Cervical spondylosis operation difficulty prediction device based on machine learning

Abstract

The invention discloses a prediction device for predicting the difficulty of a cervical spondylosis operation based on machine learning, and a cervical spondylosis operation difficulty prediction model is built in a memory and is a trained stacking model. According to the method, related historical medical records and data are selected, data processing is carried out through data integration, structuring, preprocessing and the like, and after medical variables are extracted under medical logic supervision, a multi-factor prediction model with the best effect is trained through a machine learning method; and performing weighted analysis based on the weight factors given by each independent difficulty factor to obtain a risk prediction score of the current operation, and finally obtaining the operation difficulty through the risk prediction score. By deploying the prediction model in a specific storage, a reference can be provided for a doctor to estimate the difficulty of the cervical spondylosis operation before the operation, so that the doctor can fully prepare in the perioperative period and further perform evaluation and early intervention, the operation process is smooth, and the set operation purpose is achieved.

Description

Technical Field

[0001] The present invention relates to the field of medical artificial intelligence, and in particular to a device for predicting the difficulty of cervical spondylosis surgery based on machine learning. Background Art

[0002] Degenerative cervical spondylosis refers to the chronic progressive deterioration of the bony and cartilaginous components of the cervical spine, which is usually related to age. Common causes include osteophyte formation, disc degeneration, thickening of the ligamentum flavum, and abnormal cervical curvature. These lesions can lead to spinal stenosis and spinal cord compression, ultimately leading to neuropathy and even disability in some cases. As age progresses, almost everyone has radiographic signs of degenerative changes in the cervical spine. When obvious neurological dysfunction, progressive cervical instability, or cervical deformity is observed clinically, surgical intervention may be necessary. Depending on the patient's posterior longitudinal ligament calcification, spinal cord compression, and the segment involved, combined with the patient's physical condition, one of the following three surgical approaches can be used: anterior cervical discectomy, decompression and fusion, posterior cervical laminectomy and fusion, and anterior and posterior combined decompression and fusion. Different surgical approaches bring different difficulties and challenges, so it is particularly important to conduct appropriate preoperative assessment of the difficulty of the operation. Currently, there is no recognized scoring system to determine the surgical difficulty of cervical spondylosis, which has led to young surgeons finding it difficult to safely navigate the learning curve without harming any patients.

[0003] The risk of cervical spine surgery is related to the patient's age, weight, gender, sepsis, respiratory distress, mechanical ventilation and other risk factors. The surgical risks of cervical spine patients in different countries and regions are not exactly the same, which may be related to the surgical method, medical level, etc. Therefore, when establishing a prediction model for the difficulty of cervical spine surgery, it is necessary to first conduct a case-control study to screen out independent risk factors related to surgical difficulty; secondly, calculate the relative risk of surgical risk based on the corresponding independent risk factors, and then derive the prediction model based on the calculated relative risk; finally, in order to verify the replicability (internal validity) and universality (external validity) of the model, the prediction model should also be internally and externally validated. Internal validation is to randomly select some samples from the existing specimens to test the predictive efficacy of the model, and external validation is to test the predictive efficacy of the model in other samples, but external validation often requires multicenter research, which is difficult to implement.

[0004] The use of machine learning (ML) to predict surgical difficulty is currently being used by young surgeons to better develop personalized surgical plans before surgery. Compared with traditional methods, machine learning reveals the underlying patterns in large and complex data sets, and by leveraging artificial intelligence, the latter can more effectively generate predictive models. Therefore, models based on machine learning are considered to be key tools for clinical analysis. Summary of the invention

[0005] The purpose of the present invention is to provide a method for constructing a prediction model for cervical spondylosis surgery difficulty based on machine learning to solve the problems raised in the above background technology.

