Chronic osteomyelitis prediction model training method and device based on habitat imaging and imageomics, chronic osteomyelitis prediction method and device based on habitat imaging and imageomics, electronic equipment and storage medium

Abstract

The invention relates to the technical field of medical information, in particular to a chronic osteomyelitis prediction model training method, prediction method and device based on habitat imaging and radiomics, electronic equipment and a storage medium. Comprising the following steps: respectively training multiple machine learning algorithms by using a first sample set, a second sample set, a third sample set, a fourth sample set, a fifth sample set and a sixth sample set to obtain corresponding six types of training model sets; and respectively selecting the model with the best evaluation result from the six types of training model sets to form a chronic osteomyelitis prediction model group. According to the method, the internal heterogeneity of the focus can be deeply quantified and the internal heterogeneity and microenvironment characteristics of the focus can be captured on the basis of the inhabitation characteristics, the focus can be more accurately positioned when the boundary of the focus is fuzzy, a clearer focus excision boundary is provided for surgeons, scientific excision is further converted into accurate excision, unnecessary tissue damage is reduced, and the operation efficiency is improved. And the risk of postoperative recurrence and complications is reduced.

Description

Technical Field

[0001] The present invention relates to the field of medical information technology, and is a chronic osteomyelitis prediction model training method based on habitat imaging and radiomics, a prediction method, a device, an electronic device and a storage medium. Background Art

[0002] In addition to targeted antibacterial treatment, the key to successful treatment of chronic osteomyelitis is the complete removal of all necrotic and infected bone tissue. In actual surgery, surgeons often face a severe dilemma: too little resection may lead to recurrence of infection, while excessive resection may lead to irreversible damage to the bone structure and loss of function. Moreover, due to the multi-plane osteotomy and the small exposure of the operating area involved in the operation, accurate lesion positioning becomes increasingly complex, which can easily lead to incomplete resection or excessive bone resection. Therefore, accurately judging the degree of infection and identifying the boundary between dead bone and healthy bone has become an international challenge that needs to be solved in the medical field, and it has also been a hot topic in the field of bone infection research.

[0003] Currently, machine learning has been introduced to assist decision-making in locating debridement boundaries. For example, multi-level ROI expansion is introduced for modeling, resulting in a chronic osteomyelitis prediction model set. This model set is then used to predict the original ROI features, first and second expanded ROI features of the patient being predicted, yielding corresponding original ROI prediction results, first and second expanded ROI prediction results, and then combined with the chronic osteomyelitis prediction results to determine the conditions for assisting debridement decisions for chronic osteomyelitis. However, during osteomyelitis imaging, lesions often cause abnormal enlargement and morphological changes in adjacent bone structures, manifesting as elongated and irregularly shaped extensions. This indicates abnormal changes in bone structure and the degree of osteomyelitis invasion. While this method can dynamically adjust the surgical debridement range, it cannot capture the internal heterogeneity and microenvironmental characteristics of the lesion in complex cases. Therefore, when the lesion boundary is blurred and the state of the surrounding tissue changes, it cannot accurately locate the lesion, which may still lead to uncertainty in surgical decision-making. Summary of the Invention

[0004] The present invention provides a chronic osteomyelitis prediction model training method, prediction method, device, electronic device and storage medium based on habitat imaging and radiomics, which overcomes the shortcomings of the above-mentioned existing technologies and can effectively solve the problem that the existing chronic osteomyelitis prediction methods cannot capture the heterogeneity and microenvironment characteristics inside the lesion when the lesion boundary is blurred, resulting in the inability to accurately locate the lesion.

[0005] One of the technical solutions of the present invention is achieved through the following measures: a method for training a chronic osteomyelitis prediction model based on habitat imaging and radiomics, comprising: Obtain a first sample set, a second sample set, a third sample set, a fourth sample set, a fifth sample set, and a sixth sample set, and divide each sample set into a training set and a test set in proportion, wherein the samples in the first sample set, the second sample set, and the third sample set respectively include different feature types of the first expanded ROI data and corresponding chronic osteomyelitis lesion detection identification information, and the samples in the fourth sample set, the fifth sample set, and the sixth sample set respectively include different feature types of the second expanded ROI data and corresponding chronic osteomyelitis lesion detection identification information, and the feature types include habitat features, imaging omics features, and comprehensive features; Using the training sets of the first sample set, the second sample set, the third sample set, the fourth sample set, the fifth sample set, and the sixth sample set to train multiple machine learning algorithms respectively, to obtain corresponding six types of training model sets; The test sets of the first sample set, the second sample set, the third sample set, the fourth sample set, the fifth sample set and the sixth sample set were used to test the corresponding six types of training model sets respectively, and the models with the best evaluation results were selected from the six types of training model sets to form a chronic osteomyelitis prediction model group.

[0006] The following are further optimizations and / or improvements to the above technical solutions: The obtaining of the first sample set, the second sample set, the third sample set, the fourth sample set, the fifth sample set, and the sixth sample set includes: Obtain the original MRI image data set, and outline the lesion and surrounding area for each original MRI image data to obtain the original ROI data set; Expanding the lesion area of ​​each original ROI data set of the original ROI data set to obtain a first expanded ROI data set and a second expanded ROI data set; Extracting habitat characteristics and imaging omics features of each first expanded ROI data set in the first expanded ROI data set to form a first sample set and a second sample set, and extracting multiple features therefrom to form a third sample set; The habitat characteristics and imaging omics characteristics of each second expanded ROI data in the second expanded ROI data set are extracted to form a fourth sample set and a fifth sample set, and multiple features are extracted therefrom to form a sixth sample set.

[0007] The above method extracts habitat characteristics and imaging omics features of each first expanded ROI data set in the first expanded ROI data set to form a first sample set and a second sample set, and extracts multiple features therefrom to form a third sample set, including: Extract local features from each voxel in the first expanded ROI data, and use the moving window technique to convert each local feature into a 19-dimensional feature vector as a subregion; All sub-regions in the first expanded ROI data are clustered using the K-means algorithm, and sub-regions classified into the same category are set to the same cluster ID; Merge the sub-regions with the same cluster ID to form one or more habitat regions, and extract the unique microenvironmental features representing the lesion in each habitat region to obtain the habitat features of the first expanded ROI data; Extracting geometric features, intensity features, and texture features of the first expanded ROI data, and screening the extracted features to obtain imaging omics features of the first expanded ROI data; Select multiple features from the habitat characteristics and imaging omics features of the first expanded ROI data and combine them to obtain comprehensive features; The above steps are repeated to obtain the habitat characteristics, radiomics characteristics, and comprehensive characteristics of each first expanded ROI data. The habitat characteristics of the first expanded ROI data and the corresponding chronic osteomyelitis lesion detection identification information are used as the first sample to form a first sample set. The radiomics characteristics of the first expanded ROI data and the corresponding chronic osteomyelitis lesion detection identification information are used as the second sample to form a second sample set. The comprehensive characteristics of the first expanded ROI data and the corresponding chronic osteomyelitis lesion detection identification information are used as the third sample to form a third sample set.

