Intelligent bone injury evaluation method and system based on artificial intelligence and multi-period medical images

Abstract

The present application belongs to the technical field of artificial intelligence and medical image processing, and aims to solve the problem of low efficiency and accuracy and lack of objective quantitative indicators in the existing bone injury dynamic evaluation of multi-phase medical images relying on artificial methods. The present application provides a bone injury intelligent evaluation method and system based on artificial intelligence and multi-phase medical images. The method comprises: obtaining first image data at the early stage after injury and second image data at the healing stage; registering the second image data to the spatial coordinate system of the first image data; segmenting the bone and identifying the fracture candidate region using a deep learning segmentation network; calculating the callus quantitative features; extracting the texture features and inputting them into a deep learning-based time series discrimination model to output the fresh fracture probability and the post-injury time interval; and combining the body surface injury measurement results to generate evaluation suggestions. The present application realizes objective and automated evaluation of the healing stage of bone injury through image matching, quantitative feature extraction and deep learning time series analysis.

Description

Technical Field

[0001] This invention relates to the fields of artificial intelligence and medical image processing technology, specifically to an intelligent assessment method and system for bone injuries based on artificial intelligence and multi-phase medical imaging. Background Technology

[0002] In clinical orthopedic, trauma surgery, and forensic assessments, dynamic evaluation of the fracture healing process is crucial. Currently, both clinicians and forensic pathologists primarily rely on visually comparing CT / X-ray images taken at different time points, layer by layer, to observe and compare changes in the fracture ends and surrounding bone tissue, searching for dynamic healing signs such as "periosteal reaction," "callus formation," and "blurred or sclerotic fracture lines." This method heavily depends on personal experience and lacks objective, quantifiable assessment indicators such as callus volume, dynamic CT value changes, and the probability of a fresh fracture. Furthermore, manually reviewing and comparing dozens to hundreds of CT images, repeatedly switching between two or more phases to locate the same anatomical position, is extremely time-consuming and laborious. A complex fracture assessment involving multiple phases often takes weeks or even months, severely limiting assessment efficiency. Simultaneously, the human eye is not sensitive enough to minute changes in CT values ​​(such as early fibrous callus causing only a few tens of Henle units of density increase), easily missing minute callus or microfractures, which are crucial for determining injury time and whether a fracture is fresh or old. Traditional assessment reports typically only describe qualitative and vague statements such as "no clear callus formation was observed" or "visible callus growth" without quantifiable objective indicators, making it difficult to verify and challenge the assessment conclusions.

[0003] With the development of technologies such as artificial intelligence, deep learning, and computer vision, how to use multi-phase medical image registration, bone segmentation, texture feature extraction, and time-series discrimination models to achieve automated and quantitative assessment of the bone injury healing stage has become a technical problem that urgently needs to be solved in this field.

[0004] Therefore, there is a need in this field for an intelligent assessment method and system for bone injuries based on artificial intelligence and multi-phase medical imaging to solve the above problems. Summary of the Invention

[0005] To address the aforementioned technical issues, namely the low efficiency and accuracy, and the lack of objective quantitative indicators, in the current multi-phase medical imaging dynamic assessment of bone injury, which relies on manual methods.

[0006] In a first aspect, the present invention provides an intelligent assessment method for bone injury based on artificial intelligence and multi-phase medical imaging, the method comprising the following steps: S1: Obtain the first imaging data of the person being assessed in the early post-injury period and the second imaging data during the post-injury healing period; S2: Preprocess the first image data and the second image data, and register the second image data to the spatial coordinate system of the first image data to obtain the registered image; S3: Segment the skeletal structure in the registered image and the first image data, and identify candidate fracture regions; S4: Within the fracture candidate region, calculate the callus quantification features of the second image data relative to the first image data; S5: Extract the texture features of the fracture candidate region, and input the callus quantification features and texture features into a deep learning-based temporal discrimination model to output the probability that the fracture is a fresh fracture and the inferred post-injury time interval; S6: Based on the probability of a fresh fracture, the inferred post-injury time interval, and the measurement results of surface damage, match the preset damage degree judgment rules to generate assessment suggestions.

[0007] In some preferred embodiments, registering the second image data to the spatial coordinate system of the first image data specifically includes: A non-rigid registration algorithm is used to perform spatial transformation on the second image data.