[0006] To achieve the above object, the present invention provides the following technical solutions: First, a method for constructing a cervical spondylosis surgery difficulty prediction model based on machine learning is established, including the following steps: Step 1: Select relevant historical medical records and data, and process the data through data integration, structuring, and preprocessing; Step 2: Extract medical variables under medical logic supervision and screen out independent difficulty factors that predict the difficulty of cervical spondylosis surgery; Step 3: Train the best multi-factor prediction model through machine learning methods; Step 4: Verify and test the performance and threshold of the model in an independent validation set; Furthermore, the step 2: extracting medical variables under medical logic supervision, specifically including clinical characteristics, imaging data and laboratory test results of surgical patients, and performing statistically appropriate aggregation and derivation within a corresponding appropriate time window to ultimately form variables with a certain degree of representativeness.

[0007] Furthermore, the extracted medical variables are filtered through machine learning training systems such as LASSO regression and 10-fold cross validation, and high-quality features with the best prediction effect and stable performance are automatically screened from thousands of feature pools that characterize patient information to obtain various medical variables.

[0008] Furthermore, in the step three: training the best multi-factor prediction model through machine learning methods, the algorithms involved in the experiment include but are not limited to random forests (RF), decision trees (DT), XGBoost, elastic networks (ENet), multi-layer perceptrons (MLP), supportvector machines (SVM), LightGBM, k-nearest neighbors (KNN), stacking ensemblelearning, and logistic regression algorithms.

[0009] Furthermore, step four: for each individual model in the stacking system, the cervical spondylosis patient data were randomly divided into a training set and a test set, and based on the machine learning model developed from the training set, a formula was established to calculate the score of each patient using the selected variables, and then, the performance of each model was judged in the test data set using the receiver operating curve (ROC) to summarize the performance of the model at all possible thresholds and compare the areas under the ROC, and finally, the sample predictions were intuitively interpreted through SHAP analysis to demonstrate the effectiveness of the stacking model for interpreting surgical difficulty.

[0010] Furthermore, the technical difficulty includes normal and difficult, and the division logic of the technical difficulty is: A specified threshold is set based on the risk prediction score prior assessment, where: when the risk prediction score is less than the threshold, the technical difficulty is normal; When the risk prediction score is greater than or equal to the threshold, the technical difficulty is difficult.

[0011] Specifically, when the risk prediction score is 0 to 3, the technical difficulty is normal; When the risk prediction score is 4 to 10 points, the technical difficulty is difficult.

[0012] In a second aspect, the present invention provides an electronic device, comprising: a processor and a memory, wherein the memory A computer program is stored that can be called by a processor; the processor executes to implement the method steps of the first aspect by calling the computer program stored in the memory.

[0013] In a third aspect, the present invention provides a computer program product stored on a computer-readable medium, comprising a computer-readable program, which, when executed on an electronic device, provides a user input interface to implement the method steps of the first aspect. The beneficial effects of the cervical spondylosis surgery difficulty prediction system of the present invention are: The present invention utilizes clinical information such as patient electronic medical records and test results, combined with machine learning methods, to establish a stacking model for predicting the difficulty of cervical spondylosis surgery. This model is based on easily identifiable clinical features and is easy to adopt in clinical settings. The surgical difficulty is graded to provide guidance for young endoscopists in assessing the difficulty of cervical spondylosis surgery, so that doctors can further evaluate surgical patients and intervene early, and make adequate perioperative preparations. We applied techniques such as 10-fold cross-validation to fine-tune the model performance, and the scoring system showed good discrimination ability. BRIEF DESCRIPTION OF THE DRAWINGS

[0014] Figure 1 Method flow chart Figure 2 Variable screening diagram DETAILED DESCRIPTION

[0015] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. 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 creative work are within the scope of protection of the present invention.

[0016] This embodiment provides a prediction device for the difficulty of cervical spondylosis surgery based on machine learning, comprising a computer memory, a computer processor, and a computer program stored in the computer memory and executable on the computer processor, wherein a prediction model for the difficulty of cervical spondylosis surgery is stored in the computer memory, and the prediction model is obtained online or offline through the following steps: Setting up data research sample data collection From the data sample of the study, clinical data were obtained from the HIS system according to the case report form (CRF). The CRF consists of five modules, including clinical characteristics, laboratory test results, and imaging results. The complete CRF table contains clinical data of (^) variables. The candidate predictor variables are selected from the variables in the CRF. In addition, the data extracted from the hospital data system will be checked by spinal surgeons for quality assessment based on the original data to ensure the accuracy of data collection.

