Bone defect clinical disease risk prediction method and system based on intelligent chip

Abstract

The invention discloses a bone defect clinical disease risk prediction method and system based on an intelligent chip, and the method comprises the steps: constructing a training data set containing a bone defect type, an external feature type, physiological information and a bone volume reduction rate, and distinguishing a normal state or an abnormal state according to whether a patient is accompanied with a disease affecting healing or not. Training a random forest model and a time sequence convolutional network model for a normal state and an abnormal state respectively by adopting a hybrid modeling strategy, and burning a normal / abnormal channel risk prediction model to an intelligent chip; the intelligent chip collects patient data, calculates the bone volume reduction rate, calls a corresponding model to predict future changes, and finally achieves dynamic risk early warning by comparing the deviation between a preset risk threshold value and a real-time monitoring value. According to the invention, the corresponding risk prediction model is selected based on the actual condition of the patient, and accurate and real-time risk prediction and monitoring of the bone defect healing process are realized.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of clinical disease risk data, in particular to a bone defect clinical disease risk prediction method and system based on an intelligent chip. BACKGROUND

[0002] With the continuous progress of science and technology, three-dimensional reconstruction technology in the field of medical imaging has been widely applied. For example, the application with the publication number CN118570124A and the name of hip bone defect detection and reconstruction method and system based on medical three-dimensional reconstruction can provide comprehensive and accurate hip bone defect detection for doctors by setting up a hip bone image acquisition module, a hip bone defect detection module, a hip bone three-dimensional reconstruction module, a hip bone reconstruction evaluation module and a hip bone image database module. By calculating the patient's hip bone defect index and hip bone three-dimensional bone model evaluation index, the system can quantify the degree of bone defect, provide objective and comparable evaluation data for doctors, help determine the severity of the patient's condition and develop treatment plans, evaluate the quality and accuracy of the reconstruction model, ensure that the reconstruction result conforms to the actual situation of the patient, and improve the credibility of the reconstruction model and the reliability of the clinical application.

[0003] However, in actual use, the existing hip bone defect detection and reconstruction method based on medical three-dimensional reconstruction only relies on single image data for judgment, ignores the type of bone defect and does not distinguish between physiological repair and pathological interference, and is prone to misjudgment in the judgment process by attributing accelerated repair to disease.

[0004] Therefore, it is one of the important problems to be solved to develop a multi-dimensional bone regeneration data model that combines bone defect types, external characteristics and clinical physiological information to form a dynamic monitoring system for bone defect clinical disease risk prediction method and system. SUMMARY

[0005] In view of the above problems existing in the prior art, the purpose of the present application is to provide a bone defect clinical disease risk prediction method based on an intelligent chip.

[0006] Another purpose of the present application is to provide a bone defect clinical disease risk prediction system based on an intelligent chip.

[0007] To solve the above problems, the present application adopts the following technical scheme: a bone defect clinical disease risk prediction method based on an intelligent chip, the method comprising the steps of:

[0008] Step 1, based on the type of bone defect, constructing a bone defect type set, an external sign type set and a physiological information set, and collecting patient time series bone volume data;

[0009] According to whether the patient is accompanied by other diseases affecting bone healing, the normal state and the abnormal state are defined, and based on the corresponding time series bone volume change data under the normal / abnormal state, the corresponding time series bone volume reduction rate data set is calculated; and the normal state training data set and the abnormal state training data set are constructed accordingly; wherein the data categories of the normal state training data set include the bone defect type, the external sign type, the physiological information and the corresponding normal state time series bone volume reduction rate data; the data categories of the abnormal state training data set include the bone defect type, the bone defect accompanied disease type, the external sign type, the physiological information and the corresponding abnormal state time series bone volume reduction rate data;

[0010] Step two, mixed modeling, in the normal state, the random forest algorithm is used, the bone defect type, the external sign type and the physiological information are taken as the input data, and the time series bone volume reduction rate data in the normal state is output, which is constructed as a normal channel risk prediction model; in the abnormal state, the time series convolution network algorithm is used, the bone defect accompanied disease type, the bone defect type, the external sign type and the physiological information are taken as the input data, and the time series bone volume reduction rate data in the abnormal state is output, which is constructed as an abnormal channel risk prediction model; the normal channel risk prediction model and the abnormal channel risk prediction model constructed are burned into the intelligent chip;

