in the sagittal plane, showing long T1 and long T2 signals, and fat pressing on T2WI showed obvious high signals. On axial T2 imaging, strong signals were arranged in a triangle along the midline, close to the thoracolumbar fascia. (2) The patients were defined as normal with symptoms of low back pain, but non fasciitis in lumbar magnetic resonance imaging. Exclusion Criteria for the Current Study Were as Follows （1）Image artifacts that blur the spine or any hardware associated with the spine. （2）Patients with local infection, lumbar fracture, spinal tumor or metastasis, spinal surgery, or intervertebral disc surgery and lower back trauma. （3）Recently received rehabilitation physiotherapy may lead to transient edema of fascia. (4) There was infectious edema or edema related to internal diseases. Finally, 380 patients, including 193 patients with superficial fasciitis and 187 normal people, met the standard. Then, all patients were randomly assigned to the training queue (n = 266) and the verification queue (n = 144) according to the ratio of 7: 3. Instruments and methods Inspection instrument All patients in the study underwent an MRI of the spine with a 3.0-T system (Discovery MR750, GE Healthcare, Milwaukee, WI, USA). The parameters of the sagittal T1-weighted(T1WI), sagittal T2-weighted, sagittal T2-weighted fat supression(T2WI-fs), Axial T2-weighted images were showed in the table1. Table1.The parameters of the sagittal T1-weighted images TR(ms) TE(ms) NEX Slice Thickness(mm) Spacing(mm) Sagittal T1WI 560 Min Full 4.0 4 2 Sagittal T2WI 3000 100 2.0 4 2 Sagittal T2WI-fs 1800 120 2.0 4 2 Axial T2WI 3414 120 2.0 3.5 0.5 TR:repetition time TE:echo time NEX: the number of excitations Image and clinical information acquisition Medical records were used to obtain all clinical data, including age, gender, Cobb sagittal angle, Average fat thickness. The average fat thickness was measured at the posterior superior angle of L1 vertebral body and the posterior inferior level of L5 vertebral body according to West's research. [10] Retrieve MR axial spine images from PACS and save them in Digital Imaging and Communications in Medicine (DICOM) format. All MRI examinations were handled by two radiologists with 10 years of experience in musculoskeletal imaging. In case of disagreement, the scans were diagnosed by two readers together to judge the results. Lesion segmentation and feature extraction in radiomics model The region of interest (ROI) was manually drawn on the axial MRI images for segmentation using ITK-SNAP soft-ware (version 3.6.0, www.itksnap.org) [11] . The DICOM format images of the spine MRI were imported into the soft-ware for outlining. To assess the usability of the segmentation, The ROI was plotted by two radiologists with 3 years of experience in MRI interpretation. ROI included this level of the erector and multifidus muscles as well as the fatty portion behind the muscles containing fascia, which were at the L4/5 intervertebral disc intermediate level. (Figure 1). The ROIs were verified a week later by another radiologist with 5 years of experience. Any difference was re-delineated after consultation. Before feature extraction, all images were re-sampled and normalized according to previous protocol [9] .To avoid data heterogeneity bias, the voxel size was resampled to 1 × 1 ×1 mm3 , discretization of grayscale values using 25 bin widths, and normalization of grayscale values before feature extraction. Radiomics features were extracted using AK software (Analy-sisKit, version 3.2.0, GE Healthcare, China) backend and pyradiomics (version 3.0.1, https://pyradiomics.readthedocs. io/en/latest/). Feature Selection We used two feature selection method, mRMR and LASSO to select the feature. At first, mRMR was performed to eliminate the redundant and irrelevant features, 30 features were retained. Then LASSO was conducted to choose the optimized subset of features to construct the final model. Construction and evaluation of forecasting model Rad-score is calculated for the reduced-dimension features and imported into R (version 4.0.2, www.r-project.org). Establishment and construction lasso、XGBlinear、SVMradial、SVMlinear、RF、NNet、KNNmodel based on axial T2WI MRI sequence.Ten-fold cross-validation was used for training, and receiver operating characteristic (ROC) curve was drawn. The area under the curve（AUC）, accuracy, sensitivity and specificity were calculated. We selected the best one to establish the final prediction model. (Figure 2 and Table 3) Table 2. In the Training Group and the Validation Group, a Table Showing the Diagnostic Efficacy of Radiomics, Clinics, and Nomogram Models. Characteristics Training cohort (n = 266) Test cohort (n = 114) Normal Abnormal P-values Normal Abnormal P-values Number 129 137 58 56 Gender (%) Male 52(40.3) 28(39.4) 29(50) 22 (39.3) Female 77(59.7) 47(60.6) 0.981 29(50) 34 (60.7) 0.3362 Age, y, mean ± SD 39.6±14.6 51.8±15.3 <0.0001* 37.6±15.1 51.6±14.3 <0.0001* Cobbsagittal angle,mean± SD 32.9 ±4.6 24.1 ±4.6 <0.0001* 32.4 ±3.0 22.9 ±6.5 <0.0001* Average_fat_thickness(cm),mean± SD 1.5±0.7 2.4±0.9 <0.0001* 1.5±0.7 2.3±0.8 <0.0001* Table 3 The accuracy, sensitivity, specificity of different models Model Radscore.lasso Radscore.xgblinear Radscore.svmRadial Radscore.svmLinear Radscore.rf Radscore.nnet Radscore.knn Accuracy 0.86 0.79 0.79 0.77 0.76 0.74 0.76 Sensitivity 0.76 0.60 0.52 0.68 0.48 0.72 0.44 specificity 0.94 0.94 1.0 0.77 0.92 0.69 1.0 Prediction Nomogram Build and Diagnostic Validation Clinical variables included age, sex, average fat thickness and MR sagittal Cobb angle. Clinical risk factors were assessed using univariate analysis. We constructed clinical models based on backward step wise multivariate logistic regression analysis using Akaike An Information (AIC) as the criterion, which selected the significant risk factors with p < 0.05. Additionally, the final radiomics features and independent clinical risk variables were combined to create a radiomics nomogram. We used the DeLong test to distinguish differences in receiver operating characteristic curves between models. The fit of the combined nomogram was assessed using calibration curve analysis and the Hosmer-Lemeshow test. Statistical analysis All statistical analyses for this study were performed using R (version 4.0.2, www.r-project.org). The statistical significance of all two-sided tests was set at p < 0.05. Results Clinical Characteristics The 380 patients (223 men, 157 women; mean age, 45.5±16 years; range, 15-85 years), including 187 patients with superficial fasciitis of the low back and 193 normal individuals who met the criteria were randomly assigned to the training cohort (n = 266) and the validation cohort (n = 144). We can see in Table 2 that there were differences in age, sagittal-coob angle and average fat thickness between the training set and the test set. There was no difference in gender. The AUC of the clinical model in the training group was 0.97(95% CI, 0.95-0.96) ,and the accuracy, sensitivity and specificity were 89.8% , 91.2% and 88.3% , respectively. In the validation group, the clinical model also showed good predictive performance with an AUC value of 0.96(95% CI, 0.92-1.00) and accuracy, sensitivity, and specificity of 89.47% , 82.3% , and 94.4% Rad-score Building and Diagnostic Validation Redundant and uncorrelated features were first removed by mRMR. The parameter λ was then adjusted and tested by 10-fold cross-validation with LASSO regression. The value of λ corresponding to the minimum variance model was chosen as the best value (λ = 0.014) (Figure 3). A total of 13 subsets of nonzero coefficients were selected (Figure 4), and the corresponding coefficients were calculated. Radscore was calculated by summing the selected features weighted by their coefficients and the final formula of radscore was: "Radscore = -0.551*original_firstorder_Minimum +0.088*wavelet_HLH_glcm_ClusterShade +-0.419*wavelet_LHL_glrlm_LongRunLowGrayLevelEmphasis +0.513*wavelet_HLL_glszm_GrayLevelNonUniformity +0.169*wavelet_LHL_firstorder_Skewness +-0.294*wavelet_LLL_glszm_GrayLevelNonUniformityNormalized +-0.023*original_glcm_ClusterShade +0.083*original_glcm_Autocorrelation +0.128*wavelet_LHH_glcm_ClusterShade +0*wavelet_HHL_glszm_GrayLevelNonUniformity +-0.169*wavelet_HHH_firstorder_Median +-0.02*wavelet_HLL_firstorder_Mean +0.062*wavelet_HHL_glcm_ClusterProminence + 0.069" The Construction of Radiomics Model In this study, 380 patients were randomly divided into training set (n=266) and test set (n=144) according to the ratio of 7: 3. Finally, 13 representative radiomics features were selected and applied to lasso、xgblinear、svmradial、svmlinear、RF、nnet、knn seven classifiers to build models, and 10% cross-validation was carried out. AUC, specificity, sensitivity and accuracy of seven classifier construction models in independent training sets are shown in the figure and table. AUC, specificity, sensitivity and accuracy of seven classifier construction models in independent training sets are shown in the figure2 and table3. Finally, lasso classifier is selected to construct the best radiomics model. The AUC, specificity, sensitivity and accuracy of lasso were 0.83,0.86,0.76,0.94 ( figure2 and table3).The AUC of the radiomics model was 0.83(95% CI, 0.79-0.88) in the training set, and its specificity, sensitivity and accuracy were 82.2%, 72.2% and 77.06% respectively. In the test set, the AUC was 0.83 (95% CI: 0.76-0.91), and its specificity, sensitivity and accuracy were 84.4%、64.3%、74.5%. （table4，figure6） Table 4 In the Training Group and the Validation Group, a Table Showing the Diagnostic Efficacy of Radiomics, Clinics, and Nomogram Models. Model Accuracy Accuracy Accuracy Sensitivity Specificity AUC(95%CI) p-value of DeLong-Test (%) Lower(%) Upper(%) (%) (%) (%) vs Radiomics vs Nomogram Clinics Training 89.84 85.57 93.20 91.24 88.37 (0.63-0.81) <0.001 0.064 Validation 89.47 82.33 94.44 83.92 94.82 (0.59-0.85) 0.002 0.319 Radiomics Training 77.06 71.54 81.98 72.26 82.17 (0.87-0.96) - <0.001 Validation 74.56 65.55 82.25 64.28 84.48 (0.71-0.96) - <0.001 Nomogram Training 92.48 88.62 95.34 89.05 96.12 (0.88-0.96) <0.001 - Validation 92.10 85.54 96.32 1.00 86.56 (0.73-0.96) <0.001 - Construction of Combined Nomogram Nomogram model was constructed that incorporated the radiomics signature and clinical features (Figure 5). "Nomoscore =(Intercept)*5.15088420107063 +AGE*0.0908766909591725 +cobb_sagittal_angle*0.434307849852666 +Average_fat_thickness*1.61380017073753 +Radscore*0.863471409335681" The AUC values of this Nomogram model in the training group were 0.97(95%CI, 0.96-0.99), with accuracy, sensitivity, and specificity of 92.5%, 89.0%, and 96.1%, respectively. In the validation group, the Nomogram model also showed a good prediction performance with an AUC value of 0.96 (95%CI, 0.93-1.00), accuracy, sensitivity, and specificity of 92.1%, 100.0%, and 86.5%. (Fig 6 and Table 4) The performance of the nomogram was shown in figure 5. The calibration curves were showed in figure 7. The calibration curve showed good calibration in the training set and validation set, and the p-values of the Hosmer-Lemeshow test were 0.14 and 0.25 for the training and validation cohorts respectively, which were not significantly different. Finally, we used the decision curves to assess the clinical utility of the model (Figure 8). Using the DeLong test, the AUC values between the radiomics model, and the combined nomogram model were all significantly different in the training cohort and validation cohorts with a p-value <0.001(Table 4). Discussion The purpose of this study was to analyze the changes of thoracolumbar fascia in patients with low back pain using MR Imaging radiomics based on soft tissue of the lumbar