Personalized preoperative risk assessment method for spinal tumor resection surgery

Abstract

The invention discloses a personalized preoperative risk assessment method for a spinal tumor resection operation, and the method comprises the steps: carrying out the standardization processing of features extracted from the case data of a patient and the corresponding operation plan data, and obtaining the processed features; the method comprises the following steps: preprocessing medical image data to obtain preprocessed data, performing tumor segmentation on the data through a three-dimensional attention U-Net network to obtain a tumor voxel-level probability graph, and calculating a tumor volume; obtaining the type of the tumor through the preprocessed data, the double-flow deep network and the type discrimination network; respectively carrying out standardization processing on the type, the tumor volume, the extracted wettability score and the tumor position grade to form a third feature; obtaining basic weights of the processed features and the third features from a historical database, and then dynamically adjusting the basic weights to obtain a target weight; and performing weighted summation on the target weight, the processed feature and the third feature to obtain a risk score, and matching the risk score with a preset risk score to obtain a preoperative risk assessment level.

Description

Technical Field

[0001] The present invention relates to the technical field of preoperative risk assessment, and in particular to, but is not limited to, a personalized preoperative risk assessment method for spinal cord tumor resection surgery. Background Art

[0002] Spinal cord tumor resection surgery is a high-risk and complex operation in the field of neurosurgery, with a postoperative complication rate as high as 15%-30%, which may cause neurological deficits, cerebrospinal fluid leakage, and even permanent paralysis.

[0003] In related technologies, clinical preoperative risk assessment mainly relies on the doctor's experience and judgment, but this method has the main problems of data fragmentation, strong subjectivity in assessment, and insufficient prediction accuracy.

[0004] Therefore, how to quickly and accurately conduct preoperative evaluation of spinal cord tumor resection surgery has become an urgent problem to be solved. Summary of the Invention

[0005] In view of this, an embodiment of the present invention provides a personalized preoperative risk assessment method for spinal cord tumor resection surgery, which at least solves the problem that related technologies cannot quickly and accurately perform preoperative assessment of spinal cord tumor resection surgery.

[0006] According to a first aspect of an embodiment of the present invention, a personalized preoperative risk assessment method for spinal cord tumor resection surgery is provided, comprising: Standardizing the features extracted from the patient's case data and the corresponding surgical plan data to obtain a first feature and a second feature; Preprocessing the acquired medical imaging data of the patient to obtain preprocessed data, and performing tumor segmentation on the data based on a three-dimensional attention U-Net network to obtain a voxel-level probability map of the patient's tumor; Calculating the tumor volume based on the tumor voxel-level probability map, and obtaining the tumor type based on the preprocessed data, the two-stream deep network, and the type discrimination network; The type, the tumor volume, and the infiltration score and tumor location grade extracted from the pre-processed data are respectively standardized to form a third feature; Obtaining basic weights corresponding to the first feature, the second feature, and the third feature, respectively, from a historical database, and dynamically adjusting the basic weights according to the case data, the medical imaging data, and the surgical plan data, respectively, to obtain corresponding target weights; The target weight is weighted and summed with the first feature, the second feature, and the third feature respectively to obtain a risk score, and the risk score is matched with a preset risk score to obtain the patient's preoperative risk assessment level.

[0007] According to a second aspect of an embodiment of the present invention, there is provided an electronic device comprising: a processor, a memory, a communication interface and a communication bus, wherein the processor, the memory and the communication interface communicate with each other via the communication bus; the memory is used to store at least one executable instruction, and the executable instruction enables the processor to perform an operation corresponding to the method described in the first aspect.

[0008] According to a third aspect of an embodiment of the present invention, a computer storage medium is provided, on which a computer program is stored. When the program is executed by a processor, the method according to the first aspect is implemented.

[0009] According to the solution provided by an embodiment of the present invention, features extracted from a patient's case data and corresponding surgical plan data are standardized to obtain a first feature and a second feature; the acquired medical imaging data of the patient is preprocessed to obtain preprocessed data, and the data is subjected to tumor segmentation based on a three-dimensional attention U-Net network to obtain a voxel-level probability map of the patient's tumor; the tumor volume is calculated based on the tumor voxel-level probability map, and the tumor type is obtained based on the preprocessed data, a two-stream deep network, and a type discrimination network; the type, the tumor volume, and the infiltration score and tumor location grade extracted from the preprocessed data are standardized to form a third feature; basic weights corresponding to the first, second, and third features are obtained from a historical database, and the basic weights are dynamically adjusted based on the case data, the medical imaging data, and the surgical plan data to obtain corresponding target weights; the target weight is weighted and summed with the first, second, and third features to obtain a risk score, and the risk score is matched with a preset risk score to obtain a preoperative risk assessment level for the patient. During this process, multi-dimensional information, including patient medical records, medical imaging data, and surgical plan data, is integrated to extract characteristic parameters for each dimension. Unstructured data (such as medical imaging data) and qualitative data (such as surgical plan data) are quantified to reduce the influence of subjective experience. Furthermore, the weight distribution of characteristic parameters for each dimension is automatically adjusted based on the importance of different features in the historical database to meet individual needs and conduct personalized risk assessments for patients. Furthermore, by combining characteristic parameters from the three dimensions of patient medical records, imaging information, and doctor's surgical plan data, the basic weights are adjusted based on the individual patient's situation, allowing for more accurate risk assessments. BRIEF DESCRIPTION OF THE DRAWINGS