[0011] Step three, the intelligent chip collects the current bone defect type data, the external sign type data and the multi-time series bone volume data of the patient to be predicted in real time, and calculates the historical bone volume reduction rate of the patient to be predicted according to the multi-time series bone volume data;

[0012] Step four, based on the data collected in step three, according to whether the patient to be predicted is accompanied by the accompanied disease, the corresponding normal channel or abnormal channel risk prediction model is input, and the predicted bone volume reduction rate in the future period of time is output;

[0013] Step five, based on the predicted bone volume reduction rate and the current bone defect volume data of the patient to be predicted, the future time series bone volume change data is predicted, and the risk threshold is generated based on the preset bone volume loss critical value;

[0014] Step six, compare the risk threshold with the real-time actual value deviation, which is used for monitoring or early warning.

[0015] Further, the step five includes: based on the predicted bone volume reduction rate and the current bone defect volume data of the patient to be predicted, the bone defect volume data at each future time point is calculated; and the risk threshold is generated based on the expected bone defect volume and the preset critical value.

[0016] Further, the step six is specifically to compare the calculated bone defect volume data at each future time point with the actually measured time-series bone volume data, to provide data reference for the treatment strategy.

[0017] Further, the construction of the normal channel risk prediction model comprises: collecting bone defect type, external sign type, physiological information and time-series bone volume data of patients in a normal group, calculating bone volume reduction rate based on the time-series bone volume data, and training a random forest algorithm using the bone defect type, external sign type, physiological information and the calculated bone volume reduction rate, to construct the normal channel risk prediction model.

[0018] Further, the construction of the non-normal channel risk prediction model comprises:

[0019] The bone defect accompanied disease type, bone defect type, external sign type, physiological information and non-normal bone defect volume reduction rate set of patients in a historical non-normal situation are collected, a training set and a test set are divided, and a time-series convolution network model is trained and tested to obtain a non-normal channel risk prediction model.

[0020] Further, the time-series bone volume data is obtained by periodically reconstructing a three-dimensional model through image scanning.

[0021] The application also provides a system for implementing the intelligent chip-based bone defect clinical disease risk prediction method, comprising an intelligent chip and a data acquisition device in communication connection with the intelligent chip; the intelligent chip is programmed with a trained normal channel risk prediction model and a non-normal channel risk prediction model, and is configured to:

[0022] (a) obtaining current bone defect type data, external sign type data and multi-time-series bone volume data of a patient to be predicted, and calculating bone volume reduction rate of the patient according to the multi-time-series bone volume data;

[0023] (b) calling the corresponding normal channel / non-normal channel risk prediction model according to whether the patient is accompanied by an accompanied disease, and outputting predicted bone volume reduction rate in a future period of time;

[0024] (c) predicting future bone volume change data based on the predicted bone volume reduction rate and current bone defect volume data, and generating a risk threshold based on a preset bone volume change data loss critical value;

[0025] (d) comparing the deviation between the risk threshold and the actually monitored actual bone volume data, to perform risk monitoring or early warning.

[0026] The system comprises: a patient bone defect data acquisition module to be predicted, which acquires multi-dimensional data of a current patient to be predicted, including bone defect type, external sign type, physiological information and multi-time bone volume data;

[0027] A historical patient bone defect data acquisition module acquires medical record data of historical patients for subsequent model training; these data also include bone defect type, external sign feature, physiological information and corresponding time sequence bone volume data;

[0028] A bone defect volume reduction rate mapping model construction module calculates the bone volume reduction rate of a patient based on the data collected by the historical patient bone defect data acquisition module, and constructs a "normal channel risk prediction model" (random forest) and a "non-normal channel risk prediction model" (time sequence convolution network TCN) respectively;

[0029] A normal / non-normal bone defect volume reduction rate mapping module receives data collected by the patient bone defect data acquisition module to be predicted, and calculates the current bone volume reduction rate according to the time sequence bone volume data;

[0030] A normal bone defect volume data comparison and determination module calls the "normal channel risk prediction model" for a patient in a normal state without accompanying diseases, predicts the future bone volume reduction rate, calculates the bone volume data at each time point in the future, and compares the predicted value with the actual measured value;

[0031] A non-normal bone defect volume data comparison and determination module calls the "non-normal channel risk prediction model" for a patient in a non-normal state with accompanying diseases, predicts the future bone volume reduction rate, calculates the bone volume data at each time point in the future, and compares the predicted value with the actual measured value;

[0032] A final bone defect clinical disease risk level determination module compares the risk threshold value with the actual deviation value according to the determination results of the normal bone defect volume data comparison and determination module / non-normal bone defect volume data comparison and determination module, and generates a risk level report or a warning signal.