[0008] The above-mentioned extraction of local features from each voxel in the first expanded ROI data is achieved in sequence through entropy, mean absolute deviation, joint energy, and joint entropy; or / and, the extracted features are screened in sequence through T test, correlation analysis, and LASSO regression algorithm.

[0009] The above machine learning algorithms include SVM, RandomForest, ExtraTrees, XGBoost, and LightGBM.

[0010] The second technical solution of the present invention is achieved through the following measures: a method for predicting chronic osteomyelitis based on habitat imaging and radiomics, characterized by comprising: Obtaining input data to be predicted of a patient to be predicted, wherein the input data to be predicted includes habitat characteristics, imaging omics characteristics, and comprehensive characteristics of the first expanded ROI data, and habitat characteristics, imaging omics characteristics, and comprehensive characteristics of the second expanded ROI data; Input the predicted patient's input data to the corresponding model in the chronic osteomyelitis prediction model group to obtain six corresponding prediction results, wherein the chronic osteomyelitis prediction model group is obtained according to the chronic osteomyelitis prediction model training method based on habitat imaging and radiomics; Combine the six prediction results to determine the corresponding auxiliary decision for chronic osteomyelitis debridement.

[0011] The third technical solution of the present invention is achieved by the following measures: a chronic osteomyelitis prediction model training device based on habitat imaging and radiomics, comprising: a sample acquisition unit, acquiring a first sample set, a second sample set, a third sample set, a fourth sample set, a fifth sample set, and a sixth sample set, and dividing each sample set into a training set and a test set in proportion, wherein the samples in the first sample set, the second sample set, and the third sample set respectively include different feature types of the first expanded ROI data and corresponding chronic osteomyelitis lesion detection identification information, and the samples in the fourth sample set, the fifth sample set, and the sixth sample set respectively include different feature types of the second expanded ROI data and corresponding chronic osteomyelitis lesion detection identification information, and the feature types include habitat features, imaging omics features, and comprehensive features; A training unit, using the training sets of the first sample set, the second sample set, the third sample set, the fourth sample set, the fifth sample set, and the sixth sample set to respectively train a plurality of machine learning algorithms to obtain corresponding six types of training model sets; The testing unit uses the test sets of the first sample set, the second sample set, the third sample set, the fourth sample set, the fifth sample set and the sixth sample set to test the corresponding six types of training model sets respectively, and selects the models with the best evaluation results from the six types of training model sets to form a chronic osteomyelitis prediction model group.

[0012] The fourth technical solution of the present invention is achieved by the following measures: a chronic osteomyelitis prediction device based on habitat imaging and radiomics, comprising: A prediction data acquisition unit is configured to acquire input data to be predicted of a patient to be predicted, wherein the input data to be predicted includes habitat characteristics, imaging omics characteristics, and comprehensive characteristics of the first expanded ROI data, and habitat characteristics, imaging omics characteristics, and comprehensive characteristics of the second expanded ROI data; A prediction unit, inputting the predicted input data of the patient to be predicted into the corresponding model in the chronic osteomyelitis prediction model group to obtain six corresponding prediction results, wherein the chronic osteomyelitis prediction model group is obtained according to the chronic osteomyelitis prediction model training method based on habitat imaging and radiomics; The result output unit combines the six prediction results to determine the corresponding auxiliary decision for chronic osteomyelitis debridement.

[0013] The fifth technical solution of the present invention is achieved through the following measures: an electronic device, including a processor and a memory, wherein the memory stores a computer program, and the computer program is loaded and executed by the processor to implement the steps in the chronic osteomyelitis prediction method based on habitat imaging and radiomics.

[0014] The sixth technical solution of the present invention is achieved through the following measures: a storage medium, on which a computer program that can be read by a computer is stored, and the computer program is configured to execute the steps in the chronic osteomyelitis prediction method based on habitat imaging and radiomics when it is run.

[0015] The present invention analyzes the internal structure and surrounding environment of a lesion, effectively quantifies the heterogeneity within the lesion, obtains the habitat characteristics of the expanded ROI data, and combines the habitat characteristics of the expanded ROI data with the imaging genomics characteristics to obtain comprehensive characteristics. Six types of sample sets are constructed, and model training is performed using the six types of samples based on the introduction of multiple machine learning algorithms. The model combination with the best evaluation results is selected to form a chronic osteomyelitis prediction model group. In the prediction of chronic osteomyelitis, the habitat characteristics and imaging genomics characteristics are obtained, that is, the internal heterogeneity of the lesion is deeply quantified, and the internal heterogeneity and microenvironment characteristics of the lesion are captured. When the lesion boundary is blurred, the lesion can be located more accurately, providing surgeons with a clearer lesion resection boundary, further realizing the transition from "scientific resection" to "precise resection", reducing unnecessary tissue damage, and lowering the risk of postoperative recurrence and complications. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] Attachment Figure 1 This is a schematic diagram of the implementation environment provided by the present invention.

[0017] Attachment Figure 2 This is a flow chart of the model training method provided by the present invention.

[0018] Attachment Figure 3 This is a flow chart of the sample set acquisition method provided by the present invention.

[0019] Attachment Figure 4 This is a flow chart of the method for forming the first sample set, the second sample set, and the third sample set provided by the present invention.

[0020] Attachment Figure 5 This is a flow chart of the sub-region acquisition method provided by the present invention.

[0021] Attachment Figure 6 This is a schematic diagram of the DeLong test results provided by the present invention.

[0022] Attachment Figure 7 This is a flow chart of the chronic osteomyelitis prediction method provided by the present invention.

[0023] Attachment Figure 8 This is a structural diagram of the model training device provided by the present invention.

[0024] Attachment Figure 9 This is a schematic structural diagram of the chronic osteomyelitis prediction device provided by the present invention. DETAILED DESCRIPTION

[0025] The present invention is not limited to the following embodiments, and specific implementation methods can be determined based on the technical solutions of the present invention and actual conditions.

[0026] Those skilled in the art will understand that, unless otherwise specified, a "module" or "unit" in the embodiments of the present invention refers to a computer program or portion of a computer program that has a predetermined function and works together with other related components to achieve a predetermined goal. Such a program may be implemented in whole or in part using software, hardware (e.g., processing circuitry or memory), or a combination thereof. Similarly, a processor (or multiple processors or memories) may be used to implement one or more modules or units. Furthermore, each module or unit may be part of an overall module or unit that includes the functionality of that module or unit.

[0027] In addition, in the embodiments of the present invention, “plurality” refers to two or more than two, and “first” and “second” are used for distinguishing descriptions and should not be understood as implying relative importance.

[0028] Currently, machine learning has been introduced to assist decision-making in locating debridement boundaries. However, during osteomyelitis imaging, lesions often cause abnormal enlargement and morphological changes in adjacent bone structures, manifesting as elongated and irregularly shaped extensions, which indicate abnormal changes in bone structure and the degree of osteomyelitis invasion. Although this method can dynamically adjust the scope of surgical debridement, it cannot capture the heterogeneity and microenvironmental characteristics within the lesion in complex cases. Therefore, when the lesion boundary is blurred and the state of the surrounding tissue changes, the lesion cannot be accurately located, which may still lead to uncertainty in surgical decision-making.

