A method for constructing a teaching parotid tumor enhancement model based on clustering algorithm

Abstract

The present invention relates to the field of digital teaching technology for oral medicine, and in particular to a method for constructing a parotid gland tumor enhancement model for teaching based on a clustering algorithm. The method includes S1, medical data integration; S2, cluster classification; S3, finding typical medical records; S4, invasion boundary enhancement processing; S5, synthesizing a three-dimensional parotid gland tumor model; and S6, loading a virtual oral medicine teaching platform. The present invention is based on the K-means clustering algorithm, classifies the parotid gland tumor medical records on the cloud platform, selects the most typical data points in each cluster, traces the case information and MRI scan image group represented by it, establishes a three-dimensional parotid gland tumor model of the corresponding disease and loads it into the virtual oral medicine teaching platform. Students can intuitively learn based on the three-dimensional parotid gland tumor model established based on real cases. At the same time, the clustering algorithm is also provided with an update mechanism, which can continuously adjust the model to avoid the problem of typical medical record lag in conventional teaching platforms.

Description

Technical Field

[0001] The present invention relates to the technical field of digital oral medicine teaching, and in particular to a method for constructing a parotid tumor enhancement model for teaching based on a clustering algorithm. Background Art

[0002] MRI scanning methods can be simply divided into two categories: conventional scanning and functional scanning. Conventional scanning mainly reflects anatomical morphology; functional scanning reflects functional information such as human metabolism and blood flow in different ways. Conventional scanning includes T1-weighted, T2-weighted imaging, angiography imaging, and dynamic enhancement imaging. T1-weighted (T1WI) and T2-weighted (T2WI) are the most commonly used and basic conventional scanning sequences. Almost all clinical MRI examinations will include T1-weighted and T2-weighted scans. T1 weighting can highlight the anatomical structure, while T2 weighting can highlight the lesions.

[0003] In the conventional oral medicine teaching process, teachers usually teach students based on typical cases of parotid gland tumors and MRI images shown in textbooks. However, as time goes by, the typical cases in textbooks have become lagging and cannot fully meet the needs of clinical teaching. In addition, during the teaching process, the planar information in the textbooks can hardly intuitively reflect some subtle characteristics of parotid gland tumors, which makes it difficult for beginners to learn the structure of oral tumors. Summary of the Invention

[0004] Other features and advantages of the present invention will be described in the following description, some of which will be directly apparent from the description or further understood through practice of the present invention. The objectives and other advantages of the present invention can be achieved and obtained through the specific structures clearly pointed out in the description and other drawings.

[0005] The purpose of the present invention is to overcome the above shortcomings and provide a method for constructing a teaching parotid tumor enhancement model based on a clustering algorithm. max A three-dimensional coordinate system was established, the medical records of parotid gland tumors on the cloud platform were classified, and disease clusters were constructed. The most typical data points in each cluster were selected, and the case information and MRI scan image groups represented by them were traced. According to the sequence information in the MRI scan image group, a three-dimensional maxillofacial soft tissue model and a three-dimensional bone model were established. The two were integrated into a three-dimensional parotid gland tumor model of the corresponding disease and loaded into the virtual oral medicine teaching platform. On the virtual oral medicine teaching platform, students can intuitively learn the three-dimensional parotid gland tumor model established based on real cases. At the same time, the clustering algorithm is also equipped with an update mechanism to continuously adjust the model to avoid the problem of typical medical record lag in conventional teaching platforms.

[0006] The present invention provides a method for constructing a teaching parotid tumor enhancement model based on a clustering algorithm, comprising:

[0007] S1. Medical data integration: Eligible patients with parotid gland tumors were selected and their electronic medical records were uploaded to the cloud platform via a device. The electronic medical records of each parotid gland tumor patient were at least bound to the MRI scan image set and the T1WI, ADC1 value, T2WI, and ADC2 value from the MRI test results.

[0008] S2. Cluster classification: Download electronic medical records from the cloud platform in batches, obtain T1WI, ADC1 value, T2WI and ADC2 value in the electronic medical records, and fit the maximum volume V of the tumor based on the ADC1 value and ADC2 value. max , according to T1WI, T2WI and V max The K-means clustering algorithm was used to analyze the MRI results of each parotid tumor patient and a three-dimensional coordinate system was established.