[0033] Working principle, the application builds bone defect type set and external sign type set and physiological information set, combines with the patient to be predicted, and establishes a double-channel risk prediction model under normal state / abnormal state. Random forest algorithm and time sequence convolution network algorithm are used for training respectively, bone volume change under different states is captured, and a mapping model between bone defect type information, external feature information, clinical physiological information and bone volume reduction rate is established. Normal group data: collecting historical patient bone defect type, external sign type, physiological information and time sequence bone volume data, which are used for calculating normal state bone volume reduction rate; abnormal group data: expanding the disease type set on the basis of the normal group data, which is used for calculating the non-normal state bone volume reduction rate; the current data of the patient to be predicted is collected by the intelligent chip and the bone volume reduction rate is calculated, then the corresponding prediction model is called, and the predicted bone volume reduction rate in the future period of time is output. Based on the prediction rate and the current bone defect volume of the patient, the future bone volume change can be deduced and the risk threshold can be generated. Finally, by continuously comparing the deviation of the risk threshold and the actual monitoring value, dynamic risk monitoring and early warning are realized, and quantitative data reference is provided for clinical evaluation and treatment strategy adjustment.

[0034] Compared with the prior art, the beneficial technical effects of the application are:

[0035] 1. The application realizes dynamic monitoring and prediction of disease risk of bone defect patients by integrating multi-modal data analysis model in intelligent chip; contains data acquisition layer, model construction layer and decision support layer, and adopts double model parallel mechanism (normal / abnormal state) to process different clinical scenes;

[0036] 2. Data standardization system construction: according to the bone defect type set, the feature parameters of each type are defined according to the pathological mechanism; the external sign type set contains imaging features and clinical signs; the accompanied disease type set; through multi-dimensional data acquisition, accurate disease risk judgment is realized;

[0037] 3. Time sequence data acquisition specification: volume data acquisition: continuous CT scanning; clinical information acquisition: contains biochemical indicators and treatment records; abnormal judgment standard: define the threshold of volume reduction rate mutation;

[0038] 4. Double model construction: the application sets normal channel risk prediction model and abnormal channel risk prediction model, and through intelligent condition triggering logic, it can not only avoid unnecessary imaging examination of healthy patients, but also capture early signs of malignant tumors or autoimmune diseases before the volume change reaches the clinical significant degree;

[0039] 5. Risk determination logic: normal state early warning: triggered when the predicted volume deviates from the actual volume for a certain period; non-state diagnosis: misdiagnosis determination: the disease type and the volume reduction rate pattern do not match; multiple complications detection: the volume increase rate and the single disease model prediction value differ by more than a threshold value; this application compares the future volume data set with the predicted value, and the risk determination forms a differentiated diagnosis and treatment decision tree. When the predicted volume deviates from the baseline value, the system starts the targeted review or multiple disease screening process, respectively. The hierarchical response mechanism not only optimizes the allocation of medical resources, but also breaks through the limitations of single disease diagnosis. BRIEF DESCRIPTION OF DRAWINGS

[0040] Figure 1 A flow chart of a bone defect clinical disease risk prediction method based on an intelligent chip for an embodiment of the application;

[0041] Figure 2 A structure schematic diagram of a bone defect clinical disease risk prediction system based on an intelligent chip for an embodiment of the application. DETAILED DESCRIPTION

[0042] The technical solutions of the application will be further described in detail below with reference to the embodiments and the accompanying drawings.

[0043] EMBODIMENT

[0044] As shown in Figure 1 A bone defect clinical disease risk prediction method based on an intelligent chip, the disease risk prediction method described in the application is not for disease diagnosis purposes, and the method comprises the following steps: step one, based on the type of bone defect, constructing a bone defect type set, an external sign type set, a physiological information set, and collecting time series bone volume data of patients;

[0045] According to whether the patient has other diseases that affect bone healing, normal state and non-normal state are defined, and based on the corresponding time series bone volume change data under normal / non-normal state, time series bone volume reduction rate data sets are calculated; and the normal state training data set and the non-normal state training data set are constructed accordingly; wherein the normal state training data set includes the bone defect type, the external sign type, the physiological information and the corresponding normal state time series bone volume reduction rate data; the non-normal state training data set includes the bone defect type, the bone defect accompanied disease type, the external sign type, the physiological information and the corresponding non-normal state time series bone volume reduction rate data;