[0008] In some preferred embodiments, calculating the callus quantification features of the second image data relative to the first image data specifically includes: Within a preset range around the candidate fracture area, voxels whose CT values ​​meet preset conditions are extracted, and the total volume of the voxels is counted as the callus volume. The preset range is 3-10mm around the fracture line, and the preset conditions include a CT value higher than a preset threshold and an increase in value greater than a preset change compared to the first image data.

[0009] In some preferred embodiments, the texture features include gray-level co-occurrence matrix features; the deep learning-based temporal discrimination model employs a Transformer or a Temporal Convolutional Network (TCN).

[0010] In some preferred embodiments, the method further includes: Data on skin damage is collected, and the skin damage area is automatically segmented based on a deep learning segmentation network. The true surface area of ​​the skin damage area is calculated on the three-dimensional reconstructed surface. The true surface area of ​​the skin damage area is compared with the threshold in the damage degree determination rule to obtain the measurement result of the skin damage.

[0011] In some preferred embodiments, the surface injury data includes two-dimensional photographs or three-dimensional point cloud data containing a scale.

[0012] In some preferred embodiments, the method further includes: Structured fields are extracted from the electronic medical record text using natural language processing technology, and these structured fields are then associated and mapped with the first image data and the second image data.

[0013] In some preferred embodiments, generating evaluation recommendations specifically includes: A structured damage standard knowledge base is constructed, and the extracted quantitative features are matched with the damage standard knowledge base to output damage level suggestions and confidence levels, and to suggest supplementary examination items.

[0014] In some preferred embodiments, the method further includes: The first image data, the second image data, the deformation field data generated during the registration process, the callus quantification features, and the evaluation suggestions are hashed and stored in a blockchain network.

[0015] In a second aspect, the present invention also provides an intelligent assessment system for bone injuries based on artificial intelligence and multi-phase medical imaging, the system comprising: The data acquisition module is used to acquire the first image data of the person being assessed in the early post-injury period and the second image data during the post-injury healing period. The preprocessing and registration module is used to preprocess the first image data and the second image data, and register the second image data to the spatial coordinate system of the first image data to obtain the registered image; The identification module is used to segment the skeletal structure in the registered image and the first image data, and to identify candidate fracture regions; The calculation module is used to calculate the callus quantification features of the second image data relative to the first image data within the fracture candidate region. The temporal discrimination module is used to extract the texture features of the fracture candidate region, and input the callus quantification features and the texture features into the deep learning-based temporal discrimination model to output the probability that the fracture is a fresh fracture and the inferred post-injury time interval. The assessment suggestion generation module is used to generate assessment suggestions based on the probability that the fracture is a fresh fracture, the inferred post-injury time interval, and the measurement results of surface injury, and by matching the preset injury degree judgment rules.

[0016] This invention transforms the criteria for judging the bone injury healing process from subjective qualitative descriptions to objective quantitative indicators through image matching, quantitative feature extraction, and deep learning temporal analysis. It utilizes automated multi-phase image registration, bone segmentation, fracture detection, callus quantification, and texture analysis, combined with a temporal discriminant model to output the probability of a fresh fracture and the inferred post-injury time interval. Finally, it matches the injury severity determination rules to generate assessment suggestions. The entire process eliminates the need for manual, layer-by-layer image comparison, thus reducing inconsistencies in judgments caused by differences in experience among assessors, significantly shortening the assessment cycle, and improving the detection sensitivity for early callus and microfractures. Furthermore, the probability of a fresh fracture output by this invention is a continuous value, providing a clear and verifiable assessment conclusion, replacing the conditional and vague descriptions in traditional reports, making the assessment process more transparent and evidence-based. Attached Figure Description

[0017] The features and advantages of the invention will be more clearly understood by referring to the accompanying drawings, which are schematic and should not be construed as limiting the invention in any way. In the drawings: Figure 1 This is a flowchart of an intelligent bone injury assessment method based on artificial intelligence and multi-phase medical imaging, according to an embodiment of the present invention. Figure 2 This is a schematic diagram of the structure of the intelligent bone injury assessment system based on artificial intelligence and multi-phase medical imaging, according to an embodiment of the present invention. Detailed Implementation

[0018] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0019] Based on the background art, existing forensic injury assessments rely on manual methods, resulting in low efficiency and accuracy, leading to unobjective assessment results that are difficult to verify and easily questioned. This invention provides an intelligent bone injury assessment method and system based on artificial intelligence and multi-phase medical imaging, aiming to improve the efficiency and accuracy of forensic assessments, ensure the objectivity of assessment results, and provide verifiable and traceable assessment reports.