[0017] According to the principle of at least 10 to 20 events for each candidate prediction parameter, it is planned to select about 15 clinical features to be included in the prediction model. Therefore, it is estimated that the data set used to train the model requires at least 300 cervical spine surgery patients to meet the minimum sample size requirement for developing a prediction model for robust modeling. Of course, the number of positive samples in the data set can be larger than this example, including 500, 1,000, or even more.

[0018] Assessing the difficulty of surgery The definition of difficulty of cervical fusion decompression surgery has not yet been established, so a simple modification was made to the definition of Escal et al. The difficulty of surgery was evaluated based on six criteria: operation time, surgical method, ossification of the posterior longitudinal ligament of the cervical spine, decompression segment, intraoperative blood loss, and hospital stay. Each criterion was scored with a weighted coefficient, and the total score ranged from 0 to 10. The difficulty of surgery was divided into two levels: patients with a score of <3 points were classified as the normal group; patients with a score of ≥3 points were classified as the high difficulty group.

[0019] Variable selection Since the variables were not normally distributed, continuous data and categorical data were presented as medians with interquartile ranges (IQRs) and frequencies with proportions, respectively. t-tests were used to compare continuous variables and Wilcoxon rank-sum tests were used to compare categorical variables. Variables. In order to explore the predictive factors of surgical difficulty, data variables were first preliminarily screened by univariate analysis, and then the variables were further screened by LASSO regression algorithm and 10-fold cross validation. The weight index of each variable was analyzed, and the weight index below a certain threshold was abandoned. Candidate variables were obtained. Then, clinical experts reviewed the variables in combination with published literature and their experience. According to the opinions of experts and the weights of the above variable analysis, appropriate clinical characteristics were mainly selected from key clinical states: age, BMI, ESR, RCB, MONO, AST, Ccr, ALB, diabetes, 117 ALT, cervical spine instability (CSI), hypertension, and osteoporosis differences.

[0020] Training prediction models based on machine learning algorithms Before model training, missing values ​​were filled with the mean of the current values. Model building was performed on the obtained dataset. For this model, normal surgery data and high-risk surgery data were randomly divided into training and test sets, with a ratio of 75% and 25%, respectively. Prediction model training and evaluation The types of prediction models frequently used in the current literature were selected as our algorithms. Random forests (RF), decision trees (DT), XGBoost, elastic networks(ENet), multi-layer perceptrons (MLP), support vector machines (SVM), LightGBM, k-nearest neighbors (KNN), stacking ensemble learning, and logisticregression algorithms were used to simultaneously build the machine learning process. Validation and verification of model performance and thresholds were based on the machine learning models developed from the training set in order to optimize among different models for the development of decision support programs. In the optimization process, the probability formulas constructed by each model were used to calculate the predicted probability of each patient in the test set using the selected variables. Then, the receiver-operating curve (ROC) was used to judge the performance of each model in the test data set, and the areas under the ROC were compared to summarize the performance of the model at all possible thresholds, and the appropriate threshold was selected as the division of the difficulty level of future cervical spine surgery in the model prediction. After establishing several prediction models, the method of the present invention uses a computer program to quantitatively and qualitatively evaluate the difficulty score of cervical spine surgery. Preoperative warning of cervical spine surgery meets clinical needs and helps clinicians make appropriate and timely decisions.

[0021] As a stacking model, the cervical spondylosis surgery difficulty prediction model has a prediction advantage that other machine learning methods do not have, which greatly improves the prediction probability of cervical spondylosis surgery difficulty. When the above-mentioned cervical spondylosis surgery difficulty prediction model is trained online, the clinical feature data to be predicted received by the application each time is processed as a training sample to optimize and update the surgical adverse event prediction model. In addition, the cervical spondylosis surgery difficulty prediction model can also output clinical features that affect the difficulty of cervical spondylosis surgery, which can provide more references for clinicians when predicting the possibility of high-difficulty surgery.

[0022] The computer processor in this embodiment can be any type of processor, and the memory can be a random access memory (RAM), a read-only memory (ROM), a flash memory, a first-in-first-out memory (FIFO), a first-in-last-out memory (FILO), etc.

