Prediction model for occurrence of anterior cervical approach postoperation adjacent segment degeneration of cervical spondylosis patient

Abstract

The invention discloses an anterior cervical approach postoperation adjacent segment degeneration occurrence prediction model for a cervical spondylosis patient, and belongs to the field of cervical spondylosis postoperation prediction. The invention discloses a model for predicting occurrence of anterior cervical approach postoperative adjacent segment degeneration of a cervical spondylosis patient. The model comprises a classifier and risk factors input into the classifier, the classifier is used for preprocessing and sketching a region of interest by acquiring a medical image of a target region of a patient, then extracting image omics features, inputting the image omics features into various machine learning models to calculate a C index, and selecting an optimal machine learning model as the classifier based on the C index; the risk factors are obtained through screening by performing single-factor and multi-factor logistic regression analysis on clinical features. Through the model, a clinician can be helped to avoid related risks to the greatest extent when a cervical spondylosis operation scheme is made, meanwhile, high-risk groups with ASD after an operation are recognized, and a basis is provided for implementation of targeted follow-up visit and early intervention.

Description

Technical Field

[0001] The present invention belongs to the field of postoperative prediction of cervical spondylosis, and particularly relates to a prediction model for adjacent segment degeneration after anterior cervical surgery in patients with cervical spondylosis. Background Art

[0002] Cervical spondylosis is a clinical syndrome characterized by degenerative changes in the cervical intervertebral disc, followed by pathological changes in adjacent tissue structures (such as bone hyperplasia and intervertebral disc stenosis), leading to dysfunction of surrounding tissues such as nerves and blood vessels. The patient's clinical manifestations correlate with imaging findings. Its occurrence is closely related to age-related degeneration. With aging, almost everyone develops imaging changes of cervical degeneration; however, not everyone experiences typical symptoms of neck pain or neurological deficits, which are usually caused by mechanical compression of the nerve roots, spinal cord, or blood vessels. Based on its pathological mechanisms and clinical manifestations, cervical spondylosis can be divided into radiculopathy, myelopathy, sympathetic cervical spondylosis, vertebral artery cervical spondylosis, and mixed cervical spondylosis. Cervical spondylotic myelopathy (CSM) is considered the most severe type of cervical spondylosis due to its high disability rate. It primarily results from spinal cord compression caused by factors such as cervical disc herniation, bone hyperplasia, hypertrophy of the ligamentum flavum, or ossification of the posterior longitudinal ligament, leading to spinal cord ischemia, edema, and even irreversible damage. Patients typically present with symptoms such as lower limb weakness, a feeling of walking on cotton wool, unsteady gait, numbness in the limbs, impaired fine motor skills in the hands, and hyperreflexia. Severe cases may develop quadriplegia or bladder dysfunction, often accompanied by neck pain and dizziness, significantly impacting patients' quality of life. Epidemiological studies have shown that approximately 10% of people aged 55 and older meet the clinical diagnostic criteria for CSM, and 85% of those aged 60 and older show signs of cervical degeneration on imaging studies. CSM is typically diagnosed based on a combination of clinical symptoms, physical examination, and imaging studies such as X-rays, computed tomography (CT), and magnetic resonance imaging (MRI). Researchers generally assess the neurological status of CSM patients and categorize their severity based on the modified Japanese Orthopaedic Association score (mJOA score, total score 0-17). This score is divided into three categories: mild neurological impairment (15-17 points), moderate neurological impairment (12-14 points), and severe neurological impairment (mJOA <12 points). For CSM patients with mild neurological impairment, conservative treatments such as cervical collar immobilization, medication, and physical therapy can generally be used, which can improve clinical symptoms to varying degrees. However, some patients who are refractory to conservative treatment or have moderate to severe symptoms require early surgical treatment to relieve spinal cord compression, stabilize the cervical spine, and prevent further deterioration of neurological function. Surgical approaches for CSM include anterior, posterior, and combined anterior and posterior approaches. For patients with single-segment and double-segment lesions, anterior cervical surgery is generally the preferred procedure.Anterior cervical surgery involves removing the diseased intervertebral disc, osteophytes, or vertebral body through an anterior approach to the neck, followed by the implantation of a fusion cage or titanium mesh to decompress the spinal cord and nerves and restore cervical stability. Anterior cervical discectomy and fusion (ACDF) and anterior cervical corpectomy decompression and fusion (ACCF) are the most common procedures. However, fusion of the surgical segment results in loss of range of motion (ROM) at that level, shifting the biomechanical load of the cervical spine to the adjacent, unfused segment. This can lead to long-term adjacent segment degeneration (ASD), also known as adjacent segment degeneration (CASD). This disease is generally classified into two categories: radiographic adjacent segment degeneration (RASD) and clinical adjacent segment degeneration (CASD). RASD refers to the appearance of radiographic signs of degeneration in the adjacent superior or inferior segments after cervical fusion surgery. These include X-ray findings of intervertebral disc narrowing, new osteophyte formation or increased pre-existing osteophytes, and ossification of the anterior longitudinal ligament. CT or MRI findings of adjacent disc herniation compressing the dura mater or decreasing the sagittal diameter of the spinal canal are also present. If, in addition to RASD, the patient develops new clinical symptoms and signs, such as neck pain, limb weakness, or numbness, this is considered CASD. The incidence of ASD after anterior cervical fusion surgery has varied widely in previous reports. Carrier CS et al., in a meta-analysis of 14 studies from the MEDLINE database, found that the average incidence of RASD in patients undergoing ACDF was 47.33% (range, 16%-96%), while the average incidence of CASD was 11.99% (range, 1.80-36.00%). Park JB et al. suggest that patients who develop adjacent vertebral ossification within 12 months after surgery are more likely to develop adjacent vertebral degeneration within 24 months. Several studies have shown that age, gender, smoking, preoperative adjacent segment degeneration, number of fused segments, C2-C7 sagittal vertical axis (SVA), C2-C7 Cobb angle, and T1 slope (T1S) are independent risk factors for the development of ASD after anterior fusion for cervical spondylotic myelopathy. ASD is a major cause of symptom recurrence and rehospitalization in patients after cervical spine surgery, impacting the long-term outcomes of surgery and the patient's long-term quality of life. Early identification and prediction of ASD is crucial for preventing its occurrence or taking prompt measures to avert further deterioration.