[0029] The embodiment of the present invention provides a chronic osteomyelitis prediction model training method based on habitat imaging and radiomics, a prediction method, an apparatus, an electronic device and a storage medium, which obtains a first sample set, a second sample set, a third sample set, a fourth sample set, a fifth sample set and a sixth sample set, and divides each sample set into a training set and a test set in proportion; uses the training sets of the first sample set, the second sample set, the third sample set, the fourth sample set, the fifth sample set and the sixth sample set to train multiple machine learning algorithms respectively, and obtains corresponding six types of training model sets; uses the test sets of the first sample set, the second sample set, the third sample set, the fourth sample set, the fifth sample set and the sixth sample set to train the corresponding six types of training model sets respectively. The six types of training model sets were tested, and the models with the best evaluation results were selected from the six types of training model sets to form a chronic osteomyelitis prediction model group. The input data of the patient to be predicted was input into the corresponding model in the chronic osteomyelitis prediction model group to obtain the corresponding six prediction results. The six prediction results were combined to obtain the corresponding chronic osteomyelitis debridement auxiliary decision-making, which deeply quantified the heterogeneity and microenvironment inside the lesion, and was effectively applicable to complex medical records with blurred lesion boundaries. When the lesion boundaries were blurred, the lesions could be located more accurately, providing surgeons with clearer lesion resection boundaries, and further realizing the transition from "scientific cutting" to "precise cutting", reducing unnecessary tissue damage, and reducing the risk of postoperative recurrence and complications.

[0030] Among them, the method provided by the embodiment of the present invention may involve artificial intelligence (AI) technology and can be implemented based on artificial intelligence technology, for example, using machine learning to obtain a corresponding model through sample training.

[0031] Machine Learning (ML) is a multidisciplinary field that encompasses probability theory, statistics, approximation theory, convex analysis, and algorithmic complexity theory. It specifically studies how computers can simulate or implement human learning behaviors to acquire new knowledge or skills and reorganize existing knowledge structures to continuously improve their performance. Machine learning is at the core of artificial intelligence and the fundamental way to make computers intelligent. Its applications span all areas of AI.

[0032] Deep learning (DL) specifically refers to machine learning based on deep machine learning algorithmic models and methods. It is developed based on statistical machine learning, artificial machine learning algorithms, and other algorithmic models, combined with the development of modern big data and massive computing power. The most important technical feature of deep learning is its ability to automatically extract features.

[0033] The above-mentioned machine learning and deep learning usually include technologies such as neural networks, belief networks, reinforcement learning, transfer learning, inductive learning, and self-learning.

[0034] As attached Figure 1 FIG. 1 shows a schematic diagram of an implementation environment provided by an embodiment of the present invention. The implementation environment may include: training equipment and use equipment.

[0035] Both the training device and the usage device are computer devices; optionally, the computer device is a terminal device, such as a mobile phone, tablet computer, PC (Personal Computer) and other electronic devices; or, the computer device is a server, which can be a single server, a server cluster composed of multiple servers, or a cloud computing service center, which is not limited in this embodiment of the present invention.

[0036] A training device refers to a computer device capable of training and learning a machine learning algorithm. Optionally, the training device has the ability to acquire a machine learning algorithm and train and learn it according to application requirements. For example, the training device acquires a machine learning algorithm from another device via a network, and then trains it with training samples according to application requirements, so that the machine learning algorithm has the ability to obtain a prediction result for chronic osteomyelitis. Optionally, the training device has the ability to construct a machine learning algorithm, and can independently construct a machine learning algorithm according to application requirements, and then train and learn it. For example, in order to input the predicted patient's predicted data into the corresponding model in the chronic osteomyelitis prediction model group and obtain the corresponding six prediction results, the training device independently constructs a machine learning algorithm, and then trains and learns it with samples according to application requirements.

[0037] The device used refers to a computer device that has the need to use a machine learning algorithm. Optionally, the device uses the machine learning algorithm according to application requirements and obtains the machine learning algorithm from other devices through the network. For example, if the device uses the need to obtain a prediction result of chronic osteomyelitis, it can obtain a machine learning algorithm that has completed training and learning to obtain a prediction result of chronic osteomyelitis from other devices through the network, and use the machine learning algorithm to obtain a prediction result of chronic osteomyelitis.

[0038] Based on this, the technical solution of the present invention will be introduced and explained with reference to several examples below.

[0039] Example 1: As shown in the attached Figure 2 As shown, an embodiment of the present invention discloses a method for training a chronic osteomyelitis prediction model based on habitat imaging and radiomics, comprising the following steps: Step S110: Obtain a first sample set, a second sample set, a third sample set, a fourth sample set, a fifth sample set, and a sixth sample set, and divide each sample set into a training set and a test set in proportion, wherein the samples in the first sample set, the second sample set, and the third sample set respectively include different feature types of the first expanded ROI data and corresponding chronic osteomyelitis lesion detection identification information, and the samples in the fourth sample set, the fifth sample set, and the sixth sample set respectively include different feature types of the second expanded ROI data and corresponding chronic osteomyelitis lesion detection identification information, and the feature types include habitat features, imaging genomics features, and comprehensive features.

[0040] For chronic osteomyelitis, in existing surgical debridement, an overly conservative debridement range may leave residual infection and lead to recurrence, while overly aggressive debridement may cause unnecessary bone damage. Existing chronic osteomyelitis prediction models based on imaging genomics technology and machine learning have good results in actual clinical applications, but there are still some limitations. That is, although the surgical debridement range can be dynamically adjusted, in complex cases, the blurred boundaries of the lesion and changes in the state of the surrounding tissues may still lead to uncertainty in surgical decisions. The existing chronic osteomyelitis prediction model can solve the problem of blurred lesion boundaries in complex cases. However, in some cases, surgical decisions still rely on the doctor's experience and judgment. Even though the difficulty is greatly reduced compared to before, it is difficult to achieve truly intelligent visual and precise resection.

[0041] Therefore, in this embodiment, step S110 first expands the lesion area of ​​each original ROI data to obtain corresponding first expanded ROI data and second expanded ROI data. It should also be noted that the distance between the two expansions can be determined by analyzing actual medical records.

[0042] In the above steps, the samples in the first sample set, the second sample set, and the third sample set respectively include different feature types of the first expanded ROI data and corresponding chronic osteomyelitis lesion detection identification information, and the samples in the fourth sample set, the fifth sample set, and the sixth sample set respectively include different feature types of the second expanded ROI data and corresponding chronic osteomyelitis lesion detection identification information, as follows: The first sample set includes a plurality of first samples, each of which includes habitat features of the first expanded ROI data and corresponding chronic osteomyelitis lesion detection identification information; The second sample set includes a plurality of second samples, each of which includes the radiomics features of the first expanded ROI data and corresponding chronic osteomyelitis lesion detection identification information; The third sample set includes a plurality of third samples, each of which includes comprehensive features of the first expanded ROI data and corresponding chronic osteomyelitis lesion detection identification information; The fourth sample set includes a plurality of fourth samples, each of which includes the habitat characteristics of the second expanded ROI data and corresponding chronic osteomyelitis lesion detection identification information; The fifth sample set includes a plurality of fifth samples, each of which includes the radiomics features of the second expanded ROI data and corresponding chronic osteomyelitis lesion detection identification information; The sixth sample set includes a plurality of sixth samples, each of which includes comprehensive features of the second expanded ROI data and corresponding chronic osteomyelitis lesion detection identification information; Among them, the identification information of chronic osteomyelitis lesion detection is negative result / positive result (a positive result means that chronic osteomyelitis lesions are detected within the corresponding ROI range, and a negative result means that no chronic osteomyelitis lesions are detected).