spine. The results showed that the radiomic nomogram showed good diagnostic efficiency with AUCs of 0.97 and 0.96 in the training and validation cohorts, respectively. These findings suggested that radiographic nomogram based on MRI radiomics could be a good diagnosis of low back fasciitis and provide reference for clinical decision making. Histological studies had confirmed the presence of nociceptive free nerve endings within TLF, and microdamage and inflammation of TLF may lead to low back pain. [4, 12] The superficial fascia was prone to inflammatory exudation, while stimulating the myofascial pain trigger points within the lumbar dorsal fascia, producing symptoms of low back pain. As the disease resolves, inflammation and edema subside, and changes such as degeneration, atrophy, and fibroplasia occur. [13, 14] Low back pain caused by fascia changes has attracted the attention of experts and scholars. Some researchers had found that changes in mean fascia length affect the cross-sectional area of adjacent paravertebral compartments, leading to ischemic muscle atrophy. [15, 16] In addition to length, there was an association between increased fascia thickness and decreased joint flexibility, which leaded to lower back pain in patients. Solomonw et al. [17] had found through experiments in mice that repetitive mechanical strain leads to microinjuries in the myofascial. Accurate identification of low back pain caused by fascia was of great significance for early relief of patients' pain. Dutch scholar Lambin et al. [18] proposed the concept of radiomics for the first time in 2012. It refered to the extraction of high-throughput features from medical images and, based on these features, the use of a variety of statistical analysis and data mining methods to extract truly effective features from massive features, and finally applied to the diagnosis, classification and prognosis of diseases. In previous studies [9] , we established a diagnostic model of low back fasciitis and achieved good diagnostic performance. In this study, we continued to expand the number of cases and compared the diagnostic performance of different machine learning models. In our study, 13 features were selected by LASSO logistic regression to construct the radiomics signature model to avoid over-fitting in model construction. And we found that radiomics can be used to screen patients with low back pain with altered fascia using lumbar MRI with good discriminatory performance, with AUCs of 0.83 and 0.83 in the training and validation sets, respectively. In this study, we found that age, sagittal COBB Angle, and clinical features of average fat thickness in the lower back were associated with changes in the lower back fascia. Changes in fascia are associated with lumbar instability, so it makes sense to look at radiomic features of the soft tissues of the lower back and the sagittal COBB Angle [19] 。In West’s [10] study, it was found that the lower back subcutaneous edema was related to the average fat thickness, and the thicker the person, the more prone to edema. Wilke's study [20] also showed the first indication of substantial differences in fascial thickness between healthy young and old people. Differences in age were a potential confounding factor affecting fascia. Considering the influence of clinical factors, a nomogram combining radiomic features and clinical features was established. Finally, the combined nomogram model showed a good ability to distinguish between lower back fascia changes, with the highest AUC in the training and validation cohorts at 0.97 and 0.96, respectively, higher than radiomics (training set: AUC = 0.83; Validation set: AUC = 0.83). Although the AUC of the nomogram model and the clinical model were similar, the accuracy values and specificity were not. The clinical model was not as good as the nomogram model. At present, non-steroidal anti-inflammatory drugs (NSAIDS) and skeletal muscle relaxants are often used in the treatment of low back fasciitis, or combined with medium frequency electrotherapy, infrared radiation and ultrashort wave physical therapy are also commonly used, after these methods of treatment, although the symptoms can get some relief. But through the deep understanding and the early diagnosis to the waist and back fasciitis, unceasingly carries on the deep level excavation to the pain mechanism, finds one kind of more convenient, the effective method to treat the fasciitis, alleviates the low back pain patient pain. Our study also had some limitations: the manual division of ROI for low back soft tissues was more accurate but time-consuming, and the diagnosis of low back fasciitis was highly subjective. More stringent image acquisition standards needed to be developed to obtain clearer and more compliant images. This study was retrospective and may be biased. More prospective data were needed to verify and improve the diagnostic efficacy of the Radiomics model. Conclusion In summary, we developed an nomogram model that integrated clinical models and radiolomics features to help clinicians identify and predict low back fasciitis through soft tissue magnetic resonance imaging. Declarations Author contributions S.MX and J.L are responsible for conceptualization and original draft writing. Y.H is in charge of polishing and revising. X. Q and Q.J are responsible for supervision and project management. J.C is responsible for data management and investigation. Data Availability Statement: The data that support the findings of this study are available from the corresponding author upon reasonable request. Statement: The authors declare that there are no conflicts of interest regarding the publication of this paper. And We confirm that all methods were performed in accordance with the relevant guidelines and regulations of this journal. References Hoy D, Brooks P, Blyth F, et al. The Epidemiology of low back pain. Best Pract Res Clin Rheumatol, 2010. 24 (6):769-781. Hartvigsen J, Hancock MJ, Kongsted A, et al. What low back pain is and why we need to pay attention. The Lancet, 2018. 391 (10137):2356-2367. Knezevic NN, Candido KD, Vlaeyen JWS, et al. Low back pain. The Lancet, 2021. 398 (10294):78-92. Wilke J, Schleip R, Klingler W, et al. The Lumbodorsal Fascia as a Potential Source of Low Back Pain: A Narrative Review. BioMed Research International, 2017. 2017 :5349620. Karino K, Kono M, Kono M, et al. Myofascia-dominant involvement on whole-body MRI as a risk factor for rapidly progressive interstitial lung disease in dermatomyositis. Rheumatology (Oxford), 2020. 59 (7):1734-1742. Schwarz-Nemec U, Friedrich KM, Arnoldner MA, et al. When an incidental MRI finding becomes a clinical issue : Posterior lumbar subcutaneous edema in degenerative, inflammatory, and infectious conditions of the lumbar spine. Wien Klin Wochenschr, 2020. 132 (1-2):27-34. Yan Y, Xu R, Zou T. Is thoracolumbar fascia injury the cause of residual back pain after percutaneous vertebroplasty? A prospective cohort study. Osteoporos Int, 2015. 26 (3):1119-1124. Yip SS, Aerts HJ. Applications and limitations of radiomics. Phys Med Biol, 2016. 61 (13):R150-166. Song M-x, Yang H, Yang H-q, et al. MR Imaging Radiomics Analysis Based on Lumbar Soft Tissue to Evaluate Lumbar Fascia Changes in Patients with Low Back Pain. Academic Radiology, 2023. West W, Brady-West D, West KP. A comparison of statistical associations between oedema in the lumbar fat on MRI, BMI and Back Fat Thickness (BFT). Heliyon, 2018. 4 (1). Yushkevich PA, Piven J, Hazlett HC, et al. User-guided 3D active contour segmentation of anatomical structures: significantly improved efficiency and reliability. Neuroimage, 2006. 31 (3):1116-1128. Schilder A, Hoheisel U, Magerl W, et al. Sensory findings after stimulation of the thoracolumbar fascia with hypertonic saline suggest its contribution to low back pain. Pain, 2014. Lam WWM, Chan H, Chan YL, et al. MR Imaging in Amyopathic Dermatomyositis. Acta Radiologica, 2016. 40 (1):69-72. Dalakas MC. Polymyositis, dermatomyositis and inclusion-body myositis. The New England journal of medicine 1991. 325 (21 ):1487-1498. El-Monajjed K, Driscoll M. A finite element analysis of the intra-abdominal pressure and paraspinal muscle compartment pressure interaction through the thoracolumbar fascia. Comput Methods Biomech Biomed Engin, 2020. 23 (10):585-596. Ranger TA, Teichtahl AJ, Cicuttini FM, et al. Shorter Lumbar Paraspinal Fascia Is Associated With High Intensity Low Back Pain and Disability. Spine (Phila Pa 1976), 2016. 41 (8):E489-493. Solomonow M. Neuromuscular manifestations of viscoelastic tissue degradation following high and low risk repetitive lumbar flexion. J Electromyogr Kinesiol, 2012. 22 (2):155-175. Lambin P, Rios-Velazquez E, Leijenaar R, et al. Radiomics: Extracting more information from medical images using advanced feature analysis. European Journal of Cancer, 2012. 48 (4):441-446. Jeong YM SM, Lee SH, Chung HW. Sagging posterior layer thoracolumbar fascia: can it be the cause or result of adjacent segment diseases? J Spinal Disord Tech, 2013(26(4)):E124-E129. Wilke J, Macchi V, De Caro R, et al. Fascia thickness, aging and flexibility: is there an association? J Anat, 2019. 234 (1):43-49. Additional Declarations No competing interests reported. Cite Share Download PDF Status: Posted Version 1 posted You are reading this latest preprint version Research Square lets you share your work early, gain feedback from the community, and start making changes to your manuscript prior to peer review in a journal. As a division of Research Square Company, we’re committed to making research communication faster, fairer, and more useful. We do this by developing innovative software and high quality services for the global research community. Our growing team is made up of researchers and industry professionals working together to solve the most critical problems facing scientific publishing. Also discoverable on Platform About Our Team In Review Editorial Policies Advisory Board Help Center Resources Author Services Accessibility API Access RSS feed Manage Cookie Preferences © Research Square 2026 | ISSN 2693-5015 (online) Privacy Policy Terms of Service Do Not Sell My Personal Information {"props":{"pageProps":{"initialData":{"identity":"rs-4239014","acceptedTermsAndConditions":true,"allowDirectSubmit":true,"archivedVersions":[],"articleType":"Article","associatedPublications":[],"authors":[{"id":299808889,"identity":"28b92a38-b749-46d2-bf1e-a5bfe56a2326","order_by":0,"name":"mingxin song","email":"","orcid":"","institution":"The Second Affiliated Hospital of Shandong First Medical University","correspondingAuthor":false,"prefix":"","firstName":"mingxin","middleName":"","lastName":"song","suffix":""},{"id":299808891,"identity":"fff23e0a-97d8-472e-a8d4-6bb386b3c8ae","order_by":1,"name":"ling Jiang","email":"","orcid":"","institution":"The Second Affiliated Hospital of Shandong First Medical University","correspondingAuthor":false,"prefix":"","firstName":"ling","middleName":"","lastName":"Jiang","suffix":""},{"id":299808893,"identity":"586f3733-e025-44df-940e-1b86a9b62090","order_by":2,"name":"hui Yang","email":"","orcid":"","institution":"The Second Affiliated Hospital of Shandong First Medical University","correspondingAuthor":false,"prefix":"","firstName":"hui","middleName":"","lastName":"Yang","suffix":""},{"id":299808896,"identity":"24f02134-f4f2-48bb-ba5e-2f84804a286f","order_by":3,"name":"cheng Jia","email":"","orcid":"","institution":"The Second Affiliated Hospital of Shandong First Medical University","correspondingAuthor":false,"prefix":"","firstName":"cheng","middleName":"","lastName":"Jia","suffix":""},{"id":299808899,"identity":"6c1e9b64-3d34-4478-89bb-c2d37ada7343","order_by":4,"name":"Jian