[0003] Radiomics is an analytical method based on medical imaging data. Its core process involves extracting high-throughput quantitative features from CT and MRI images, integrating machine learning and data mining techniques to analyze potential pathological information in the images to assist in disease diagnosis, prognosis prediction, and treatment decision-making. The core of this technology lies in converting medical images into quantifiable features that can reflect the biological characteristics of the disease or their correlation with clinical outcomes. Microscopic heterogeneity analysis provides a multidimensional basis for lesion characterization. Radiomics, through non-invasive, quantitative, and automated analysis, can provide accurate and comprehensive disease information and has significant clinical application value in areas such as early lesion identification, personalized treatment strategy development, and disease progression monitoring. Existing studies have demonstrated that predictive models based on radiomics have significantly improved diagnostic accuracy compared to traditional methods when applied to the early identification of diseases such as breast cancer, colorectal cancer, hepatocellular carcinoma, bladder cancer, lung cancer, and glioma. These studies provide empirical support for the translational value of radiomics in tumor diagnosis and prognostic assessment.

[0004] However, current research on imaging genomics in the field of cervical degenerative diseases and adjacent segment degeneration is still relatively limited. Summary of the Invention

[0005] In view of the deficiencies in the prior art, the purpose of the present invention is to provide a prediction model for adjacent segment degeneration after anterior cervical surgery in patients with cervical spondylosis, thereby solving the problems in the prior art.

[0006] The purpose of the present invention can be achieved through the following technical solutions:

[0007] A prediction model for adjacent segment degeneration after anterior cervical surgery in patients with cervical spondylosis, including:

[0008] Classifier: Medical images of the patient's target area are acquired, preprocessed, and the region of interest is delineated. Radiomics features are then extracted and fed into multiple machine learning models to calculate the C-index. Based on the C-index, the optimal machine learning model is selected as the classifier.

[0009] and risk factors input into the classifier; screened by univariate and multivariate logistic regression analysis of clinical characteristics.

[0010] Furthermore, the preprocessing includes intensity normalization, format conversion, fixed resolution resampling, N4 magnetic field correction and noise removal for medical images.

[0011] Furthermore, the imaging omics features include: geometric shape features, first-order statistical features, texture features and high-order features.

