Biological material implantation effect prediction system based on deep learning

Abstract

The invention relates to the technical field of biomedicine and artificial intelligence, in particular to a biological material implantation effect prediction system based on deep learning. According to the method, the deep learning model is combined with the convolutional neural network and the long-short-term memory network, multiple data types are effectively fused, the accuracy and robustness of implantation effect prediction are improved, potential operation risks and complications can be identified, and a scientific treatment decision basis is provided for doctors. And the signal matching and feedback module compares the prediction result with historical case data to generate feedback signals including high matching, a re-evaluation scheme and a risk warning signal. The system dynamically adjusts a treatment scheme according to feedback to ensure efficient intervention. Through intelligent prediction, subjective judgment is reduced, the utilization efficiency of medical resources is improved, invalid treatment and excessive treatment are avoided, and meanwhile, a treatment scheme is optimized by referring to historical cases.

Description

Technical Field

[0001] The present invention relates to the field of biomedicine and artificial intelligence technology, and in particular to a biomaterial implantation effect prediction system based on deep learning. Background Art

[0002] Biomaterial implantation surgery has important applications in medical treatment and is widely used in orthopedics, cardiovascular, neurosurgery, tumor treatment and other fields. Although current medical technology has made significant progress, the individual differences in the effects of biomaterial implantation are large. How to accurately predict the implantation effect and reduce postoperative complications remains a difficult problem in clinical treatment.

[0003] Traditional implant effect evaluation mainly relies on the doctor's experience and postoperative observation, and it is often difficult to accurately quantify individual differences among patients. At the same time, with the increase in the amount of information such as patient physiological characteristics, historical surgical data, and medical imaging, how to effectively integrate these multi-dimensional data and improve the accuracy of prediction has become a challenge in medical research and clinical practice.

[0004] The rise of deep learning technology has provided new solutions for the processing and analysis of biomedical data, especially showing great potential in medical image analysis, time series data processing and multimodal data fusion.

[0005] Although some current methods have been applied to predict the effects of biomaterial implantation, the following problems and shortcomings still exist:

[0006] First, existing technologies often process physiological characteristic data, medical imaging data, historical surgical data, and biomaterial property data separately, lacking an effective data fusion mechanism. Traditional methods do not fully utilize the complementarity of multimodal data, resulting in limited prediction results and difficulty in comprehensively evaluating the implant effect of patients.

[0007] Secondly, most current prediction models rely on a single data source or simple machine learning methods, which are unable to deeply explore the complex relationships in the data. Traditional implant effect prediction systems often rely on physician experience and postoperative follow-up, lack scientific prediction tools, and result in low accuracy and robustness of prediction results.

[0008] In response to the above problems, it is necessary to propose a biomaterial implantation effect prediction system based on deep learning. Summary of the invention

[0009] The purpose of the present invention is to solve the problems existing in the background technology and to propose a biomaterial implantation effect prediction system based on deep learning.

[0010] The purpose of the present invention can be achieved through the following technical solutions:

[0011] The present invention provides a prediction system for the implantation effect of biomaterials based on deep learning, including a data acquisition module, a data preprocessing module, a deep learning and prediction module, a comparison database, and a signal matching and feedback module.

[0012] The data acquisition module is responsible for collecting biomedical data of the target patient for biomaterial implantation, including the physiological characteristics, historical surgical data, and medical imaging data of the target patient. Collect the chemical and physical property data of the biomedical material to be implanted to prepare data for subsequent effect prediction modeling.

[0013] The specifically collected data includes:

[0014] Physiological characteristic data, including heart rate x11, blood pressure x12, body temperature x13, blood oxygen saturation x14, and respiratory rate x15.

[0015] Historical surgical data: surgical type identifier X21, surgical date X22, and postoperative complication identifier X23.