[0043] Step S120 , using the training sets of the first sample set, the second sample set, the third sample set, the fourth sample set, the fifth sample set and the sixth sample set to train multiple machine learning algorithms respectively, to obtain corresponding six types of training model sets.

[0044] In this embodiment, the types of various machine learning algorithms are selected as needed, including but not limited to SVM, RandomForest, ExtraTrees, XGBoost, and LightGBM.

[0045] Step S130, using the test sets of the first sample set, the second sample set, the third sample set, the fourth sample set, the fifth sample set and the sixth sample set to test the corresponding six types of training model sets respectively, and selecting the models with the best evaluation results from the six types of training model sets to form a chronic osteomyelitis prediction model group.

[0046] In this embodiment, one or more evaluation indicators can be introduced to evaluate all test results in each type of training model set, and the model with the best evaluation results can be selected. The evaluation indicators here can include but are not limited to Accuracy, AUC, 95%CI, Sensitivity, Specificity, PPV, NPV, and Cohort.

[0047] The present invention discloses a chronic osteomyelitis prediction model training method based on habitat imaging and radiomics. The method analyzes the internal structure and surrounding environment of the lesion, effectively quantifies the heterogeneity within the lesion, obtains the habitat characteristics of the expanded ROI data, and combines the habitat characteristics and radiomics characteristics of the expanded ROI data to obtain comprehensive characteristics. Six types of sample sets are constructed, and the six types of samples are used to train models separately based on the introduction of multiple machine learning algorithms. The model combination with the best evaluation results is selected to form a chronic osteomyelitis prediction model group. The chronic osteomyelitis prediction model group can be effectively applied to lesions with blurred boundaries. When the lesion boundaries are blurred, the habitat characteristics and radiomics characteristics can be combined to more accurately locate the lesions, providing surgeons with clearer lesion resection boundaries, further realizing the transition from "scientific cutting" to "precise cutting", reducing unnecessary tissue damage, and reducing the risk of postoperative recurrence and complications.

[0048] Example 2: As shown in the attached Figure 3 As shown, the embodiment of the present invention is a further optimization of the above embodiment, wherein obtaining the first sample set, the second sample set, the third sample set, the fourth sample set, the fifth sample set and the sixth sample set includes: Step S210 , obtaining an original MRI image data set, and outlining the lesion and the surrounding area for each original MRI image data to obtain an original ROI data set.

[0049] The above-mentioned original MRI image dataset is obtained. Specifically, the original MRI image dataset includes a plurality of original MRI image data. Each original MRI image data is acquired by an MR scanner. Specifically, the patients all receive conventional T1-weighted sequence and T2-weighted sequence, and STIR sequence scanning is added to the T2 sequence. During the scanning, coronal, sagittal, and axial planes are used according to the location of the lesion.

[0050] It should also be noted that the original MRI image dataset can also be screened, and the screening can be performed through manual evaluation. Specifically, since osteomyelitis appears as low signal intensity on T1WI and high signal on T2WI and short Tau inversion recovery images, the above criteria are used for evaluation and screening, provided that the image sequence is complete and free of artifact interference.

[0051] For example, a German Siemens 1.5T MR scanner was used for patient data acquisition. The main scanning parameters of the patients were T1WI TR 600ms TE9.5ms, T2WI TR 3000ms TE 88ms, and FS T2WI TR 3600ms TE83ms. The FOV was 320mm and the matrix size was 256×256. All patients received conventional T1-weighted sequence and T2-weighted sequence, and fat suppression sequence (STIR sequence) was added to the T2 sequence. The coronal, sagittal, and axial planes were used for scanning according to the location of the lesion, with a layer thickness of 4mm and a layer spacing of 0.4mm. The voxel spacing of all volumes was standardized to 1mm×1mm×1mm by fixed-resolution resampling.

[0052] The above process delineates the lesion and surrounding area on each raw MRI image data to obtain the original ROI dataset. The region of interest (ROI) of each raw MRI image data is delineated based on ITK-SNAP to obtain the corresponding raw ROI data. It should also be noted that manual verification and correction can be performed after delineation.

[0053] Step S220 , performing lesion area expansion on each original ROI data set of the original ROI data set to obtain a first expanded ROI data set and a second expanded ROI data set.

[0054] The mask filling tool on the OnekeyAI platform can be used to systematically expand the ROI mask at different radii for each original ROI, forming a first and second expanded ROI dataset. The expansion distance can be set to 5mm and 10mm. It should also be noted that each expanded area can also be manually corrected.

[0055] Step S230, extracting habitat features and radiomic features of each first expanded ROI data set in the first expanded ROI data set to form a first sample set and a second sample set, and extracting multiple features therefrom to form a third sample set; Step S240 , extracting habitat features and radiomics features of each second expanded ROI data set in the second expanded ROI data set to form a fourth sample set and a fifth sample set, and extracting multiple features therefrom to form a sixth sample set.

[0056] The implementation method of the above steps S230 and S240 is the same, wherein as shown in the attached Figure 4 As shown, the habitat characteristics and imaging omics characteristics of each first expanded ROI data set in the first expanded ROI data set are extracted to form a first sample set and a second sample set, and multiple features are extracted therefrom to form a third sample set, including: Step S231 , extracting local features from each voxel in the first expanded ROI data, and converting each local feature into a 19-dimensional feature vector using a moving window technique as a subregion; The moving window here can be but is not limited to a 5×5×5 moving window. The 19-dimensional feature vector includes shape descriptors, texture features and first-order statistical attributes, specifically including firstorder_Entropy, firstorder_MeanAbsoluteDeviation, firstorder_Median, glcm_DifferenceAverage, glcm_DifferenceEntropy, glcm_DifferenceVariance, glcm_Imc1, glcm_Imc2, glcm_InverseVariance, glcm_JointEnergy, glcm_JointEntropy, glcm_SumEntropy, grllm_LongRunEmphasis, grllm_RunEntropy, grllm_RunVariance, glszm_SizeZoneNonUniformityNormalized, glszm_SmallAreaHighGrayLevelEmphasis, ngtdm_Contrast, and ngtdm_Strength.