[0012] Furthermore, the machine learning models include: logistic regression model, naive Bayes model, support vector machine model, K-nearest neighbor model, random forest model, extreme random tree model, extreme gradient boosting model, lightweight gradient boosting computing model, gradient boosting model, adaptive enhancement model and multi-layer perception machine learning model.

[0013] Furthermore, the clinical characteristics include: age at surgery, smoking, hypertension, number of surgical segments, preoperative T1S, and ΔC2-C7 Cobb angle.

[0014] Furthermore, the risk factors include: smoking, hypertension, number of surgical segments and ΔC2-C7 Cobb angle.

[0015] Furthermore, the classifier is a multi-layer perception machine learning model.

[0016] A method for constructing a prediction model for adjacent segment degeneration after anterior cervical surgery in patients with cervical spondylosis comprises the following steps:

[0017] S1, preprocessing medical images by delineating regions of interest from medical images, extracting and screening radiomics features, and then inputting them into multiple machine learning models to calculate the C index, thereby selecting the optimal radiomics model as the classifier;

[0018] S2, by performing univariate and multivariate logistic regression analysis on clinical characteristics, risk factors were screened; and the risk factors were combined with the classifier to construct a prediction model.

[0019] A device for predicting adjacent segment degeneration after anterior cervical surgery in patients with cervical spondylosis, equipped with the above-mentioned prediction model.

[0020] A system for predicting adjacent segment degeneration after anterior cervical surgery in patients with cervical spondylosis, comprising:

[0021] CT scanning module: used to obtain medical images of the patient's target area:

[0022] Image segmentation module: used for preprocessing the medical image and outlining the region of interest;

[0023] Feature extraction module: used to extract radiomics features from the delineated region of interest;

[0024] Classifier screening module: Input radiomics features into multiple machine learning models, calculate the C index, and select the optimal model as the classifier;

[0025] Risk factor screening module: Univariate and multivariate logistic regression analysis of clinical characteristics was performed to screen risk factors;

[0026] Prediction module: input risk factors into the classifier and output prediction results.

[0027] Beneficial effects of the present invention:

[0028] This study uses high-throughput radiomics feature analysis methods, combined with clinically relevant risk factors, to construct a joint clinical-radiological prediction model. The aim is to assess the incidence of adjacent segment degeneration after anterior cervical surgery in patients with cervical spondylotic myelopathy and to validate the model's predictive efficacy. This model can help clinicians minimize associated risks when formulating surgical plans for cervical spondylosis, while also identifying high-risk groups for developing ASD after surgery, providing a basis for targeted follow-up and early intervention. BRIEF DESCRIPTION OF THE DRAWINGS

[0029] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, for ordinary technicians in this field, other drawings can be obtained based on these drawings without any creative work.

[0030] Figure 1 is a workflow diagram of model construction in an embodiment of the present invention;

[0031] Figure 2 is a distribution diagram of the imaging omics feature types of the present invention;

[0032] Figure 3 The present invention uses lasso regression for feature screening;

[0033] Figure 4 is the comparison of the AUC values ​​of the radiomics model constructed by the classifier of the present invention in the training set and the validation set respectively;

[0034] Figure 5 is the comparison of the AUC values ​​of the optimal radiomics model, clinical model, and combined model of the present invention in the training set and validation set respectively;

[0035] Figure 6 is a nomogram model of the combined model of the present invention;

[0036] Figure 7 is a comparison of the calibration curves of the optimal radiomics model, clinical model, and combined model of the present invention in the training set and validation set, respectively;

[0037] Figure 8 It is a comparison of the decision curves of the optimal radiomics model, clinical model and combined model of the present invention in the training set and the validation set respectively. DETAILED DESCRIPTION

[0038] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making any creative efforts shall fall within the scope of protection of the present invention.

[0039] Example 1

[0040] This example describes the construction of a prediction model for adjacent segment degeneration after anterior cervical surgery in patients with cervical spondylosis, the construction of a data set, and the evaluation of the model.

[0041] 1. Materials and methods

[0042] This is a single-center, retrospective, observational study. Clinical, imaging, and follow-up data were retrospectively collected from patients with cervical spondylotic myelopathy who underwent anterior cervical surgery at Zhongda Hospital, Southeast University, between January 1, 2015, and December 31, 2022. A total of 152 patients with CSM who underwent anterior cervical surgery were enrolled in this study based on the inclusion criteria. All CSM cases were randomly divided into a training set (n = 101) and a validation set (n = 51) using a 2:1 stratified sampling method. The study protocol was approved by the Ethics Committee of Zhongda Hospital, Southeast University (No. 2024ZDSYLL189-P01).

