Chronic osteomyelitis prediction model training method and device based on radiomics, chronic osteomyelitis prediction method and device, electronic equipment and storage medium
By constructing a radiomics-based chronic osteomyelitis prediction model and using multiple sample sets and various machine learning algorithm training, the problems of subjectivity and individual differences in radiological visual judgment were solved, achieving more accurate lesion assessment and debridement auxiliary decision-making, and reducing the risk of bone injury.
Patent Information
- Application Number
- CN202510632529.3
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-05-16
- Publication Date
- 2025-09-19
AI Technical Summary
In existing technologies, locating the debridement boundary of chronic osteomyelitis based on experience and visual judgment of imaging is limited by subjectivity and individual differences. It is difficult to find the optimal balance between thorough debridement and minimizing bone damage, resulting in inaccurate mapping of the region of interest, which may miss lesions or mix with healthy tissue.
By obtaining multiple sample sets and dividing them into training sets and test sets, a chronic osteomyelitis prediction model group was formed using multiple machine learning algorithms. Combined with radiomics feature extraction and filtering methods, prediction models of the original ROI, the first expanded ROI, and the second expanded ROI were constructed to provide auxiliary debridement decision-making.
It improves the accuracy and comprehensiveness of the assessment of the extent of chronic osteomyelitis lesions, provides a safe balance between thorough debridement and minimizing bone damage, reduces the risk of recurrence and reduces bone damage.
Smart Images
Figure CN120674027A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the field of medical information technology, and is a radiomics-based chronic osteomyelitis prediction model training method, prediction method, device, electronic equipment and storage medium. Background Art
[0002] Chronic osteomyelitis is believed to be treated by the surgeon relying on experience and visual judgment based on imaging to determine the debridement boundaries. However, this approach is subject to subjectivity and individual differences. Even experienced specialists sometimes struggle to accurately distinguish the boundary between infection and healthy tissue, and to find the optimal balance between "thorough debridement" and "minimizing bone damage." The difficulty of treating stubborn infectious bone diseases poses a significant challenge to orthopedic surgeons. Surgical debridement plays a crucial role in combating chronic osteomyelitis. However, the success of surgery often hangs by a thread: an overly conservative debridement may leave residual infection and lead to recurrence, while an overly aggressive debridement will inevitably cause unnecessary bone damage.
[0003] Currently, surgeons often rely on experience and visual judgment based on imaging to determine the debridement boundary. However, this method is subject to subjectivity and individual variability. Even experienced specialists sometimes struggle to accurately distinguish the boundary between infected and healthy tissue, and find the optimal balance between "thorough debridement" and "minimizing bone damage." Therefore, accurately mapping the region of interest (ROI) of chronic osteomyelitis lesions to the resection area is a difficult task. If the ROI is too small, critical lesions may be missed, while if it is too large, healthy tissue may be lost.
[0004] With the rapid development of artificial intelligence, how to combine machine learning to accurately map the region of interest (ROI) of chronic osteomyelitis lesions to the resection range, provide effective auxiliary decision-making for chronic osteomyelitis debridement, and provide surgeons with more comprehensive lesion assessment and clearer surgical margin selection has become a key issue. Summary of the Invention
[0005] The present invention provides a radiomics-based chronic osteomyelitis prediction model training method, prediction method, device, electronic device and storage medium, which overcomes the shortcomings of the above-mentioned existing technologies. It can effectively solve the problem that the existing method of locating the debridement boundary in chronic osteomyelitis debridement relies on experience and visual judgment of imaging, which is limited by subjectivity and individual differences and is difficult to comprehensively evaluate the scope of chronic osteomyelitis lesions.
[0006] One of the technical solutions of the present invention is achieved through the following measures: a radiomics-based chronic osteomyelitis prediction model training method, comprising: Obtain a first sample set, a second sample set, and a third sample set, and divide each sample set into a training set and a test set in proportion, wherein the first sample set includes a plurality of first samples, each first sample includes an original ROI feature and corresponding chronic osteomyelitis lesion detection identification information, the second sample set includes a plurality of second samples, each second sample includes a first expanded ROI feature and corresponding chronic osteomyelitis lesion detection identification information, and the third sample set includes a plurality of third samples, each third sample includes a second expanded ROI feature and corresponding chronic osteomyelitis lesion detection identification information; Using the training sets in the first sample set, the second sample set, and the third sample set to train multiple machine learning algorithms respectively, to obtain corresponding original ROI training model sets, first expanded ROI training model sets, and second expanded ROI training model sets; The original ROI training model set, the first expanded ROI training model set, and the second expanded ROI training model set were tested using the test sets in the first sample set, the second sample set, and the third sample set, respectively. The models with the best evaluation results were selected as the original ROI prediction model, the first expanded ROI prediction model, and the second expanded ROI prediction model to form a chronic osteomyelitis prediction model group.
[0007] The following are further optimizations and / or improvements to the above technical solutions: The above process of obtaining the first sample set, the second sample set, and the third sample set includes: Acquire an original MRI image dataset, wherein the original MRI image dataset includes a plurality of original MRI image data; Each MRI image data is delineated for the lesion and surrounding area, and the original MRI image data set is converted into the original ROI data set, the first expanded ROI data set, and the second expanded ROI data set based on the delineation results; Feature extraction and screening are performed on the original ROI dataset, the first expanded ROI dataset, and the second expanded ROI dataset respectively to form the first sample set, the second sample set, and the third sample set, and the chronic osteomyelitis lesion detection identification information of all samples in each sample set is determined.
[0008] The above-mentioned feature extraction and screening processes for the original ROI dataset, the first expanded ROI dataset, and the second expanded ROI dataset are the same, wherein the feature extraction and screening of the original ROI dataset include: Select a filtering method and extract the radiomics features in each raw ROI data in combination with the selected filtering method; The imaging features of each original ROI data were screened once using T test and LASSO regression algorithm; The mRMR classifier was used to perform a secondary screening of the primary screening results, and multiple features with the highest correlation with the chronic osteomyelitis lesion detection results and the lowest correlation between features were obtained.
[0009] Each first expanded ROI data in the first expanded ROI data set is ROI data expanded outward by 5 mm from the corresponding original ROI data, and each second expanded ROI data in the second expanded ROI data set is ROI data expanded outward by 10 mm from the corresponding original ROI data.
[0010] The above-mentioned machine learning algorithms include KNN algorithm, SVM algorithm, RF algorithm, and Logistic Regression algorithm.