[0046] Step two, mixed modeling, under normal conditions, using random forest algorithm, taking bone defect type, external sign type and physiological information as input data, outputting time series bone volume reduction rate data in normal state, constructing a normal channel risk prediction model; under abnormal conditions, using time series convolution network algorithm, taking bone defect accompanied by disease type, bone defect type, external sign type and physiological information as input data, outputting time series bone volume reduction rate data in abnormal state, constructing an abnormal channel risk prediction model; the constructed normal channel risk prediction model and abnormal channel risk prediction model are burned into the intelligent chip; in some preferred embodiments, the intelligent chip uses a low-power ARM Cortex-M7 core and supports TensorFlow Lite Micro inference framework, and the model occupies a storage of <512KB after INT8 quantization;

[0047] Step three, the intelligent chip collects current bone defect type data, external sign type data and multi-time series bone volume data of the patient to be predicted in real time, and calculates the historical bone volume reduction rate of the patient to be predicted according to the multi-time series bone volume data;

[0048] Step four, based on the data collected in step three, according to whether the patient to be predicted is accompanied by an accompanied disease, inputting the corresponding normal channel or abnormal channel risk prediction model, outputting the predicted bone volume reduction rate in the future period of time;

[0049] Step five, based on the predicted bone volume reduction rate and the current bone defect volume data of the patient to be predicted, predicting future time series bone volume change data, and generating a risk threshold based on a preset bone volume loss critical value;

[0050] Step six, comparing the risk threshold with the real-time actual value deviation for monitoring or early warning.

[0051] The present application forms a multi-dimensional, dynamic monitoring system by constructing a multi-dimensional data model, combining bone defect type, external sign type and clinical physiological information. In clinical practice, bone defects can be divided into main types such as traumatic defects, pathological defects, infectious defects and congenital defects according to causes, positions and morphological characteristics. Each type corresponds to different healing mechanisms and complication risks, for example, traumatic defects are usually accompanied by surrounding soft tissue injury, while pathological defects may reflect systemic bone diseases.

[0052] Meanwhile, the external sign type set is refined through imaging and clinical observation, including key index data such as edge hardening degree, density change in defect area, and periosteal reaction. The index data and the bone defect type form a complementary relationship: for example, an infectious defect often shows an irregular edge and surrounding bone dissolution, while a congenital defect shows a clear boundary of developmental abnormalities. The application establishes a bone defect type-external feature correlation matrix to capture the biological behavior differences of different defect modes and provide structured classification basis for subsequent time series data analysis. The bone defect type set and the external sign type set provided by the application are consistent with the clinical diagnosis logic of orthopedics and provide quantifiable input features for machine learning models.

[0053] In some preferred embodiments, for a patient to be predicted, multi-dimensional data are synchronously collected, including: bone defect type data, specifically obtained through imaging examination and medical records to determine the clinical bone injury classification of the defect;

[0054] External sign type data, specifically CT / MRI-dependent imaging analysis to quantify defect edge morphology, density gradient and other parameters;

[0055] Clinical physiological information, including age, underlying diseases, nutrition indicators and other variable data affecting healing.

[0056] Time series bone volume reduction rate data, which is obtained by reconstructing a three-dimensional model through regular imaging scans to accurately track the spatial changes of the defect.

[0057] In some embodiments, the patient time series bone volume data is collected in a dual-path acquisition mode. The data of patients in a normal state is used to establish a normal channel risk prediction model, and the data of patients in an abnormal state includes cases affected by diseases such as infection and osteoporosis. The application separates the above two data sets to ensure the purity of the normal healing trajectory and enhance the identification ability of disease interference factors.

[0058] In some cases, for the same patient to be predicted at different healing stages, the application also provides a dynamic acquisition strategy, which compares the differences in bone volume change rates between normal and abnormal groups to provide negative samples with clinical significance for subsequent model training.

[0059] Specifically, the application establishes a dynamic mapping relationship between bone defect characteristics and bone volume reduction rate changes; for patients with bone defects under normal circumstances, the method described in the application uses a random forest model to construct a bone volume reduction rate prediction system. The model analyzes the bone defect type, external type and clinical physiological information in the historical data to capture the synergistic effect of multiple factors on the healing rate. For example, traumatic defects show a faster volume reduction trend in young patients, while patients with diabetes may experience delayed healing due to microcirculation disorders.