[0043] Inclusion criteria were as follows:

[0044] (1) Based on the patient's clinical symptoms, signs, and imaging examinations, a senior neurosurgeon's spine and spinal cord subspecialist diagnosed CSM; the diagnostic criteria for CSM are based on the expert consensus on the classification, diagnosis, and non-surgical treatment of cervical spondylosis (2018): 1. Clinical manifestations of typical cervical spinal cord damage, mainly limb movement disorders, sensory and reflex abnormalities. 2. Imaging examinations show clear signs of spinal cord compression, which correspond to clinical symptoms. 3. Exclude amyotrophic lateral sclerosis, intraspinal mass, acute spinal cord injury, subacute combined degeneration of the spinal cord, syringomyelia, chronic polyneuropathy, etc.

[0045] (2) All patients had undergone cervical spine X-ray, CT, and MRI examinations before and after surgery, and their imaging, demographic, and medical records were complete. Preoperative CT examinations must have been performed within 2 weeks before surgery;

[0046] (3) After admission, the patient received anterior cervical spine surgery, with the surgical segment being single or double; (Anterior cervical spine surgery includes ACDF, ACCF, and TDR.)

[0047] (4) Age ≥18 years;

[0048] (5) Follow-up time was 12 months to 24 months after surgery (inclusive);

[0049] (6) Follow-up by telephone obtained verbal informed consent and ethics authorization, and agreed to participate in this project.

[0050] Exclusion criteria:

[0051] (1) do not meet the diagnostic criteria for CSM;

[0052] (2) history of cervical spine surgery;

[0053] (3) more than two surgical levels;

[0054] (4) lack of complete demographic, imaging, and case data;

[0055] (5) Age <18 years old;

[0056] (6) Postoperative follow-up time <12 months;

[0057] (7) Do not agree to participate in this project.

[0058] All CSM patients underwent a conventional plain CT scan of the cervical spine before surgery using a GE Revolution CT, Philips Spectral CT, Siemens SOMATOM Force CT, or other CT scanners. Scanning parameters included a tube voltage of 120 kV, a tube current of 315 mA, an image acquisition matrix size of 512 × 512, a slice thickness of 5 mm, and a slice spacing of 5 mm.

[0059] Different scanning parameters of the equipment and images from different scanning devices will lead to image inconsistencies and feature differences. Image data preprocessing will directly affect the accuracy of ROI delineation, the repeatability of features and the generalization ability of the model. In this embodiment, a fixed resolution resampling method is used to ensure that images from different CT scanners or different scanning parameters have a uniform spatial resolution. All images are resampled and the voxel spacing is standardized to 333 mm. At the same time, the Z-score normalization method is used to avoid the influence of image intensity differences on feature extraction, the Gaussian filter method is used for noise removal, and the N4 correction method is used for image correction.

[0060] 2. Outline the area of ​​interest

[0061] In this embodiment, the vertebrae from the first cervical vertebra (the vertebrae are composed of the vertebral body, vertebral arch and seven processes) to the first thoracic vertebra (C1-T1) are used as the image segmentation targets. After the patient completes the CT scan, the patient's cervical spine CT tissue window image is exported from the Picture Archiving Communication System (PACS) in DICOM format, and the original image's DICOM format is converted into a software-recognizable NIfTI format using Mango software. A neurosurgeon with 10 years of work experience in spinal cord subspecialty uses 3D Slicer software (https: / / www.slicer.org; Version 5.6.1) to perform ROI semi-manual segmentation (such as Figure 2 As shown, Figure 2 (A and B are 3D images of the ROI, and C, D, and E are schematic diagrams of the ROI delineation on cervical spine CT images.) A radiologist with five years of experience in spinal CT diagnostics assessed the ROI for all patients. In case of disagreement, the ROI was re-delineated with the senior chief physician for consultation.