[0057] Further, as attached Figure 5 As shown, the above-mentioned extraction of local features from each voxel in the first expanded ROI data is achieved by entropy, mean absolute deviation, joint energy, and joint entropy in sequence, as follows: Entropy is the uncertainty / randomness in the image values ​​and is calculated as: The mean absolute deviation (MAD) is the average distance between all intensity values ​​and the mean of the image array. The calculation formula of MAD is: Difference entropy is a measure of the randomness / variability in the differences in neighborhood intensity values. The calculation formula for difference entropy is: The variance is a measure of heterogeneity that gives higher weights to pairs of different intensity levels. The variance is calculated as: Joint energy is a measure of uniform patterns in an image. The calculation formula for joint energy is: Joint entropy is a measure of the randomness / variability of neighborhood strength values. The calculation formula for joint entropy is: Step S232: clustering all sub-regions in the first expanded ROI data using the K-means algorithm, and setting the sub-regions classified into the same category to the same cluster ID; Here, the K-means algorithm was used to cluster all subregions in the first expanded ROI data, that is, all voxels and their related features were clustered. Multiple cluster centers ranging from 3 to 10 were analyzed to classify different habitats within the lesion. Subregions classified into the same category were set to the same cluster ID, and the effectiveness of the clustering was evaluated using the Calinski-Harabasz score.

[0058] The principle of the K-means algorithm is to divide the data into K different clusters and iteratively update the centroids of these clusters to minimize the sum of squares within each cluster. The core of the K-means algorithm is the objective function, which is optimized to achieve effective clustering. The objective function is expressed as: in, J is the objective function; N is the number of data points; K is the number of clusters; W ik is a binary indicator (if the data point i In clustering k If yes, it is 1, otherwise it is 0); x i It is i data points; u k It is clustering k The center of mass; is a data point i and centroid k The squared Euclidean distance between them.

[0059] In step S233 , the sub-regions with the same cluster ID are merged to form one or more habitat regions, and the unique microenvironmental features representing the lesion in each habitat region are extracted to obtain the habitat features of the first expanded ROI data.

[0060] Furthermore, to address the non-clustered areas caused by the unsupervised nature of the clustering algorithm, the KNN method was used to ensure label consistency between the various habitats. In addition, feature extraction was performed using the pyradiomics tool.

[0061] The above steps S231 to S233 realize independent analysis of lesion heterogeneity. The obtained analysis results are the habitat characteristics composed of the microenvironmental characteristics of each habitat area in the expanded ROI data, which improves the quantification of intralesion heterogeneity and lays the foundation for more accurate lesion localization.

[0062] Specifically, during osteomyelitis imaging, lesions often cause abnormal enlargement and morphological changes in adjacent bone structures, manifesting as elongated, irregular extensions, indicating abnormal bone structural changes and the extent of osteomyelitis invasion. Among texture features, the high grayscale continuity (HighGrayLevelRunEmphasis) of the GLCM feature and the largeAreaHighGrayLevelEmphasis of the wavelet feature were emphasized. The surface model relies heavily on features in high grayscale regions for prediction, particularly within the peripheral 5mm region. These features reflect large areas of high density or high signal intensity, suggesting bone destruction and spread of inflammation caused by infection or inflammation. Furthermore, the high grayscale zone emphasis of the LBP feature describes the heterogeneity of the size of regions with the same grayscale value. Its high negative weight suggests that in the peripheral region of osteomyelitis, a more uniform grayscale distribution may be associated with less active lesions or inflammation. Conversely, higher heterogeneity is more indicative of disease activity because it indicates that the region may contain more irregular imaging features, helping to exclude homogeneous tissue structures with less heterogeneity.

[0063] Therefore, habitat imaging shows significant potential in the clinical diagnosis and treatment of chronic osteomyelitis. When a patient comes to the hospital, a standard MRI scan can be performed first to obtain imaging data for habitat imaging analysis. The lesion area is then divided into multiple "habitats" or sub-regions. The sub-regions are identified by algorithms such as K-means clustering, which can capture the heterogeneity and microenvironmental characteristics within the lesions. The personalized imaging genomics features derived from these sub-regions can more accurately assess the aggressiveness and treatment response of the lesions. During the surgical planning stage, the detailed information provided by habitat imaging is used to formulate a personalized surgical plan, which can achieve a high probability of complete removal of the lesion while protecting healthy bone tissue to the greatest extent. This strategy not only helps to completely eliminate the source of infection, but also effectively protects the structural integrity and function of the bones, minimizing the patient's postoperative recovery time and complication rate.

[0064] Step S234: extracting geometric features, intensity features, and texture features of the first expanded ROI data, and screening the extracted features to obtain radiomic features of the first expanded ROI data; Specifically included here: (1) Extraction of radiomic features Radiomic features were extracted for each subregion within each first / second expanded ROI. These features included geometric, intensity, and texture features. Geometric features were used to assess the shape and spatial dimensions of the lesion. Intensity features measured the brightness levels of voxels. Texture features explored spatial patterns within the lesion using sophisticated techniques (such as GLCM, GLRLM, GLSZM, and NGTDM). These radiomic features were extracted using the pyradiomics tool (version 3.0.1) and adhered to the strict protocols established by the Imaging Biomarker Standardization Initiative (IBSI).

[0065] (2) Feature screening The imaging features of each first expansion ROI / second expansion ROI were screened using T-test, correlation analysis and LASSO regression algorithm. After screening, 10 imaging features of each first expansion ROI / second expansion ROI were finally retained, including Elongation, MajorAxisLength, HighGrayLevelRunEmphasis, LargeAreaHighGrayLevelEmphasis, HighGrayLevelZoneEmphasis, etc.

[0066] Specifically: The feature distributions were normalized using the mean and standard deviation of the training cohort, and statistical evaluation was performed using t-tests with a significance level set at P < 0.05. Only features showing statistical significance were retained.

[0067] In the correlation analysis, the Pearson correlation coefficient was used to identify and remove highly correlated features, with a threshold set at 0.9. This process was further optimized using the minimum redundancy maximum correlation algorithm, which optimized the feature set to 32 features by balancing correlation and redundancy.

[0068] The final feature selection is implemented using LASSO regression, which imposes a penalty on the regression coefficients to simplify the model and effectively eliminate irrelevant features. The optimal regularization parameter λ is determined through 10-fold cross-validation to ensure that the most predictive features are selected.

[0069] The above-mentioned integrated approach from intraclass correlation coefficient (ICC) filtering to LASSO regression can establish a robust and predictive radiomics signature.

[0070] Step S235 : selecting a plurality of features from the habitat features and radiomics features of the first expanded ROI data and combining them to obtain a comprehensive feature.

[0071] Step S236: Loop through the above steps to obtain the habitat characteristics, radiomics characteristics, and comprehensive characteristics of each first expanded ROI data. The habitat characteristics of the first expanded ROI data and the corresponding chronic osteomyelitis lesion detection identification information are used as the first sample to form a first sample set. The radiomics characteristics of the first expanded ROI data and the corresponding chronic osteomyelitis lesion detection identification information are used as the second sample to form a second sample set. The comprehensive characteristics of the first expanded ROI data and the corresponding chronic osteomyelitis lesion detection identification information are used as the third sample to form a third sample set.

[0072] Example 3: The present invention verifies the effectiveness of the radiomics-based chronic osteomyelitis prediction model training method disclosed in the above examples, as follows: (1) Obtaining a sample set Patients with a high clinical suspicion of chronic osteomyelitis requiring surgical intervention were selected. Diagnosis was based on intraoperative histopathological testing, culture of the same pathogen from at least two infected sites, or a clear sinus tract directly connected to the bone. Cases caused by unusual pathogens (such as mycobacteria) were excluded. Pregnant or lactating patients, those with metallic implants, a diagnosis of acute osteomyelitis, Charcot arthropathy, diabetes, or chronic osteomyelitis in non-long bone sites were excluded.