[0009] S3. Finding typical medical records: Set k cluster centers and calculate the center point of each cluster center, perform basic classification of parotid gland tumor symptoms, find the data point closest to the cluster center, and use the medical record corresponding to this data point as the typical medical record of the disease category. Trace the MRI scan image group in the electronic medical record information;

[0010] S4. Invasion boundary enhancement processing: The MRI scan image group is subjected to noise reduction and smoothing processing, and edge detection and edge enhancement are performed to enhance the invasion boundary and show the tumor morphology, thereby obtaining an enhanced image group;

[0011] S5. Synthesize a 3D parotid gland tumor model: Synthesize a 3D craniomaxillofacial soft tissue model based on the conventional MRI scan images in the enhanced image group. Then, synthesize a virtual CT image based on the ultrashort echo sequence MRI scan images in the enhanced image group. Generate a 3D craniomaxillofacial skeletal model based on the virtual CT image. Integrate the 3D craniomaxillofacial skeletal model and the 3D craniomaxillofacial soft tissue model to form a 3D parotid gland tumor model.

[0012] S6. Load the virtual oral medicine teaching platform: Load the generated three-dimensional parotid gland tumor model into the virtual oral medicine teaching platform. Different types of parotid gland tumor models are classified into corresponding directories. The teacher can click the directory or enter the coordinate point to call out the corresponding model.

[0013] In some embodiments, in step S1, the ADC1 value is the longitudinal cross-sectional area S1 of the tumor obtained during T1WI imaging, and the ADC2 value is the transverse cross-sectional area S2 of the tumor obtained during T2WI imaging;

[0014] In step S2, the maximum volume V of the tumor is fitted based on the ADC1 and ADC2 values.max The specific steps are:

[0015] The longitudinal cross-sectional area S1 of the tumor was obtained based on the ADC1 value, and the transverse cross-sectional area S2 of the tumor was obtained based on the ADC2 value;

[0016] An ellipsoid was fitted based on the longitudinal cross-sectional area S1 and the transverse cross-sectional area S2 to represent the shape of the tumor;

[0017] The volume V of the ellipsoid is calculated based on S1 and S2. The specific formula is:

[0018]

[0019] Since the volume calculation of an ellipsoid requires at least three cross-sectional area calculations, multiple results will be obtained by only calculating two cross-sectional area calculations. The maximum volume value V is selected from the multiple results. max As a parameter entering the clustering algorithm.

[0020] In some embodiments, based on T1WI, T2WI and V max The specific steps to establish a three-dimensional coordinate axis are:

[0021] According to the signal distribution of T1WI, the X-axis is divided into three segments: low, medium, and high signal. Each signal segment is further divided into three segments: uniform, uneven, and mixed, for a total of 9 segments.

[0022] Similar to T1WI signals, T2WI signals also divide the Y axis into 9 segments according to the distribution;

[0023] Z axis according to V max The data of the value establishes the coordinates;

[0024] For the T1WI signal and T2WI signal in the MRI detection results, the midpoint of each segment is used as the coordinate information of the signal actually located in the three-dimensional coordinate axis.

[0025] In some embodiments, in step S3, the specific steps of classifying parotid gland tumor conditions using the Mahalanobis distance K-means clustering algorithm are:

[0026] S31, randomly select 6 data points as cluster centers;

[0027] S32. Assigning each data point to the nearest cluster center based on Mahalanobis distance, wherein the Mahalanobis distance is calculated using the inverse of the covariance;

[0028] S33, recalculate the centroid of each cluster as the new cluster center so that its total square distance is minimized;

[0029] S34, repeat steps S32-S33 until convergence;

[0030] S35. Obtain six clusters. Based on the clusters, under the guidance of professional doctors, identify the parotid gland tumor type in each cluster. Specific types include: Warthin tumor, pleomorphic adenoma, malignant tumor, basal cell adenoma, parotid lymphoma, and abnormal medical records. Abnormal medical records will be excluded and not included in the data collection.

[0031] S36. Calculate the cluster centers of the five clusters excluding the abnormal medical records, calculate the data point closest to each cluster center using the Mahalanobis distance, use the data point as the typical data of this type of parotid gland tumor disease, record the data point coordinates, and trace back the MRI scan image group in the electronic medical record information.

[0032] In some embodiments, in step S36, a manual screening step is required for the nearest data point under each disease classification, and a professional doctor identifies the nearest data point found. If it does not have typical tumor characteristics or is a complex medical record of a collection of multiple types of tumors or even a classification error, the next nearest data point is selected sequentially until the data point is available.

[0033] In some embodiments, in step S4, the MRI scan image group includes conventional MRI scan images and ultra-short echo sequence MRI scan images. The original scan image data, T1WI, ADC1 value, T2WI and ADC2 value can be obtained from the conventional MRI scan images, which are used to establish a three-dimensional craniofacial soft tissue model in step S5; the ultra-short echo sequence MRI scan images are synthesized into a virtual CT image in step S5 through UTE scan image data and converted into a three-dimensional craniofacial bone model.