Qin","email":"","orcid":"","institution":"The Second Affiliated Hospital of Shandong First Medical University","correspondingAuthor":false,"prefix":"","firstName":"Jian","middleName":"","lastName":"Qin","suffix":""},{"id":299808902,"identity":"9e9f90a3-0efb-43c8-b7c7-f9d949c1e6fc","order_by":5,"name":"Qiang Xiao","email":"data:image/png;base64,iVBORw0KGgoAAAANSUhEUgAAAZAAAAAyAQMAAABI0h/eAAAABlBMVEX///8AAABVwtN+AAAACXBIWXMAAA7EAAAOxAGVKw4bAAAAqElEQVRIiWNgGAWjYBAC+xv536R5Kmx4+NkbiNTCJpHDJs1zJk1GsucAsVp4zjAbzmw7bGNww4FYLew9jA8+tp3nYbjBwPjhYw4xWpj5DxxIbLvNwzi7gVly5jaitPAwgLUwyxxgY+YlWsvHbed42CQSSNBycOa8Azw8JGk5zHMmmUeC52Az8X45zFNlZ29/vPngh4/EaEECjA2kqR8Fo2AUjIJRgBsAALcjMf5iBU7NAAAAAElFTkSuQmCC","orcid":"","institution":"The Second Affiliated Hospital of Shandong First Medical University","correspondingAuthor":true,"prefix":"","firstName":"Qiang","middleName":"","lastName":"Xiao","suffix":""}],"badges":[],"createdAt":"2024-04-09 01:29:17","currentVersionCode":1,"declarations":"","doi":"10.21203/rs.3.rs-4239014/v1","doiUrl":"https://doi.org/10.21203/rs.3.rs-4239014/v1","draftVersion":[],"editorialEvents":[],"editorialNote":"","failedWorkflow":false,"files":[{"id":56399291,"identity":"1d394410-eeb8-468f-aa81-58ef81f0a7ca","added_by":"auto","created_at":"2024-05-13 16:05:35","extension":"png","order_by":1,"title":"Figure 1","display":"","copyAsset":false,"role":"figure","size":293437,"visible":true,"origin":"","legend":"\u003cp\u003eAxial T2-weighted L4/5-disc intermediate level was used to map the region of interest (ROI). This includes the erector spinae and multifidus at this level and the part of the fat behind the muscle that contains the fascia\u003c/p\u003e","description":"","filename":"image1.png","url":"https://assets-eu.researchsquare.com/files/rs-4239014/v1/1a58dd02da9954f455d4351a.png"},{"id":56398786,"identity":"27674d06-657f-44e7-8cbd-903b3f456f5c","added_by":"auto","created_at":"2024-05-13 15:57:35","extension":"png","order_by":2,"title":"Figure 2","display":"","copyAsset":false,"role":"figure","size":225322,"visible":true,"origin":"","legend":"\u003cp\u003eAUC curves for the seven machine-learning models.\u003c/p\u003e","description":"","filename":"image2.png","url":"https://assets-eu.researchsquare.com/files/rs-4239014/v1/a27038486c71821588a16b24.png"},{"id":56398784,"identity":"e1fb42bd-8683-47e2-80ae-704586fac6ac","added_by":"auto","created_at":"2024-05-13 15:57:35","extension":"png","order_by":3,"title":"Figure 3","display":"","copyAsset":false,"role":"figure","size":284827,"visible":true,"origin":"","legend":"\u003cp\u003eSelection of the hyperparameters (λ) in the LASSO model by 10-fold cross-validation based on the minimum criterion of binomial deviation. The binomial deviation is plotted as a function of log (λ). The optimal λ value of 0.0142 was chosen. LASSO: Least Absolute Shrinkage and Selection Operators.\u003c/p\u003e","description":"","filename":"image3.png","url":"https://assets-eu.researchsquare.com/files/rs-4239014/v1/887acffbbb1f378a3cc94d73.png"},{"id":56398783,"identity":"f59f623c-1ce7-4297-9003-48556e73c3c1","added_by":"auto","created_at":"2024-05-13 15:57:35","extension":"png","order_by":4,"title":"Figure 4","display":"","copyAsset":false,"role":"figure","size":266157,"visible":true,"origin":"","legend":"\u003cp\u003eHistogram of the Rad-score: the y-axis represents the 13 selected radiomics, and the x-axis represents the coefficients of the radiomics.\u003c/p\u003e","description":"","filename":"image4.png","url":"https://assets-eu.researchsquare.com/files/rs-4239014/v1/8d313b98622337997dd32e32.png"},{"id":56399636,"identity":"90c8efb0-4605-4082-a590-8c6eb1e48d69","added_by":"auto","created_at":"2024-05-13 16:13:35","extension":"png","order_by":5,"title":"Figure 5","display":"","copyAsset":false,"role":"figure","size":154748,"visible":true,"origin":"","legend":"\u003cp\u003eRadiomics nomogram model combining the values of age, sagittal cobb angle, average fat thickness and rad-scores developed in the training cohort.\u003c/p\u003e","description":"","filename":"image5.png","url":"https://assets-eu.researchsquare.com/files/rs-4239014/v1/10e1937e462a1eacdf7c38d6.png"},{"id":56398789,"identity":"52f5b3a3-2583-44af-bc8a-a38a69904039","added_by":"auto","created_at":"2024-05-13 15:57:35","extension":"png","order_by":6,"title":"Figure 6","display":"","copyAsset":false,"role":"figure","size":275596,"visible":true,"origin":"","legend":"\u003cp\u003eAUC of clinics, radiomics and nomogram models in the training cohort and validation cohort\u003c/p\u003e","description":"","filename":"image6.png","url":"https://assets-eu.researchsquare.com/files/rs-4239014/v1/ef810814599d49579bcf6345.png"},{"id":56398791,"identity":"d1db1353-e338-4a83-a5ef-87c233674619","added_by":"auto","created_at":"2024-05-13 15:57:35","extension":"png","order_by":7,"title":"Figure 7","display":"","copyAsset":false,"role":"figure","size":185779,"visible":true,"origin":"","legend":"\u003cp\u003eThe nomogram calibration curve of the training cohort (left) and the validation cohort (right) showing the relationship between the predictive and true values. The distance between the dotted line and the solid line indicated the prediction ability of the model, with an inverse relationship between the distance and prediction ability\u003c/p\u003e","description":"","filename":"image7.png","url":"https://assets-eu.researchsquare.com/files/rs-4239014/v1/bb8a01fbfc6ff45c6034dad2.png"},{"id":56398788,"identity":"33448355-0189-4637-a11d-7f2d007f84f6","added_by":"auto","created_at":"2024-05-13 15:57:35","extension":"png","order_by":8,"title":"Figure 8","display":"","copyAsset":false,"role":"figure","size":161422,"visible":true,"origin":"","legend":"\u003cp\u003eDecision curve analysis (DCA) of radiomics nomogram. DCA shows that the application of radiomics nomogram predicts the changes in lumbar fascia better than clinical models. The red and blue lines represent the net benefit of the radiographic and clinical models, respectively. The green line represents the hypothesis that all patients have fascial alterations. The black line represents the hypothesis that no patients have altered fascia.\u003c/p\u003e","description":"","filename":"image8.png","url":"https://assets-eu.researchsquare.com/files/rs-4239014/v1/1deb5de3f5212a9a378544b2.png"},{"id":60902913,"identity":"1dc0bd96-af47-4919-ba6f-5016c1a96099","added_by":"auto","created_at":"2024-07-23 11:19:23","extension":"pdf","order_by":0,"title":"","display":"","copyAsset":false,"role":"manuscript-pdf","size":2667644,"visible":true,"origin":"","legend":"","description":"","filename":"manuscript.pdf","url":"https://assets-eu.researchsquare.com/files/rs-4239014/v1/e50e77ec-9c20-451c-ac3f-c264ce50e020.pdf"}],"financialInterests":"No competing interests reported.","formattedTitle":"Prediction of low back fasciitis by machine learning method based on radiomics","fulltext":[{"header":"Introduction","content":"\u003cp\u003eLow back pain is a very common problem that has a huge impact on individuals, families and businesses around the world. Globally, disability due to low back pain has increased by 54% in 25 years.\u003csup\u003e[1, 2]\u003c/sup\u003e Low back pain can be caused by any one of a number of anatomical structures. Bones, discs, joints, ligaments, muscles, etc. are mostly considered to be one of the common causes of low back pain.\u003csup\u003e[3]\u003c/sup\u003e Fascia, however, is often overlooked. Recent studies have shown that the fascia is a potential source of lower back pain, and that inflammation and micro-damage to the fascia may contribute to lower back pain.\u003csup\u003e[4]\u003c/sup\u003e Therefore, it is necessary to accurately diagnose the low back pain caused by fascia and to prevent and treat it early.\u003c/p\u003e\n\u003cp\u003eIt has been proved that conventional MRI（magnetic resonance imaging) can better observe the anatomical structure and change of fascia.\u003csup\u003e[5-7]\u003c/sup\u003e Compared with CT, MR is more distinguishable in the anatomy of fascial structures and adjacent structures, and more reliable in the identification of subtle morphological differences. However, MRI diagnosis is often subjective based on the experience of the radiologist. This makes it difficult to make early and detailed observations of fascia. To overcome the influence of examiner\u0026apos;s experience and subjectivity on diagnostic results, it is a challenge for imaging doctors to find a more objective, stable and efficient diagnostic method.\u003c/p\u003e\n\u003cp\u003eRadiomics extracts quantitative features from medical images with high throughput, and makes diagnosis and prediction through quantitative analysis. It is objective, non-invasive and exploitable.\u003csup\u003e[8]\u003c/sup\u003e We have achieved good diagnostic efficacy in previous studies by establishing an Radiomics nomogram model.\u003csup\u003e[9]\u003c/sup\u003e\u0026nbsp;\u003c/p\u003e\n\u003cp\u003eIn this study, we continued to expand the number of cases and add clinical factors. And we used seven different machine learning models, lasso、XGBlinear(Extreme gradient boosting)、SVMradial(support vector machine)、SVMlinear、RF(Random forest)、NNet(Neural Network)、KNN(K Nearest Neighbors) model,respectively. The final radiomics model with the best performance were selected.basing on the two-dimensional MRI imagesThe radiomics features were extracted and the optimal model was selected to establish the diagnostic model of lumbar fasciitis, which would reduce the missed diagnosis rate and provide a basis for the selection of treatmen.\u003c/p\u003e"},{"header":"Materials And Methods","content":"\u003ch3\u003ePatients\u003c/h3\u003e\n\u003cp\u003eThe institutional review board approved this retrospective study, and informed consent was waived. The primary cohort for this study was retrieved from the Institutional Picture Archiving and Communication System (PACS) database for patients who underwent lumbar MRI due to low back pain at our institution in September 2020 and September 2023.\u003c/p\u003e\n\u003ch4\u003eInclusion Criteria for the Current Study Were as Follows\u003c/h4\u003e\n\u003cp\u003e(1) Patients with low back pain symptoms confirmed by clinical and MRI diagnosis of shallow lumbar fasciitis, defined as abnormal. The definition standard of superficial fasciitis of the back was as follows：\u003c/p\u003e\n\u003cp\u003eThe clinical manifestation was lumbar tenderness with local soft tissue stiffness. Some can touch the granular induration in the painful part. MRI showed striated or lamellar abnormal signals in the superficial lumbar fascia in the sagittal plane, showing long T1 and long T2 signals, and fat pressing on T2WI showed obvious high signals. On axial T2 imaging, strong signals were arranged in a triangle along the midline, close to the thoracolumbar fascia.\u003c/p\u003e\n\u003cp\u003e(2) The patients were defined as normal with symptoms of low back pain, but non fasciitis in lumbar magnetic resonance imaging.\u003c/p\u003e\n\u003ch4\u003eExclusion Criteria for the Current Study Were as Follows\u003c/h4\u003e\n\u003cp\u003e（1）Image artifacts that blur the spine or any hardware associated with the spine.\u003c/p\u003e\n\u003cp\u003e（2）Patients with local infection, lumbar fracture, spinal tumor or metastasis, spinal surgery, or intervertebral disc surgery and lower back trauma.\u003c/p\u003e\n\u003cp\u003e（3）Recently received rehabilitation physiotherapy may lead to transient edema of fascia.\u003c/p\u003e\n\u003cp\u003e\u0026nbsp;(4) There was infectious edema or edema related to internal diseases.\u003c/p\u003e\n\u003cp\u003eFinally, 380 patients, including 193 patients with superficial fasciitis and 187 normal people, met the standard. Then, all patients were randomly assigned to the training queue (n = 266) and the verification queue (n = 144) according to the ratio of 7: 3.