[0020] like Figure 1 As shown, the intelligent assessment method for bone injury based on artificial intelligence and multi-phase medical imaging provided by this invention includes the following steps: S1: Obtain the first imaging data of the person being assessed in the early post-injury period and the second imaging data during the post-injury healing period.

[0021] In the above, the first and second image data can be medical image data such as CT, MRI, or X-ray, specifically in DICOM format. The initial first image data can be image data within 12 hours, 24 hours, or 48 hours after the injury, and the second image data during the healing period can be image data 1-3 weeks or 2-6 weeks after the injury. Those skilled in the art can flexibly set the specific time of the initial post-injury period and the healing period in practical applications, or configure different specific time of the initial post-injury period and the healing period according to different injury conditions. For example, the first image data is the baseline image collected within 24 hours after the injury, and the second image data is the follow-up image collected 2-6 weeks after the injury.

[0022] S2: Preprocess the first image data and the second image data, and register the second image data to the spatial coordinate system of the first image data to obtain the registered image.

[0023] In the above, the preprocessing operation includes resampling DICOM files from different devices to a unified voxel space, such as 1mm×1mm×1mm, and normalizing CT values ​​to a standard bone window range, such as a window width of 2000HU and a window level of 300HU. The registration operation uses a non-rigid registration algorithm, such as B-spline registration based on mutual information or the deep learning registration network VoxelMorph, to transform the second image data space to the coordinate system of the first image data and generate the registered image.

[0024] For example, the system calls a pre-trained VoxelMorph network, taking the first and second image data as inputs. The network directly outputs a dense deformation field from the second image data to the first image data, where each voxel contains a three-dimensional displacement vector (x, y, z). Subsequently, this deformation field is used to spatially resample the second image data, generating a registered image spatially aligned to the first image data voxel by voxel. Another example is the system calling a pre-trained B-spline registration algorithm, which maximizes the mutual information between the two images by optimizing the positions of control points, thereby obtaining a smooth deformation field. In practical applications, other types of non-rigid registration methods can also be selected for the registration algorithm. This embodiment of the invention does not limit this. Non-rigid registration algorithms simulate the elastic deformation of soft tissue, calculating a personalized displacement for each voxel in the second image data, thus enabling it to accurately align with the corresponding anatomical structure in the first image data, compensating for the positional differences caused by non-rigid motion between the two scans. This method ensures precise pixel-level correspondence between multiple images, laying a solid foundation for subsequent comparison of CT value changes and accurate calculation of callus volume at the same spatial location, and avoiding quantification errors caused by inaccurate registration.

[0025] S3: Segment the skeletal structure in the registered image and the first image data, and identify candidate fracture regions.

[0026] Skeletal segmentation can be automatically completed using deep learning segmentation networks such as 3D U-Net or nnUNet. Fracture candidate regions can be identified using 3D target detection networks, such as 3D RetinaNet or Faster R-CNN 3D, marking suspicious regions on the registered images and the first image data.

[0027] S4: Within the fracture candidate region, calculate the callus quantification features of the second image data relative to the first image data.

[0028] Preferably, calculating the callus quantification features of the second image data relative to the first image data specifically includes: extracting voxels whose CT values ​​meet preset conditions within a preset range around the candidate fracture area, and calculating the total volume of the voxels as the callus volume; wherein, the preset range is 3-10mm around the fracture line, and the preset conditions include that the CT value is higher than a preset threshold and the increase compared to the first image data exceeds a preset change amount.

[0029] For example, the callus quantification feature involves extracting voxels with a CT value higher than 300 HU and an increase of more than 100 HU compared to the corresponding area in the first image data within a 5mm radius around the fracture candidate region, and calculating the total volume of these voxels to obtain the callus volume. Specifically, based on the fracture candidate region identified in step S3, the center point or centerline of the fracture line is automatically located. Then, a ring-shaped region is expanded outward in three-dimensional space with the fracture line as the center. The radius of this region is set to 5mm, meaning all voxels within a 5mm radius around the fracture line are extracted. Within this region, the system compares the second image data and the first image data voxel by voxel. The system counts all voxels that simultaneously meet the above two conditions and calculates the total volume of these voxels in cubic centimeters (cm). 3 This volume value is the callus volume. This quantitative and repeatable method of callus measurement can capture the formation of tiny callus that is difficult to detect with the naked eye, and describe the healing process using the objective indicator of volume. Compared with the existing subjective human judgment of presence or absence, this greatly improves the sensitivity and accuracy of early callus detection.