[0011] The second technical solution of the present invention is achieved through the following measures: a method for predicting chronic osteomyelitis based on radiomics, comprising: Obtaining the original ROI feature, the first expanded ROI feature, and the second expanded ROI feature of the patient to be predicted; Inputting the original ROI feature, the first expanded ROI feature, and the second expanded ROI feature of the patient to be predicted into the corresponding model in the chronic osteomyelitis prediction model group to obtain the corresponding original ROI prediction result, the first expanded ROI prediction result, and the second expanded ROI prediction result, wherein the chronic osteomyelitis prediction model group is obtained according to the method according to any one of claims 1 to 5; Combined with the chronic osteomyelitis prediction results, the conditions are determined to obtain the corresponding chronic osteomyelitis debridement auxiliary decision-making, where the chronic osteomyelitis prediction results include: (a) If the original ROI prediction model predicts a positive result, and both the first and second expanded ROI prediction results are negative, the debridement auxiliary decision for chronic osteomyelitis is to perform debridement with minimal bone damage; (b) If the prediction result of the first expanded ROI is positive, the auxiliary decision for debridement of chronic osteomyelitis is to appropriately expand the debridement based on the original osteomyelitis resection range; (c) If the prediction result of the second expanded ROI is positive, the auxiliary decision for debridement of chronic osteomyelitis is complete debridement; Among them, positive means that there are chronic osteomyelitis lesions in the area, and negative means that there are no chronic osteomyelitis lesions in the area.
[0012] The third technical solution of the present invention is achieved by the following measures: a chronic osteomyelitis prediction model training device based on radiomics, comprising: a sample acquisition unit, acquiring a first sample set, a second sample set, and a third sample set, and dividing each sample set into a training set and a test set in proportion, wherein the first sample set includes a plurality of first samples, each of which includes an original ROI feature and corresponding chronic osteomyelitis lesion detection identification information; the second sample set includes a plurality of second samples, each of which includes a first expanded ROI feature and corresponding chronic osteomyelitis lesion detection identification information; and the third sample set includes a plurality of third samples, each of which includes a second expanded ROI feature and corresponding chronic osteomyelitis lesion detection identification information; A model training unit, which uses the training sets in the first sample set, the second sample set, and the third sample set to train multiple machine learning algorithms respectively to obtain corresponding original ROI training model sets, first expanded ROI training model sets, and second expanded ROI training model sets; The model selection unit uses the test sets in the first sample set, the second sample set and the third sample set to test the original ROI training model set, the first expanded ROI training model set and the second expanded ROI training model set respectively, and selects the model with the best evaluation result as the original ROI prediction model, the first expanded ROI prediction model and the second expanded ROI prediction model respectively to form a chronic osteomyelitis prediction model group.
[0013] The fourth technical solution of the present invention is achieved by the following measures: a chronic osteomyelitis prediction device based on radiomics, comprising: A prediction data acquisition unit is configured to acquire the original ROI feature, the first expanded ROI feature, and the second expanded ROI feature of the patient to be predicted; A prediction unit, inputting the original ROI feature, the first expanded ROI feature, and the second expanded ROI feature of the patient to be predicted into a corresponding model in a chronic osteomyelitis prediction model group, and obtaining a corresponding original ROI prediction result, the first expanded ROI prediction result, and the second expanded ROI prediction result, wherein the chronic osteomyelitis prediction model group is obtained by the method according to any one of claims 1 to 5; The analysis unit determines the conditions based on the chronic osteomyelitis prediction results to obtain the corresponding chronic osteomyelitis debridement auxiliary decision, wherein the chronic osteomyelitis prediction results include: (a) If the original ROI prediction model predicts a positive result, and both the first and second expanded ROI prediction results are negative, the debridement auxiliary decision for chronic osteomyelitis is to perform debridement with minimal bone damage; (b) If the prediction result of the first expanded ROI is positive, the auxiliary decision for debridement of chronic osteomyelitis is to appropriately expand the debridement based on the original osteomyelitis resection range; (c) If the prediction result of the second expanded ROI is positive, the auxiliary decision for debridement of chronic osteomyelitis is complete debridement; Among them, positive means that there are chronic osteomyelitis lesions in the area, and negative means that there are no chronic osteomyelitis lesions in the area.
[0014] 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 imaging omics.
[0015] 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 imaging omics when it is run.
[0016] The present invention uses the original ROI as a sample for subsequent model training, expands it twice, constructs three types of samples, and uses the three types of samples to perform model training respectively based on the introduction of multiple machine learning algorithms. The model with the best evaluation result is selected as the original ROI prediction model, the first expanded ROI prediction model, and the second expanded ROI prediction model to form a chronic osteomyelitis prediction model group. In this way, the comprehensiveness and accuracy of the model prediction results are improved by utilizing the step-by-step expansion strategy and algorithm optimization, so that effective auxiliary decision-making for chronic osteomyelitis debridement can be obtained when applied, providing support for surgeons to obtain more comprehensive lesion assessment and clearer surgical margin selection, transforming from "one-size-fits-all" to "scientific cutting", and providing accurate and effective auxiliary data support for finding a relatively safe balance between "thorough debridement" and "minimizing bone damage", thereby reducing bone damage on the basis of reducing the risk of recurrence and promoting the formulation of more comprehensive and accurate surgical plans. BRIEF DESCRIPTION OF THE DRAWINGS
[0017] Attachment Figure 1 A schematic diagram of an implementation environment provided for one embodiment of the present invention.
[0018] Attachment Figure 2 A flowchart of a model training method provided by one embodiment of the present invention.
[0019] Attachment Figure 3 A schematic flow chart of a method for obtaining a sample set according to an embodiment of the present invention.
[0020] Attachment Figure 4 A schematic diagram of a model training workflow provided for one embodiment of the present invention.
[0021] Attachment Figure 5 A schematic diagram of using 3Dslicer to outline osteomyelitis lesions is provided in accordance with one embodiment of the present invention.
[0022] Attachment Figure 6A schematic diagram of the application effect of using the t-test and LASSO regression algorithm in selecting texture features within a lesion, provided in one embodiment of the present invention.
[0023] Attachment Figure 7 A schematic diagram of the features and their weights finally selected after secondary screening provided by one embodiment of the present invention.
[0024] Attachment Figure 8 This is a heat map of the correlation matrix of the features finally selected after the secondary screening provided by one embodiment of the present invention.
[0025] Attachment Figure 9 A schematic diagram of the ROC curve of a model provided by one embodiment of the present invention.
[0026] Attachment Figure 10 A schematic diagram of a model CIC curve provided by one embodiment of the present invention.
[0027] Attachment Figure 11 A flowchart of a method for predicting chronic osteomyelitis provided by one embodiment of the present invention.