[0062] 3. Extract and filter radiomics features

[0063] After ROI delineation and image preprocessing were completed, radiomics features were extracted from all patient ROIs using Pyradiomics. These features included geometric shape features, first-order statistical features, texture features, and higher-order features. Texture features were extracted using various methods, including the Gray Level Co-occurrence Matrix (GLCM), Gray Level Size Zone Matrix (GLSZM), Gray Level Run Length Matrix (GLRLM), Neighbouring Gray Tone Difference Matrix (NGTDM), and Gray Level Dependence Matrix (GLDM). Detailed information on feature names and mathematical formulas can be found in the Pyradiomics documentation.

[0064] The extracted radiomic features were statistically examined using the Levene t test, and only those with a p-value less than 0.05 were retained. The Spearman rank correlation coefficient was used to calculate the correlation between features, and any feature with a correlation coefficient exceeding 0.9 between any two attributes was retained. For feature filtering, a recursive elimination method was used, which removed the most redundant features in the current set during each iteration. Subsequently, multivariate selection was performed using the least absolute shrinkage and selection operator (LASSO) regression method to select the most appropriate lambda value and, therefore, the most valuable feature subset. The dataset was randomly distributed in a 2:1 ratio, assigning it to either the training or test dataset. The training set was used to train model parameters using known data, while the test set was used for independent evaluation of model performance.

[0065] 4. Build radiomics models, clinical models, and combined clinical-radiological models

[0066] In this embodiment, 11 radiomics models were used, including: Logistic Regression (LR), Naive Bayes (NB), Support Vector Machines (SVM), K-Nearest Neighbor (KNN), Random Forest (RF), ExtraTress, eXtreme Gradient Boosting (XGBoost), Lightweight Gradient Boosting (LightGBM), Gradient Boosting (GBDT), Adaptive Boosting (AdaBoost), Multilayer Perceptron (MLP) Machine Learning (ML). To evaluate the predictive performance of these 11 models, ROC curves were generated for the training and validation cohorts. The area under the ROC curve (AUC), accuracy, sensitivity, and specificity were calculated. (classifier), and the results showed that the MLP imaging omics model performed well in the training set (AUC = 0.768, 95% CI: 0.66-0.87) and the validation set (AUC = 0.765, 95% CI: 0.62-0.91).

[0067] Univariate and multivariate logistic regression (LR) analyses were performed, with the occurrence of ASD after anterior cervical surgery in CSM patients as the dependent variable. Factors with a P value < 0.05 were selected from the univariate LR analysis and set as independent variables. Multivariate LR analysis was then performed to identify risk factors for ASD after anterior cervical surgery in CSM patients. Results: Univariate logistic regression analysis of the clinical data of CSM patients revealed six factors with P values ​​< 0.05 (Tables 1-4): age at surgery, smoking, hypertension, number of surgical levels, preoperative T1S, and ΔC2-C7 Cobb angle. These six factors were associated with the development of ASD after anterior cervical surgery in CSM patients. Multivariate logistic regression analysis was performed with age at surgery, smoking, hypertension, number of surgical segments, preoperative T1S and ΔC2-C7 Cobb angle as independent variables. After adjusting for potential confounding factors, it was found that smoking, hypertension, number of surgical segments and ΔC2-C7 Cobb angle were independent risk factors for ASD after anterior cervical surgery in patients with CSM. These four risk factors were selected.

[0068] The results of univariate logistic regression analysis are shown in Table 1 below:

[0069] Table 1 Univariate logistic regression analysis of clinical characteristics

[0070]

[0071] Multivariate logistic regression analysis is shown in Table 2

[0072] Table 2 Multivariate logistic regression analysis of clinical characteristics

[0073]

[0074] Finally, a joint clinical-radiological model was constructed by combining the radiomics model (optimal classifier) ​​with clinical risk factors, and the model's effectiveness was verified through validation. The four identified risk factors were then input into the optimal classifier to construct a joint clinical prognosis prediction model (i.e., a model for predicting adjacent segment degeneration after anterior cervical surgery in patients with cervical spondylosis).

[0075] 5. Model Evaluation

[0076] This study used receiver operating characteristic (ROC) analysis to validate the performance of radiomics, clinical, and combined clinical-radiological models. The predictive performance of each model in the training and validation sets was assessed by measuring the area under the curve (AUC). AUC values ​​range from 0 to 1, with higher values ​​indicating better model performance. The sensitivity, specificity, and 95% confidence interval (95% CI) of each model were calculated. Decision curve analysis (DCA) was used to quantify the clinical net benefit of each model at different risk thresholds to comprehensively evaluate its clinical application potential. A visual nomogram was constructed for the ASD prediction model: an integral conversion system was established using multivariate regression coefficients. The contribution of each predictor variable was quantified into a standardized score. The accumulated total score was then mapped to a predicted probability scale, enabling the quantitative conversion of biological indicators into clinical endpoints. This integral model objectively quantifies the weighted influence of each factor on the outcome variable. Finally, the predictive calibration of the nomogram was further verified by the calibration curve (CC), and its clinical decision-making assistance value was evaluated in combination with DCA.