[0010] In order to more clearly illustrate the technical solutions in the embodiments of the present invention, the following briefly introduces the drawings required for describing the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. Those skilled in the art can also derive other drawings based on these drawings without inventive work, among which: Figure 1 A schematic flow chart of a personalized preoperative risk assessment method for spinal cord tumor resection surgery provided by an embodiment of the present invention; Figure 2 Schematic diagrams of the first, second and third features provided in embodiments of the present invention; Figure 3 A schematic structural diagram of an electronic device provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0011] In order to make the purpose, technical solutions and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, rather than all the embodiments. The following embodiments are used to illustrate the present invention, but are not used to limit the scope of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of the present invention.

[0012] In the following description, reference is made to “some embodiments”, which describes a subset of all possible embodiments, but it will be understood that “some embodiments” may be the same subset or different subsets of all possible embodiments and may be combined with each other without conflict.

[0013] It should be pointed out that the terms "first\second\third" involved in the embodiments of the present invention are only used to distinguish similar objects and do not represent a specific ordering of the objects. It can be understood that "first\second\third" can be interchanged with a specific order or sequence where permitted, so that the embodiments of the present invention described here can be implemented in an order other than that illustrated or described here.

[0014] Those skilled in the art will understand that, unless otherwise defined, all terms used herein (including technical and scientific terms) have the same meaning as commonly understood by those skilled in the art in the art to which the embodiments of the present invention pertain. It should also be understood that terms such as those defined in common dictionaries should be understood to have meanings consistent with their meanings in the context of the prior art and, unless specifically defined as herein, should not be interpreted in an idealized or overly formal sense.

[0015] Figure 1This is a flow chart of a personalized preoperative risk assessment method for spinal cord tumor resection surgery provided in an embodiment of the present invention. The personalized preoperative risk assessment method for spinal cord tumor resection surgery provided in an embodiment of the present invention can be executed by an electronic device, such as a computer, a server, etc.

[0016] like Figure 1 As shown in the following, a personalized preoperative risk assessment approach for spinal cord tumor resection surgery is presented, including: S101 , respectively standardize features extracted from the patient's case data and the corresponding surgical plan data to obtain a first feature and a second feature.

[0017] In an embodiment of the present invention, patient case data can be obtained from a hospital's HIS (Hospital Information System). Case data is a collection of information collected during medical treatment, including information about the patient's health status, disease diagnosis, treatment process, and outcomes. This data can include various types of information. A patient's surgical plan data typically includes a series of detailed plans and information designed to ensure the safety and effectiveness of the surgery. After obtaining the case data, features are extracted, including the patient's age, gender, BMI, previous surgical history, underlying diseases, allergies, comorbidity index, American Society of Anesthesiologists Physical Ability Score (ASA Score), and Spinal Injury Association Neurological Function Score (ASIA Neurological Function Score). These extracted features are then standardized and integrated to form the first feature. After obtaining the surgical plan data, features are extracted, including information about the surgical approach, resection extent, auxiliary techniques, and estimated surgical duration. These extracted features are then standardized and integrated to form the second feature.

[0018] For example, the patient's age, gender, BMI index, previous surgical history, underlying disease, allergy history, comorbidity index, ASA score, and ASIA neurological function score are standardized to obtain the standardized age. ,gender , BMI index , previous surgical history , underlying diseases History of allergies , ASA score , ASIA neurological function score , comorbidity index , integrating the above standardized features, we get the first feature, as follows: After standardizing the surgical approach, resection range, auxiliary techniques and estimated operation time, the standardized surgical approach was obtained. , resection range / , assistive technology , Estimated duration of surgery The above features after standardization are integrated to obtain the second feature, as follows: Among them, when performing standardization, the Min-Max normalization method can be used for standardization, or it can be standardized through one-hot encoding, or it can be standardized by Z-score standardization.

[0019] S102. Preprocess the acquired medical imaging data of the patient to obtain preprocessed data, and perform tumor segmentation on the preprocessed data based on a three-dimensional attention U-Net network to obtain a voxel-level probability map of the patient's tumor.