[0060] For patients under abnormal circumstances, a non-normal channel risk prediction model is used to analyze bone volume reduction rate data to identify the unique fluctuating reduction pattern in infectious defects or the abnormal expansion phenomenon caused by malignant proliferation in pathological defects, in order to handle the complex interference brought by accompanying diseases.

[0061] The structure of the normal / abnormal channel risk prediction model is shown in Table 1 Normal / Abnormal State Risk Prediction Model Structure Example:

[0062] Table 1 Normal / Abnormal State Risk Prediction Model Structure Example Table

[0063]

[0064] Normal data has strong regularity, random forest is good at processing structured data, and has strong ability to capture feature interaction and nonlinear relationship; and high interpretability, which is convenient for clinicians to understand the prediction logic; while abnormal data is accompanied by disease interference, which needs to capture the timing dynamics and key features; the dilated convolution of TCN can extract long-range dependencies, and the attention mechanism can weight important features;

[0065] The normal / abnormal channel risk prediction model described in the application corresponds to the physiological and pathological states of bone defect healing through differentiated algorithm design, and their synergistic effect constitutes the dual prediction ability of the system.

[0066] In use, the doctor inputs the current bone data of the patient to obtain the personalized bone volume change prediction curve of the patient, providing a quantitative basis for clinical decision-making.

[0067] The application also provides a bone defect disease risk prediction system based on an intelligent chip, such as Figure 2As shown, the system is practically applied in dynamic monitoring and risk warning of patient healing process. The system constructs a normal channel / abnormal channel risk prediction model, constructs a bone defect type set and an external sign type set, physiological information and time series bone volume data based on the bone defect type, calculates normal state time series bone volume reduction rate data based on the time series bone volume data, establishes a normal channel risk prediction model, constructs a corresponding bone defect type set, bone defect accompanied disease type, constructs an external sign type set, physiological information and time series bone volume data based on the bone defect type, calculates normal state time series bone volume reduction rate data based on the time series bone volume data, and establishes an abnormal channel risk prediction model.

[0068] Among them, the historical bone defect accompanied disease type data, bone defect type data, bone defect external sign type data, clinical physiological information and volume reduction rate data of the patient group in the abnormal situation are used to construct a mapping model between the bone defect accompanied disease type data, the bone defect type data, the bone defect external sign type data, the clinical physiological information and the bone defect volume reduction rate, and an abnormal channel risk prediction model is obtained.

[0069] The bone defect data of the patient to be predicted is collected by the bone defect data of the patient to be predicted data collection module.

[0070] The historical patient bone defect data collection module is used to collect the historical data of the patient to be predicted. It should be noted that the physiological information includes gender, age, weight, height and body mass index, etc. Specifically, the clinical physiological information can be obtained by investigating the medical record information system. The bone defect type data and the bone defect external sign type data can be obtained by collecting the X-ray and CT imaging data of the patient before and after the operation during the follow-up period. The CT shows that there is a relatively independent low-density bone defect area in the vertebral body within 3 days after the operation. Then, two orthopedic doctors with more than 5 years of work experience identify the bone defect. If there is a difference, another vice director identifies the bone defect to determine the final result.

[0071] As described in the present application, the time series bone volume reduction rate data is obtained by periodically scanning the three-dimensional model, for example, the bone defect volume data can be obtained by collecting the original CT image data (layer thickness 0.625 mm) of the patient after the internal fixation surgery to the discharge. Specifically, the three-dimensional reconstruction image is obtained by importing the Dicom format into the Mimics19.0 software (Materialise company), and the bone CT threshold mask is selected by the threshold segmentation method. In the threshold segmentation option, the Edit Mask and Region Growing tools are used to edit the bone defect mask to obtain an accurate independent bone defect mask. After three-dimensional reconstruction of the bone defect, the volume value can be seen by viewing the attribute.