[0034] In some embodiments, in step S4, the specific steps of performing invasion boundary enhancement processing on the conventional MRI scan image and the ultrashort echo sequence MRI scan image are:

[0035] S41, taking the conventional MRI scan image and the ultrashort echo sequence MRI scan image as a first grayscale image group;

[0036] S42, performing noise reduction and smoothing processing on the first grayscale image group to obtain a second grayscale image group;

[0037] S43, performing edge detection on the second grayscale image group using a prewitt operator to obtain an edge-detected image group;

[0038] S44. Perform edge enhancement on the edge detection image group through the edge function to enhance the invasion boundary and show the tumor morphology, thereby obtaining an enhanced image group to facilitate the establishment of a three-dimensional model.

[0039] In some embodiments, during the process of establishing a three-dimensional craniofacial soft tissue model, data of the parotid gland region and tumor boundary are extracted based on the enhanced invasion boundary to generate maxillofacial soft tissue data containing texture features; during the process of establishing a three-dimensional craniofacial skeletal model, bone tissue is differentiated by segmenting the upper and lower jaws and adjacent bony structures to generate three-dimensional maxillofacial hard tissue data. In some embodiments, in step S5, the three-dimensional craniofacial skeletal model and the three-dimensional craniofacial soft tissue model are imported into a medical imaging processing platform and integrated to form a three-dimensional parotid gland tumor model. If there is overlap in some regional models, the three-dimensional craniofacial skeletal model is used as a reference, and the maxillofacial soft tissue is adjusted based on T1WI, ADC1 value, T2WI, and ADC2 value.

[0040] In some embodiments, in step S6, the electronic medical record information on the cloud platform is updated at regular intervals, and the clustering algorithm synchronously updates the position of the data point in the three-dimensional coordinate system to find the nearest point. If the data point changes, a new three-dimensional parotid tumor model is established and loaded into the corresponding parotid tumor disease type directory of the virtual oral medicine teaching platform according to its classification, and the coordinates of the nearest data point are updated.

[0041] By adopting the above technical solution, the beneficial effects of the present invention are:

[0042] The present invention is based on K-means clustering algorithm, according to T1WI, T2WI and V max A three-dimensional coordinate system was established, the medical records of parotid gland tumors on the cloud platform were classified, and disease clusters were constructed. The most typical data points in each cluster were selected, and the case information and MRI scan image groups represented by them were traced. According to the sequence information in the MRI scan image group, a three-dimensional maxillofacial soft tissue model and a three-dimensional bone model were established. The two were integrated into a three-dimensional parotid gland tumor model of the corresponding disease and loaded into the virtual oral medicine teaching platform. On the virtual oral medicine teaching platform, students can intuitively learn the three-dimensional parotid gland tumor model established based on real cases. At the same time, the clustering algorithm is also equipped with an update mechanism to continuously adjust the model to avoid the problem of typical medical record lag in conventional teaching platforms.

[0043] It is to be understood that the foregoing general description and the following detailed description are exemplary and explanatory only and are not restrictive of the disclosure.

[0044] Undoubtedly, these and other objects of the present invention will become more apparent after the following detailed description of the preferred embodiment is described with reference to the various figures and drawings.

[0045] In order to make the above and other objects, features and advantages of the present invention more obvious and easy to understand, one or more preferred embodiments are given below and described in detail with reference to the accompanying drawings. BRIEF DESCRIPTION OF THE DRAWINGS

[0046] The accompanying drawings are used to provide further understanding of the present invention and constitute a part of the specification. Together with the embodiments of the present invention, they are used to explain the present invention but do not constitute a limitation of the present invention.

[0047] In the drawings, like components are given like reference numerals, and the drawings are schematic and not necessarily drawn to scale.

[0048] 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, the drawings described below are only one or several embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on such drawings without paying any creative work.

[0049] Figure 1 A schematic diagram of the process of constructing a parotid tumor enhancement model in some embodiments of the present invention;

[0050] Figure 2 Schematic diagram of T1WI and ADC1 values ​​obtained by MRI T1-weighted imaging in some embodiments of the present invention;

[0051] Figure 3 Schematic diagram of T2WI and ADC2 values ​​obtained by MRI T2-weighted imaging in some embodiments of the present invention;

[0052] Figure 4 Schematic diagram of the position of Warthin's tumor data points in a three-dimensional coordinate system in some embodiments of the present invention;

[0053] Figure 5 A schematic diagram of the classification of parotid gland tumors in some embodiments of the present invention;

[0054] Figure 6 Schematic diagram of the overall process of generating a three-dimensional parotid gland tumor model from an MRI scan image set in some embodiments of the present invention. DETAILED DESCRIPTION

[0055] In order to make the purpose, technical solutions and advantages of the present invention more clearly understood, the present invention is further described in detail below in conjunction with specific embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention, but are not intended to limit the present invention.