[0020] In embodiments of the present invention, a patient's medical imaging data refers to information about the patient's internal structure and function obtained through various imaging techniques. This data is crucial for diagnosing diseases, formulating treatment plans (such as surgical procedures), and evaluating treatment outcomes. Medical imaging data consists of data from different modalities, including T1-weighted imaging data (MRI T1), T2-weighted imaging data (MRI T2), and CT data. Preprocessing can include processes such as registration of the T1-weighted imaging data, T2-weighted imaging data, and CT data, to produce preprocessed data. A tumor voxel-level probability map provides the probability that each voxel belongs to a specific class (e.g., tumor or non-tumor tissue). A deep learning segmentation network, which can be a 3D attention U-Net network, is proposed. The 3D attention U-Net network is used to segment tumor regions on the preprocessed data, i.e., the preprocessed medical imaging data, ultimately outputting a sigmoid-activated voxel-level probability map.

[0021] S103. Calculate the tumor volume based on the tumor voxel-level probability map, and obtain the tumor type based on the preprocessed data, the two-stream deep network, and the type discrimination network.

[0022] In an embodiment of the present invention, a tumor volume calculation formula is proposed. The tumor voxel-level probability map and the tumor volume calculation formula can be combined to obtain the tumor volume. A two-stream deep network is used to fuse the multiple types of data contained in the data. The type discrimination network can determine the tumor type based on the fused data.

[0023] S104 , performing standardization processing on the type, tumor volume, and the infiltration score and tumor location grade extracted from the pre-processed data to form the third feature.

[0024] In an embodiment of the present invention, feature extraction is first performed on the preprocessed data to obtain the tumor location grade and infiltration score, and then the type, tumor volume, tumor location grade and infiltration score are standardized to obtain standardized features. Finally, the above standardized features are integrated to obtain the third feature.

[0025] For example, the tumor volume, tumor type, tumor location grade, infiltration score, and bone density are standardized to obtain the standardized tumor volume. , tumor type , Position classification , infiltration score , and finally perform standardized data integration to obtain the third feature: like Figure 2 As stated, Figure 2 Schematic diagram of the first feature, the second feature and the third feature provided in the embodiment of the present invention. Figure 2 In the algorithm, age, gender, and underlying diseases are extracted from case data to form the first feature, tumor volume, tumor type, and location grade are extracted from medical imaging data to form the second feature, and surgical approach and resection range are extracted from surgical plan data to form the third feature.

[0026] S105. Obtain basic weights corresponding to the first feature, the second feature, and the third feature from the historical database, and dynamically adjust the basic weights according to the case data, medical imaging data, and surgical plan data to obtain corresponding target weights.

[0027] In an embodiment of the present invention, the historical database integrates multi-source heterogeneous data from the hospital information system (HIS), picture archiving and communication system (PACS), surgical record system, and follow-up database, while enhancing rare medical record data and monitoring performance data. The above data include but are not limited to various features such as patient age, gender, BMI index, previous surgical history, underlying diseases, allergy history, comorbidity index, ASA score, ASIA neurological function score, tumor volume, tumor type, tumor location grade, infiltration score, surgical approach, resection range, and auxiliary technology. The correlation between the features is calculated, and the basic weight corresponding to each feature is obtained based on the correlation result. Finally, the basic weight corresponding to each feature contained in the first feature, the second feature, and the third feature is obtained from the historical database.

[0028] Furthermore, the original value corresponding to each feature is obtained from the case data, medical imaging data and surgical plan data, and the basic weight is adjusted based on the value to obtain the target weight corresponding to each feature included.

[0029] For example, the basic weights corresponding to the features included in the first feature, the second feature, and the third feature are obtained from the historical database as follows: Patient medical record data weights: (age weight 0.048, gender 0.022, BMI 0.032, previous surgical history 0.022, underlying disease 0.043, ASA 0.081, ASIA 0.043, comorbidity 0.037) Medical imaging data weights: (tumor volume 0.134, location grade 0.089, infiltration score 0.089) Surgical plan data weight: (resection range 0.083, operation duration 0.083) Among them, the sum of the weights of patient medical record data, medical imaging data, and surgical plan data is 1.

[0030] From the current case data of the patient, the patient's age is 20 years old. According to the following age adjustment formula, the age adjustment value is 1.0. The age adjustment formula is as follows: Where x is the patient, Adjust the value for the patient's age.

[0031] Furthermore, after the age adjustment value is obtained, the age adjustment value and the basic weight corresponding to the age are multiplied to obtain the target weight corresponding to the age.

[0032] The adjustment formula for body mass index is as follows: In the above formula, is the adjusted value for the patient's body mass index.

[0033] The adjustment formula for tumor volume is as follows: In the above formula, is the adjusted value for the patient's tumor volume.

[0034] The adjustment formula for tumor type is as follows: In the above formula, Adjusted values ​​for the patient's tumor type.