[0028] Attachment Figure 12 A schematic diagram of a case flow diagram provided for an embodiment of the present invention.
[0029] Attachment Figure 13 A schematic diagram of the structure of a model training device provided in one embodiment of the present invention.
[0030] Attachment Figure 14 A schematic structural diagram of a chronic osteomyelitis prediction device provided by one embodiment of the present invention. DETAILED DESCRIPTION
[0031] 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.
[0032] Those skilled in the art will understand that, unless otherwise specified, a "module" or "unit" in the embodiments of the present application 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 (such as 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.
[0033] In addition, the “plurality” in the embodiments of the present application refers to two or more, and “first” and “second” are used to distinguish descriptions and should not be understood as implying relative importance.
[0034] Currently, surgeons often rely on experience and visual judgment based on imaging to determine the debridement boundary. However, this method is subject to subjectivity and individual variability. Even experienced specialists sometimes struggle to accurately distinguish the boundary between infected and healthy tissue, and find the optimal balance between "thorough debridement" and "minimizing bone damage." Therefore, accurately mapping the region of interest (ROI) of chronic osteomyelitis lesions to the resection area is a difficult task. If the ROI is too small, critical lesions may be missed, while if it is too large, healthy tissue may be lost.
[0035] The embodiment of the present application provides a chronic osteomyelitis prediction model training method based on radiomics, a prediction method, an apparatus, an electronic device and a storage medium, which obtains a first sample set, a second sample set and a third sample set, and divides each sample set into a training set and a test set in proportion; uses the training sets in the first sample set, the second sample set and the third sample set to train a plurality of machine learning algorithms respectively, and obtains the corresponding original ROI training model set, the first expanded ROI training model set and the second expanded ROI training model set; uses the test sets in the first sample set, the second sample set and the third sample set to train the original ROI training model set, the first expanded ROI training model set and the second expanded ROI training model set respectively. The expanded ROI training model set was tested, and the model with the best evaluation results was selected as the original ROI prediction model, the first expanded ROI prediction model, and the second expanded ROI prediction model to form a chronic osteomyelitis prediction model group. The original ROI prediction results, the first expanded ROI prediction results, and the second expanded ROI prediction results of the patient to be predicted were obtained through the chronic osteomyelitis prediction model group. The corresponding chronic osteomyelitis debridement auxiliary decision was obtained in combination with the prediction results, providing accurate and effective auxiliary data support for finding a relatively safe balance between "thorough debridement" and "minimizing bone damage", and supporting surgeons to obtain more comprehensive lesion assessment and clearer surgical margin selection.
[0036] Among them, the method provided in the embodiment of the present application 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.
[0037] 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.
[0038] 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.
[0039] 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.
[0040] As attached Figure 1 FIG. 1 shows a schematic diagram of an implementation environment provided by an embodiment of the present application. The implementation environment may include: training equipment and user equipment.
[0041] 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 to this embodiment of the present application.
[0042] 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 the original ROI prediction result, the first expanded ROI prediction result, and the second expanded ROI prediction result. Optionally, the training device has the ability to construct a machine learning algorithm, and it can independently construct a machine learning algorithm according to application requirements, and then train and learn it. For example, in order to obtain the original ROI prediction result, the first expanded ROI prediction result, and the second expanded ROI prediction result based on the original ROI features, the first expanded ROI features, and the second expanded ROI features of the patient to be predicted, the training device independently constructs a machine learning algorithm, and then trains and learns it with samples according to application requirements.
[0043] The using device refers to a computer device that has the need to use a machine learning algorithm. Optionally, the using device obtains the machine learning algorithm from other devices through the network according to application requirements. For example, the using device has the need to obtain the original ROI prediction result, the first expanded ROI prediction result, and the second expanded ROI prediction result. It can obtain the machine learning algorithm that has completed training and learning to obtain the original ROI prediction result, the first expanded ROI prediction result, and the second expanded ROI prediction result from other devices through the network, and use the machine learning algorithm to obtain the original ROI prediction result, the first expanded ROI prediction result, and the second expanded ROI prediction result.
[0044] Based on this, the technical solution of this application will be introduced and explained with reference to several examples below.
[0045] Example 1: As shown in the attached Figure 2 As shown, the embodiment of the present invention discloses a method for training a chronic osteomyelitis prediction model based on radiomics, comprising: Step S110: Obtain a first sample set, a second sample set, and a third sample set, and divide each sample set into a training set and a test set in proportion, wherein the first sample set includes multiple first samples, each first sample includes the original ROI feature and the corresponding chronic osteomyelitis lesion detection identification information, the second sample set includes multiple second samples, each second sample includes the first expanded ROI feature and the corresponding chronic osteomyelitis lesion detection identification information, and the third sample set includes multiple third samples, each third sample includes the second expanded ROI feature and the corresponding chronic osteomyelitis lesion detection identification information.
[0046] For chronic osteomyelitis, in existing surgical debridement, an overly conservative debridement range may leave residual infection and lead to recurrence of the disease, while an overly aggressive debridement may cause unnecessary bone damage. Therefore, in order to build a chronic osteomyelitis prediction model that can balance "thorough debridement" and "minimizing bone damage", step S110 in this embodiment not only uses the original ROI as a sample for subsequent model training, but also expands it twice, thereby obtaining the corresponding first sample set, second sample set, and third sample set for subsequent training.
[0047] It should also be noted that the distance between the two expansions can be determined by analyzing the actual medical records.
[0048] Step S120 , using the training sets in the first sample set, the second sample set, and the third sample set to train multiple machine learning algorithms respectively, to obtain the corresponding original ROI training model set, the first expanded ROI training model set, and the second expanded ROI training model set.
[0049] In the above steps, the training sets in the first sample set, the second sample set, and the third sample set are used to train multiple machine learning algorithms respectively, specifically: Using the training set of the first sample set to train multiple machine learning algorithms respectively, to obtain multiple original ROI training models, forming an original ROI training model set; Using the training set of the second sample set to train multiple machine learning algorithms respectively, to obtain multiple first expanded ROI training model sets, forming a first expanded ROI training model set; The training set of the third sample set is used to train multiple machine learning algorithms respectively to obtain multiple second expanded ROI training models, thereby forming a second expanded ROI training model set.
[0050] It should also be noted that various machine learning algorithms can be selected as needed, including but not limited to KNN algorithm, SVM algorithm, RF algorithm, and Logistic Regression algorithm.