[0056] In the present invention, unless otherwise clearly specified and limited, a first feature "above" or "below" a second feature may be that the first and second features are in direct contact, or the first and second features are in indirect contact through an intermediate medium. In the description of this specification, the description with reference to the terms "one embodiment", "some embodiments", "example", "specific example", or "some examples" means that the specific features, structures, materials or characteristics described in conjunction with the embodiment or example are included in at least one embodiment or example of the present invention. In this specification, the schematic representation of the above terms does not necessarily refer to the same embodiment or example. Moreover, the specific features, structures, materials or characteristics described may be combined in an appropriate manner in any one or more embodiments or examples.

[0057] Reference Figure 1 , Figure 1 Schematic diagram of the process of constructing a parotid tumor enhancement model in some embodiments of the present invention.

[0058] According to some embodiments of the present invention, the present invention provides a method for constructing a teaching parotid tumor enhancement model based on a clustering algorithm, comprising:

[0059] S1. Medical Data Integration: Eligible patients with parotid gland tumors are selected and their electronic medical records are uploaded to the cloud platform via a device. Each patient's electronic medical record contains at least an MRI scan image set and the T1WI, ADC1, T2WI, and ADC2 values ​​from the MRI test results. Specifically, the ADC1 value is the longitudinal cross-sectional area S1 of the tumor obtained during T1WI imaging, and the ADC2 value is the transverse cross-sectional area S2 of the tumor obtained during T2WI imaging.

[0060] After obtaining the patient's informed consent and data desensitization, the uploading party will actively select appropriate medical records and transmit them to the cloud platform. For example, if a doctor at Hospital A encounters a patient with Warthin's tumor, he or she will actively upload the patient's medical record information to the cloud platform. Before uploading, the electronic medical record will be anonymized, and direct identifiers such as the patient's name, ID number, and contact information will be deleted, while the diagnosis code and imaging data will be retained. This operation is not mandatory and is actively uploaded by the hospital and complies with relevant legal requirements. Specifically, the process of encrypting electronic medical record information and uploading it to the cloud storage platform through equipment such as computers, mobile phones, MRI equipment, etc. has been widely used in major dental hospitals and will not be described in detail here. For related technologies, please refer to the existing technologies with publication numbers CN116522415A and CN117275647A. However, it should be understood that the examples of public technical documents are only used to indicate that the technology is well known in the art, and does not mean that the specific implementation method of the present invention is the same as the content of the public documents.

[0061] Reference Figure 2-Figure 3 , Figure 2Schematic diagram of T1WI and ADC1 values ​​obtained by MRI T1-weighted imaging in some embodiments of the present invention; Figure 3 Schematic diagram of T2WI and ADC2 values ​​obtained by MRI T2-weighted imaging in some embodiments of the present invention.

[0062] S2. Cluster classification: Download electronic medical records from the cloud platform in batches, obtain T1WI, ADC1 value, T2WI and ADC2 value in the electronic medical records, and fit the maximum volume V of the tumor based on the ADC1 value and ADC2 value. max , according to T1WI, T2WI and V max The K-means clustering algorithm was used to analyze the MRI results of each parotid tumor patient and a three-dimensional coordinate system was established.

[0063] The maximum volume V of the tumor is fitted based on the ADC1 and ADC2 values. max The specific steps are:

[0064] The longitudinal cross-sectional area S1 of the tumor was obtained based on the ADC1 value, and the transverse cross-sectional area S2 of the tumor was obtained based on the ADC2 value;

[0065] An ellipsoid was fitted based on the longitudinal cross-sectional area S1 and the transverse cross-sectional area S2 to represent the shape of the tumor;

[0066] The volume V of the ellipsoid is calculated based on S1 and S2. The specific formula is:

[0067]

[0068] Since the volume calculation of an ellipsoid requires at least three cross-sectional area calculations, multiple results will be obtained by only calculating two cross-sectional area calculations. The maximum volume value V is selected from the multiple results. max As a parameter entering the clustering algorithm.

[0069] For example, Figure 2-3 As shown in the figure, the T1WI signal obtained during T1-weighted imaging of Warthin tumor is a uniform medium signal, and the ADC1 value is 0.97cm 2 The T2WI signal obtained during T2-weighted imaging was uniformly high signal, and the ADC2 value was 0.66 cm 2 ;

[0070] Assuming that the tumor is an elongated ellipsoid, the volume is obtained by substituting it into the ellipsoid volume calculation formula. (Take two decimal places), assuming that the tumor is a flat ellipsoid, substitute it into the ellipsoid volume calculation formula to get the volume (round to two decimal places);

[0071] Considering that in the process of virtual oral medicine teaching, larger tumors are easier for students to observe and learn, when multiple tumor volume results are obtained, the maximum value V is uniformly taken. max As parameters that go into the clustering algorithm, in this case, .

[0072] According to T1WI, T2WI and V max The specific steps to establish a three-dimensional coordinate axis are:

[0073] According to the signal distribution of T1WI, the X-axis is divided into three segments: low, medium, and high signal. Each signal segment is further divided into three segments: uniform, uneven, and mixed, for a total of 9 segments.