[0035] The adjustment formula for the infiltration score is as follows: In the above formula, The adjusted value for the patient's infiltration score.

[0036] The formula for adjusting the estimated duration of surgery is as follows: In the above formula, The adjusted value for the patient's estimated surgery duration.

[0037] The adjustment formula for the comorbidity index is as follows: In the above formula, is the adjusted value of the patient's comorbidity index.

[0038] The above formula is only a partial formula, and the adjustment of other basic weights is set according to different situations.

[0039] S106. Perform a weighted summation of the target weight and the first feature, the second feature, and the third feature to obtain a risk score, and match the risk score with a preset risk score to obtain a preoperative risk assessment level of the patient.

[0040] In an embodiment of the present invention, the target weight is weighted and summed with the first feature, the second feature, and the third feature to obtain a total risk score.

[0041] For example: Feature vector of medical record data: Feature vectors of medical imaging data: Feature vector of surgical plan data: Target weights for medical record data: (age weight 0.048, gender 0.022, BMI 0.032, previous surgical history 0.022, underlying disease 0.043, ASA 0.081, ASIA 0.043, comorbidity 0.037) Target weights for medical imaging data: (tumor volume 0.134, location grade 0.089, infiltration score 0.089) Target weights for surgical plan data: (resection extent 0.083, surgical duration 0.083) Weighted sum: Among them, RiskScore is the risk score.

[0042] Furthermore, the preset risk scores are as follows: Low risk: RiskScore ≤ 0.35 Medium risk: 0.35 <RiskScore≤0.65 High risk: RiskScore>0.65 The risk score calculated above ( ) is matched with the preset risk score to finally obtain the patient's preoperative risk assessment level.

[0043] It is understood that in the embodiments of the present invention, on the one hand, multi-dimensional information such as patient medical records, medical imaging data, and surgical plan data is integrated to extract characteristic parameters in each dimension; unstructured data (such as medical imaging data) and qualitative data (such as surgical plan data) are quantified to reduce the influence of subjective experience, and the weight distribution of characteristic parameters in each dimension is automatically adjusted based on the importance of different features in the historical database to meet individual needs and conduct personalized risk assessments for patients. On the other hand, by combining the characteristic parameters in the three dimensions of patient medical records, imaging information, and doctor's surgical plan data, the basic weights are adjusted according to the individual patient's situation to conduct a more accurate risk assessment.

[0044] In some embodiments of the present invention, preprocessing the acquired medical imaging data of the patient in S102 to obtain preprocessed data can be achieved through S1021 to S1022, which is explained in the following steps.

[0045] S1021 , respectively processing the T1-weighted imaging data, the T2-weighted imaging data, and the CT data through resampling, bias field correction, and grayscale normalization to obtain first T1-weighted imaging data, first T2-weighted imaging data, and first CT data.

[0046] In some embodiments of the present invention, the purpose of resampling is to adjust the spatial resolution of the image so that images acquired from different sources or at different time points have the same voxel size. The bias field effect refers to the phenomenon that the brightness gradient of the image changes due to the inhomogeneity of the magnetic field. Bias field correction is to eliminate the impact of this inhomogeneity so that the brightness in the image can truly reflect the properties of the tissue. Grayscale normalization is to make image data from different batches or sources comparable, which is achieved by adjusting the grayscale value distribution of the image. First, the T1-weighted imaging data, T2-weighted imaging data, and CT data are resampled, then bias field correction is performed, and finally grayscale normalization is performed to obtain the first T1-weighted imaging data, the first T2-weighted imaging data, and the first CT data.

[0047] S1022. Perform rigid registration on the first T1-weighted imaging data, the first T2-weighted imaging data, and the first CT data using a Powell algorithm based on mutual information to obtain first registered data; and obtain preprocessed data based on the first registered data. In embodiments of the present invention, the mutual information-based Powell algorithm is a commonly used technique for medical image registration. It uses mutual information as a similarity metric and employs an optimization algorithm to find optimal transformation parameters, thereby achieving registration between images of different modalities. Rigid registration involves finding one or more transformations (such as translations and rotations) that align multiple images without altering their shapes (i.e., without deformations such as scaling or shearing). The mutual information-based Powell algorithm performs rigid registration on the first T1-weighted imaging data, the first T2-weighted imaging data, and the first CT data, generating registered first data. This registered first data incorporates the first T1-weighted imaging data, the first T2-weighted imaging data, and the first CT data. The registered first data is then further processed to obtain preprocessed data.

[0048] In the embodiment of the present invention, S1023 can be implemented through S301 to S302, which is explained through the following steps.

[0049] S301 , performing elastic deformation correction on the registered first data according to a multi-resolution B-spline elastic registration algorithm to obtain corrected first data.