[0030] S5: Extract texture features of the candidate fracture region, and input the callus quantification features and texture features into a deep learning-based temporal discrimination model to output the probability that the fracture is a fresh fracture and the inferred post-injury time interval.

[0031] The aforementioned texture features include the contrast, correlation, energy, and homogeneity of the gray-level co-occurrence matrix. The temporal discrimination model can use Transformer or Temporal Convolutional Network (TCN). Its input is a sequence of changes in callus volume and CT value at multiple time points, and its output is the probability of fresh fracture and the time interval after injury. The probability of fresh fracture is a continuous value from 0 to 100%, and the time interval after injury is less than 1 week, 1 to 2 weeks, 2 to 4 weeks, more than 1 month, and old fracture.

[0032] Specifically, on the CT images of the fracture candidate region, the system first extracts gray-level co-occurrence matrix (GLCM) features. A 5×5 sliding window is selected, and all voxels within the region of interest are traversed. The GLCM is calculated at 0°, 45°, 90°, and 135°, and four key features are derived from these: contrast reflects image sharpness and texture groove depth; correlation reflects linear dependence of gray levels; energy reflects texture uniformity and the uniformity of gray-level distribution; and homogeneity reflects the magnitude of local texture variations. These features together constitute a texture feature vector. Simultaneously, the system acquires callus volume sequences at multiple time points for the same fracture point. These callus volume, CT value change temporal features, and texture feature vector spatial features are input into a deep learning-based temporal discrimination model. This model employs a Transformer encoder structure, utilizing its self-attention mechanism to capture long-distance dependencies between features at different time points; or it uses a temporal convolutional network, extracting multi-scale temporal features through dilated convolutions. The model's output layer is a Softmax classifier, outputting the probability distribution of the fracture belonging to different time categories. By fusing texture features and temporal features, the model can more comprehensively and precisely describe the dynamic process of fracture healing, significantly improving the accuracy of distinguishing between fresh and old fractures. Especially when the change in callus volume is not significant, texture features provide key discriminative information.

[0033] S6: Based on the probability of a fresh fracture, the inferred post-injury time interval, and the measurement results of surface damage, match the preset damage degree judgment rules to generate assessment suggestions.

[0034] Optionally, the injury severity determination rule is a structured knowledge base of the "Standards for the Identification of Human Injury Severity". The matching engine compares quantitative features such as the probability of freshness of fracture, number of fractures, and abrasion area with the knowledge base clauses to generate possible injury levels and confidence levels.

[0035] This invention automatically acquires multiple images, uses registration technology to achieve spatial alignment, and then uses a deep learning model to accurately segment bones, identify fractures, and quantify the volume and texture changes of callus formation. Finally, these objective quantitative indicators are input into a time-series model for inference to obtain a probabilistic judgment of the age of the fracture. Combined with the accurate measurement results of surface injuries, it automatically matches legal standards to form auxiliary assessment suggestions, reducing manual operation steps, improving the efficiency of bone injury assessment, and ensuring the accuracy and objectivity of the results.

[0036] Preferably, the evaluation recommendations are generated, specifically including: A structured injury standard knowledge base is constructed. Extracted quantitative features are matched with this knowledge base to output injury level suggestions and confidence levels, and to suggest necessary supplementary examinations. In other words, legal text rules are transformed into machine-executable logical judgment rules. An automated rule engine precisely compares the system's quantitative analysis results with standard clauses to derive assessment suggestions and intelligently suggests supplementary directions when information is insufficient. For example, all clauses in the "Standards for the Identification of the Degree of Human Injury" are structured and encoded to construct a knowledge base. Specifically, the clause "Two or more rib fractures constitute minor injury level two" is encoded as: Injury type = "fracture", anatomical location = "rib", counting condition = "≥2", injury level = "minor injury level two". The matching engine in the assessment suggestion generation module receives quantitative features from other modules: number of fractures = 3, probability of fresh fracture = 90%, abrasion area = 1.21cm². 2 The matching engine uses these features as query conditions to search and compare them in the knowledge base. For example, if the rule "more than two rib fractures" is matched, the output injury level suggestion is "minor injury level 2" with a confidence level of 95%. At the same time, if the injured person has both fractures and internal organ damage, but the system lacks data on internal organ damage, the matching engine will prompt: "Currently, abdominal CT images are missing. It is recommended to supplement the examination to assess whether there is internal organ damage, thereby completing the full match of the serious injury clause."