[0074] Similar to T1WI signals, T2WI signals also divide the Y axis into 9 segments according to the distribution;

[0075] Z axis according to V max The data of the value establishes the coordinates;

[0076] For the T1WI signal and T2WI signal in the MRI detection results, the midpoint of each segment is used as the coordinate information of the signal actually located in the three-dimensional coordinate axis.

[0077] Reference Figure 4 , Figure 4 Schematic diagram of the position of Warthin's tumor data points in a three-dimensional coordinate system in some embodiments of the present invention.

[0078] For example, in the case of Warthin's tumor, if the T1WI signal result is a uniform medium signal, then its actual X-axis coordinate is the midpoint of the uniform medium signal segment; if the T2WI signal result is a uniform high signal, then its actual Y-axis coordinate is the midpoint of the uniform high signal segment; combined with the height of the ADC value on the Z axis, the position of the Warthin's tumor case record in the three-dimensional coordinate axis can be determined, and the coordinate point is (uniform medium signal, uniform high signal, 0.63). According to the actual situation of the T1WI signal and T2WI signal, the operation of specifying the coordinate point for each signal is conducive to the rapid classification of the clustering algorithm, saving computing power and improving the running speed;

[0079] S3. Finding typical medical records: Set k cluster centers and calculate the center point of each cluster center, perform basic classification of parotid gland tumor symptoms, find the data point closest to the cluster center, and use the medical record corresponding to this data point as the typical medical record of the disease category. Trace the MRI scan image group in the electronic medical record information;

[0080] The specific steps for classifying parotid gland tumors using the Mahalanobis distance K-means clustering algorithm are as follows:

[0081] S31, randomly select 6 data points as cluster centers;

[0082] S32. Assigning each data point to the nearest cluster center based on Mahalanobis distance, wherein the Mahalanobis distance is calculated using the inverse of the covariance;

[0083] S33, recalculate the centroid of each cluster as the new cluster center so that its total square distance is minimized;

[0084] S34. Repeat steps S32-S33 until convergence.

[0085] Reference Figure 5 , Figure 5 Schematic diagram of parotid gland tumor classification in some embodiments of the present invention.

[0086] S35. Obtain six clusters. Based on the clusters, under the guidance of professional doctors, identify the parotid gland tumor type in each cluster. Specific types include: Warthin tumor, pleomorphic adenoma, malignant tumor, basal cell adenoma, parotid lymphoma, and abnormal medical records. Abnormal medical records will be excluded and not included in the data collection.

[0087] S36. Calculate the cluster centers of the five clusters excluding the abnormal medical records, calculate the data point closest to each cluster center using Mahalanobis distance, use the data point as the typical data of the parotid gland tumor of this type, record the coordinates of the data point, and trace back the MRI scan image group in the electronic medical record information;

[0088] For the nearest data point under each disease classification, a manual screening step is required, and professional doctors will identify the nearest data point found. If it does not have typical tumor characteristics or is a complex medical record of multiple types of tumors or even a classification error, the next nearest data point will be selected sequentially until the data point is available.

[0089] Reference Figure 6 , Figure 6 Schematic diagram of the overall process of generating a three-dimensional parotid gland tumor model from an MRI scan image set in some embodiments of the present invention.

[0090] S4. Invasion boundary enhancement processing: The MRI scan image group is subjected to noise reduction and smoothing processing, and edge detection and edge enhancement are performed to enhance the invasion boundary and show the tumor morphology, thereby obtaining an enhanced image group;

[0091] The MRI scan image group includes conventional MRI scan images and ultra-short echo sequence MRI scan images. The original scan image data, T1WI, ADC1 value, T2WI and ADC2 value can be obtained from the conventional MRI scan images, which are used to establish a three-dimensional craniomaxillofacial soft tissue model in step S5. The ultra-short echo sequence MRI scan images are synthesized into a virtual CT image in step S5 through the UTE scan image data and converted into a three-dimensional craniomaxillofacial bone model.

[0092] The UTE ultrashort echo sequence is used to acquire MRI data for hard tissue. In conventional MRI sequences, cortical bone appears as a very short, difficult-to-detect T2 signal, making it difficult to distinguish bone tissue from air. The head, however, contains many small, complex air cavities and thin bones, which is why it is difficult to synthesize CT using conventional MRI sequences. The UTE ultrashort echo sequence can be used to capture the short T2 signal of bone, improving the contrast between bone and surrounding air or soft tissue. After obtaining soft tissue sequence images through scanning, the UTE sequence can be combined with the Dixon sequence, which reflects water-fat contrast, to obtain the UTE / Dixon sequence for head and neck scanning, further improving the resolution between tissues. Based on the above sequence, image post-processing technology is used to successfully synthesize virtual CT images. The technical solution for synthesizing a three-dimensional craniomaxillofacial soft tissue model and a three-dimensional craniomaxillofacial skeletal model using conventional MRI scan images and ultrashort echo sequence MRI scan images, respectively, can be found in the patent application document with publication number CN111583221A, which details the specific steps for constructing a three-dimensional image and will not be elaborated here.