[0050] In some embodiments of the present invention, a multi-resolution B-spline elastic registration algorithm is primarily used to align or match multiple images. By performing detailed elastic deformation correction on the image data, highly accurate image alignment can be achieved even in the presence of complex and non-uniform deformations. Elastic deformation correction is further performed on the registered first data using the multi-resolution B-spline elastic registration algorithm to obtain corrected first data.

[0051] S302 : Perform subvoxel alignment processing on the corrected first data according to a bicubic spline interpolation algorithm to obtain preprocessed data.

[0052] In some embodiments of the present invention, bicubic spline interpolation can be applied in two directions to generate a smooth image scaling, rotation, or deformation effect. Subvoxel alignment is performed on the corrected first data according to the bicubic spline interpolation algorithm to obtain preprocessed data.

[0053] In some embodiments of the present invention, performing tumor segmentation on the data based on a three-dimensional attention U-Net network in S102 to obtain a voxel-level probability map of the patient's tumor can be achieved through S102A, which is explained through the following steps.

[0054] S102A. Input the preprocessed data into the encoder to obtain a latent feature map, and input the latent feature map into the transposed convolution layer and the spatial attention layer to obtain a voxel-level probability map; the output of the transposed convolution layer is the input of the spatial attention layer.

[0055] In some embodiments of the present invention, a 3D attention U-Net network includes an encoder and a decoder. The decoder includes a transposed convolutional layer and a spatial attention layer. The transposed convolutional layer increases the spatial size (i.e., height and width) of the input while adjusting the number of channels. Preprocessed data is first input into the encoder to generate a latent feature map. This latent feature map is then input into the transposed convolutional layer to generate a larger-dimensional latent feature map. This larger-dimensional latent feature map is then input into the spatial attention layer to generate a voxel-level probability map.

[0056] In some embodiments of the present invention, calculating the tumor volume based on the tumor voxel-level probability map in S103 can be implemented through S1031 to S1032, which is explained in the following steps.

[0057] S1031. Binarize the tumor voxel-level probability map using a set probability threshold to obtain an image mask; the image mask includes positive voxels and negative voxels.

[0058] S1032. Count the number of positive voxels and obtain the tumor volume based on the number and the tumor volume calculation formula.

[0059] In some embodiments of the present invention, a probability threshold is set to binarize the tumor voxel-level probability map, generating an image mask containing only 0 and 1 (1 represents a positive voxel predicted to be a tumor, and 0 represents a negative voxel). The number of voxels with a value of 1 in the image mask is counted, i.e., the number of positive voxels, and the number is substituted into the tumor volume calculation formula to calculate the tumor volume. The tumor volume calculation formula is as follows: in, represents the number of positive voxels in the mask; Indicates the voxel resolution in the x, y, and z directions in millimeters (mm); represents the standard deviation of the probability of boundary voxels.

[0060] In some embodiments of the present invention, obtaining the tumor type based on the preprocessed data, the dual-stream deep network, and the type discrimination network in S103 can be implemented through S103A to S103C, which is explained in the following steps.

[0061] S103A. Input the preprocessed data into a two-stream deep network for feature extraction to obtain a fourth feature; the fourth feature includes the tumor shape, tumor boundary, and internal structure of the tumor.

[0062] In some embodiments of the present invention, a two-stream deep network is constructed based on the 3D ResNet-50 algorithm, and the preprocessed data is input into the two-stream deep network to extract three-dimensional spatial features (tumor shape, boundary, internal structure) to obtain the fourth feature.

[0063] S103B. Performing a time series analysis on the enhanced first T1-weighted imaging data and the first T2-weighted imaging data based on a dynamic contrast enhancement sequence to obtain an enhanced dynamic feature; and splicing the fourth feature and the enhanced dynamic feature to obtain a spliced ​​feature.

[0064] In some embodiments of the present invention, the enhanced first T1-weighted imaging data and the first T2-weighted imaging data primarily refer to the use of a contrast agent, which can shorten the T1 and T2 relaxation times of surrounding tissues, thereby producing brighter signals on the image. The contrast agent is used to improve the quality of the first T1-weighted imaging data and the first T2-weighted imaging data, thereby enabling better observation of specific anatomical structures or pathological conditions. A time series analysis of the enhanced first T1-weighted imaging data and the first T2-weighted imaging data is performed based on a dynamic contrast enhancement sequence to obtain an enhanced dynamic feature. The fourth feature and the enhanced dynamic feature can be spliced ​​in a specific dimension to obtain a spliced ​​feature.

[0065] S103C. Input the concatenated features into a type discrimination network to obtain a probability distribution of tumor types, and obtain a predicted type based on the type probability distribution.

[0066] In some embodiments of the present invention, the type discrimination network can be a tumor type decision tree, and the tumor type can be schwannoma, meningioma, ependymoma, and hemangioblastoma. The spliced ​​features are input into the type discrimination network to obtain the tumor type probability distribution, and the tumor type with the highest probability can be used as the predicted tumor type of the patient.