[0037] The intelligent bone injury assessment method based on artificial intelligence and multi-phase medical imaging of the present invention also includes: Data on skin damage is collected, and the skin damage area is automatically segmented based on a deep learning segmentation network. The actual surface area of ​​the skin damage area is calculated on the 3D reconstructed surface. The actual surface area of ​​the skin damage area is compared with the threshold in the damage degree judgment rule to obtain the measurement result of skin damage.

[0038] In one specific implementation, a handheld 3D structured light scanner is first used to scan the damaged area from at least three different angles to acquire 3D point cloud data of the damaged area containing color information. The system then meshes the point cloud data and uses a Poisson surface reconstruction algorithm to generate a high-precision 3D surface model. Next, a pre-trained deep learning segmentation network, such as 2D U-Net, is invoked to map the colored point cloud onto a 2D plane for initial segmentation, or a network such as PointNet++ is used directly on the 3D mesh for segmentation. The boundaries of abrasion and contusion areas are automatically identified. Forensic experts can also manually correct the segmentation results through an interactive interface. Finally, the system integrates and accumulates the areas of all triangular facets within the segmented damaged area on the reconstructed 3D surface to calculate the true surface area, with the output result accurate to square millimeters (mm). 2 The system automatically compares the area value with the area threshold in the structured knowledge base of the "Standards for the Assessment of the Degree of Human Injury" built into the system, and outputs a conclusion on whether a certain level of injury has been reached, such as abrasion area ≥ 10.0 cm². 2 It was a minor injury.

[0039] Alternatively, a regular digital camera or smartphone can be used. A calibration plate or scale with known geometric dimensions can be placed next to the damaged area, and one or more two-dimensional photographs containing the scale can be taken. The system utilizes photogrammetry to determine the proportional relationship between the pixels of the photograph and the actual size by identifying the size of the scale. Then, combined with the triangulation principle of multi-view photographs, a three-dimensional model of the damaged area is reconstructed. In practical applications, other methods can also be used to collect surface injury data, such as laser scanning, which is not limited in this embodiment of the invention.

[0040] Preferably, the intelligent assessment method for bone injury based on artificial intelligence and multi-phase medical imaging of the present invention further includes: The system extracts structured fields from electronic medical record text using natural language processing (NLP) technology and maps these structured fields to first and second image data. First, the system accesses the hospital information system or receives uploaded electronic medical record documents. Then, it calls a deep learning-based Chinese NLP model, such as the BERT pre-trained model, to perform named entity recognition and relation extraction on the medical record text. This model can identify predefined entities from unstructured text, such as "injury location" (e.g., "right 6th rib"), "injury type" (e.g., "fracture"), "occurrence time" (e.g., "March 24, 2026"), and "diagnosis conclusion" (e.g., "right scapular fracture"). The extracted entities and their relationships are filled into a predefined structured table, forming structured fields. Subsequently, the system logically associates and maps key fields such as "occurrence time" and "injury location" in the medical record with the first and second image data obtained in step S1 (the DICOM header file contains examination time and examination location), establishing a data link between the medical record text and the image data. This approach enables the automatic conversion from text medical records to structured information, breaking down information silos between text and image data. It provides richer and more comprehensive data support for multi-dimensional automated assisted assessment, reducing the workload and error rate of manual data entry and verification.

[0041] Preferably, the intelligent assessment method for bone injury based on artificial intelligence and multi-phase medical imaging of the present invention further includes: The first image data, the second image data, the deformation field data generated during the registration process, the callus quantification characteristics and evaluation suggestions are hashed and stored in the blockchain network. By utilizing the distributed ledger, encrypted hash chain and timestamp technology of the blockchain, once the data fingerprint is recorded, no one can unilaterally tamper with the original data corresponding to the stored hash value. By comparing the real-time hash value of the original data with the hash value stored on the chain, it can be verified whether the data has been tampered with. For example, after the draft assessment report is generated, the system automatically collects key data objects generated throughout the assessment process, including: the original first and second image data files, the deformation field matrix or file generated in the registration step, the callus volume and other callus quantitative feature values ​​calculated in step S4, the fresh fracture probability value output in step S5, and the generated structured assessment suggestions. The system serializes these data objects and then uses the SHA-256 secure hash algorithm to calculate a unique hash value. This hash value is like a digital fingerprint of the data; any slight modification to the original data will result in a completely different hash value. Subsequently, the system connects to a judicial blockchain network, such as a regional judicial assessment consortium chain, through an application programming interface and packages the hash value and corresponding timestamps, case identifiers, and other metadata into a transaction, which is then submitted to nodes on the blockchain network for consensus and on-chain storage.