[0093] The specific steps for performing invasion boundary enhancement processing on conventional MRI scan images and ultrashort echo sequence MRI scan images are as follows:

[0094] S41, taking the conventional MRI scan image and the ultrashort echo sequence MRI scan image as a first grayscale image group;

[0095] S42, performing noise reduction and smoothing processing on the first grayscale image group to obtain a second grayscale image group;

[0096] S43, performing edge detection on the second grayscale image group using a prewitt operator to obtain an edge-detected image group;

[0097] S44, performing edge enhancement on the edge detection image group by using an edge function, thereby enhancing the invasion boundary and displaying the tumor morphology, and obtaining an enhanced image group, so as to facilitate the establishment of a three-dimensional model;

[0098] The richness of texture information contained in different regions is different. For example, the background area basically contains no texture information. The richness of texture information contained in the tumor area and tissue bone area is also different. The richness of texture information contained in different regions can be reflected by the gradient amplitude distribution of pixel points in the region. Therefore, the existing prewitt edge detection operator is first used to calculate the gradient amplitude of each pixel point in the grayscale image, and then the distribution of the gradient amplitude of the pixel points in each segmented region is analyzed. The obtained texture richness reflects the richness of texture information contained in the segmented region. The greater the texture richness, the more texture information is contained in the segmented region. Subsequently, the area to be enhanced, namely the tumor area and tissue bone area, can be screened out based on the texture richness. Then, based on the texture richness, the different areas to be enhanced are enhanced to different degrees through the edge function to ensure that the texture information in the area to be enhanced is more prominent, which facilitates the establishment of a three-dimensional model.

[0099] S5. Synthesize a 3D parotid gland tumor model: Synthesize a 3D craniomaxillofacial soft tissue model based on the conventional MRI scan images in the enhanced image group. Then, synthesize a virtual CT image based on the ultrashort echo sequence MRI scan images in the enhanced image group. Generate a 3D craniomaxillofacial skeletal model based on the virtual CT image. Integrate the 3D craniomaxillofacial skeletal model and the 3D craniomaxillofacial soft tissue model to form a 3D parotid gland tumor model.

[0100] In the process of establishing a three-dimensional craniofacial soft tissue model, data of the parotid gland region and tumor boundary are extracted based on the enhanced invasion boundary to generate maxillofacial soft tissue data containing texture features; in the process of establishing a three-dimensional craniofacial skeletal model, bone tissue is distinguished by segmenting the upper and lower jaws and adjacent bony structures to generate three-dimensional maxillofacial hard tissue data; it is understandable that the generation of maxillofacial soft tissue data containing texture features and the generation of three-dimensional maxillofacial hard tissue data based on the invention technical solution with publication number CN111583221A are prior arts and will not be elaborated on here;

[0101] The 3D craniofacial skeleton model and the 3D craniofacial soft tissue model were imported into the medical imaging processing platform and integrated to form a 3D parotid gland tumor model. If there was any overlap in the models in some areas, the 3D craniofacial skeleton model was used as the benchmark, and the maxillofacial soft tissue was adjusted according to T1WI, ADC1 value, T2WI, and ADC2 value.

[0102] Based on the established 3D craniomaxillofacial skeletal model and 3D craniomaxillofacial soft tissue model, the models were imported into 3D Slicer software. The two were combined to form a 3D parotid gland tumor model that can simultaneously display the skeletal shape and soft tissue characteristics. The model was then gradually fine-tuned based on the actual MRI scan results to make it fit the actual case situation. The above operation method can be easily implemented by those skilled in the art by operating 3D Slicer software. It is a prior art and the specific adjustment steps are not expanded here to avoid redundancy.

[0103] It is understandable that the generated 3D parotid tumor model is only used for teaching demonstration and cannot replace clinical diagnosis or surgical navigation, and is not used in actual treatment. Therefore, in order to make the model structure reasonable or facilitate student observation, fine-tuning is performed on the model or the generated model results have certain deviations from the actual case, which are all within an acceptable range and do not affect the actual teaching results. Accordingly, the specific operation method of the 3D Slicer software can refer to the existing technology and will not be elaborated here.

[0104] S6. Loading the virtual oral medicine teaching platform: Load the generated 3D parotid gland tumor model into the virtual oral medicine teaching platform. Different types of parotid gland tumor models are classified into corresponding directories. The teacher can call up the corresponding model by clicking the directory or entering the coordinates.