[0051] In step S130, the original ROI training model set, the first expanded ROI training model set, and the second expanded ROI training model set are tested using the test sets in the first sample set, the second sample set, and the third sample set, and the models with the best evaluation results are selected as the original ROI prediction model, the first expanded ROI prediction model, and the second expanded ROI prediction model to form a chronic osteomyelitis prediction model group.
[0052] In the above steps, the process of obtaining the original ROI prediction model includes: using the test set of the first sample set to train all the original ROI training models in the original ROI training model set, so that multiple test results can be obtained, one or more evaluation indicators are introduced to evaluate all the test results, and the model with the best evaluation result is selected as the original ROI prediction model. The evaluation indicators here can include but are not limited to AUC, accuracy, specificity and sensitivity.
[0053] The steps of obtaining the first outward-expanding ROI prediction model and the second outward-expanding ROI prediction model are the same as above and will not be described in detail.
[0054] The present invention discloses a radiomics-based chronic osteomyelitis prediction model training method. On the basis of using the original ROI as a sample for subsequent model training, it is also expanded twice to construct three types of samples. Based on the introduction of multiple machine learning algorithms, the three types of samples are used to train the models respectively. The model with the best evaluation result is selected as the original ROI prediction model, the first expanded ROI prediction model, and the second expanded ROI prediction model to form a chronic osteomyelitis prediction model group, which ensures the accuracy of osteomyelitis prediction and provides accurate and effective auxiliary data support for orthopedic physicians to find a relatively safe balance between "thorough debridement" and "minimizing bone damage."
[0055] 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 the process of obtaining the first sample set, the second sample set, and the third sample set includes: Step S210 : acquiring an original MRI image dataset, wherein the original MRI image dataset includes a plurality of original MRI image data.
[0056] In the above steps, the raw MRI image dataset includes several raw MRI images, each acquired using an MRI scanner. Specifically, patients received conventional T1-weighted and T2-weighted sequences, and STIR sequences were added to the T2 sequence. Scans were performed using coronal, sagittal, and axial planes, depending on the lesion location.
[0057] 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.
[0058] Step S220 : delineating the lesion and surrounding area for each MRI image data, and converting the original MRI image data set into an original ROI data set, a first expanded ROI data set, and a second expanded ROI data set based on the delineation results.
[0059] In the above steps, the process of outlining the lesion and surrounding area for each MRI image data includes: (1) 3D Slicer (version 5.3.0) software was used to accurately delineate the lesion and simultaneously annotate the outer contour of the lesion and its bounding box. The T2-weighted STIR sequence was used as a reference to carefully delineate the ROI on the coronal plane.
[0060] (2) The original ROI, the first expanded ROI, and the second expanded ROI were initially delineated using automated tools, and then manually adjusted to confirm the accuracy of the delineated range to ensure that it did not exceed the bone structure.
[0061] (3) Outline the plane of each lesion one by one, taking care to avoid necrotic and hemorrhagic areas to ensure comprehensive coverage of the lesion tissue.
[0062] (3) For lesions with unclear boundaries, focus on outlining the obvious high-signal areas.
[0063] (4) For multiple lesions, only the largest lesion should be outlined.
[0064] The distances of the above two expansions can be determined by analyzing actual medical records. In this embodiment, based on actual historical data, it is found that the surgical debridement recommendation of "at least 5 mm expansion" has certain effectiveness. However, in complex cases, the boundary of the lesion and the state of the surrounding tissue may vary due to individual differences. Therefore, although the debridement recommendation of "at least 5 mm expansion" clarifies the lower limit of osteotomy, it fails to provide the upper limit of osteotomy. In other words, the expected debridement range recommendation is not achieved, and bone damage cannot be reduced. At the same time, the lesion edge range for complete debridement of the osteomyelitis lesion is not given. Therefore, in this embodiment, the two expansions can be set to 5 mm and 10 mm.
[0065] In step S230 , feature extraction and screening are performed on the original ROI dataset, the first expanded ROI dataset, and the second expanded ROI dataset, respectively, to form a first sample set, a second sample set, and a third sample set, and chronic osteomyelitis lesion detection identification information of all samples in each sample set is determined.
[0066] The above process of extracting and filtering features of the original ROI dataset, the first expanded ROI dataset, and the second expanded ROI dataset includes: Step S231 : Select a filtering method, and extract the radiomics features in each original ROI data in combination with the selected filtering method.
[0067] In this step, a filtering method is selected, and the image data in the original ROI dataset, the first expanded ROI dataset, and the second expanded ROI dataset are filtered using the selected filtering method to facilitate feature extraction.
[0068] Specifically, the extraction of radiomic features in each original ROI data was completed based on the pyradiomics module of 3D Slicer software.
[0069] Specifically, the filtering method may be selected based on the data characteristics of the original ROI data set, the first expanded ROI data set, and the second expanded ROI data set.
[0070] For example, you can choose wavelet filtering and LoG filter. Wavelet filtering iteratively decomposes the original image to obtain image information at different scales. LoG filtering is used as an edge enhancement filter to highlight areas with obvious grayscale changes. The setting of the sigma parameter value in the LoG filter can adjust the prominence of the texture. Smaller sigma values can be used to enhance complex and fine texture details, while larger values emphasize texture features at larger scales.
[0071] Step S232 , using T-test and LASSO regression algorithm to screen the imaging features of each original ROI data.
[0072] In step S233 , the mRMR classifier is used to perform a secondary screening on the primary screening results to obtain multiple features with the highest correlation with the chronic osteomyelitis lesion detection results and the lowest correlation between features.
[0073] In order to avoid feature redundancy and improve model accuracy, the mRMR classifier was introduced to reduce the dimensionality of all features and complete the secondary screening.
[0074] In this embodiment, each original ROI, each expanded 5mm ROI, and each expanded 10mm ROI can ultimately retain 10 features, as follows: The characteristics of the original ROI include originalfirstorderMinimum, originalfirstorderSkewness, originalshapeSurfaceVolumeRatio, originalglDmDependenceVariance, originalshapeMajorAxisLength, wavelet-HLHglDmDepend enceNonUniformity, wavelet-HLLfirstorderMedian, wavelet-LHHglDmLargeDependenceHighGrayLevelEmphasis, log-sigma-4-0-mm-3DfirstorderEnergy and log-sigma-5-0-mm-3Dfirstorder90Percentile; The characteristics of the outer 5mmROI include wavelet-HLHglcmMCC, wavelet-LLHglcmClusterShade, wavelet-HLHfirstorderSkewness, wavelet-HLHglcmCorrelation, originalglcmMCC, wavelet-LLLfirstorder Minimum, originalfirstorderKurtosis, log-sigma-5-0-mm-3DglszmLargeAreaHighGrayLevelEmphasis, wavelet-HLHfirstorderMean and wavelet-HLHglrlmRunLengthNonUniformity; The features of the 10mm ROI include wavelet-LLHglcmCorrelation, wavelet-LLHfirstorderSkewness, wavelet-HHHglDmSmallDependenceHighGrayLevelEmphasis, wavelet-HHHglDmSmallDependenceLowGrayLevelEmphasis, log-si gma-5-0-mm-3DfirstorderSkewness, wavelet-HLHglcmMCC, wavelet-LHLglcmCorrelation, originalglDmSmallDependenceHighGrayLevelEmphasis, originalshapeMajorAxisLength and originalshapeSphericity.