[0067] Among them, the probability distribution expression of the four types of tumors calculated in the type discrimination network is: Wherein, c represents the tumor type; represents the transpose of the weight vector corresponding to the cth class; Indicates the strengthening dynamic characteristics; Indicates converting the linear combination result into a probability distribution, Represents the fourth feature, ensuring that the sum of all class probabilities is 1.

[0068] The final type discrimination network output is as follows: , respectively representing the probability of being a schwannoma, meningioma, ependymoma, and hemangioblastoma.

[0069] In some embodiments of the present invention, S104 also includes S201 to S204, which are explained through the following steps.

[0070] S201. Construct a spinal canal coordinate system using the preprocessed data, and calculate the average minimum distance between the tumor and the spinal cord based on the spinal canal coordinate system.

[0071] S202 : Classify the tumor based on the average minimum distance and a preset threshold to obtain a tumor location classification.

[0072] In some embodiments of the present invention, the patient's vertebral body is obtained from preprocessed data, and a spinal canal coordinate system is constructed with the center of the lower endplate of the vertebral body as the origin; the left direction as the positive X-axis; the ventrodorsal direction as the Y-axis, with the ventral side as the positive Y-axis; and the craniocaudal direction as the Z-axis, with the cranial side as the positive Z-axis. The total number of positive voxels in the tumor and spinal cord regions is further obtained, and the average minimum distance is calculated based on the total number of positive voxels. The preset thresholds are as follows: Substitute the calculated average minimum distance into the above-mentioned preset threshold formula to obtain the tumor location grade. To meet the minimum average distance ( ) is greater than This indicates that the tumor is located outside the spinal dura mater and is classified as grade 1. To satisfy the minimum average distance greater than Less than Indicates that the tumor appears , judged as level 2 To meet the minimum average distance less than Description appears in , judged as level 3.

[0073] S203 , dividing the pre-processed data into a marginal area and a core area according to the existing contour information of the tumor in the pre-processed data; and performing feature extraction on the marginal area to obtain multi-scale texture features.

[0074] In some embodiments of the present invention, the core region is defined as the interior of the tumor outline, and the edge region may extend 3 mm outward from the tumor boundary. Based on the existing tumor contour information in the preprocessed data, the preprocessed data is divided into the edge region and the core region. Feature extraction is performed on the edge region to obtain multi-scale texture features including the grayscale run length matrix (short run dominance) and the Haar wavelet energy spectrum.

[0075] S204. Estimate the surface fractal dimension of the edge area using the differential box counting method and multi-scale texture features.

[0076] S205 , calculating the entropy values ​​corresponding to the edge area and the core area respectively based on the multi-scale texture features, and calculating the wetness score based on the entropy values ​​and the surface fractal dimension.

[0077] In some embodiments of the present invention, surface fractal dimension is often used to describe the complexity and irregularity of an object's boundary or edge. Multi-scale texture features can be grayscale run length matrix (short run dominance) and Haar wavelet energy spectrum. The surface fractal dimension of the edge region is first estimated using differential box counting and multi-scale texture features. Then, based on the multi-scale texture features, the entropy values ​​corresponding to the edge region and the core region are calculated.

[0078] Among them, the infiltration scoring formula is proposed as follows: In the above formula, Scoring for infiltration, Represents the entropy value corresponding to the edge area, represents the entropy value corresponding to the core area, and D represents the surface fractal dimension.

[0079] Furthermore, the entropy value and the surface fractal dimension are substituted into the above-mentioned wettability score formula to obtain the wettability score.

[0080] Example 1: A personalized preoperative risk assessment method for spinal cord tumor resection 1. Application scenarios A 56-year-old male patient was diagnosed with intramedullary ependymoma at the T8 level of the thoracic spine and was scheduled for complete resection via a posterior median approach. The system of the present invention was used to perform a preoperative risk assessment.