[0105] The electronic medical record information on the cloud platform is updated at regular intervals. The clustering algorithm synchronously updates the position of the data points in the three-dimensional coordinate system and searches for the nearest point. If the data points change, a new three-dimensional parotid tumor model is established and loaded into the corresponding parotid tumor disease type directory of the virtual oral medicine teaching platform according to its classification. The coordinates of the nearest data point are also updated. Preferably, the update time is three months, and the longest case data retention time does not exceed 5 years.

[0106] It should be understood that the embodiments disclosed herein are not limited to the specific processing steps or materials disclosed herein, but should extend to equivalent substitutions of such features understood by those skilled in the relevant art. It should also be understood that the terminology used herein is for the purpose of describing specific embodiments only and is not intended to be limiting.

[0107] The "embodiment" mentioned in the specification means that a particular feature or characteristic described in conjunction with the embodiment is included in at least one embodiment of the present invention. Therefore, the phrase or "embodiment" appearing in various places throughout the specification does not necessarily refer to the same embodiment.

[0108] Furthermore, the described features or characteristics may be combined in any other suitable manner into one or more embodiments. In the above description, some specific details, such as thickness, quantity, etc., are provided to provide a comprehensive understanding of the embodiments of the present invention. However, those skilled in the relevant art will appreciate that the present invention may be implemented without one or more of the above specific details or may be implemented using other methods, components, materials, etc.

Claims

1. A method for constructing a teaching parotid tumor enhancement model based on a clustering algorithm, characterized in that: include S1. Medical Data Integration: Eligible patients with parotid gland tumors were selected and their electronic medical records were uploaded to the cloud platform via a device. Each patient's electronic medical record included at least an MRI scan image set and the T1WI, ADC1, T2WI, and ADC2 values ​​from the MRI test results. The ADC1 value represents the longitudinal cross-sectional area S1 of the tumor obtained during T1WI imaging, and the ADC2 value represents the transverse cross-sectional area S2 of the tumor obtained during T2WI imaging. S2. Cluster classification: Download electronic medical records from the cloud platform in batches, obtain T1WI, ADC1 value, T2WI and ADC2 value in the electronic medical records, and fit the maximum volume V of the tumor based on the ADC1 value and ADC2 value. max , according to T1WI, T2WI and V max The K-means clustering algorithm was used to analyze the MRI results of each parotid tumor patient and a three-dimensional coordinate system was established. The maximum volume V of the tumor is fitted based on the ADC1 and ADC2 values. max The specific steps are: The longitudinal cross-sectional area S1 of the tumor was obtained based on the ADC1 value, and the transverse cross-sectional area S2 of the tumor was obtained based on the ADC2 value; An ellipsoid was fitted based on the longitudinal cross-sectional area S1 and the transverse cross-sectional area S2 to represent the shape of the tumor; The volume V of the ellipsoid is calculated based on S1 and S2. The specific formula is: Since the volume calculation of an ellipsoid requires at least three cross-sectional area calculations, multiple results will be obtained by only calculating two cross-sectional area calculations. The maximum volume value V is selected from the multiple results. max As parameters entering the clustering algorithm; S3. Finding typical medical records: Set k cluster centers and calculate the center point of each cluster center, perform basic classification of parotid gland tumor symptoms, find the data point closest to the cluster center, and use the medical record corresponding to this data point as the typical medical record of the disease category. Trace the MRI scan image group in the electronic medical record information; S4. Invasion boundary enhancement processing: The MRI scan image group is subjected to noise reduction and smoothing processing, and edge detection and edge enhancement are performed to enhance the invasion boundary and show the tumor morphology, thereby obtaining an enhanced image group; S5. Synthesize a 3D parotid gland tumor model: Synthesize a 3D craniomaxillofacial soft tissue model based on the conventional MRI scan images in the enhanced image group. Then, synthesize a virtual CT image based on the ultrashort echo sequence MRI scan images in the enhanced image group. Generate a 3D craniomaxillofacial skeletal model based on the virtual CT image. Integrate the 3D craniomaxillofacial skeletal model and the 3D craniomaxillofacial soft tissue model to form a 3D parotid gland tumor model. S6. Load the virtual oral medicine teaching platform: Load the generated three-dimensional parotid gland tumor model into the virtual oral medicine teaching platform. Different types of parotid gland tumor models are classified into corresponding directories. The teacher can click the directory or enter the coordinate point to call out the corresponding model.

2. The method for constructing a teaching parotid tumor enhancement model based on a clustering algorithm according to claim 1, characterized in that: According to T1WI, T2WI and V max The specific steps to establish a three-dimensional coordinate axis are: According to the signal distribution of T1WI, the X-axis is divided into three segments: low, medium, and high signal. Each signal segment is further divided into three segments: uniform, uneven, and mixed, for a total of 9 segments. Similar to T1WI signals, T2WI signals also divide the Y axis into 9 segments according to the distribution; Z axis according to V max The data of the value establishes the coordinates; For the T1WI signal and T2WI signal in the MRI detection results, the midpoint of each segment is used as the coordinate information of the signal actually located in the three-dimensional coordinate axis.