[0072] AsFigure 2 As shown, the application provides a bone defect clinical disease risk prediction system based on an intelligent chip, which comprises an intelligent chip and a data acquisition device in communication connection with the intelligent chip; the intelligent chip has a trained normal channel risk prediction model and an abnormal channel risk prediction model burned therein and is configured to:

[0073] (a) acquire current bone defect type data, external sign type data and multi-time sequence bone volume data of a patient to be predicted, and calculate a bone volume reduction rate of the patient according to the multi-time sequence bone volume data;

[0074] (b) according to whether the patient is accompanied by a concomitant disease, call the corresponding normal channel / abnormal channel risk prediction model, and output a predicted bone volume reduction rate in a future period of time;

[0075] (c) predict future bone volume change data based on the predicted bone volume reduction rate and current bone defect volume data, and generate a risk threshold based on a preset bone volume change data loss critical value;

[0076] (d) compare the deviation between the risk threshold and the actual bone volume data monitored in real time to perform risk monitoring or early warning.

[0077] The software functions of the system include the following several modules working in cooperation:

[0078] A bone defect data acquisition module of a patient to be predicted acquires multi-dimensional data of a patient to be predicted, including bone defect type, external sign type, physiological information and multi-time sequence bone volume data;

[0079] A historical patient bone defect data acquisition module acquires medical record data of a historical patient for subsequent model training. These data also include bone defect type, external sign feature, physiological information and corresponding time sequence bone volume data;

[0080] A bone defect volume reduction rate mapping model construction module calculates the bone volume reduction rate of a patient based on the data collected by the historical patient bone defect data acquisition module, and constructs a “normal channel risk prediction model” (random forest) and a “non-normal channel risk prediction model” (time sequence convolution network TCN) respectively; for example, by analyzing medical image data (mainly CT or cone beam CT) of a patient at different time points (for example, 1 month, 3 months and 6 months after surgery), a model capable of dynamically and quantitatively displaying and analyzing the new bone growth speed and spatial distribution law of the bone defect area is constructed, and the time sequence medical image is converted into bone regeneration dynamics information.

[0081] Normal / abnormal bone defect volume reduction rate mapping module, receiving the data collected by the patient bone defect data acquisition module to be predicted, and calculating the current bone volume reduction rate according to the time sequence bone volume data;

[0082] Normal bone defect volume data comparison determination module, for patients in normal state without accompanying diseases, calling the "normal channel risk prediction model", predicting the future bone volume reduction rate, calculating the future bone volume data at each time point, and comparing the predicted value with the actual measured value;

[0083] Abnormal bone defect volume data comparison determination module, for patients in abnormal state with accompanying diseases, calling the "abnormal channel risk prediction model", predicting the future bone volume reduction rate, calculating the future bone volume data at each time point, and comparing the predicted value with the actual measured value;

[0084] Final bone defect clinical disease risk level determination module, according to the determination results of the normal bone defect volume data comparison determination module / abnormal bone defect volume data comparison determination module, comparing the risk threshold value with the actual deviation value, and generating a risk level report or a warning signal.

[0085] When the system evaluates that the patient is in a normal healing state, the intelligent chip generates a prediction curve based on the normal channel risk prediction model, showing the expected defect volume data at each future time point. If the actual monitoring data deviates from the predicted trajectory, for example, the volume reduction rate suddenly slows down or expands abnormally, the system will automatically trigger the accompanying disease examination suggestion, suggesting that there may be unidentified infection or metabolic abnormalities.

[0086] For patients who have been diagnosed with accompanying diseases, the abnormal channel risk prediction model judges the effectiveness of the current treatment plan by comparing the time sequence difference between the predicted volume and the actual volume. If the predicted value is consistently higher than the actual measured value, the accuracy of the initial diagnosis needs to be considered; on the contrary, if the predicted value is significantly lower than the actual value, it may indicate the presence of multiple concurrent diseases or treatment resistance.

[0087] The dynamic feedback mechanism enables clinicians to adjust treatment strategies in a timely manner, transforming traditional passive observation into active intervention. The real-time computing capability of the intelligent chip is particularly suitable for postoperative follow-up scenarios. After analyzing the image data taken by the patient regularly, the risk level report can be generated, significantly improving the efficiency and accuracy of chronic bone defect management.