\u003c/p\u003e\n\u003ch2\u003eInstruments and methods\u003c/h2\u003e\n\u003ch4\u003eInspection instrument\u003c/h4\u003e\n\u003cp\u003eAll patients in the study underwent an MRI of the spine with a 3.0-T system (Discovery MR750, GE Healthcare, Milwaukee, WI, USA). The parameters of the sagittal T1-weighted(T1WI), sagittal T2-weighted, sagittal T2-weighted fat supression(T2WI-fs), Axial T2-weighted images were showed in the table1.\u0026nbsp;\u003c/p\u003e\n\u003cp\u003eTable1.The parameters of the sagittal T1-weighted images\u003c/p\u003e\n\u003ctable border=\"1\" cellspacing=\"0\" cellpadding=\"0\" width=\"616\"\u003e\n \u003ctbody\u003e\n \u003ctr\u003e\n \u003ctd width=\"20.48780487804878%\" valign=\"top\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"13.821138211382113%\" valign=\"top\"\u003e\n \u003cp\u003eTR(ms)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"12.195121951219512%\" valign=\"top\"\u003e\n \u003cp\u003eTE(ms)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"12.032520325203253%\" valign=\"top\"\u003e\n \u003cp\u003eNEX\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"25.040650406504064%\" valign=\"top\"\u003e\n \u003cp\u003eSlice Thickness(mm)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"16.422764227642276%\" valign=\"top\"\u003e\n \u003cp\u003eSpacing(mm)\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd width=\"20.48780487804878%\" valign=\"top\"\u003e\n \u003cp\u003eSagittal T1WI\u003c/p\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"13.821138211382113%\" valign=\"top\"\u003e\n \u003cp\u003e560\u003c/p\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"12.195121951219512%\" valign=\"top\"\u003e\n \u003cp\u003eMin Full\u003c/p\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"12.032520325203253%\" valign=\"top\"\u003e\n \u003cp\u003e4.0\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"25.040650406504064%\" valign=\"top\"\u003e\n \u003cp\u003e4\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"16.422764227642276%\" valign=\"top\"\u003e\n \u003cp\u003e2\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd width=\"20.48780487804878%\"\u003e\n \u003cp\u003eSagittal\u0026nbsp;T2WI\u003c/p\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"13.821138211382113%\"\u003e\n \u003cp\u003e3000\u003c/p\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"12.195121951219512%\"\u003e\n \u003cp\u003e100\u003c/p\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"12.032520325203253%\" valign=\"top\"\u003e\n \u003cp\u003e2.0\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"25.040650406504064%\" valign=\"top\"\u003e\n \u003cp\u003e4\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"16.422764227642276%\" valign=\"top\"\u003e\n \u003cp\u003e2\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd width=\"20.48780487804878%\"\u003e\n \u003cp\u003eSagittal\u0026nbsp;T2WI-fs\u003c/p\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"13.821138211382113%\"\u003e\n \u003cp\u003e1800\u003c/p\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"12.195121951219512%\"\u003e\n \u003cp\u003e120\u003c/p\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"12.032520325203253%\" valign=\"top\"\u003e\n \u003cp\u003e2.0\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"25.040650406504064%\" valign=\"top\"\u003e\n \u003cp\u003e4\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"16.422764227642276%\" valign=\"top\"\u003e\n \u003cp\u003e2\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd width=\"20.48780487804878%\"\u003e\n \u003cp\u003eAxial T2WI\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"13.821138211382113%\"\u003e\n \u003cp\u003e3414\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"12.195121951219512%\"\u003e\n \u003cp\u003e120\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"12.032520325203253%\" valign=\"top\"\u003e\n \u003cp\u003e2.0\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"25.040650406504064%\" valign=\"top\"\u003e\n \u003cp\u003e3.5\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"16.422764227642276%\" valign=\"top\"\u003e\n \u003cp\u003e0.5\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003c/tbody\u003e\n\u003c/table\u003e\n\u003cp\u003eTR:repetition time TE:echo time NEX: the number of excitations\u0026nbsp;\u003c/p\u003e\n\u003ch4\u003eImage and clinical information acquisition\u003c/h4\u003e\n\u003cp\u003eMedical records were used to obtain all clinical data, including age, gender, Cobb sagittal angle, Average fat thickness. The average fat thickness was measured at the posterior superior angle of L1 vertebral body and the posterior inferior level of L5 vertebral body according to West\u0026apos;s research.\u003csup\u003e[10]\u003c/sup\u003e Retrieve MR axial spine images from PACS and save them in Digital Imaging and Communications in Medicine (DICOM) format.\u003c/p\u003e\n\u003cp\u003eAll MRI examinations were handled by two radiologists with 10 years of experience in musculoskeletal imaging. In case of disagreement, the scans were diagnosed by two readers together to judge the results.\u0026nbsp;\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eLesion segmentation and feature extraction in radiomics model\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThe region of interest (ROI) was manually drawn on the axial MRI images for segmentation using ITK-SNAP soft-ware (version 3.6.0, www.itksnap.org) \u003csup\u003e[11]\u003c/sup\u003e. The DICOM format images of the spine MRI were imported into the soft-ware for outlining. To assess the usability of the segmentation, The ROI was plotted by two radiologists with 3 years of experience in MRI interpretation. ROI included this level of the erector and multifidus muscles as well as the fatty portion behind the muscles containing fascia, which were at the L4/5 intervertebral disc intermediate level. (Figure 1). The ROIs were verified a week later by another radiologist with 5 years of experience. Any difference was re-delineated after consultation.\u003c/p\u003e\n\u003cp\u003eBefore feature extraction, all images were re-sampled and normalized according to previous protocol\u003csup\u003e[9]\u003c/sup\u003e .To avoid data heterogeneity bias, the voxel size was resampled to 1 \u0026times; 1 \u0026times;1 mm3 , discretization of grayscale values using 25 bin widths, and normalization of grayscale values before feature extraction. Radiomics features were extracted using AK software (Analy-sisKit, version 3.2.0, GE Healthcare, China) backend and pyradiomics (version 3.0.1, https://pyradiomics.readthedocs. io/en/latest/).\u0026nbsp;\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eFeature Selection\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eWe used two feature selection method, mRMR and LASSO to select the feature. At first, mRMR was performed to eliminate the redundant and irrelevant features, 30 features were retained. Then LASSO was conducted to choose the optimized subset of features to construct the final model.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eConstruction and evaluation of forecasting model\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eRad-score is calculated for the reduced-dimension features and imported into R (version 4.0.2, www.r-project.org). Establishment and construction lasso、XGBlinear、SVMradial、SVMlinear、RF、NNet、KNNmodel based on axial T2WI MRI sequence.Ten-fold cross-validation was used for training, and receiver operating characteristic (ROC) curve was drawn.\u003c/p\u003e\n\u003cp\u003eThe area under the curve（AUC）, accuracy, sensitivity and specificity were calculated. We selected the best one to establish the final prediction model. (Figure 2 and Table 3)\u0026nbsp;\u003c/p\u003e\n\u003cp\u003eTable 2. In the Training Group and the Validation Group, a Table Showing the Diagnostic Efficacy of Radiomics, Clinics, and Nomogram Models.\u003c/p\u003e\n\u003ctable border=\"1\" cellspacing=\"0\" cellpadding=\"0\" width=\"671\"\u003e\n \u003ctbody\u003e\n \u003ctr\u003e\n \u003ctd width=\"21.162444113263785%\" colspan=\"2\" valign=\"top\"\u003e\n \u003cp\u003eCharacteristics\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"38.152011922503725%\" colspan=\"3\"\u003e\n \u003cp\u003eTraining cohort (n = 266)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"40.68554396423249%\" colspan=\"3\" valign=\"top\"\u003e\n \u003cp\u003eTest cohort (n = 114)\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd width=\"18.330849478390462%\" valign=\"top\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"14.157973174366617%\" colspan=\"2\" valign=\"top\"\u003e\n \u003cp\u003eNormal\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"14.008941877794337%\" valign=\"top\"\u003e\n \u003cp\u003eAbnormal\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"12.816691505216095%\" valign=\"top\"\u003e\n \u003cp\u003eP-values\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"12.816691505216095%\" valign=\"top\"\u003e\n \u003cp\u003eNormal\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"14.157973174366617%\" valign=\"top\"\u003e\n \u003cp\u003eAbnormal\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"13.710879284649776%\" valign=\"top\"\u003e\n \u003cp\u003eP-values\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd width=\"18.330849478390462%\" valign=\"top\"\u003e\n \u003cp\u003eNumber\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"14.157973174366617%\" colspan=\"2\" valign=\"top\"\u003e\n \u003cp\u003e129\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"14.008941877794337%\" valign=\"top\"\u003e\n \u003cp\u003e137\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"12.816691505216095%\" valign=\"top\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"12.816691505216095%\" valign=\"top\"\u003e\n \u003cp\u003e58\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"14.157973174366617%\" valign=\"top\"\u003e\n \u003cp\u003e56\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"13.710879284649776%\" valign=\"top\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd width=\"18.330849478390462%\" valign=\"top\"\u003e\n \u003cp\u003eGender (%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"14.157973174366617%\" colspan=\"2\" valign=\"top\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"14.008941877794337%\" valign=\"top\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"12.816691505216095%\" valign=\"top\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"12.816691505216095%\" valign=\"top\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"14.157973174366617%\" valign=\"top\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"13.710879284649776%\" valign=\"top\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd width=\"18.330849478390462%\" valign=\"top\"\u003e\n \u003cp\u003eMale\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"14.157973174366617%\" colspan=\"2\" valign=\"top\"\u003e\n \u003cp\u003e52(40.3)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"14.008941877794337%\" valign=\"top\"\u003e\n \u003cp\u003e28(39.4)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"12.816691505216095%\" valign=\"top\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"12.816691505216095%\" valign=\"top\"\u003e\n \u003cp\u003e29(50)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"14.157973174366617%\" valign=\"top\"\u003e\n \u003cp\u003e22 (39.3)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"13.710879284649776%\" valign=\"top\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd width=\"18.330849478390462%\" valign=\"top\"\u003e\n \u003cp\u003eFemale\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"14.157973174366617%\" colspan=\"2\" valign=\"top\"\u003e\n \u003cp\u003e77(59.7)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"14.008941877794337%\" valign=\"top\"\u003e\n \u003cp\u003e47(60.6)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"12.816691505216095%\" valign=\"top\"\u003e\n \u003cp\u003e0.981\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"12.816691505216095%\" valign=\"top\"\u003e\n \u003cp\u003e29(50)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"14.157973174366617%\" valign=\"top\"\u003e\n \u003cp\u003e34 (60.7)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"13.710879284649776%\" valign=\"top\"\u003e\n \u003cp\u003e0.3362\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd width=\"18.330849478390462%\" valign=\"top\"\u003e\n \u003cp\u003eAge, y,\u003c/p\u003e\n \u003cp\u003emean\u0026nbsp;\u0026plusmn;\u0026nbsp;SD\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"14.157973174366617%\" colspan=\"2\" valign=\"top\"\u003e\n \u003cp\u003e39.6\u0026plusmn;14.6\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"14.008941877794337%\" valign=\"top\"\u003e\n \u003cp\u003e51.8\u0026plusmn;15.3\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"12.816691505216095%\" valign=\"top\"\u003e\n \u003cp\u003e\u0026lt;0.0001*\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"12.816691505216095%\" valign=\"top\"\u003e\n \u003cp\u003e37.6\u0026plusmn;15.1\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"14.157973174366617%\" valign=\"top\"\u003e\n \u003cp\u003e51.6\u0026plusmn;14.3\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"13.710879284649776%\" valign=\"top\"\u003e\n \u003cp\u003e\u0026lt;0.0001*\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd width=\"18.330849478390462%\" valign=\"top\"\u003e\n \u003cp\u003eCobbsagittal angle,mean\u0026plusmn;\u0026nbsp;SD\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"14.157973174366617%\" colspan=\"2\" valign=\"top\"\u003e\n \u003cp\u003e32.9\u0026nbsp;\u0026plusmn;4.6\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"14.008941877794337%\" valign=\"top\"\u003e\n \u003cp\u003e24.1\u0026nbsp;\u0026plusmn;4.6\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"12.816691505216095%\" valign=\"top\"\u003e\n \u003cp\u003e\u0026lt;0.0001*\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"12.816691505216095%\" valign=\"top\"\u003e\n \u003cp\u003e32.4\u0026nbsp;\u0026plusmn;3.0\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"14.157973174366617%\" valign=\"top\"\u003e\n \u003cp\u003e22.9\u0026nbsp;\u0026plusmn;6.5\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"13.710879284649776%\" valign=\"top\"\u003e\n \u003cp\u003e\u0026lt;0.0001*\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd width=\"18.330849478390462%\" valign=\"top\"\u003e\n \u003cp\u003eAverage_fat_thickness(cm),mean\u0026plusmn; SD\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"14.157973174366617%\" colspan=\"2\" valign=\"top\"\u003e\n \u003cp\u003e1.5\u0026plusmn;0.7\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"14.008941877794337%\" valign=\"top\"\u003e\n \u003cp\u003e2.4\u0026plusmn;0.9\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"12.816691505216095%\" valign=\"top\"\u003e\n \u003cp\u003e\u0026lt;0.0001*\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"12.816691505216095%\" valign=\"top\"\u003e\n \u003cp\u003e1.5\u0026plusmn;0.7\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"14.157973174366617%\" valign=\"top\"\u003e\n \u003cp\u003e2.3\u0026plusmn;0.8\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"13.710879284649776%\" valign=\"top\"\u003e\n \u003cp\u003e\u0026lt;0.0001*\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003c/tbody\u003e\n\u003c/table\u003e\n\u003cp\u003e\u003cbr\u003e\u003c/p\u003e\n\u003cp\u003eTable 3 The accuracy, sensitivity, specificity of different models\u003c/p\u003e\n\u003ctable border=\"0\" cellspacing=\"0\" cellpadding=\"0\" align=\"left\" width=\"690\"\u003e\n \u003ctbody\u003e\n \u003ctr\u003e\n \u003ctd width=\"12.318840579710145%\" valign=\"top\"\u003e\n \u003cp\u003eModel\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"2.318840579710145%\" valign=\"top\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"11.304347826086957%\" valign=\"top\"\u003e\n \u003cp\u003eRadscore.lasso\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"12.46376811594203%\" valign=\"top\"\u003e\n \u003cp\u003eRadscore.xgblinear\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"12.46376811594203%\" valign=\"top\"\u003e\n \u003cp\u003eRadscore.svmRadial\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"12.318840579710145%\" valign=\"top\"\u003e\n \u003cp\u003eRadscore.svmLinear\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"11.73913043478261%\" valign=\"top\"\u003e\n \u003cp\u003eRadscore.rf\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"11.014492753623188%\" valign=\"top\"\u003e\n \u003cp\u003eRadscore.nnet\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"14.057971014492754%\" valign=\"top\"\u003e\n \u003cp\u003eRadscore.knn\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd width=\"12.318840579710145%\"\u003e\n \u003cp\u003eAccuracy\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"2.318840579710145%\" valign=\"top\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"11.304347826086957%\" valign=\"top\"\u003e\n \u003cp\u003e0.86\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"12.46376811594203%\" valign=\"top\"\u003e\n \u003cp\u003e0.79\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"12.46376811594203%\" valign=\"top\"\u003e\n \u003cp\u003e0.79\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"12.318840579710145%\" valign=\"top\"\u003e\n \u003cp\u003e0.77\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"11.73913043478261%\" valign=\"top\"\u003e\n \u003cp\u003e0.76\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"11.014492753623188%\" valign=\"top\"\u003e\n \u003cp\u003e0.74\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"14.057971014492754%\" valign=\"top\"\u003e\n \u003cp\u003e0.76\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd width=\"12.318840579710145%\"\u003e\n \u003cp\u003eSensitivity\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"2.318840579710145%\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"11.304347826086957%\"\u003e\n \u003cp\u003e0.76\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"12.46376811594203%\"\u003e\n \u003cp\u003e0.60\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"12.46376811594203%\"\u003e\n \u003cp\u003e0.52\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"12.318840579710145%\"\u003e\n \u003cp\u003e0.68\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"11.73913043478261%\"\u003e\n \u003cp\u003e0.48\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"11.014492753623188%\"\u003e\n \u003cp\u003e0.72\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"14.057971014492754%\"\u003e\n \u003cp\u003e0.44\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd width=\"12.318840579710145%\"\u003e\n \u003cp\u003especificity\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"2.318840579710145%\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"11.304347826086957%\"\u003e\n \u003cp\u003e0.94\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"12.46376811594203%\"\u003e\n \u003cp\u003e0.94\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"12.46376811594203%\"\u003e\n \u003cp\u003e1.0\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"12.318840579710145%\"\u003e\n \u003cp\u003e0.77\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"11.73913043478261%\"\u003e\n \u003cp\u003e0.92\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"11.014492753623188%\"\u003e\n \u003cp\u003e0.69\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"14.057971014492754%\"\u003e\n \u003cp\u003e1.0\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003c/tbody\u003e\n\u003c/table\u003e\n\u003cp\u003e\u003cbr\u003e\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003ePrediction Nomogram Build and Diagnostic Validation\u0026nbsp;\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eClinical variables included age, sex, average fat thickness and MR sagittal Cobb angle. Clinical risk factors were assessed using univariate analysis. We constructed clinical models based on backward step wise multivariate logistic regression analysis using Akaike An Information (AIC) as the criterion, which selected the significant risk factors with p \u0026lt; 0.05. Additionally, the final radiomics features and independent clinical risk variables were combined to create a radiomics nomogram. We used the DeLong test to distinguish differences in receiver operating characteristic curves between models. The fit of the combined nomogram was assessed using calibration curve analysis and the Hosmer-Lemeshow test.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eStatistical analysis \u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eAll statistical analyses for this study were performed using R (version 4.0.2, www.r-project.org). The statistical significance of all two-sided tests was set at p \u0026lt; 0.05.\u003c/p\u003e"},{"header":"Results","content":"\u003ch3\u003eClinical Characteristics\u003c/h3\u003e\n\u003cp\u003eThe 380 patients (223 men, 157 women; mean age, 45.5\u0026plusmn;16 years; range, 15-85 years), including 187 patients with superficial fasciitis of the low back and 193 normal individuals who met the criteria were randomly assigned to the training cohort (n = 266) and the validation cohort (n = 144). We can see in Table 2 that there were differences in age, sagittal-coob angle and average fat thickness between the training set and the test set. There was no difference in gender.\u003c/p\u003e\n\u003cp\u003eThe AUC of the clinical model in the training group was 0.97(95% CI, 0.95-0.96) ,and the accuracy, sensitivity and specificity were 89.8% , 91.2% and 88.3% , respectively. In the validation group, the clinical model also showed good predictive performance with an AUC value of 0.96(95% CI, 0.92-1.00) and accuracy, sensitivity, and specificity of 89.47% , 82.3% , and 94.4%\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eRad-score Building and Diagnostic Validation\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eRedundant and uncorrelated features were first removed by mRMR. The parameter \u0026lambda; was then adjusted and tested by 10-fold cross-validation with LASSO regression. The value of \u0026lambda; corresponding to the minimum variance model was chosen as the best value (\u0026lambda; = 0.014) (Figure 3). A total of 13 subsets of nonzero coefficients were selected (Figure 4), and the corresponding coefficients were calculated. Radscore was calculated by summing the selected features weighted by their coefficients and the final formula of radscore was:\u003c/p\u003e\n\u003cp\u003e\u0026quot;Radscore = \u0026nbsp;-0.551*original_firstorder_Minimum\u003c/p\u003e\n\u003cp\u003e+0.088*wavelet_HLH_glcm_ClusterShade\u003c/p\u003e\n\u003cp\u003e+-0.419*wavelet_LHL_glrlm_LongRunLowGrayLevelEmphasis\u003c/p\u003e\n\u003cp\u003e+0.513*wavelet_HLL_glszm_GrayLevelNonUniformity\u003c/p\u003e\n\u003cp\u003e+0.169*wavelet_LHL_firstorder_Skewness\u003c/p\u003e\n\u003cp\u003e+-0.294*wavelet_LLL_glszm_GrayLevelNonUniformityNormalized\u003c/p\u003e\n\u003cp\u003e+-0.023*original_glcm_ClusterShade\u003c/p\u003e\n\u003cp\u003e+0.083*original_glcm_Autocorrelation\u003c/p\u003e\n\u003cp\u003e+0.128*wavelet_LHH_glcm_ClusterShade\u003c/p\u003e\n\u003cp\u003e+0*wavelet_HHL_glszm_GrayLevelNonUniformity\u003c/p\u003e\n\u003cp\u003e+-0.169*wavelet_HHH_firstorder_Median\u003c/p\u003e\n\u003cp\u003e+-0.02*wavelet_HLL_firstorder_Mean\u003c/p\u003e\n\u003cp\u003e+0.062*wavelet_HHL_glcm_ClusterProminence + 0.069\u0026quot;\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eThe Construction of Radiomics Model\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003e\u0026nbsp; In this study, 380 patients were randomly divided into training set (n=266) and test set (n=144) according to the ratio of 7: 3. Finally, 13 representative radiomics features were selected and applied to lasso、xgblinear、svmradial、svmlinear、RF、nnet、knn seven classifiers to build models, and 10% cross-validation was carried out. AUC, specificity, sensitivity and accuracy of seven classifier construction models in independent training sets are shown in the figure and table. AUC, specificity, sensitivity and accuracy of seven classifier construction models in independent training sets are shown in the figure2 and table3.