[0073] A total of 92 patients with confirmed chronic osteomyelitis and 77 patients without chronic osteomyelitis were enrolled. Data were acquired using a Siemens 1.5T MR scanner. The original MRI dataset was generated. The primary scanning parameters were T1-weighted TR (T1-weighted image) 600ms, TE (time per second) 9.5ms; T2-weighted TR (T2-weighted image) 3000ms, TE (time per second) 88ms; and FS T2-weighted TR (T2-weighted image) 3600ms, TE (time per second) 83ms. A field of view (FOV) of 320 mm and a matrix size of 256 × 256 were used. All patients underwent conventional T1-weighted and T2-weighted sequences, with a fat-suppressed STIR sequence added to the T2 sequence. Coronal, sagittal, and axial views were used depending on lesion location, with a slice thickness of 4 mm and a slice spacing of 0.4 mm. Voxel spacing across all volumes was normalized to 1 mm × 1 mm × 1 mm using a fixed-resolution resampling method.

[0074] The lesion and the surrounding area are delineated on each original MRI image data to obtain an original ROI data set, and the lesion area is expanded on each original ROI data set to obtain a first expanded ROI data set and a second expanded ROI data set.

[0075] Obtaining a corresponding first sample set, a second sample set, a third sample set, a fourth sample set, a fifth sample set, and a sixth sample set based on the first expanded ROI data set and the second expanded ROI data set; The first sample set includes a plurality of first samples, each of which includes habitat features of the first expanded ROI data and corresponding chronic osteomyelitis lesion detection identification information; The second sample set includes a plurality of second samples, each of which includes the radiomics features of the first expanded ROI data and corresponding chronic osteomyelitis lesion detection identification information; The third sample set includes a plurality of third samples, each of which includes comprehensive features of the first expanded ROI data and corresponding chronic osteomyelitis lesion detection identification information; The fourth sample set includes a plurality of fourth samples, each of which includes the habitat characteristics of the second expanded ROI data and corresponding chronic osteomyelitis lesion detection identification information; The fifth sample set includes a plurality of fifth samples, each of which includes the radiomics features of the second expanded ROI data and corresponding chronic osteomyelitis lesion detection identification information; The sixth sample set includes a plurality of sixth samples, each of which includes comprehensive features of the second expanded ROI data and corresponding chronic osteomyelitis lesion detection identification information.

[0076] (2) Model training The training sets of the first sample set, the second sample set, the third sample set, the fourth sample set, the fifth sample set and the sixth sample set are used to train multiple machine learning algorithms respectively, and the machine learning algorithms include SVM, RandomForest, ExtraTrees, XGBoost and LightGBM.

[0077] (3) Model testing The test sets and validation sets of the first, second, third, fourth, fifth and sixth sample sets were used to test and verify the corresponding six types of training model sets, and Accuracy, AUC, 95%CI, Sensitivity, Specificity, PPV, NPV and Cohort were introduced for evaluation.

[0078] For a machine learning algorithm, the results of training, testing, and validation were evaluated using Accuracy, AUC, 95% CI, Sensitivity, Specificity, PPV, NPV, and Cohort for a sample set including radiomic features of the expanded ROI data, a sample set including comprehensive features of the expanded ROI data, a sample set including habitat features of the first expanded ROI data, and a sample set including habitat features of the second expanded ROI data. The evaluation results are shown in Table 1: Table 1 Evaluation results of training, testing and validation Signature Accuracy AUC 95% CI Sensitivity Specificity PPV NPV Cohort Radiomics 0.846 0.923 0.8690-0.9774 0.861 0.836 0.775 0.902 Train P5_H1 0.813 0.933 0.8838-0.9819 0.861 0.782 0.721 0.896 Train P5_H2 0.780 0.900 0.8377-0.9628 0.833 0.745 0.682 0.872 Train P5_H3 0.956 0.983 0.9575-1.0000 0.972 0.945 0.921 0.981 Train P10_H1 0.758 0.858 0.7815-0.9347 0.833 0.709 0.652 0.867 Train P10_H2 0.824 0.878 0.8071-0.9489 0.750 0.873 0.794 0.842 Train P10_H3 0.835 0.889 0.8143-0.9635 0.750 0.891 0.818 0.845 Train Combined 0.967 0.984 0.9534-1.0000 0.972 0.964 0.946 0.981 Train Radiomics 0.757 0.874 0.7645-0.9826 0.800 0.706 0.762 0.750 val P5_H1 0.730 0.818 0.6815-0.9538 0.850 0.588 0.708 0.769 val P5_H2 0.703 0.729 0.5563-0.9025 0.700 0.706 0.737 0.667 val P5_H3 0.703 0.894 0.7908-0.9974 0.550 0.882 0.846 0.625 val P10_H1 0.811 0.847 0.7132-0.9809 0.950 0.647 0.760 0.917 val P10_H2 0.676 0.743 0.5773-0.9080 0.600 0.765 0.750 0.619 val P10_H3 0.649 0.856 0.7316-0.9802 0.500 0.824 0.769 0.583 val Combined 0.757 0.909 0.8091-1.0000 0.600 0.941 0.923 0.667 val Radiomics 0.744 0.846 0.7228-0.9685 0.870 0.600 0.714 0.800 test P5_H1 0.651 0.671 0.4992-0.8421 0.696 0.600 0.667 0.632 test P5_H2 0.651 0.773 0.6310-0.9146 0.609 0.700 0.700 0.609 test P5_H3 0.767 0.854 0.7307-0.9780 0.739 0.800 0.810 0.727 test P10_H1 0.605 0.650 0.4828-0.8172 0.739 0.450 0.607 0.600 test P10_H2 0.744 0.747 0.5940-0.8995 0.696 0.800 0.800 0.696 test P10_H3 0.698 0.722 0.5676-0.8759 0.609 0.800 0.778 0.640 test Combined 0.767 0.857 0.7408-0.9723 0.739 0.800 0.810 0.727 test Among them, P5_H1, P5_H2, and P5_H3 are the habitat characteristics of three habitat areas in the first expanded ROI data with an outward expansion of 5 mm, and P10_H1, P10_H2, and P10_H3 are the habitat characteristics of three habitat areas in the second expanded ROI data with an outward expansion of 10 mm.

[0079] The P5_H3 model achieved an AUC of 0.983 (95% CI: 0.9575 to 1.0000) in the training set and 0.894 (95% CI: 0.7908 to 0.9974) in the validation set, demonstrating robustness across diverse datasets. Furthermore, the P5_H3-based model outperformed other models in training, testing, and validation. This suggests that the region indicated by P5_H3 may harbor chronic osteomyelitis lesions that are difficult to identify with the naked eye. This provides more precise guidance for surgical debridement, aiming to minimize damage to healthy bone tissue and achieve personalized treatment strategies.

[0080] Furthermore, the DeLong test results of the above model after testing are shown in the attached figure. Figure 6 As shown in the figure, the P5_H3 model and the comprehensive feature model show significant improvement compared to other models during training.