[0075] Example 3: The present invention verifies the effectiveness of the radiomics-based chronic osteomyelitis prediction model training method disclosed in the above embodiment. Figure 4 The model training process of this embodiment is as follows: (1) Obtaining the first, second, and third sample sets (1) As attached Figure 4 In the MRI scan part, obtain the original MRI image dataset as follows: Patients with a high clinical suspicion of chronic osteomyelitis requiring surgical intervention were selected. These patients had persistent bone marrow infection for more than 10 weeks. 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 not in the long bones were excluded.
[0076] A total of 57 patients diagnosed with chronic osteomyelitis and 45 patients without chronic osteomyelitis were selected for data acquisition using a Siemens 1.5T MR scanner to generate a raw MRI image dataset. The main scanning parameters for the patients were T1WITR 600ms TE 9.5ms; T2WI TR 3000ms TE 88ms; FS T2WI TR 3600ms TE 83ms, a FOV of 320mm, and a matrix size of 256×256. All patients underwent conventional T1-weighted and T2-weighted sequences, with STIR sequences added to the T2 sequence. Coronal, sagittal, and axial scans were performed based on the location of the lesion, with a slice thickness of 4mm and a slice spacing of 0.4mm.
[0077] (2) As attached Figure 4 In the lesion segmentation part, the lesion and surrounding area are delineated for each original MRI image data. Based on the delineation results, the original ROI dataset, the expanded 5mm ROI dataset, and the expanded 10mm ROI dataset are obtained, as follows: As attached Figure 5 As shown in the figure, 3D Slicer (version 5.3.0) software was used to accurately delineate the lesion, and the outer contour of the lesion and its bounding box were annotated at the same time. The original region of interest (original ROI), the region of interest expanded 5 mm outward (5-mm expanded ROI), and the region of interest expanded 10 mm outward (10-mm expanded ROI) were preliminarily delineated using an automated tool, and then manual adjustments were made to confirm the accuracy of the delineation range to ensure that it did not exceed the bone structure.
[0078] After delineation, the delineation results of all original MRI image data were classified to obtain the original ROI dataset, the expanded 5mm ROI dataset, and the expanded 10mm ROI dataset.
[0079] (3) As attached Figure 4In the Feature Selection section, feature extraction and screening are performed on the original ROI dataset, the expanded 5mm ROI dataset, and the expanded 10mm ROI dataset to obtain the optimal features of the original ROI, the expanded 5mm ROI, and the expanded 10mm ROI, as follows: Wavelet filtering and LoG filter were applied, and the pyradiomics module of 3D Slicer software was used to extract features from each original ROI, each expanded 5mm ROI, and each expanded 10mm ROI. 1037 features were extracted from each original ROI, each expanded 5mm ROI, and each expanded 10mm ROI, including shape, firstorder, GLCM, GLSZM, GLDM, GLRLM, and Neighboring Gray Level Dependence Matrix (NGLDM).
[0080] Through the T test and LASSO regression algorithm to perform a feature screening, the original ROI retains 82 features, the 5mm expanded ROI retains 70 features, and the 10mm expanded ROI retains 226 features. Figure 6 The application effect of using t-test and LASSO regression algorithm in texture feature selection within the lesion is demonstrated. In the attached figure, A is the texture feature selection of the original ROI; B is the texture feature selection of the expanded 5mm ROI; and C is the texture feature selection of the expanded 10mm ROI.
[0081] To further reduce feature dimensionality and build an efficient model, the mRMR classifier was used for secondary screening. The core principle of this algorithm is to find the feature combinations with the highest correlation with the output results (Max-Relevance) and the lowest correlation between features (Minimum Redundancy) from the original feature set. Ultimately, each original ROI, each 5mm expanded ROI, and each 10mm expanded ROI retained 10 features, as follows: The features of the original ROI include originalfirstorderMinimum, originalfirstorderSkewness, originalshapeSurfaceVolumeRatio, originalglDmDependenceVariance, originalshapeMajorAxisLength, wavelet-HLHglDmDependenceNonUniformity, wavelet-HLLfirstorderMedian, wavelet-LHHglDmLargeDependenceHighGrayLevelEmphasis, log-sigma-4-0-mm-3DfirstorderEnergy, and log-sigma-5-0-mm-3Dfirstorder90Percentile; The features of the ROI expanded by 5 mm include wavelet-HLHglcmMCC, wavelet-LLHglcmClusterShade, wavelet-HLHfirstorderSkewness, wavelet-HLHglcmCorrelation, originalglcmMCC, wavelet-LLLfirstorderMinimum, originalfirstorderKurtosis, log-sigma-5-0-mm-3DglszmLargeAreaHighGrayLevelEmphasis, wavelet-HLHfirstorderMean, and wavelet-HLHglrlmRunLengthNonUniformity; The features of the 10mm ROI include wavelet-LLHglcmCorrelation, wavelet-LLHfirstorderSkewness, wavelet-HHHglDmSmallDependenceHighGrayLevelEmphasis, wavelet-HHHglDmSmallDependenceLowGrayLevelEmphasis, log-si gma-5-0-mm-3DfirstorderSkewness, wavelet-HLHglcmMCC, wavelet-LHLglcmCorrelation, originalglDmSmallDependenceHighGrayLevelEmphasis, originalshapeMajorAxisLength and originalshapeSphericity.
[0082] Attachment Figure 7 The final filtered features and their weights in the original ROI, the expanded 5mm ROI, and the expanded 10mm ROI are shown. A is the feature weight in the original ROI, B is the feature weight in the expanded 5mm ROI, and C is the feature weight in the expanded 10mm ROI. Further, as attached Figure 8 As shown in the figure, a correlation matrix heat map is used to display the correlation between the final selected features in the original ROI, the 5mm expanded ROI, and the 10mm expanded ROI. Among them, the total correlation coefficient of all feature groups is less than 0.7, indicating that there is no collinearity problem between the selected features and confirming the independence of the feature sets used for analysis.