\u003c/p\u003e\n\u003cp\u003eFinally, lasso classifier is selected to construct the best radiomics model. The AUC, specificity, sensitivity and accuracy of lasso were 0.83,0.86,0.76,0.94 ( figure2 and table3).The AUC of the radiomics model was 0.83(95% CI, 0.79-0.88) in the training set, and its specificity, sensitivity and accuracy were 82.2%, 72.2% and 77.06% respectively. In the test set, the AUC was 0.83 (95% CI: 0.76-0.91), and its specificity, sensitivity and accuracy were 84.4%、64.3%、74.5%.\u0026nbsp;（table4，figure6）\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eTable 4 \u003c/strong\u003eIn the Training Group and the Validation Group, a Table Showing the Diagnostic Efficacy of Radiomics, Clinics, and Nomogram Models.\u003c/p\u003e\n\u003ctable border=\"0\" cellspacing=\"0\" cellpadding=\"0\" width=\"748\"\u003e\n \u003ctbody\u003e\n \u003ctr\u003e\n \u003ctd width=\"10.843373493975903%\" valign=\"top\"\u003e\n \u003cp\u003e\u003cstrong\u003eModel\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"9.63855421686747%\" valign=\"top\"\u003e\n \u003cp\u003e\u003cstrong\u003e\u0026nbsp;\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"8.835341365461847%\" valign=\"top\"\u003e\n \u003cp\u003e\u003cstrong\u003eAccuracy\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"8.96921017402945%\" valign=\"top\"\u003e\n \u003cp\u003e\u003cstrong\u003eAccuracy\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"8.96921017402945%\" valign=\"top\"\u003e\n \u003cp\u003e\u003cstrong\u003eAccuracy\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"9.504685408299865%\" valign=\"top\"\u003e\n \u003cp\u003e\u003cstrong\u003eSensitivity\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"9.63855421686747%\" valign=\"top\"\u003e\n \u003cp\u003e\u003cstrong\u003eSpecificity\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"10.977242302543507%\" valign=\"top\"\u003e\n \u003cp\u003e\u003cstrong\u003eAUC(95%CI)\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"22.623828647925034%\" colspan=\"3\" valign=\"top\"\u003e\n \u003cp\u003e\u003cstrong\u003ep-value of DeLong-Test\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd width=\"10.843373493975903%\" valign=\"top\"\u003e\n \u003cp\u003e\u003cstrong\u003e\u0026nbsp;\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"9.63855421686747%\" valign=\"top\"\u003e\n \u003cp\u003e\u003cstrong\u003e\u0026nbsp;\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"8.835341365461847%\" valign=\"top\"\u003e\n \u003cp\u003e\u003cstrong\u003e(%)\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"8.96921017402945%\" valign=\"top\"\u003e\n \u003cp\u003e\u003cstrong\u003eLower(%)\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"8.96921017402945%\" valign=\"top\"\u003e\n \u003cp\u003e\u003cstrong\u003eUpper(%)\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"9.504685408299865%\" valign=\"top\"\u003e\n \u003cp\u003e\u003cstrong\u003e(%)\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"9.63855421686747%\" valign=\"top\"\u003e\n \u003cp\u003e\u003cstrong\u003e(%)\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"10.977242302543507%\" valign=\"top\"\u003e\n \u003cp\u003e\u003cstrong\u003e(%)\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"9.772423025435074%\" valign=\"top\"\u003e\n \u003cp\u003e\u003cstrong\u003evs Radiomics\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"2.0080321285140563%\" valign=\"top\"\u003e\n \u003cp\u003e\u003cstrong\u003e\u0026nbsp;\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"10.843373493975903%\" valign=\"top\"\u003e\n \u003cp\u003e\u003cstrong\u003evs Nomogram\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd width=\"10.843373493975903%\"\u003e\n \u003cp\u003e\u003cstrong\u003eClinics\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"9.63855421686747%\" valign=\"top\"\u003e\n \u003cp\u003e\u003cstrong\u003e\u0026nbsp;\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"8.835341365461847%\" valign=\"top\"\u003e\n \u003cp\u003e\u003cstrong\u003e\u0026nbsp;\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"8.96921017402945%\" valign=\"top\"\u003e\n \u003cp\u003e\u003cstrong\u003e\u0026nbsp;\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"8.96921017402945%\" valign=\"top\"\u003e\n \u003cp\u003e\u003cstrong\u003e\u0026nbsp;\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"9.504685408299865%\" valign=\"top\"\u003e\n \u003cp\u003e\u003cstrong\u003e\u0026nbsp;\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"9.63855421686747%\" valign=\"top\"\u003e\n \u003cp\u003e\u003cstrong\u003e\u0026nbsp;\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"10.977242302543507%\" valign=\"top\"\u003e\n \u003cp\u003e\u003cstrong\u003e\u0026nbsp;\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"9.772423025435074%\" valign=\"top\"\u003e\n \u003cp\u003e\u003cstrong\u003e\u0026nbsp;\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"2.0080321285140563%\" valign=\"top\"\u003e\n \u003cp\u003e\u003cstrong\u003e\u0026nbsp;\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"10.843373493975903%\" valign=\"top\"\u003e\n \u003cp\u003e\u003cstrong\u003e\u0026nbsp;\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd width=\"10.843373493975903%\"\u003e\n \u003cp\u003e\u003cstrong\u003e\u0026nbsp;\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"9.63855421686747%\"\u003e\n \u003cp\u003e\u003cstrong\u003eTraining\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"8.835341365461847%\"\u003e\n \u003cp\u003e\u003cstrong\u003e89.84\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"8.96921017402945%\"\u003e\n \u003cp\u003e\u003cstrong\u003e85.57\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"8.96921017402945%\"\u003e\n \u003cp\u003e\u003cstrong\u003e93.20\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"9.504685408299865%\"\u003e\n \u003cp\u003e\u003cstrong\u003e91.24\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"9.63855421686747%\"\u003e\n \u003cp\u003e\u003cstrong\u003e88.37\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"10.977242302543507%\"\u003e\n \u003cp\u003e\u003cstrong\u003e(0.63-0.81)\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"9.772423025435074%\"\u003e\n \u003cp\u003e\u003cstrong\u003e\u0026lt;0.001\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"2.0080321285140563%\" valign=\"top\"\u003e\n \u003cp\u003e\u003cstrong\u003e\u0026nbsp;\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"10.843373493975903%\"\u003e\n \u003cp\u003e\u003cstrong\u003e0.064\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd width=\"10.843373493975903%\"\u003e\n \u003cp\u003e\u003cstrong\u003e\u0026nbsp;\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"9.63855421686747%\"\u003e\n \u003cp\u003e\u003cstrong\u003eValidation\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"8.835341365461847%\"\u003e\n \u003cp\u003e\u003cstrong\u003e89.47\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"8.96921017402945%\"\u003e\n \u003cp\u003e\u003cstrong\u003e82.33\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"8.96921017402945%\"\u003e\n \u003cp\u003e\u003cstrong\u003e94.44\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"9.504685408299865%\"\u003e\n \u003cp\u003e\u003cstrong\u003e83.92\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"9.63855421686747%\"\u003e\n \u003cp\u003e\u003cstrong\u003e94.82\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"10.977242302543507%\"\u003e\n \u003cp\u003e\u003cstrong\u003e\u0026nbsp;(0.59-0.85)\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"9.772423025435074%\"\u003e\n \u003cp\u003e\u003cstrong\u003e0.002\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"2.0080321285140563%\" valign=\"top\"\u003e\n \u003cp\u003e\u003cstrong\u003e\u0026nbsp;\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"10.843373493975903%\"\u003e\n \u003cp\u003e\u003cstrong\u003e0.319\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd width=\"10.843373493975903%\"\u003e\n \u003cp\u003e\u003cstrong\u003eRadiomics\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"9.63855421686747%\"\u003e\n \u003cp\u003e\u003cstrong\u003e\u0026nbsp;\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"8.835341365461847%\"\u003e\n \u003cp\u003e\u003cstrong\u003e\u0026nbsp;\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"8.96921017402945%\"\u003e\n \u003cp\u003e\u003cstrong\u003e\u0026nbsp;\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"8.96921017402945%\"\u003e\n \u003cp\u003e\u003cstrong\u003e\u0026nbsp;\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"9.504685408299865%\"\u003e\n \u003cp\u003e\u003cstrong\u003e\u0026nbsp;\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"9.63855421686747%\"\u003e\n \u003cp\u003e\u003cstrong\u003e\u0026nbsp;\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"10.977242302543507%\"\u003e\n \u003cp\u003e\u003cstrong\u003e\u0026nbsp;\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"9.772423025435074%\"\u003e\n \u003cp\u003e\u003cstrong\u003e\u0026nbsp;\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"2.0080321285140563%\" valign=\"top\"\u003e\n \u003cp\u003e\u003cstrong\u003e\u0026nbsp;\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"10.843373493975903%\"\u003e\n \u003cp\u003e\u003cstrong\u003e\u0026nbsp;\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd width=\"10.843373493975903%\"\u003e\n \u003cp\u003e\u003cstrong\u003e\u0026nbsp;\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"9.63855421686747%\"\u003e\n \u003cp\u003e\u003cstrong\u003eTraining\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"8.835341365461847%\"\u003e\n \u003cp\u003e\u003cstrong\u003e77.06\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"8.96921017402945%\"\u003e\n \u003cp\u003e\u003cstrong\u003e71.54\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"8.96921017402945%\"\u003e\n \u003cp\u003e\u003cstrong\u003e81.98\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"9.504685408299865%\"\u003e\n \u003cp\u003e\u003cstrong\u003e72.26\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"9.63855421686747%\"\u003e\n \u003cp\u003e\u003cstrong\u003e82.17\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"10.977242302543507%\"\u003e\n \u003cp\u003e\u003cstrong\u003e\u0026nbsp;(0.87-0.96)\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"9.772423025435074%\"\u003e\n \u003cp\u003e\u003cstrong\u003e-\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"2.0080321285140563%\" valign=\"top\"\u003e\n \u003cp\u003e\u003cstrong\u003e\u0026nbsp;\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"10.843373493975903%\"\u003e\n \u003cp\u003e\u003cstrong\u003e\u0026lt;0.001\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd width=\"10.843373493975903%\"\u003e\n \u003cp\u003e\u003cstrong\u003e\u0026nbsp;\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"9.63855421686747%\"\u003e\n \u003cp\u003e\u003cstrong\u003eValidation\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"8.835341365461847%\"\u003e\n \u003cp\u003e\u003cstrong\u003e74.56\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"8.96921017402945%\"\u003e\n \u003cp\u003e\u003cstrong\u003e65.55\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"8.96921017402945%\"\u003e\n \u003cp\u003e\u003cstrong\u003e82.25\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"9.504685408299865%\"\u003e\n \u003cp\u003e\u003cstrong\u003e64.28\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"9.63855421686747%\"\u003e\n \u003cp\u003e\u003cstrong\u003e84.48\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"10.977242302543507%\"\u003e\n \u003cp\u003e\u003cstrong\u003e(0.71-0.96)\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"9.772423025435074%\"\u003e\n \u003cp\u003e\u003cstrong\u003e-\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"2.0080321285140563%\" valign=\"top\"\u003e\n \u003cp\u003e\u003cstrong\u003e\u0026nbsp;\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"10.843373493975903%\"\u003e\n \u003cp\u003e\u003cstrong\u003e\u0026lt;0.001\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd width=\"10.843373493975903%\"\u003e\n \u003cp\u003e\u003cstrong\u003eNomogram\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"9.63855421686747%\"\u003e\n \u003cp\u003e\u003cstrong\u003e\u0026nbsp;\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"8.835341365461847%\"\u003e\n \u003cp\u003e\u003cstrong\u003e\u0026nbsp;\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"8.96921017402945%\"\u003e\n \u003cp\u003e\u003cstrong\u003e\u0026nbsp;\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"8.96921017402945%\"\u003e\n \u003cp\u003e\u003cstrong\u003e\u0026nbsp;\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"9.504685408299865%\"\u003e\n \u003cp\u003e\u003cstrong\u003e\u0026nbsp;\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"9.63855421686747%\"\u003e\n \u003cp\u003e\u003cstrong\u003e\u0026nbsp;\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"10.977242302543507%\"\u003e\n \u003cp\u003e\u003cstrong\u003e\u0026nbsp;\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"9.772423025435074%\"\u003e\n \u003cp\u003e\u003cstrong\u003e\u0026nbsp;\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"2.0080321285140563%\" valign=\"top\"\u003e\n \u003cp\u003e\u003cstrong\u003e\u0026nbsp;\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"10.843373493975903%\"\u003e\n \u003cp\u003e\u003cstrong\u003e\u0026nbsp;\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd width=\"10.843373493975903%\"\u003e\n \u003cp\u003e\u003cstrong\u003e\u0026nbsp;\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"9.63855421686747%\"\u003e\n \u003cp\u003e\u003cstrong\u003eTraining\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"8.835341365461847%\"\u003e\n \u003cp\u003e\u003cstrong\u003e92.48\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"8.96921017402945%\"\u003e\n \u003cp\u003e\u003cstrong\u003e88.62\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"8.96921017402945%\"\u003e\n \u003cp\u003e\u003cstrong\u003e95.34\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"9.504685408299865%\"\u003e\n \u003cp\u003e\u003cstrong\u003e89.05\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"9.63855421686747%\"\u003e\n \u003cp\u003e\u003cstrong\u003e96.12\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"10.977242302543507%\"\u003e\n \u003cp\u003e\u003cstrong\u003e(0.88-0.96)\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"9.772423025435074%\"\u003e\n \u003cp\u003e\u003cstrong\u003e\u0026lt;0.001\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"2.0080321285140563%\" valign=\"top\"\u003e\n \u003cp\u003e\u003cstrong\u003e\u0026nbsp;\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"10.843373493975903%\"\u003e\n \u003cp\u003e\u003cstrong\u003e-\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd width=\"10.843373493975903%\"\u003e\n \u003cp\u003e\u003cstrong\u003e\u0026nbsp;\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"9.63855421686747%\"\u003e\n \u003cp\u003e\u003cstrong\u003eValidation\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"8.835341365461847%\"\u003e\n \u003cp\u003e\u003cstrong\u003e92.10\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"8.96921017402945%\"\u003e\n \u003cp\u003e\u003cstrong\u003e85.54\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"8.96921017402945%\"\u003e\n \u003cp\u003e\u003cstrong\u003e96.32\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"9.504685408299865%\"\u003e\n \u003cp\u003e\u003cstrong\u003e1.00\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"9.63855421686747%\"\u003e\n \u003cp\u003e\u003cstrong\u003e86.56\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"10.977242302543507%\"\u003e\n \u003cp\u003e\u003cstrong\u003e(0.73-0.96)\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"9.772423025435074%\"\u003e\n \u003cp\u003e\u003cstrong\u003e\u0026lt;0.001\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"2.0080321285140563%\" valign=\"top\"\u003e\n \u003cp\u003e\u003cstrong\u003e\u0026nbsp;\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"10.843373493975903%\"\u003e\n \u003cp\u003e\u003cstrong\u003e-\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003c/tbody\u003e\n\u003c/table\u003e\n\u003cp\u003e\u003cbr\u003e\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eConstruction of Combined Nomogram\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eNomogram model was constructed that incorporated the radiomics signature and clinical features (Figure 5).\u0026nbsp;\u003c/p\u003e\n\u003cp\u003e\u0026quot;Nomoscore =(Intercept)*5.15088420107063\u003c/p\u003e\n\u003cp\u003e+AGE*0.0908766909591725\u003c/p\u003e\n\u003cp\u003e+cobb_sagittal_angle*0.434307849852666\u003c/p\u003e\n\u003cp\u003e+Average_fat_thickness*1.61380017073753\u003c/p\u003e\n\u003cp\u003e+Radscore*0.863471409335681\u0026quot;\u003c/p\u003e\n\u003cp\u003eThe AUC values of this Nomogram model in the training group were 0.97(95%CI, 0.96-0.99), with accuracy, sensitivity, and specificity of 92.5%, 89.0%, and 96.1%, respectively. In the validation group, the Nomogram model also showed a good prediction performance with an AUC value of 0.96 (95%CI, 0.93-1.00), accuracy, sensitivity, and specificity of 92.1%, 100.0%, and 86.5%. (Fig 6 and Table 4)\u003c/p\u003e\n\u003cp\u003eThe performance of the nomogram was shown in figure 5. The calibration curves were showed in figure 7. The calibration curve showed good calibration in the training set and validation set, and the p-values of the Hosmer-Lemeshow test were 0.14 and 0.25 for the training and validation cohorts respectively, which were not significantly different.\u003c/p\u003e\n\u003cp\u003eFinally, we used the decision curves to assess the clinical utility of the model (Figure 8). Using the DeLong test, the AUC values between the radiomics model, and the combined nomogram model were all significantly different in the training cohort and validation cohorts with a p-value \u0026lt;0.001(Table 4).\u003c/p\u003e"},{"header":"Discussion","content":"\u003cp\u003eThe purpose of this study was to analyze the changes of thoracolumbar fascia in patients with low back pain using MR Imaging radiomics based on soft tissue of the lumbar spine. The results showed that the radiomic nomogram showed good diagnostic efficiency with AUCs of 0.97 and 0.96 in the training and validation cohorts, respectively. These findings suggested that radiographic nomogram based on MRI radiomics could be a good diagnosis of low back fasciitis and provide reference for clinical decision making.\u0026nbsp;\u003c/p\u003e\n\u003cp\u003eHistological studies had confirmed the presence of nociceptive free nerve endings within TLF, and microdamage and inflammation of TLF may lead to low back pain.\u003csup\u003e[4, 12]\u003c/sup\u003e The superficial fascia was prone to inflammatory exudation, while stimulating the myofascial pain trigger points within the lumbar dorsal fascia, producing symptoms of low back pain. As the disease resolves, inflammation and edema subside, and changes such as degeneration, atrophy, and fibroplasia occur.\u003csup\u003e[13, 14]\u003c/sup\u003e\u003c/p\u003e\n\u003cp\u003eLow back pain caused by fascia changes has attracted the attention of experts and scholars. Some researchers had found that changes in mean fascia length affect the cross-sectional area of adjacent paravertebral compartments, leading to ischemic muscle atrophy.\u003csup\u003e[15, 16]\u003c/sup\u003eIn addition to length, there was an association between increased fascia thickness and decreased joint flexibility, which leaded to lower back pain in patients. Solomonw et al. \u003csup\u003e[17]\u0026nbsp;\u003c/sup\u003ehad found through experiments in mice that repetitive mechanical strain leads to microinjuries in the myofascial. Accurate identification of low back pain caused by fascia was of great significance for early relief of patients\u0026apos; pain.\u003c/p\u003e\n\u003cp\u003eDutch scholar Lambin et al. \u003csup\u003e[18]\u0026nbsp;\u003c/sup\u003eproposed the concept of radiomics for the first time in 2012. It refered to the extraction of high-throughput features from medical images and, based on these features, the use of a variety of statistical analysis and data mining methods to extract truly effective features from massive features, and finally applied to the diagnosis, classification and prognosis of diseases. \u0026nbsp;In previous studies\u003csup\u003e[9]\u003c/sup\u003e, we established a diagnostic model of low back fasciitis and achieved good diagnostic performance. In this study, we continued to expand the number of cases and compared the diagnostic performance of different machine learning models. In our study, 13 features were selected by LASSO logistic regression to construct the radiomics signature model to avoid over-fitting in model construction. And we found that radiomics can be used to screen patients with low back pain with altered fascia using lumbar MRI with good discriminatory performance, with AUCs of 0.83 and 0.83 in the training and validation sets, respectively.\u003c/p\u003e\n\u003cp\u003eIn this study, we found that age, sagittal COBB Angle, and clinical features of average fat thickness in the lower back were associated with changes in the lower back fascia. Changes in fascia are associated with lumbar instability, so it makes sense to look at radiomic features of the soft tissues of the lower back and the sagittal COBB Angle\u003csup\u003e[19]\u003c/sup\u003e。In West\u0026rsquo;s \u003csup\u003e[10]\u003c/sup\u003estudy, it was found that the lower back subcutaneous edema was related to the average fat thickness, and the thicker the person, the more prone to edema. Wilke\u0026apos;s study\u003csup\u003e[20]\u003c/sup\u003e also showed the first indication of substantial differences in fascial thickness between healthy young and old people. Differences in age were a potential confounding factor affecting fascia. Considering the influence of clinical factors, a nomogram combining radiomic features and clinical features was established. Finally, the combined nomogram model showed a good ability to distinguish between lower back fascia changes, with the highest AUC in the training and validation cohorts at 0.97 and 0.96, respectively, higher than radiomics (training set: AUC = 0.83; Validation set: AUC = 0.83). Although the AUC of the nomogram model and the clinical model were similar, the accuracy values and specificity were not. The clinical model was not as good as the nomogram model.