[0081] 2. Data input and standardization 2.1 Patient medical record data input and standardization Age: 56 years old → Calculated according to the segmented normalization formula: Gender: Male → coded as −1 BMI: 28 → normalized to level 3 (25 ≤ BMI < 30) Previous surgical history: Lumbar discectomy 5 years ago (weighted ): Underlying diseases: hypertension (partially controlled, level=0.9) and diabetes (HbA1c=7.5, level=1.0): 2.2 Image Data Input and Standardization Tumor volume: The volume calculated by 3D Attention U-Net segmentation is 12.5ml, and the historical database average is , standard deviation , after standardization: Tumor type: ependymoma → one-hot encoding is [0,0,1,0] Location classification: intramedullary ( ) → Ordinal code is 2 Infiltration score: , the minimum value of the historical database is 0.2, and the maximum value is 0.9. After normalization: 2.3 Surgical plan input and standardization Surgical approach: posterior median approach → one-hot encoding [1,0,0] Resection range: full resection → one-hot encoding [1,0,0] Assistive technology: intraoperative ultrasound → one-hot encoding [0,1,0] Estimated surgery duration: 4 hours, historical mean 3.5 hours, standard deviation 0.8 hours, after standardization: 3. Dynamic weight calculation and risk assessment 3.1 Feature Vector Integration Feature vector of medical record data: Feature vectors of medical imaging data: Feature vector of surgical plan data: 3.2 Dynamic Weight Allocation According to the weight of historical database: Patient medical history weight: 0.35 (age weight 0.048, gender 0.022, BMI 0.032, previous surgical history 0.022, underlying disease 0.043, ASA 0.081, ASIA 0.043, comorbidity 0.037) Image data weight: 0.40 (tumor volume 0.134, location grade 0.089, infiltration score 0.089) Surgical plan weight: 0.25 (extremity of resection 0.083, duration of surgery 0.083) 3.3 Comprehensive score calculation Calculate the total VaR by weighted summation: 3.4 Risk Level Output Classification by threshold: Low risk: RiskScore ≤ 0.35 Medium risk: 0.35 <RiskScore≤0.65 High risk: RiskScore>0.65 The total risk value in this case was 0.623, and the system output was medium risk, suggesting the need to optimize the surgical plan (such as adding neuroelectrophysiological monitoring) and strengthen intraoperative monitoring.

[0082] Reference Figure 3 , shows a schematic structural diagram of an electronic device according to an embodiment of the present invention. The specific embodiment of the present invention does not limit the specific implementation of the electronic device.

[0083] like Figure 3 As shown, the electronic device may include: a processor (processor) 502, a communications interface (Communications Interface 504), a memory (memory) 506, and a communication bus 508.

[0084] in: The processor 502 , the communication interface 504 , and the memory 506 communicate with each other via a communication bus 508 .

[0085] The communication interface 504 is used to communicate with other electronic devices or servers.

[0086] The processor 502 is configured to execute the program 510 , and specifically may execute the relevant steps in the above method embodiment.

[0087] Specifically, the program 510 may include program codes, which include computer operation instructions.

[0088] Processor 502 may be a central processing unit (CPU), an application-specific integrated circuit (ASIC), or one or more integrated circuits configured to implement embodiments of the present invention. The one or more processors included in a smart device may be of the same type, such as one or more CPUs, or different types, such as one or more CPUs and one or more ASICs.

[0089] The memory 506 is used to store the program 510. The memory 506 may include a high-speed RAM memory, and may also include a non-volatile memory (non-volatile memory), such as at least one disk storage.

[0090] The program 510 may be specifically configured to enable the processor 502 to execute operations corresponding to the methods described in the above method embodiments.

[0091] The specific implementation of each step in program 510 can be found in the corresponding descriptions of the corresponding steps and units in the above-mentioned method embodiments, and will not be repeated here. Those skilled in the art will clearly understand that for the convenience and brevity of description, the specific working processes of the above-mentioned devices and modules can refer to the corresponding process descriptions in the above-mentioned method embodiments, and will not be repeated here.

[0092] It should be pointed out that, according to the needs of implementation, the various components / steps described in the embodiments of the present invention can be split into more components / steps, or two or more components / steps or partial operations of components / steps can be combined into new components / steps to achieve the purpose of the embodiments of the present invention.

[0093] The methods according to the embodiments of the present invention described above can be implemented in hardware, firmware, or as software or computer code that can be stored on a recording medium (such as a CD ROM, RAM, floppy disk, hard disk, or magneto-optical disk), or as computer code originally stored on a remote recording medium or non-transitory machine-readable medium downloaded over a network and then stored on a local recording medium. Thus, the methods described herein can be processed by such software stored on a recording medium using a general-purpose computer, a dedicated processor, or programmable or dedicated hardware (such as an ASIC or FPGA). It will be understood that a computer, processor, microprocessor controller, or programmable hardware includes a storage component (e.g., RAM, ROM, flash memory, etc.) that can store or receive software or computer code. When the software or computer code is accessed and executed by the computer, processor, or hardware, the methods described herein are implemented. Furthermore, when a general-purpose computer accesses the code for implementing the methods described herein, the execution of the code transforms the general-purpose computer into a dedicated computer for performing the methods described herein.

[0094] Those skilled in the art will appreciate that the units and method steps of each example described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are performed in hardware or software depends on the specific application and design constraints of the technical solution. Professionals and technicians can use different methods to implement the described functions for each specific application, but such implementation should not be considered to be beyond the scope of the embodiments of the present invention.