[0083] In the 5mm expanded ROI, wavelet transform-related MCC and correlation features significantly depict the texture characteristics of the suspected osteomyelitis area after ROI expansion. Both MCC and correlation features belong to the gray-level co-occurrence matrix feature. MCC represents the complexity of the texture, while correlation reflects the linear dependence and regularity of the image texture. The 5mm expanded ROI encompasses a wider area than the original ROI, resulting in relatively more complex and irregular texture features. Therefore, the MCC feature exhibits a high positive weight, while the correlation feature exhibits a high negative weight. These features indicate that the texture within the expanded ROI becomes more complex and uneven, reflecting the involvement of a wider range of pathology.
[0084] During a 10mm expansion, the shape-related sphericity feature and the wavelet transform-related HighGrayLevelEmphasis feature became key to describing suspected osteomyelitis areas. Sphericity quantifies the roundness of the lesion relative to a sphere, and its high positive weight may be due to the smoothing and rounding of the ROI edges after excessive expansion. Conversely, HighGrayLevelEmphasis, a measure of small-dependence distribution, indicates high grayscale levels and small grayscale variations. When this feature exhibits a high negative weight, it indicates that the covered area has large grayscale variations, which is consistent with the inclusion of more healthy bone tissue with low signal intensity on MRI images after a 10mm expansion.
[0085] (4) Determine the chronic osteomyelitis lesion detection identification information of each original ROI, each 5mm expanded ROI, and each 10mm expanded ROI to obtain the first sample set, the second sample set, and the third sample set.
[0086] (2) As attached Figure 4 In the feature selection (Model Construction) part, model training is performed as follows: The KNN algorithm, SVM algorithm, RF algorithm and Logistic Regression algorithm were trained using the training sets in the first sample set, the second sample set and the third sample set, respectively, thereby obtaining four original ROI training models, four expanded 5mm ROI training models and four 10mm ROI training models.
[0087] GridSearchCV is used to optimize hyperparameters during training. The hyperparameters selected for the above KNN algorithm, SVM algorithm, RF algorithm, and LogisticRegression algorithm are shown in Table 1: Table 1 Hyperparameter selection table (3) As attached Figure 4 In the Model Performance section, perform model testing as follows: Four original ROI training models, four expanded 5mm ROI training models, and four 10mm ROI training models were tested using the test sets from the first, second, and third sample sets, respectively. CIC curves, ROC curves, and model performance (AUC, accuracy, specificity, and sensitivity) were introduced for evaluation. Specifically: (1) The performance of the model after evaluation is shown in Table 2.
[0088] Table 2 Model performance table It can be seen that the Logistic Regression algorithm performs best among the four machine learning algorithms, with the highest AUC, accuracy, and sensitivity.
[0089] (2) ROC curve (Receiver Operating Characteristic Curve), the receiver operating characteristic curve, is used to evaluate the discrimination ability of the binary classification model, intuitively measure the model performance and select the optimal classification threshold. The ROC curve of the model tested in this embodiment is shown in the attached figure. Figure 9 As shown in the results, when the expansion range was 5 mm, the four models of KNN, RF, SVM and Logistic regression all showed high AUC, accuracy, specificity and sensitivity. However, when the expansion range increased to 10 mm, the accuracy, sensitivity and specificity of all models decreased. The DeLong test showed that there were significant differences in the Logistic regression model among the three expansion ranges (0 mm, 5 mm, 10 mm) (P < 0.05). Figure 9 The results show the key impact of the expansion range on model performance. Specifically, the model trained with a 5mm ROI expansion performed best in terms of AUC value, suggesting that moderate expansion of the osteomyelitis area visible on MRI can effectively improve the diagnostic efficacy in clinical practice, while excessive expansion may lead to a decrease in diagnostic efficacy. It is very important to select the optimal expansion range to maximize the diagnostic accuracy in the diagnosis of osteomyelitis based on radiomics.
[0090] The CIC curve (Clinical Impact Curve) is used to evaluate the actual clinical application value of the prediction model. It helps determine the optimal application range of the model by showing the number of positive cases and the number of correctly classified cases under different classification probability thresholds. The CIC curve of the model obtained by the test in this embodiment is shown in the attached figure. Figure 10 As shown, attached Figure 10 When the high-risk threshold for the 5mm ROI training model exceeded 68%, the predicted number of positive cases was highly consistent with the actual number of positive cases. In contrast, the high-risk threshold corresponding to the optimal clinical benefit of other models required reaching above 80%. These results suggest that in clinical practice, the 5mm ROI training model may provide a better role in evaluating treatment efficacy and may facilitate more accurate treatment decisions.
[0091] Example 4: As shown in the attached Figure 11 As shown, the embodiment of the present invention discloses a method for predicting chronic osteomyelitis based on radiomics, comprising: Step S310, obtaining the original ROI feature, the first expanded ROI feature, and the second expanded ROI feature of the patient to be predicted; Step S320, inputting the original ROI feature, the first expanded ROI feature, and the second expanded ROI feature of the patient to be predicted into the corresponding model in the chronic osteomyelitis prediction model group to obtain the corresponding original ROI prediction result, the first expanded ROI prediction result, and the second expanded ROI prediction result, wherein the chronic osteomyelitis prediction model group is obtained according to the method described in the above embodiment; Step S330: Determine the conditions based on the chronic osteomyelitis prediction result to obtain a corresponding auxiliary decision for chronic osteomyelitis debridement, wherein the chronic osteomyelitis prediction result includes: (a) If the original ROI prediction model predicts a positive result, and both the first and second expanded ROI prediction results are negative, the debridement auxiliary decision for chronic osteomyelitis is to perform debridement with minimal bone damage; (b) If the prediction result of the first expanded ROI is positive, the auxiliary decision for debridement of chronic osteomyelitis is to appropriately expand the debridement based on the original osteomyelitis resection range; (c) If the prediction result of the second expanded ROI is positive, the auxiliary decision for debridement of chronic osteomyelitis is complete debridement; Among them, positive means that there are chronic osteomyelitis lesions in the area, and negative means that there are no chronic osteomyelitis lesions in the area.
[0092] An embodiment of the present invention discloses a radiomics-based chronic osteomyelitis prediction method. The method uses a chronic osteomyelitis prediction model trained with a first sample set, a second sample set, and a third sample set formed by an original ROI dataset, a first expanded ROI dataset, and a second expanded ROI dataset to predict the original ROI features, the first expanded ROI features, and the second expanded ROI features of the predicted patient, respectively, to obtain corresponding auxiliary decision-making for chronic osteomyelitis debridement. This method provides support for surgeons to obtain more comprehensive lesion assessments and clearer surgical margin selection, and provides accurate and effective auxiliary data support for the transition from "one-size-fits-all" to "scientific cutting." It finds a relatively safe balance between "thorough debridement" and "minimizing bone damage," thereby reducing bone damage while reducing the risk of recurrence and promoting the formulation of more comprehensive and accurate surgical plans.