\u003c/p\u003e\n\u003cp\u003eAt present, non-steroidal anti-inflammatory drugs (NSAIDS) and skeletal muscle relaxants are often used in the treatment of low back fasciitis, or combined with medium frequency electrotherapy, infrared radiation and ultrashort wave physical therapy are also commonly used, after these methods of treatment, although the symptoms can get some relief. But through the deep understanding and the early diagnosis to the waist and back fasciitis, unceasingly carries on the deep level excavation to the pain mechanism, finds one kind of more convenient, the effective method to treat the fasciitis, alleviates the low back pain patient pain.\u003c/p\u003e\n\u003cp\u003eOur study also had some limitations: the manual division of ROI for low back soft tissues was more accurate but time-consuming, and the diagnosis of low back fasciitis was highly subjective. More stringent image acquisition standards needed to be developed to obtain clearer and more compliant images. This study was retrospective and may be biased. More prospective data were needed to verify and improve the diagnostic efficacy of the Radiomics model.\u003c/p\u003e"},{"header":"Conclusion","content":"\u003cp\u003eIn summary, we developed an nomogram model that integrated clinical models and radiolomics features to help clinicians identify and predict low back fasciitis through soft tissue magnetic resonance imaging.\u003c/p\u003e"},{"header":"Declarations","content":"\u003ch2\u003eAuthor contributions\u003c/h2\u003e\n\u003cp\u003eS.MX and J.L are responsible for conceptualization and original draft writing. Y.H is in charge of polishing and revising. X. Q and Q.J are responsible for supervision and project management. J.C is responsible for data management and investigation.\u003c/p\u003e\n\u003ch2\u003eData Availability Statement:\u003c/h2\u003e\n\u003cp\u003eThe data that support the findings of this study are available from the corresponding author upon reasonable request.\u003c/p\u003e\n\u003ch2\u003eStatement:\u003c/h2\u003e\n\u003cp\u003eThe authors declare that there are no conflicts of interest regarding the publication of this paper. And We confirm that all methods were performed in accordance with the relevant guidelines and regulations of this journal.\u003c/p\u003e"},{"header":"References","content":"\u003col\u003e\n\u003cli\u003eHoy D, Brooks P, Blyth F, et al. \u003cem\u003eThe Epidemiology of low back pain.\u003c/em\u003e Best Pract Res Clin Rheumatol, 2010. \u003cstrong\u003e24\u003c/strong\u003e(6):769-781.\u003c/li\u003e\n\u003cli\u003eHartvigsen J, Hancock MJ, Kongsted A, et al. \u003cem\u003eWhat low back pain is and why we need to pay attention.\u003c/em\u003e The Lancet, 2018. \u003cstrong\u003e391\u003c/strong\u003e(10137):2356-2367.\u003c/li\u003e\n\u003cli\u003eKnezevic NN, Candido KD, Vlaeyen JWS, et al. \u003cem\u003eLow back pain.\u003c/em\u003e The Lancet, 2021. \u003cstrong\u003e398\u003c/strong\u003e(10294):78-92.\u003c/li\u003e\n\u003cli\u003eWilke J, Schleip R, Klingler W, et al. \u003cem\u003eThe Lumbodorsal Fascia as a Potential Source of Low Back Pain: A Narrative Review.\u003c/em\u003e BioMed Research International, 2017. \u003cstrong\u003e2017\u003c/strong\u003e:5349620.\u003c/li\u003e\n\u003cli\u003eKarino K, Kono M, Kono M, et al. \u003cem\u003eMyofascia-dominant involvement on whole-body MRI as a risk factor for rapidly progressive interstitial lung disease in dermatomyositis.\u003c/em\u003e Rheumatology (Oxford), 2020. \u003cstrong\u003e59\u003c/strong\u003e(7):1734-1742.\u003c/li\u003e\n\u003cli\u003eSchwarz-Nemec U, Friedrich KM, Arnoldner MA, et al. \u003cem\u003eWhen an incidental MRI finding becomes a clinical issue : Posterior lumbar subcutaneous edema in degenerative, inflammatory, and infectious conditions of the lumbar spine.\u003c/em\u003e Wien Klin Wochenschr, 2020. \u003cstrong\u003e132\u003c/strong\u003e(1-2):27-34.\u003c/li\u003e\n\u003cli\u003eYan Y, Xu R, Zou T. \u003cem\u003eIs thoracolumbar fascia injury the cause of residual back pain after percutaneous vertebroplasty? A prospective cohort study.\u003c/em\u003e Osteoporos Int, 2015. \u003cstrong\u003e26\u003c/strong\u003e(3):1119-1124.\u003c/li\u003e\n\u003cli\u003eYip SS, Aerts HJ. \u003cem\u003eApplications and limitations of radiomics.\u003c/em\u003e Phys Med Biol, 2016. \u003cstrong\u003e61\u003c/strong\u003e(13):R150-166.\u003c/li\u003e\n\u003cli\u003eSong M-x, Yang H, Yang H-q, et al. \u003cem\u003eMR Imaging Radiomics Analysis Based on Lumbar Soft Tissue to Evaluate Lumbar Fascia Changes in Patients with Low Back Pain.\u003c/em\u003e Academic Radiology, 2023.\u003c/li\u003e\n\u003cli\u003eWest W, Brady-West D, West KP. \u003cem\u003eA comparison of statistical associations between oedema in the lumbar fat on MRI, BMI and Back Fat Thickness (BFT).\u003c/em\u003e Heliyon, 2018. \u003cstrong\u003e4\u003c/strong\u003e(1).\u003c/li\u003e\n\u003cli\u003eYushkevich PA, Piven J, Hazlett HC, et al. \u003cem\u003eUser-guided 3D active contour segmentation of anatomical structures: significantly improved efficiency and reliability.\u003c/em\u003e Neuroimage, 2006. \u003cstrong\u003e31\u003c/strong\u003e(3):1116-1128.\u003c/li\u003e\n\u003cli\u003eSchilder A, Hoheisel U, Magerl W, et al. \u003cem\u003eSensory findings after stimulation of the thoracolumbar fascia with hypertonic saline suggest its contribution to low back pain.\u003c/em\u003e Pain, 2014.\u003c/li\u003e\n\u003cli\u003eLam WWM, Chan H, Chan YL, et al. \u003cem\u003eMR Imaging in Amyopathic Dermatomyositis.\u003c/em\u003e Acta Radiologica, 2016. \u003cstrong\u003e40\u003c/strong\u003e(1):69-72.\u003c/li\u003e\n\u003cli\u003eDalakas MC. \u003cem\u003ePolymyositis, dermatomyositis and inclusion-body myositis.\u003c/em\u003e The New England journal of medicine 1991.\u003cstrong\u003e 325\u003c/strong\u003e(21 ):1487-1498.\u003c/li\u003e\n\u003cli\u003eEl-Monajjed K, Driscoll M. \u003cem\u003eA finite element analysis of the intra-abdominal pressure and paraspinal muscle compartment pressure interaction through the thoracolumbar fascia.\u003c/em\u003e Comput Methods Biomech Biomed Engin, 2020. \u003cstrong\u003e23\u003c/strong\u003e(10):585-596.\u003c/li\u003e\n\u003cli\u003eRanger TA, Teichtahl AJ, Cicuttini FM, et al. \u003cem\u003eShorter Lumbar Paraspinal Fascia Is Associated With High Intensity Low Back Pain and Disability.\u003c/em\u003e Spine (Phila Pa 1976), 2016. \u003cstrong\u003e41\u003c/strong\u003e(8):E489-493.\u003c/li\u003e\n\u003cli\u003eSolomonow M. \u003cem\u003eNeuromuscular manifestations of viscoelastic tissue degradation following high and low risk repetitive lumbar flexion.\u003c/em\u003e J Electromyogr Kinesiol, 2012. \u003cstrong\u003e22\u003c/strong\u003e(2):155-175.\u003c/li\u003e\n\u003cli\u003eLambin P, Rios-Velazquez E, Leijenaar R, et al. \u003cem\u003eRadiomics: Extracting more information from medical images using advanced feature analysis.\u003c/em\u003e European Journal of Cancer, 2012. \u003cstrong\u003e48\u003c/strong\u003e(4):441-446.\u003c/li\u003e\n\u003cli\u003eJeong YM SM, Lee SH, Chung HW. \u003cem\u003eSagging posterior layer thoracolumbar fascia: can it be the cause or result of adjacent segment diseases?\u003c/em\u003e J Spinal Disord Tech, 2013(26(4)):E124-E129.\u003c/li\u003e\n\u003cli\u003eWilke J, Macchi V, De Caro R, et al. \u003cem\u003eFascia thickness, aging and flexibility: is there an association?\u003c/em\u003e J Anat, 2019. \u003cstrong\u003e234\u003c/strong\u003e(1):43-49.\u003c/li\u003e\n\u003c/ol\u003e"}],"fulltextSource":"","fullText":"","funders":[],"hasAdminPriorityOnWorkflow":false,"hasManuscriptDocX":true,"hasOptedInToPreprint":true,"hasPassedJournalQc":"","hasAnyPriority":false,"hideJournal":true,"highlight":"","institution":"","isAcceptedByJournal":false,"isAuthorSuppliedPdf":false,"isDeskRejected":"","isHiddenFromSearch":false,"isInQc":false,"isInWorkflow":false,"isPdf":false,"isPdfUpToDate":true,"isWithdrawnOrRetracted":false,"journal":{"display":true,"email":"[email protected]","identity":"researchsquare","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":true,"externalIdentity":"","sideBox":"","snPcode":"","submissionUrl":"/submission","title":"Research Square","twitterHandle":"researchsquare","acdcEnabled":true,"dfaEnabled":false,"editorialSystem":"","reportingPortfolio":"","inReviewEnabled":false,"inReviewRevisionsEnabled":true},"keywords":"Fascia, Radiomics, Low Back Pain, Nomogram, MRI","lastPublishedDoi":"10.21203/rs.3.rs-4239014/v1","lastPublishedDoiUrl":"https://doi.org/10.21203/rs.3.rs-4239014/v1","license":{"name":"CC BY 4.0","url":"https://creativecommons.org/licenses/by/4.0/"},"manuscriptAbstract":"\u003cp\u003e\u003cstrong\u003ePurpose: \u003c/strong\u003eThe aim of this study was to analyze lumbar fascial changes in patients with lower back pain using lumbar spine soft tissue-based MR imaging radiomics.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eMethods:\u003c/strong\u003e We retrospectively analyzed the lumbar MRI of 380 patients with low back pain. Patients were randomly assigned to either the training (n = 187) or validation (n = 193) cohorts. Seven different machine learning algorithms were used to establish classification prediction models, and their prediction efficiency were evaluated with the best performance was selected as the final classification prediction model. Multivariate logistic regression analysis was used to create radiomics model and combined nomogram model and their predictive performance were evaluated using receiver operating characteristic (ROC) curves.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eResults:\u003c/strong\u003eAmong the seven machine learning models, the Lasso model exhibited the highest diagnostic efficiency with an AUC of 0.835. The radiomics nomogram integrated clinical and radiomics signature features, demonstrating strong performance in both the training and validation sets with AUC values of 0.97 and 0.96, respectively. AUC and DCA indicated that the radiomics nomogram model effectively diagnosed lumbar fasciitis.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eConclusion:\u003c/strong\u003e We developed an nomogram model that integrated clinical and radiomics features to help clinicians identify and predict low back fasciitis through soft tissue magnetic resonance.\u003c/p\u003e","manuscriptTitle":"Prediction of low back fasciitis by machine learning method based on radiomics","msid":"","msnumber":"","nonDraftVersions":[{"code":1,"date":"2024-05-13 15:57:30","doi":"10.21203/rs.3.rs-4239014/v1","editorialEvents":[{"type":"communityComments","content":0}],"status":"published","journal":{"display":true,"email":"[email protected]","identity":"researchsquare","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":true,"externalIdentity":"","sideBox":"","snPcode":"","submissionUrl":"/submission","title":"Research Square","twitterHandle":"researchsquare","acdcEnabled":true,"dfaEnabled":false,"editorialSystem":"","reportingPortfolio":"","inReviewEnabled":false,"inReviewRevisionsEnabled":true}}],"origin":"","ownerIdentity":"8635633f-f018-4a0a-9157-1993ca392032","owner":[],"postedDate":"May 13th, 2024","published":true,"recentEditorialEvents":[],"rejectedJournal":[],"revision":"","amendment":"","status":"posted","subjectAreas":[{"id":31632321,"name":"Health sciences/Diseases/Trauma"},{"id":31632322,"name":"Physical sciences/Physics/Information theory and computation"}],"tags":[],"updatedAt":"2024-07-23T11:11:15+00:00","versionOfRecord":[],"versionCreatedAt":"2024-05-13 15:57:30","video":"","vorDoi":"","vorDoiUrl":"","workflowStages":[]},"version":"v1","identity":"rs-4239014","journalConfig":"researchsquare"},"__N_SSP":true},"page":"/article/[identity]/[[...version]]","query":{"redirect":"/article/rs-4239014","identity":"rs-4239014","version":["v1"]},"buildId":"8U1c8b4HqxoKbykW_rLl7","isFallback":false,"isExperimentalCompile":false,"dynamicIds":[84888],"gssp":true,"scriptLoader":[]}