[0088] Taking the prediction and intervention process of a traumatic tibial defect patient as an example, as follows:

[0089] Patient basic information: Current bone defect data: Defect type: Traumatic cortical bone defect (AO classification 42-B2); Defect volume: 1200mm³ (CT three-dimensional reconstruction measurement); External sign type: Tibia second degree depression (determined by X-ray and CT imaging data);

[0090] Clinical physiological information 33: 32-year-old male, BMI 24.3, serum calcium 2.4mmol / L, ALP 85U / L;

[0091] Normal state prediction path: Input mapping model: input the above data into the normal channel risk prediction model (random forest model), output the predicted bone defect volume reduction rate is 1.8mm³ / day; Future volume prediction: based on the current volume 1200mm³, calculate the volume in the next 30 days, as shown in the following table 1-1:

[0092] Table 1-1

[0093]

[0094] Abnormal trigger check: If the actual monitoring finds that the volume on the 15th day is 1190mm³ (> predicted value 1185mm³), and the subsequent 25th day is 1175mm³ (> predicted value 1167mm³), the system determines that the accompanying disease (such as occult osteomyelitis) needs to be checked;

[0095] Non-normal state prediction path: Accompanying disease input: diabetes (HbA1c 7.8%), vasculitis (CRP 15mg / L);

[0096] Non-normal model prediction (TCN + attention mechanism): Output bone volume reduction rate 0.6mm³ / day (due to metabolic disease inhibiting bone regeneration); Function: TCN's dilated convolution structure is good at capturing long-range dependencies, and the attention mechanism can dynamically weight important time points and physiological indicators, both of which complement each other to improve the sensitivity of the model to pathological states; Structure: attention layer is located after TCN;

[0097] Calculate the predicted volume in the next 30 days, as shown in the following table 1-2:

[0098] Table 1-2

[0099]

[0100] Dynamic determination:

[0101] Case 1: Actual volume on the 20th day is 1180mm³ (< predicted value 1188mm³) and continues to be low, then trigger diabetes control review (exclude insulin resistance aggravation);

[0102] Case 2: Actual volume of 1195 mm3 on day 20 (> predicted value of 1188 mm3) and consistently high, check for hyperparathyroidism (abnormal blood calcium).

[0103] Risk level output: Final decision: If the examination confirms that the blood vessels are in the active stage, the risk level is upgraded to level III (high risk of infection); trigger the following interventions: biological treatment (TNF-α inhibitor), customized bone transport device parameter adjustment (pulling speed reduced to 0.5 mm / day).

[0104] Finally, it should be pointed out that the above is only the preferred specific embodiment of the present application, but the protection scope of the present application is not limited thereto, any person skilled in the art can easily think of changes or replacements within the technical scope disclosed by the present application, which should be covered within the protection scope of the present application.

Claims

1. A method for predicting the clinical risk of bone defects based on a smart chip, characterized in that, The method includes the following steps: Step 1: Based on the type of bone defect, construct a set of bone defect types, a set of external signs, and a set of physiological information, and collect patients' time-series bone volume data; Based on whether the patient has other diseases that affect bone healing, normal and abnormal states are defined. Based on the time-series bone volume change data corresponding to the normal / abnormal states, the corresponding time-series bone volume reduction rate datasets are calculated. Accordingly, normal state training datasets and abnormal state training datasets are constructed; wherein, the normal state training dataset includes the aforementioned bone defect type, external sign type, physiological information and corresponding normal state time-series bone volume reduction rate data; the abnormal state training dataset includes the aforementioned bone defect type, bone defect comorbidity disease type, external sign type, physiological information and corresponding abnormal state time-series bone volume reduction rate data. Step 2: Hybrid modeling. Under normal conditions, a random forest algorithm is used, taking bone defect type, external sign type, and physiological information as input data, and outputting time-series bone volume reduction rate data under normal conditions to construct a normal channel risk prediction model. Under abnormal conditions, a temporal convolutional network algorithm is used, taking bone defect comorbidity type, bone defect type, external sign type, and physiological information as input data, and outputting time-series bone volume reduction rate data under abnormal conditions to construct an abnormal channel risk prediction model. The completed normal channel risk prediction model and abnormal channel risk prediction model are then burned into a smart chip. Step 3: The intelligent chip collects the current bone defect type data, external sign type data and multi-time series bone volume data of the patient to be predicted in real time, and calculates the historical bone volume reduction rate of the patient to be predicted based on the multi-time series bone volume data. Step 4: Based on the data collected in Step 3, and depending on whether the patient to be predicted has any comorbid diseases, input the corresponding normal or abnormal channel risk prediction model, and output the predicted rate of bone volume reduction in the future. Step 5: Based on the predicted rate of bone volume reduction and the current bone defect volume data of the patient to be predicted, predict the future time-series bone volume change data, and generate a risk threshold based on the preset bone volume loss threshold. Step 6: Compare the deviation between the risk threshold and the real-time actual value for monitoring or early warning.