[0093] Example 5: Based on a real case, the effectiveness of Example 4 is verified as follows: As attached Figure 12 As shown, the design process includes (A) MR image of the patient after admission; (B) CT image of the patient after admission; (C)-(D) Based on the lesion delineation for chronic osteomyelitis debridement decision-making assistance given in Example 4, the original ROI was determined to have an outer extension of 7 mm and an outer extension of 14 mm; (E) Mark the area of the infected bone to be resected on the patient's body surface; (F)-(G) External fixation, debridement, segmental bone resection, and antibiotic bone cement filling were performed during the operation; (H)-(I) Before and after bone cement removal and tibial osteotomy before lengthening; (J) Lengthening is complete, the extended bone segment is mineralizing, and there is no recurrence of infection.
[0094] This embodiment applies the method disclosed in Example 4 to provide an auxiliary decision for debridement of chronic osteomyelitis. Taking this auxiliary decision for debridement of chronic osteomyelitis into consideration and performing debridement, it not only helps doctors find a relatively safe balance between "thorough debridement" and "minimizing bone damage" and provides accurate and effective auxiliary data support, but also ensures that there is no infection recurrence during debridement to the greatest extent.
[0095] Example 6: As shown in the attached Figure 13 As shown, an embodiment of the present invention discloses a chronic osteomyelitis prediction model training device based on radiomics, comprising: a sample acquisition unit, acquiring a first sample set, a second sample set, and a third sample set, and dividing each sample set into a training set and a test set in proportion, wherein the first sample set includes a plurality of first samples, each of which includes an original ROI feature and corresponding chronic osteomyelitis lesion detection identification information; the second sample set includes a plurality of second samples, each of which includes a first expanded ROI feature and corresponding chronic osteomyelitis lesion detection identification information; and the third sample set includes a plurality of third samples, each of which includes a second expanded ROI feature and corresponding chronic osteomyelitis lesion detection identification information; A model training unit, which uses the training sets in the first sample set, the second sample set, and the third sample set to train multiple machine learning algorithms respectively to obtain corresponding original ROI training model sets, first expanded ROI training model sets, and second expanded ROI training model sets; The model selection unit uses the test sets in the first sample set, the second sample set and the third sample set to test the original ROI training model set, the first expanded ROI training model set and the second expanded ROI training model set respectively, and selects the model with the best evaluation result as the original ROI prediction model, the first expanded ROI prediction model and the second expanded ROI prediction model respectively to form a chronic osteomyelitis prediction model group.
[0096] The specific execution methods of each module in this embodiment are the same as those in the above embodiments 1 to 3 and will not be repeated here.
[0097] Example 7: As shown in the attached Figure 14 As shown, an embodiment of the present invention discloses a chronic osteomyelitis prediction device based on radiomics, comprising: A prediction data acquisition unit is configured to acquire the original ROI feature, the first expanded ROI feature, and the second expanded ROI feature of the patient to be predicted; A prediction unit, inputting the original ROI feature, the first expanded ROI feature, and the second expanded ROI feature of the patient to be predicted into a corresponding model in a chronic osteomyelitis prediction model group, and obtaining a corresponding original ROI prediction result, the first expanded ROI prediction result, and the second expanded ROI prediction result, wherein the chronic osteomyelitis prediction model group is obtained by the method according to any one of claims 1 to 5; The analysis unit determines the conditions based on the chronic osteomyelitis prediction results to obtain the corresponding chronic osteomyelitis debridement auxiliary decision, wherein the chronic osteomyelitis prediction results include: (a) If the original ROI prediction model predicts a positive result, and both the first and second expanded ROI prediction results are negative, the debridement auxiliary decision for chronic osteomyelitis is to perform debridement with minimal bone damage; (b) If the prediction result of the first expanded ROI is positive, the auxiliary decision for debridement of chronic osteomyelitis is to appropriately expand the debridement based on the original osteomyelitis resection range; (c) If the prediction result of the second expanded ROI is positive, the auxiliary decision for debridement of chronic osteomyelitis is complete debridement; Among them, positive means that there are chronic osteomyelitis lesions in the area, and negative means that there are no chronic osteomyelitis lesions in the area.
[0098] The specific execution methods of each module in this embodiment are the same as those in the above embodiments 4 to 5 and will not be repeated here.
[0099] Example 8: The embodiment of the present invention discloses a storage medium, on which a computer program that can be read by a computer is stored. The computer program is configured to execute a radiomics-based chronic osteomyelitis prediction method when running.
[0100] 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.
[0101] Example 9: 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 radiomics.
[0102] 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 disclosure of this application. 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.
[0103] Those skilled in the art will appreciate that the embodiments of the present application may be provided as methods, systems, or computer program products. Therefore, the present application may take the form of a fully hardware embodiment, a fully software embodiment, or an embodiment combining software and hardware. Furthermore, the present application may take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code. The solutions in the embodiments of the present application may be implemented in various computer languages, such as the object-oriented programming language Java and the interpreted scripting language JavaScript.
[0104] The present application is described with reference to the flowcharts and / or block diagrams of the methods, devices (systems), and computer program products according to the embodiments of the present application. It should be understood that each process and / or block in the flowchart and / or block diagram, as well as the combination of processes and / or blocks in the flowchart and / or block diagram, 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 flowchart and / or block diagram. 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.
[0105] 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.
[0106] The above contents are merely specific implementation methods of the present application, each of which has strong adaptability and implementation effect, but the scope of protection of the present application is not limited thereto. Any technician familiar with the technical field can easily think of changes or replacements within the technical scope disclosed in the present application, which should be included in the scope of protection of the present application. Therefore, equivalent changes made in accordance with the claims of the present application are still within the scope covered by the present application.