[0042] like Figure 2 As shown, the intelligent bone injury assessment system based on artificial intelligence and multi-phase medical imaging of the present invention includes: The data acquisition module is used to acquire the first image data of the person being assessed in the early post-injury period and the second image data during the post-injury healing period. The preprocessing and registration module is used to preprocess the first image data and the second image data, and register the second image data to the spatial coordinate system of the first image data to obtain the registered image. The recognition module is used to segment the skeletal structure in the registered image and the first image data, and to identify candidate fracture regions. The calculation module is used to calculate the callus quantification features of the second image data relative to the first image data within the fracture candidate region. The temporal discrimination module is used to extract the texture features of the fracture candidate region and input the callus quantification features and texture features into the deep learning-based temporal discrimination model, and output the probability that the fracture is a fresh fracture and the inferred post-injury time interval. The assessment suggestion generation module is used to generate assessment suggestions based on the probability that the fracture is a fresh fracture, the inferred post-injury time interval, and the measurement results of surface injury, and by matching the preset injury degree judgment rules.

[0043] The data acquisition module can be a unified interface that supports importing image data such as CT and MRI from DICOM files, local storage, or hospital PACS systems. It also supports real-time acquisition of body surface data from devices such as USB cameras and structured light scanners. The preprocessing and registration module integrates image resampling, normalization, and deep learning-based VoxelMorph registration algorithm libraries. The recognition module loads a pre-trained nnUNet bone segmentation model and a 3D RetinaNet fracture detection model. The calculation module implements a callus volume statistics algorithm. The temporal discrimination module deploys a trained Transformer model to output the probability of fresh fractures. The assessment suggestion generation module includes a structured injury standard knowledge base and a rule matching engine. The modules communicate with each other through a predefined application programming interface to collaboratively complete the automated processing flow from raw data to assessment suggestions.

[0044] Specifically, during operation, the system employs a data acquisition module to acquire multi-source input, a preprocessing and registration module to standardize data and perform spatial alignment, an identification and calculation module to extract lesions and quantify their physiological changes, a time-series discrimination module to intelligently analyze trends and provide probabilistic conclusions, and a final assessment suggestion generation module to match the quantitative results with legal standards and output auxiliary decision-making information. By integrating the core steps of bone injury assessment into an automated and intelligent software system, a complete solution from data entry to report generation is achieved. This invention's system covers multiple dimensions, including surface and internal injuries, morphological and physiological changes, significantly improving the overall objectivity, efficiency, and intelligence of bone injury assessment.

[0045] It should be noted that the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitation, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes said element.

[0046] The various embodiments in this invention are described in a progressive manner. Similar or identical parts between embodiments can be referred to mutually. Each embodiment focuses on describing the differences from other embodiments. In particular, the system embodiments are basically similar to the method embodiments, so the description is relatively simple; relevant parts can be referred to the descriptions in the method embodiments.

[0047] Those skilled in the art should understand that the discussion of any of the above embodiments is merely exemplary and is not intended to imply that the scope of this disclosure is limited to these examples; within the framework of this disclosure, the technical features of the above embodiments or different embodiments can also be combined, the steps can be implemented in any order, and there are many other variations of different aspects of one or more embodiments of the present invention as described above, which are not provided in detail for the sake of brevity.

[0048] Although this disclosure has been described in conjunction with specific embodiments thereof, many substitutions, modifications and variations of these embodiments will be apparent to those skilled in the art from the foregoing description.

[0049] One or more embodiments of the present invention are intended to cover all such substitutions, modifications, and variations falling within the broad scope of the appended claims. Therefore, any omissions, modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of one or more embodiments of the present invention should be included within the scope of this disclosure.