[0077] Example 2

[0078] In this embodiment, specific experiments are conducted to verify the superiority of the prediction model proposed in the present invention.

[0079] 1. Randomized data set clinical data

[0080] Based on the inclusion and exclusion criteria, this study enrolled 152 patients with CSM. They were randomly divided into a training set (n = 101) and a validation set (n = 51) using a 2:1 ratio using stratified sampling. The training set consisted of 68 patients in the ASD-free group and 33 patients in the ASD group; the validation set consisted of 34 patients in the ASD-free group and 17 patients in the ASD group. Demographic and imaging data are detailed in Table 3.

[0081] Table 3 Clinical data baseline

[0082]

[0083] Table 3 Clinical data baseline table

[0084]

[0085] Table 3 Clinical data baseline table

[0086]

[0087]

[0088] 2. Extraction and screening of radiomics features

[0089] A total of 1834 radiomic features were extracted from the ROIs of 152 CSM patients using Pyradiomics, of which GLCM accounted for 24.0%, first-order features accounted for 19.6%, GLSZM accounted for 17.4%, GLRLM accounted for 17.4%, GLDM accounted for 15.3%, NGTDM accounted for 5.5%, and geometric state features accounted for 0.8% (e.g. Figure 2 shown).

[0090] All extracted imaging features were preliminarily screened based on the Levene-t test, and features with correlation were retained (p<0.05). Subsequently, the Spearman rank correlation coefficient was used to evaluate the correlation between features. If the correlation coefficient between any two features exceeded 0.9, one of the features was retained, and features with higher variance contribution were retained first to reduce multicollinearity. Then, the LASSO regression method was used to optimize the regularization parameter λ (Lambda) using ten-fold cross validation to screen key features with non-zero regression coefficients. Finally, 8 imaging features were screened out, such as Figure 3 As shown, Figure 3 A in is the regression cross validation curve, Figure 3 B in is the regression coefficient path diagram, Figure 3 C in the figure is the distribution map of the eight best radiomics features corresponding to the alpha value after LASSO dimensionality reduction.

[0091] 3. Construction of radiomics model

[0092] The model was constructed based on the screened radiomics features, and 11 ML algorithms were selected for comparison, including LR, NB, SVM, KNN, RF, ExtraTress, XGBoost, LightGBM, GBDT, AdaBoost, and MLP. The results are shown in the figure. Figure 4 As shown, Figure 4 A and B in the figure represent the comparison of each imaging omics model in the training set and validation set, respectively. The results show that the MLP imaging omics model has a better overall performance in the training set (AUC = 0.768, 95% CI: 0.66-0.87) and validation set (AUC = 0.765, 95% CI: 0.62-0.91).

[0093] 4. Comparison and verification of the effectiveness of the three models

[0094] The radiomics model, clinical model, and combined model were evaluated for predicting ASD. The results showed that the radiomics model achieved an AUC of 0.768 and 0.765 in the training and test sets, respectively, while the corresponding AUCs for the clinical model were 0.821 and 0.734. The predictive ability of the combined clinical-radiological model was further improved in the training set (AUC = 0.856) and validation set (AUC = 0.813).

[0095] in, Figure 5 A and B in the figure represent the comparison of each model in the training set and validation set, respectively. The results show that the AUC of the clinical-radiological omics joint model is significantly higher than that of the single model in both types of data sets.

[0096] In order to enhance the clinical practicality of the model, this example constructs a Nomogram to visualize the prediction results. Figure 6 As shown in the results, the model score was positively correlated with the risk of ASD after anterior cervical surgery.

[0097] Figure 7 is the calibration curve, Figure 7 A and B in the figure represent the comparison of each model in the training set and validation set, respectively. Through calibration curve analysis, the consistency between the predicted probability and the actual observation value of the joint model is better than that of a single model.