[0081] In summary, the introduction of habitat characteristics of the habitat area and the model obtained by training can more accurately locate the lesion, thereby providing more precise guidance for surgical debridement and reducing the damage to healthy bone tissue.

[0082] Example 4: As shown in the attached Figure 7 As shown, an embodiment of the present invention discloses a method for predicting chronic osteomyelitis based on habitat imaging and radiomics, comprising: Step S310, obtaining input data to be predicted of the patient to be predicted, wherein the input data to be predicted includes habitat characteristics, radiomics characteristics, and comprehensive characteristics of the first expanded ROI data, and habitat characteristics, radiomics characteristics, and comprehensive characteristics of the second expanded ROI data; Step S320 , inputting the input data of the patient to be predicted into the corresponding model in the chronic osteomyelitis prediction model group to obtain six corresponding prediction results, wherein the chronic osteomyelitis prediction model group is obtained according to the method for constructing a chronic osteomyelitis model based on habitat imaging and radiomics as described in the above embodiment; Step S330 , combining the six prediction results to determine the corresponding auxiliary decision for debridement of chronic osteomyelitis.

[0083] In determining the extent of mature debridement, the maximum predicted positive interval across all models is generally used as the recommended resection margin. This ensures surgical thoroughness, reduces the risk of residual or recurrent chronic osteomyelitis, and embodies the principle of "thorough debridement." In clinical situations where bone preservation is clearly prioritized, precise bone preservation can be achieved based on the negative interval across all models to minimize bone damage. Specific margin decisions are based on considerations of multiple model assessments, patient characteristics, and intraoperative assessments, achieving a personalized balance.

[0084] An embodiment of the present invention discloses a method for predicting chronic osteomyelitis based on habitat imaging and radiomics. The method is based on a chronic osteomyelitis prediction model obtained by a method for constructing a chronic osteomyelitis model based on habitat imaging and radiomics, and on the basis of obtaining habitat characteristics and radiomics characteristics, the method deeply quantifies the internal heterogeneity of the lesion. When the lesion boundary is blurred, the method can more accurately locate the lesion, provide surgeons with a clearer lesion resection boundary, and further realize the transition from "scientific cutting" to "precise cutting", thereby reducing unnecessary tissue damage and lowering the risk of postoperative recurrence and complications.

[0085] Example 5: As shown in the attached Figure 8 As shown, an embodiment of the present invention discloses a chronic osteomyelitis prediction model training device based on habitat imaging and radiomics, comprising: a sample acquisition unit, acquiring a first sample set, a second sample set, a third sample set, a fourth sample set, a fifth sample set, and a sixth sample set, and dividing each sample set into a training set and a test set in proportion, wherein the samples in the first sample set, the second sample set, and the third sample set respectively include different feature types of the first expanded ROI data and corresponding chronic osteomyelitis lesion detection identification information, and the samples in the fourth sample set, the fifth sample set, and the sixth sample set respectively include different feature types of the second expanded ROI data and corresponding chronic osteomyelitis lesion detection identification information, and the feature types include habitat features, imaging omics features, and comprehensive features; A training unit, using the training sets of the first sample set, the second sample set, the third sample set, the fourth sample set, the fifth sample set, and the sixth sample set to respectively train a plurality of machine learning algorithms to obtain corresponding six types of training model sets; The testing unit uses the test sets of the first sample set, the second sample set, the third sample set, the fourth sample set, the fifth sample set and the sixth sample set to test the corresponding six types of training model sets respectively, and selects the models with the best evaluation results from the six types of training model sets to form a chronic osteomyelitis prediction model group.

[0086] Example 6: As shown in the attached Figure 9 As shown, an embodiment of the present invention discloses a chronic osteomyelitis prediction device based on habitat imaging and radiomics, comprising: A prediction data acquisition unit is configured to acquire input data to be predicted of a patient to be predicted, wherein the input data to be predicted includes habitat characteristics, imaging omics characteristics, and comprehensive characteristics of the first expanded ROI data, and habitat characteristics, imaging omics characteristics, and comprehensive characteristics of the second expanded ROI data; A prediction unit, inputting input data of a patient to be predicted into a corresponding model in a chronic osteomyelitis prediction model group to obtain six corresponding prediction results, wherein the chronic osteomyelitis prediction model group is obtained by the method according to any one of claims 1 to 5; The result output unit combines the six prediction results to determine the corresponding auxiliary decision for chronic osteomyelitis debridement.

[0087] Example 7: The embodiment of the present invention discloses a storage medium, on which a computer program readable by a computer is stored, and the computer program is configured to execute a method for predicting chronic osteomyelitis based on habitat imaging and radiomics when running.

[0088] The above storage medium may include, but is not limited to, various media that can store computer programs, such as a USB flash drive, a read-only memory, a mobile hard disk, a magnetic disk, or an optical disk.

[0089] Example 8: An embodiment of the present invention discloses an electronic device, including a processor and a memory, wherein the memory stores a computer program, and the computer program is loaded and executed by the processor to implement a chronic osteomyelitis prediction method based on habitat imaging and radiomics.

[0090] The processor may be a central processing unit (CPU), a general-purpose processor, a digital signal processor (DSP), an ASIC, an FPGA, or other programmable logic device, a transistor logic device, a hardware component, or any combination thereof. It may implement or execute the various exemplary logic blocks, modules, and circuits described in conjunction with the present disclosure. It may also be a combination that implements computing functions, such as a combination of one or more microprocessors, a combination of a DSP and a microprocessor, and so on. Memory may include, but is not limited to, various media capable of storing computer programs, such as USB flash drives, read-only memories, removable hard drives, magnetic disks, or optical disks.

[0091] Those skilled in the art will appreciate that embodiments of the present invention may be provided as methods, systems, or computer program products. Thus, the present invention may take the form of an entirely hardware embodiment, an entirely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention may take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk drives, CD-ROMs, optical storage devices, etc.) containing computer-usable program code. The solutions in the embodiments of the present invention may be implemented using various computer languages, such as the object-oriented programming language Java and the interpreted scripting language JavaScript.

[0092] The present invention is described with reference to flowcharts and / or block diagrams of methods, devices (systems), and computer program products according to embodiments of the present invention. It should be understood that each process and / or block in the flowcharts and / or block diagrams, as well as combinations of processes and / or blocks in the flowcharts and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowcharts and / or block diagrams. Figure 1 a process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0093] These computer program instructions may also be stored in a computer readable memory that can direct a computer or other programmable data processing device to work in a specific manner, so that the instructions stored in the computer readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 a process or multiple processes and / or boxes Figure 1 The function specified in one or more boxes.

[0094] The above content is only a specific implementation method of the present invention, which has strong adaptability and implementation effect, but the scope of protection of the present invention is not limited to this. Any technician familiar with this technical field can easily think of changes or replacements within the technical scope disclosed by the present invention, which should be covered by the scope of protection of the present invention. Therefore, equivalent changes made according to the claims of the present invention are still within the scope covered by the present invention.