2. The method for predicting the clinical risk of bone defects based on intelligent chips according to claim 1, characterized in that, Step five includes: calculating bone defect volume data at future time points based on the predicted rate of bone volume reduction and the current bone defect volume data of the patient to be predicted; the risk threshold is generated based on the expected bone defect volume and a preset critical value.

3. The method for predicting the clinical risk of bone defects based on intelligent chips according to claim 2, characterized in that, Step six specifically involves comparing the calculated bone defect volume data at each future time point with the actual measured time-series bone volume data to provide data reference for treatment strategies.

4. The method for predicting the clinical risk of bone defects based on a smart chip according to claim 1, characterized in that, The construction of the normal pathway risk prediction model includes: collecting bone defect type, external sign type, physiological information and time-series bone volume data of normal patients; calculating the bone volume reduction rate based on the time-series bone volume data; and training a random forest algorithm using the bone defect type, external sign type, physiological information and the calculated bone volume reduction rate to construct the normal pathway risk prediction model.

5. The method for predicting the clinical disease risk of bone defects based on intelligent chips according to claim 1, characterized in that, The construction of the abnormal channel risk prediction model includes: We collected data on bone defects in patients under abnormal conditions, including associated disease types, defect types, external signs, physiological information, and the rate of reduction in abnormal bone defect volume. We then divided these data into training and testing sets to train and test a temporal convolutional network model, thereby obtaining a risk prediction model for abnormal pathways.

6. The method for predicting the clinical disease risk of bone defects based on intelligent chips according to claim 1, characterized in that, The time-series bone volume data is obtained by reconstructing a three-dimensional model through periodic image scanning.

7. A system for implementing the smart chip-based clinical disease risk prediction method for bone defects as described in claim 1, characterized in that, It includes a smart chip and a data acquisition device that communicates with the smart chip; the smart chip has a trained normal channel risk prediction model and an abnormal channel risk prediction model programmed into it, and is configured as follows: (a) Obtain current bone defect type data, external sign type data and multi-temporal bone volume data of the patient to be predicted, and calculate the rate of bone volume reduction of the patient based on the multi-temporal bone volume data; (b) Depending on whether the patient has comorbidities, call the corresponding normal / abnormal channel risk prediction model and output the predicted rate of bone volume reduction over a future period of time. (c) Based on the predicted rate of bone volume reduction and the current bone defect volume data, predict future bone volume change data, and generate a risk threshold based on a preset bone volume change data loss threshold. (d) Compare the deviation between the risk threshold and the actual bone volume data monitored in real time for risk monitoring or early warning.

8. The clinical disease risk prediction system for bone defects based on a smart chip according to claim 7, characterized in that, The system includes: a bone defect data acquisition module for patients to be predicted, which collects multi-dimensional data of the current patients to be predicted, including bone defect type, external sign type, physiological information and multi-time series bone volume data; The historical patient bone defect data acquisition module collects historical patient medical record data for subsequent model training; this data also includes bone defect type, external characteristics, physiological information and corresponding time-series bone volume data. The bone defect volume reduction rate mapping model construction module calculates the bone volume reduction rate of patients based on the data collected by the historical patient bone defect data acquisition module, and constructs a "normal channel risk prediction model" (random forest) and an "abnormal channel risk prediction model" (temporal convolutional network TCN). The normal / abnormal bone defect volume reduction rate mapping module receives data collected by the bone defect data acquisition module of the patient to be predicted, and calculates the current bone volume reduction rate based on the time-series bone volume data. The normal bone defect volume data comparison and judgment module, for patients in a normal state without accompanying diseases, calls the "normal channel risk prediction model" to predict the future rate of bone volume reduction, calculates bone volume data at each future time point, and compares the predicted value with the actual measured value. The abnormal bone defect volume data comparison and judgment module, for patients in abnormal states with comorbid diseases, calls the "abnormal channel risk prediction model" to predict the future rate of bone volume reduction, calculates bone volume data at each future time point, and compares the predicted value with the actual measured value. The final clinical risk level determination module for bone defects compares the risk threshold with the actual deviation value based on the judgment results of the normal bone defect volume comparison module / abnormal bone defect volume comparison module, and generates a risk level report or warning signal.

Images