Claims
1. A method for training a chronic osteomyelitis prediction model based on radiomics, characterized in that: include: Obtain a first sample set, a second sample set, and a third sample set, and divide each sample set into a training set and a test set in proportion, wherein the first sample set includes a plurality of first samples, each first sample includes an original ROI feature and corresponding chronic osteomyelitis lesion detection identification information, the second sample set includes a plurality of second samples, each second sample includes a first expanded ROI feature and corresponding chronic osteomyelitis lesion detection identification information, and the third sample set includes a plurality of third samples, each third sample includes a second expanded ROI feature and corresponding chronic osteomyelitis lesion detection identification information; Using the training sets in the first sample set, the second sample set, and the third sample set to train multiple machine learning algorithms respectively, to obtain corresponding original ROI training model sets, first expanded ROI training model sets, and second expanded ROI training model sets; The original ROI training model set, the first expanded ROI training model set, and the second expanded ROI training model set were tested using the test sets in the first sample set, the second sample set, and the third sample set, respectively. The models with the best evaluation results were selected as the original ROI prediction model, the first expanded ROI prediction model, and the second expanded ROI prediction model to form a chronic osteomyelitis prediction model group.
2. The radiomics-based chronic osteomyelitis prediction model training method according to claim 1, characterized in that: The process of obtaining the first sample set, the second sample set, and the third sample set includes: Acquire an original MRI image dataset, wherein the original MRI image dataset includes a plurality of original MRI image data; Each MRI image data is delineated for the lesion and surrounding area, and the original MRI image data set is converted into the original ROI data set, the first expanded ROI data set, and the second expanded ROI data set based on the delineation results; Feature extraction and screening are performed on the original ROI dataset, the first expanded ROI dataset, and the second expanded ROI dataset respectively to form the first sample set, the second sample set, and the third sample set, and the chronic osteomyelitis lesion detection identification information of all samples in each sample set is determined.
3. The radiomics-based chronic osteomyelitis prediction model training method according to claim 2, characterized in that: The feature extraction and screening processes for the original ROI dataset, the first expanded ROI dataset, and the second expanded ROI dataset are the same, wherein the feature extraction and screening of the original ROI dataset include: Select a filtering method and extract the radiomics features in each raw ROI data in combination with the selected filtering method; The imaging features of each original ROI data were screened once using T test and LASSO regression algorithm; The mRMR classifier was used to perform a secondary screening of the primary screening results, and multiple features with the highest correlation with the chronic osteomyelitis lesion detection results and the lowest correlation between features were obtained.
4. The radiomics-based chronic osteomyelitis prediction model training method according to claim 2 or 3, characterized in that: Each first expanded ROI data in the first expanded ROI data set is ROI data expanded outward by 5 mm from the corresponding original ROI data, and each second expanded ROI data in the second expanded ROI data set is ROI data expanded outward by 10 mm from the corresponding original ROI data.
5. The method for training a chronic osteomyelitis prediction model based on radiomics according to any one of claims 1 to 4, characterized in that: The multiple machine learning algorithms include KNN algorithm, SVM algorithm, RF algorithm, and LogisticRegression algorithm.
6. A method for predicting chronic osteomyelitis based on radiomics, characterized in that: include: Obtaining the original ROI feature, the first expanded ROI feature, and the second expanded ROI feature of the patient to be predicted; Inputting the original ROI feature, the first expanded ROI feature, and the second expanded ROI feature of the patient to be predicted into the corresponding model in the chronic osteomyelitis prediction model group to obtain the corresponding original ROI prediction result, the first expanded ROI prediction result, and the second expanded ROI prediction result, wherein the chronic osteomyelitis prediction model group is obtained according to the method according to any one of claims 1 to 5; Combined with the chronic osteomyelitis prediction results, the conditions are determined to obtain the corresponding chronic osteomyelitis debridement auxiliary decision-making, where the chronic osteomyelitis prediction results include: (a) If the original ROI prediction model predicts a positive result, and both the first and second expanded ROI prediction results are negative, the debridement auxiliary decision for chronic osteomyelitis is to perform debridement with minimal bone damage; (b) If the prediction result of the first expanded ROI is positive, the auxiliary decision for debridement of chronic osteomyelitis is to appropriately expand the debridement based on the original osteomyelitis resection range; (c) If the prediction result of the second expanded ROI is positive, the auxiliary decision for debridement of chronic osteomyelitis is complete debridement; Among them, positive means that there are chronic osteomyelitis lesions in the area, and negative means that there are no chronic osteomyelitis lesions in the area.
7. A radiomics-based chronic osteomyelitis prediction model training device 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, and a third sample set, and dividing each sample set into a training set and a test set in proportion, wherein the first sample set includes a plurality of first samples, each of which includes an original ROI feature and corresponding chronic osteomyelitis lesion detection identification information; the second sample set includes a plurality of second samples, each of which includes a first expanded ROI feature and corresponding chronic osteomyelitis lesion detection identification information; and the third sample set includes a plurality of third samples, each of which includes a second expanded ROI feature and corresponding chronic osteomyelitis lesion detection identification information; A model training unit, which uses the training sets in the first sample set, the second sample set, and the third sample set to train multiple machine learning algorithms respectively to obtain corresponding original ROI training model sets, first expanded ROI training model sets, and second expanded ROI training model sets; The model selection unit uses the test sets in the first sample set, the second sample set and the third sample set to test the original ROI training model set, the first expanded ROI training model set and the second expanded ROI training model set respectively, and selects the model with the best evaluation result as the original ROI prediction model, the first expanded ROI prediction model and the second expanded ROI prediction model respectively to form a chronic osteomyelitis prediction model group.
8. A radiomics-based chronic osteomyelitis prediction device using the method according to claim 6, characterized in that: include: A prediction data acquisition unit is configured to acquire the original ROI feature, the first expanded ROI feature, and the second expanded ROI feature of the patient to be predicted; A prediction unit, inputting the original ROI feature, the first expanded ROI feature, and the second expanded ROI feature of the patient to be predicted into a corresponding model in a chronic osteomyelitis prediction model group, and obtaining a corresponding original ROI prediction result, the first expanded ROI prediction result, and the second expanded ROI prediction result, wherein the chronic osteomyelitis prediction model group is obtained by the method according to any one of claims 1 to 5; The analysis unit determines the conditions based on the chronic osteomyelitis prediction results to obtain the corresponding chronic osteomyelitis debridement auxiliary decision, wherein the chronic osteomyelitis prediction results include: (a) If the original ROI prediction model predicts a positive result, and both the first and second expanded ROI prediction results are negative, the debridement auxiliary decision for chronic osteomyelitis is to perform debridement with minimal bone damage; (b) If the prediction result of the first expanded ROI is positive, the auxiliary decision for debridement of chronic osteomyelitis is to appropriately expand the debridement based on the original osteomyelitis resection range; (c) If the prediction result of the second expanded ROI is positive, the auxiliary decision for debridement of chronic osteomyelitis is complete debridement; Among them, positive means that there are chronic osteomyelitis lesions in the area, and negative means that there are no chronic osteomyelitis lesions in the area.
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.