[0098] Figure 8 is the decision curve (DCA), Figure 8 A and B in the figure represent the comparison of each model in the training set and validation set, respectively. The decision curve further verifies its clinical advantage. The combined model shows a higher net benefit value and accuracy within the threshold probability range, suggesting that it has important clinical application potential in ASD risk stratification.

[0099] Throughout this specification, references to terms such as "one embodiment," "example," or "specific example" indicate that the specific features, structures, materials, or characteristics described in conjunction with that embodiment or example are included in at least one embodiment or example of the present invention. In this specification, schematic representations of these terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in any one or more embodiments or examples.

[0100] The basic principles, main features, and advantages of the present invention are shown and described above. Those skilled in the art should understand that the present invention is not limited to the above embodiments. The above embodiments and descriptions are merely illustrative of the principles of the present invention. Various changes and modifications may be made to the present invention without departing from the spirit and scope of the present invention, and such changes and modifications fall within the scope of the invention as claimed.

Claims

1. A prediction model for adjacent segment degeneration after anterior cervical surgery in patients with cervical spondylosis, characterized by: include: Classifier: Medical images of the patient's target area are acquired, preprocessed, and the region of interest is delineated. Radiomics features are then extracted and fed into multiple machine learning models to calculate the C-index. Based on the C-index, the optimal machine learning model is selected as the classifier. and risk factors input into the classifier; screened by univariate and multivariate logistic regression analysis of clinical characteristics.

2. The prediction model for adjacent segment degeneration after anterior cervical surgery for patients with cervical spondylosis according to claim 1, characterized in that: The preprocessing includes intensity normalization, format conversion, fixed resolution resampling, N4 magnetic field correction and noise removal for medical images.

3. The prediction model for adjacent segment degeneration after anterior cervical surgery for patients with cervical spondylosis according to claim 1, characterized in that: The imaging omics features include: geometric shape features, first-order statistical features, texture features and high-order features.

4. The prediction model for adjacent segment degeneration after anterior cervical surgery for patients with cervical spondylosis according to claim 1, characterized in that: The machine learning models include: logistic regression model, naive Bayes model, support vector machine model, K-nearest neighbor model, random forest model, extreme random tree model, extreme gradient boosting model, lightweight gradient boosting calculation model, gradient boosting model, adaptive enhancement model and multi-layer perception machine learning model.

5. The prediction model for adjacent segment degeneration after anterior cervical surgery for patients with cervical spondylosis according to claim 1, characterized in that: The clinical characteristics include: age at surgery, smoking, hypertension, number of surgical segments, preoperative T1S, and ΔC2-C7 Cobb angle.

6. A prediction model for adjacent segment degeneration after anterior cervical surgery in patients with cervical spondylosis according to claim 1 or 5, characterized in that: The risk factors include smoking, hypertension, number of surgical segments and ΔC2-C7 Cobb angle.

7. The prediction model for adjacent segment degeneration after anterior cervical surgery for patients with cervical spondylosis according to claim 1, characterized in that: The classifier is a multi-layer perceptron machine learning model.

8. A method for constructing a prediction model for adjacent segment degeneration after anterior cervical surgery in patients with cervical spondylosis, characterized in that: The following steps are involved: S1, preprocessing medical images by delineating regions of interest from medical images, extracting and screening radiomics features, and then inputting them into multiple machine learning models to calculate the C index, thereby selecting the optimal radiomics model as the classifier; S2, by performing univariate and multivariate logistic regression analysis on clinical characteristics, risk factors were screened; and the risk factors were combined with the classifier to construct a prediction model.

9. A device for predicting adjacent segment degeneration after anterior cervical surgery in patients with cervical spondylosis, characterized in that: Equipped with the prediction model according to any one of claims 1 to 7.

10. A system for predicting adjacent segment degeneration after anterior cervical surgery in patients with cervical spondylosis, characterized by: include: CT Scan Module: used to obtain medical images of the patient's target area: Image segmentation module: used for preprocessing the medical image and outlining the region of interest; Feature extraction module: used to extract radiomics features from the delineated region of interest; Classifier screening module: Input radiomics features into multiple machine learning models, calculate the C index, and select the optimal model as the classifier; Risk factor screening module: Univariate and multivariate logistic regression analysis of clinical characteristics was performed to screen risk factors; Prediction module: input risk factors into the classifier and output prediction results.

Images