Claims

1. A method for intelligent assessment of bone injury based on artificial intelligence and multi-phase medical imaging, characterized in that, The method includes the following steps: S1: Obtain the first imaging data of the person being assessed in the early post-injury period and the second imaging data during the post-injury healing period; S2: Preprocess the first image data and the second image data, and register the second image data to the spatial coordinate system of the first image data to obtain the registered image; S3: Segment the skeletal structure in the registered image and the first image data, and identify candidate fracture regions; S4: Within the fracture candidate region, calculate the callus quantification features of the second image data relative to the first image data; S5: Extract the texture features of the fracture candidate region, and input the callus quantification features and texture features into a deep learning-based temporal discrimination model to output the probability that the fracture is a fresh fracture and the inferred post-injury time interval; S6: Based on the probability of a fresh fracture, the inferred post-injury time interval, and the measurement results of surface damage, match the preset damage degree judgment rules to generate assessment suggestions.

2. The intelligent assessment method for bone injury based on artificial intelligence and multi-phase medical imaging according to claim 1, characterized in that, Registering the second image data to the spatial coordinate system of the first image data specifically includes: A non-rigid registration algorithm is used to perform spatial transformation on the second image data.

3. The intelligent assessment method for bone injury based on artificial intelligence and multi-phase medical imaging according to claim 1, characterized in that, Calculating the callus quantification features of the second image data relative to the first image data specifically includes: Within a preset range around the candidate fracture area, voxels whose CT values ​​meet preset conditions are extracted, and the total volume of the voxels is counted as the callus volume. The preset range is 3-10mm around the fracture line, and the preset conditions include a CT value higher than a preset threshold and an increase in value greater than a preset change compared to the first image data.

4. The method according to claim 1, characterized in that, The texture features include gray-level co-occurrence matrix features; the deep learning-based temporal discrimination model employs Transformer or Temporal Convolutional Network (TCN).

5. The intelligent assessment method for bone injury based on artificial intelligence and multi-phase medical imaging according to claim 1, characterized in that, The method further includes: Data on skin damage is collected, and the skin damage area is automatically segmented based on a deep learning segmentation network. The true surface area of ​​the skin damage area is calculated on the three-dimensional reconstructed surface. The true surface area of ​​the skin damage area is compared with the threshold in the damage degree determination rule to obtain the measurement result of the skin damage.

6. The intelligent assessment method for bone injury based on artificial intelligence and multi-phase medical imaging according to claim 5, characterized in that, The surface injury data includes two-dimensional photographs or three-dimensional point cloud data containing a scale.

7. The intelligent assessment method for bone injury based on artificial intelligence and multi-phase medical imaging according to claim 1, characterized in that, The method further includes: Structured fields are extracted from the electronic medical record text using natural language processing technology, and these structured fields are then associated and mapped with the first image data and the second image data.

8. The intelligent assessment method for bone injury based on artificial intelligence and multi-phase medical imaging according to claim 1, characterized in that, Generate assessment recommendations, specifically including: A structured damage standard knowledge base is constructed, and the extracted quantitative features are matched with the damage standard knowledge base to output damage level suggestions and confidence levels, and to suggest supplementary examination items.

9. The intelligent assessment method for bone injury based on artificial intelligence and multi-phase medical imaging according to claim 1, characterized in that, The method further includes: The first image data, the second image data, the deformation field data generated during the registration process, the callus quantification features, and the evaluation suggestions are hashed and stored in a blockchain network.

10. A bone injury intelligent assessment system based on artificial intelligence and multi-phase medical imaging, characterized in that, The system includes: The data acquisition module is used to acquire the first image data of the person being assessed in the early post-injury period and the second image data during the post-injury healing period. The preprocessing and registration module is used to preprocess the first image data and the second image data, and register the second image data to the spatial coordinate system of the first image data to obtain the registered image; The identification module is used to segment the skeletal structure in the registered image and the first image data, and to identify candidate fracture regions; The calculation module is used to calculate the callus quantization features of the second image data relative to the first image data within the fracture candidate region. The temporal discrimination module is used to extract the texture features of the fracture candidate region, and input the callus quantification features and the texture features into the deep learning-based temporal discrimination model to output the probability that the fracture is a fresh fracture and the inferred post-injury time interval. The assessment suggestion generation module is used to generate assessment suggestions based on the probability that the fracture is a fresh fracture, the inferred post-injury time interval, and the measurement results of surface injury, and by matching the preset injury degree judgment rules.

Images