3. The method for constructing a teaching parotid tumor enhancement model based on a clustering algorithm according to claim 2, characterized in that: In step S3, the specific steps for classifying parotid gland tumors using the Mahalanobis distance K-means clustering algorithm are as follows: S31, randomly select 6 data points as cluster centers; S32. Assigning each data point to the nearest cluster center based on Mahalanobis distance, wherein the Mahalanobis distance is calculated using the inverse of the covariance; S33, recalculate the centroid of each cluster as the new cluster center so that its total square distance is minimized; S34, repeat steps S32-S33 until convergence; S35. Obtain six clusters. Based on the clusters, under the guidance of professional doctors, identify the parotid gland tumor type in each cluster. Specific types include: Warthin tumor, pleomorphic adenoma, malignant tumor, basal cell adenoma, parotid lymphoma, and abnormal medical records. Abnormal medical records will be excluded and not included in the data collection. S36. Calculate the cluster centers of the five clusters excluding the abnormal medical records, calculate the data point closest to each cluster center using the Mahalanobis distance, use the data point as the typical data of this type of parotid gland tumor disease, record the data point coordinates, and trace back the MRI scan image group in the electronic medical record information.

4. The method for constructing a teaching parotid tumor enhancement model based on a clustering algorithm according to claim 3, characterized in that: In step S36, for the nearest data point under each disease classification, a manual screening step is required, and professional doctors identify the nearest data point found. If it does not have typical tumor characteristics or is a complex medical record of a collection of multiple types of tumors or even a classification error, the next nearest data point will be selected in sequence until the data point is available.

5. The method for constructing a teaching parotid tumor enhancement model based on a clustering algorithm according to claim 1, characterized in that: In step S4, the MRI scan image group includes conventional MRI scan images and ultrashort echo sequence MRI scan images. The original scan image data, T1WI, ADC1 value, T2WI and ADC2 value can be obtained from the conventional MRI scan images, which are used to establish a three-dimensional craniomaxillofacial soft tissue model in step S5; the ultrashort echo sequence MRI scan images are synthesized into a virtual CT image in step S5 through the UTE scan image data and converted into a three-dimensional craniomaxillofacial bone model.

6. The method for constructing a teaching parotid tumor enhancement model based on a clustering algorithm according to claim 5, characterized in that: In step S4, the specific steps of performing invasion boundary enhancement processing on the conventional MRI scan image and the ultrashort echo sequence MRI scan image are as follows: S41, taking the conventional MRI scan image and the ultrashort echo sequence MRI scan image as a first grayscale image group; S42, performing noise reduction and smoothing processing on the first grayscale image group to obtain a second grayscale image group; S43, performing edge detection on the second grayscale image group using a prewitt operator to obtain an edge-detected image group; S44. Perform edge enhancement on the edge detection image group through the edge function to enhance the invasion boundary and show the tumor morphology, thereby obtaining an enhanced image group to facilitate the establishment of a three-dimensional model.

7. The method for constructing a teaching parotid tumor enhancement model based on a clustering algorithm according to claim 6, characterized in that: In the process of establishing the 3D craniofacial soft tissue model, the data of the parotid gland area and tumor boundary were extracted based on the enhanced invasion boundary to generate the maxillofacial soft tissue data containing texture features; In the process of establishing a three-dimensional craniofacial skeleton model, the upper and lower jaws and adjacent bony structures are segmented to distinguish the bone tissue and generate three-dimensional maxillofacial hard tissue data.

8. The method for constructing a teaching parotid tumor enhancement model based on a clustering algorithm according to claim 1, characterized in that: In step S5, the three-dimensional craniofacial skeleton model and the three-dimensional craniofacial soft tissue model are imported into the medical image processing platform and integrated to form a three-dimensional parotid gland tumor model. If there is overlap in some regional models, the three-dimensional craniofacial skeleton model is used as a reference, and the maxillofacial soft tissue is adjusted according to T1WI, ADC1 value, T2WI and ADC2 value.

9. The method for constructing a teaching parotid tumor enhancement model based on a clustering algorithm according to claim 1, characterized in that: In step S6, the electronic medical record information on the cloud platform is updated at regular intervals. The clustering algorithm synchronously updates the position of the data points in the three-dimensional coordinate system and searches for the nearest point. If the data points change, a new three-dimensional parotid tumor model is established and loaded into the corresponding parotid tumor disease type directory of the virtual oral medicine teaching platform according to its classification, and the coordinates of the nearest data point are updated.

Images