Claims

1. A method for training a chronic osteomyelitis prediction model based on habitat imaging and radiomics, characterized in that: include: Obtain a first sample set, a second sample set, a third sample set, a fourth sample set, a fifth sample set, and a sixth sample set, and divide each sample set into a training set and a test set in proportion, wherein the samples in the first sample set, the second sample set, and the third sample set respectively include different feature types of the first expanded ROI data and corresponding chronic osteomyelitis lesion detection identification information, and the samples in the fourth sample set, the fifth sample set, and the sixth sample set respectively include different feature types of the second expanded ROI data and corresponding chronic osteomyelitis lesion detection identification information, and the feature types include habitat features, imaging omics features, and comprehensive features; Using the training sets of the first sample set, the second sample set, the third sample set, the fourth sample set, the fifth sample set, and the sixth sample set to train multiple machine learning algorithms respectively, to obtain corresponding six types of training model sets; The test sets of the first sample set, the second sample set, the third sample set, the fourth sample set, the fifth sample set and the sixth sample set were used to test the corresponding six types of training model sets respectively, and the models with the best evaluation results were selected from the six types of training model sets to form a chronic osteomyelitis prediction model group.

2. The method for training a chronic osteomyelitis prediction model based on habitat imaging and radiomics according to claim 1, characterized in that: The obtaining of the first sample set, the second sample set, the third sample set, the fourth sample set, the fifth sample set, and the sixth sample set includes: Obtain the original MRI image data set, and outline the lesion and surrounding area for each original MRI image data to obtain the original ROI data set; Expanding the lesion area of ​​each original ROI data set of the original ROI data set to obtain a first expanded ROI data set and a second expanded ROI data set; Extracting habitat characteristics and imaging omics features of each first expanded ROI data set in the first expanded ROI data set to form a first sample set and a second sample set, and extracting multiple features therefrom to form a third sample set; The habitat characteristics and imaging omics characteristics of each second expanded ROI data in the second expanded ROI data set are extracted to form a fourth sample set and a fifth sample set, and multiple features are extracted therefrom to form a sixth sample set.

3. The method for training a chronic osteomyelitis prediction model based on habitat imaging and radiomics according to claim 2, characterized in that: The step of extracting habitat characteristics and imaging omics characteristics of each first expanded ROI data set in the first expanded ROI data set to form a first sample set and a second sample set, and extracting multiple features therefrom to form a third sample set includes: Extract local features from each voxel in the first expanded ROI data, and use the moving window technique to convert each local feature into a 19-dimensional feature vector as a subregion; All sub-regions in the first expanded ROI data are clustered using the K-means algorithm, and sub-regions classified into the same category are set to the same cluster ID; Merge the sub-regions with the same cluster ID to form one or more habitat regions, and extract the unique microenvironmental features representing the lesion in each habitat region to obtain the habitat features of the first expanded ROI data; Extracting geometric features, intensity features, and texture features of the first expanded ROI data, and screening the extracted features to obtain imaging omics features of the first expanded ROI data; Select multiple features from the habitat characteristics and imaging omics features of the first expanded ROI data and combine them to obtain comprehensive features; The above steps are repeated to obtain the habitat characteristics, radiomics characteristics, and comprehensive characteristics of each first expanded ROI data. The habitat characteristics of the first expanded ROI data and the corresponding chronic osteomyelitis lesion detection identification information are used as the first sample to form a first sample set. The radiomics characteristics of the first expanded ROI data and the corresponding chronic osteomyelitis lesion detection identification information are used as the second sample to form a second sample set. The comprehensive characteristics of the first expanded ROI data and the corresponding chronic osteomyelitis lesion detection identification information are used as the third sample to form a third sample set.

4. The method for training a chronic osteomyelitis prediction model based on habitat imaging and radiomics according to claim 3, characterized in that: The extracting of local features from each voxel in the first expanded ROI data is achieved in sequence by entropy, mean absolute deviation, joint energy, and joint entropy; or / and, the screening of the extracted features is achieved in sequence by T-test, correlation analysis, and LASSO regression algorithm.

5. The method for training a chronic osteomyelitis prediction model based on habitat imaging and radiomics according to any one of claims 1 to 4, characterized in that: The machine learning algorithms include SVM, RandomForest, ExtraTrees, XGBoost, and LightGBM.

6. A method for predicting chronic osteomyelitis based on habitat imaging and radiomics, characterized in that: include: Obtaining input data to be predicted of a patient to be predicted, wherein the input data to be predicted includes habitat characteristics, imaging omics characteristics, and comprehensive characteristics of the first expanded ROI data, and habitat characteristics, imaging omics characteristics, and comprehensive characteristics of the second expanded ROI data; Inputting the predicted input data of the patient to be predicted into the corresponding model in the chronic osteomyelitis prediction model group to obtain six corresponding prediction results, wherein the chronic osteomyelitis prediction model group is obtained by the method according to any one of claims 1 to 5; Combine the six prediction results to determine the corresponding auxiliary decision for chronic osteomyelitis debridement.

7. A chronic osteomyelitis prediction model training device based on habitat imaging and radiomics using the method according to any one of claims 1 to 5, characterized in that: include: a sample acquisition unit, acquiring a first sample set, a second sample set, a third sample set, a fourth sample set, a fifth sample set, and a sixth sample set, and dividing each sample set into a training set and a test set in proportion, wherein the samples in the first sample set, the second sample set, and the third sample set respectively include different feature types of the first expanded ROI data and corresponding chronic osteomyelitis lesion detection identification information, and the samples in the fourth sample set, the fifth sample set, and the sixth sample set respectively include different feature types of the second expanded ROI data and corresponding chronic osteomyelitis lesion detection identification information, and the feature types include habitat features, imaging omics features, and comprehensive features; A training unit, using the training sets of the first sample set, the second sample set, the third sample set, the fourth sample set, the fifth sample set, and the sixth sample set to respectively train a plurality of machine learning algorithms to obtain corresponding six types of training model sets; The testing unit uses the test sets of the first sample set, the second sample set, the third sample set, the fourth sample set, the fifth sample set and the sixth sample set to test the corresponding six types of training model sets respectively, and selects the models with the best evaluation results from the six types of training model sets to form a chronic osteomyelitis prediction model group.

8. A chronic osteomyelitis prediction device based on habitat imaging and radiomics using the method of claim 6, characterized in that: include: A prediction data acquisition unit is configured to acquire input data to be predicted of a patient to be predicted, wherein the input data to be predicted includes habitat characteristics, imaging omics characteristics, and comprehensive characteristics of the first expanded ROI data, and habitat characteristics, imaging omics characteristics, and comprehensive characteristics of the second expanded ROI data; A prediction unit, inputting input data of a patient to be predicted into a corresponding model in a chronic osteomyelitis prediction model group to obtain six corresponding prediction results, wherein the chronic osteomyelitis prediction model group is obtained by the method according to any one of claims 1 to 5; The result output unit combines the six prediction results to determine the corresponding auxiliary decision for chronic osteomyelitis debridement.

9. An electronic device, characterized in that: The method comprises a processor and a memory, wherein a computer program is stored in the memory, and the computer program is loaded and executed by the processor to implement the steps in the method according to claim 6.

10. A storage medium, characterized in that: The storage medium stores a computer program that can be read by a computer, and the computer program is configured to execute the steps of the method according to claim 6 when run.

Images