[0023] It will be apparent to those skilled in the art that the invention is not limited to the details of the exemplary embodiments described above and that the invention can be implemented in other specific forms without departing from the spirit or essential features of the invention. Therefore, the embodiments should be considered exemplary and non-limiting in all respects, and the scope of the invention is defined by the appended claims rather than the foregoing description, and it is intended that all variations falling within the meaning and scope of the equivalent elements of the claims be included in the invention. Any reference numeral in a claim should not be considered as limiting the claim to which it relates.

[0024] In addition, it should be understood that although the present specification is described according to implementation modes, not every implementation mode contains only one independent technical solution. This description of the specification is only for the sake of clarity. Those skilled in the art should regard the specification as a whole. The technical solutions in each embodiment may also be appropriately combined to form other implementation modes that can be understood by those skilled in the art.

Claims

1. A device for predicting the difficulty of cervical spondylosis surgery based on machine learning, comprising a computer memory, a computer processor, and a computer program stored in the computer memory and executable on the computer processor. It is characterized by: The computer memory stores a cervical spondylosis surgery difficulty prediction model, which includes a trained stacking model; When the computer processor executes the computer program, the following steps are implemented: Step 1: Select and receive relevant historical medical records and clinical data, and process the data through data integration, structuring, and preprocessing; Step 2: Extract medical variables under medical logic supervision and screen out independent difficulty factors that predict the difficulty of cervical spondylosis surgery; Step 3: Train the best multi-factor prediction model through machine learning methods; Step 4: Verify and test the performance and threshold of the model in an independent validation set.

2. The method for predicting the difficulty of cervical spondylosis surgery based on machine learning according to claim 1, characterized in that: The second step is to extract medical variables under the supervision of medical logic, including diagnosis, medical history, tests, examinations, chief complaints, physical signs, and medical advice, and to form variables with certain representativeness by statistically appropriately aggregating and deriving them within a corresponding appropriate time window.

3. The method for predicting the difficulty of cervical spondylosis surgery based on machine learning according to claim 2, characterized in that: The medical variables are extracted by a machine learning training system, which automatically selects high-quality features with the best prediction effect and stable performance from thousands of feature pools that characterize patient information to obtain various medical variables.

4. The method for predicting the difficulty of cervical spondylosis surgery based on machine learning according to claim 1, characterized in that: Step three: train the best multi-factor prediction model through machine learning methods and stack it into an integrated learning model.

5. The method for predicting the difficulty of cervical spondylosis surgery based on machine learning according to claim 1, characterized in that: In step 4, for each model, the patient data were randomly divided into a training set and a test set. Based on the machine learning model developed from the training set, a formula was established to calculate the score for each patient using the selected variables. Then, the performance of each model was judged in the test data set using the receiver operating curve (ROC) to summarize the performance of the model at all possible thresholds and compare the areas under the ROC. Finally, the sample predictions were intuitively interpreted through SHAP analysis to demonstrate the effectiveness of the stacked model for interpreting surgical difficulty.

6. The cervical spondylosis surgery difficulty prediction system according to claim 1, characterized in that: The technical difficulty includes normal and difficult, and the division logic of the technical difficulty is: A specified threshold is set based on the risk prediction score prior assessment, where: when the risk prediction score is less than the threshold, the technical difficulty is normal; When the risk prediction score is greater than or equal to the threshold, the technical difficulty is difficult; Specifically, when the risk prediction score is 0 to 3, the technical difficulty is normal; When the risk prediction score is 4 to 10 points, the technical difficulty is difficult.

7. An electronic device, characterized in that: include: A processor and a memory, wherein the memory stores a computer program that can be called by the processor; The processor calls the computer program stored in the memory to execute and implement the cervical spondylosis surgery difficulty prediction system as described in any one of claims 1 to 6.

8. A computer program product stored on a computer readable medium, characterized in that: It comprises a computer-readable program, which, when executed on an electronic device, provides a user input interface to implement a cervical spondylosis surgery difficulty prediction system as described in any one of claims 1 to 6.

Images