[0095] The above implementation methods are only used to illustrate the embodiments of the present invention, and are not intended to limit the embodiments of the present invention. Ordinary technicians in the relevant technical field may make various changes and modifications without departing from the spirit and scope of the embodiments of the present invention. Therefore, all equivalent technical solutions also fall within the scope of the embodiments of the present invention, and the scope of patent protection of the embodiments of the present invention should be defined by the claims.

Claims

1. A personalized preoperative risk assessment method for spinal cord tumor resection surgery, characterized in that: include: Standardizing the features extracted from the patient's case data and the corresponding surgical plan data to obtain a first feature and a second feature; Preprocessing the acquired medical imaging data of the patient to obtain preprocessed data, and performing tumor segmentation on the preprocessed data based on a three-dimensional attention U-Net network to obtain a voxel-level probability map of the patient's tumor; Calculating the tumor volume based on the tumor voxel-level probability map, and obtaining the tumor type based on the preprocessed data, the two-stream deep network, and the type discrimination network; The type, the tumor volume, and the infiltration score and tumor location grade extracted from the pre-processed data are respectively standardized to form a third feature; Obtaining basic weights corresponding to the first feature, the second feature, and the third feature, respectively, from a historical database, and dynamically adjusting the basic weights according to the case data, the medical imaging data, and the surgical plan data, respectively, to obtain corresponding target weights; The target weight is weighted and summed with the first feature, the second feature, and the third feature respectively to obtain a risk score, and the risk score is matched with a preset risk score to obtain the patient's preoperative risk assessment level.

2. The method according to claim 1, characterized in that The medical imaging data includes T1-weighted imaging data, T2-weighted imaging data and CT data; Preprocessing the acquired medical imaging data of the patient to obtain preprocessed data includes: The T1-weighted imaging data, the T2-weighted imaging data, and the CT data are processed respectively by resampling, bias field correction, and grayscale normalization to obtain first T1-weighted imaging data, first T2-weighted imaging data, and first CT data; The first T1-weighted imaging data, the first T2-weighted imaging data, and the first CT data are rigidly registered using a Powell algorithm based on mutual information to obtain registered first data; and the preprocessed data is acquired based on the registered first data.

3. The method according to claim 2, characterized in that The acquiring the preprocessed data based on the registered first data includes: performing elastic deformation correction on the registered first data according to a multi-resolution B-spline elastic registration algorithm to obtain corrected first data; Subvoxel-level alignment processing is performed on the corrected first data according to a bicubic spline interpolation algorithm to obtain the preprocessed data.

4. The method according to claim 1, wherein The three-dimensional attention U-Net network includes an encoder and a decoder, and the decoder includes a transposed convolution layer and a spatial attention layer; The step of performing tumor segmentation on the data based on a three-dimensional attention U-Net network to obtain a voxel-level probability map of the patient's tumor includes: The preprocessed data is input into the encoder to obtain a latent feature map, and the latent feature map is input into the transposed convolution layer and the spatial attention layer to obtain a voxel-level probability map; the output of the transposed convolution layer is the input of the spatial attention layer.

5. The method according to claim 1, wherein Calculating the tumor volume based on the tumor voxel-level probability map includes: Binarizing the tumor voxel-level probability map using a set probability threshold to obtain an image mask; the image mask includes positive voxels and negative voxels; The number of the positive voxels is counted, and the tumor volume is obtained based on the number and a tumor volume calculation formula.

6. The method according to claim 1, characterized in that Obtaining the type of tumor based on the preprocessed data, the dual-stream deep network, and the type discrimination network includes: Inputting the preprocessed data into the two-stream deep network for feature extraction to obtain a fourth feature; the fourth feature includes the tumor shape, tumor boundary and internal structure of the tumor; performing a time series analysis on the enhanced first T1-weighted imaging data and the first T2-weighted imaging data based on a dynamic contrast enhancement sequence to obtain an enhanced dynamic feature; and splicing the fourth feature and the enhanced dynamic feature to obtain a spliced ​​feature; The spliced ​​features are input into the type discrimination network to obtain the type probability distribution of the tumor, and the predicted type is obtained based on the type probability distribution.

7. The method according to claim 1, characterized in that Before normalizing the type, the tumor volume, and the infiltration score and tumor location grade extracted from the pre-processed data to form the third feature, the method further includes: constructing a spinal canal coordinate system using the preprocessed data, and calculating an average minimum distance between the tumor and the spinal cord based on the spinal canal coordinate system; Classifying the tumor based on the average minimum distance and a preset threshold to obtain the tumor location classification; According to the existing tumor contour information in the preprocessed data, the preprocessed data is divided into a marginal area and a core area; and features are extracted from the marginal area to obtain multi-scale texture features; estimating the surface fractal dimension of the edge region by using a differential box counting method and the multi-scale texture feature; The entropy values ​​corresponding to the edge region and the core region are calculated based on the multi-scale texture features, and the wettability score is calculated based on the entropy values ​​and the surface fractal dimension.

Images