[0016] Among them, the surgical type identifier represents different preset surgical types through specific numerical values, including orthopedic surgery, cardiovascular surgery, oncology surgery, neurosurgery, transplantation surgery, ophthalmic surgery, thoracic surgery, gastrointestinal surgery, urological surgery, and gynecological surgery;

[0017] Among them, the postoperative complication identifier represents different preset postoperative complication types through specific numerical values, including postoperative infection, hematoma or bleeding, deep vein thrombosis, pulmonary complications, cardiovascular complications, renal failure, nerve injury, poor wound healing, nausea and vomiting, and drug allergic reactions.

[0018] Medical imaging data, medical images Pi at the implantation position collected by medical imaging devices based on MRI, CT, and X-ray. Where i is the medical image identifier, i = 1, 2, 3,..., n; where n is the total number of medical images.

[0019] The chemical and physical property data of the biomedical material, including the mechanical properties, chemical stability, and cell affinity data of the biomaterial. Specifically includes elastic modulus x31, tensile strength x32, half-maximal inhibitory concentration x33, degradation time x34, and cell attachment rate x35.

[0020] The data preprocessing module is responsible for cleaning and normalizing the collected data, converting parameters with different dimensions into dimensionless parameters with similar numerical ranges, and unifying them to the same scale to provide high-quality input data for the subsequent deep learning model.

[0021] The specific process of data cleaning is as follows:

[0022] For numerical parameters: physiological feature data, remove invalid data, duplicate data, and outliers, and remove abnormal numerical values that exceed the preset threshold.

[0023] For the collected medical image data, use the wavelet transform denoising algorithm to suppress noise to improve the image quality. The wavelet transform denoising algorithm is specifically as follows:

[0024] Perform wavelet transform on the color value fi(x, y) of the pixels in the original image Pi:

[0025] where fi(x, y) is the color value of the pixels in each medical image Pi before filtering, and (x, y) is the pixel coordinate; Wi(a, b) is the wavelet transform coefficient corresponding to the pixels in each medical image Pi obtained after filtering; where a and b are the preset scale parameter and translation parameter respectively, and the preset (a, b) combination obtains different wavelet transform coefficients Wi(a, b) by controlling the different extraction precisions of high-frequency noise and low-frequency noise. Among them is the wavelet function; among them and are the scale parameters of the wavelet transform in the x-axis and y-axis directions respectively; among them and are the translation parameters of the wavelet transform in the x-axis and y-axis directions respectively;

[0026] Perform hard threshold processing on the obtained wavelet transform coefficients Wi(a, b) to obtain the wavelet transform coefficients after threshold processing ;

[0027] For the wavelet transform coefficients after threshold processing, then restore them back to the denoised version of the original image through inverse wavelet transform to obtain the pixel color value . The inverse wavelet transform formula is:

[0028] ; among them is the inverse wavelet transform function.

[0029] Arrange the pixel color values of each biomedical image Pi after wavelet transform denoising according to the pixel coordinates (x, y), and compress them to the preset image resolution w×h to generate the pixel color value matrix Ui = U1, U2,..., Un of the biomedical image Pi.

[0030] The specific process of normalization is as follows:

[0031] For the chemical and physical property data of the biomedical material to be implanted and the physiological characteristic data after data cleaning, the data is normalized using a standardization formula to make its mean 0 and standard deviation 1, and at the same time, all numerical data is converted into dimensionless parameters. The formula is as follows:

[0032] ; where X is the physiological characteristic data before normalization, including the heart rate x11, blood pressure x12, body temperature x13, blood oxygen saturation x14, and respiratory rate x15 after data cleaning; Z(X) is the physiological characteristic data after normalization processing. Among them and are the mean and standard deviation of the physiological characteristic data X respectively.

[0033] The deep learning and prediction module includes a CNN (Convolutional Neural Network) for medical image feature extraction and an LSTM (Long Short-Term Memory Network) for numerical data and associated feature extraction. All the data output by the data preprocessing module is subjected to data fusion, and the prediction of the implant effect of the biomedical material is carried out through a deep learning model. A hybrid model combining a CNN (Convolutional Neural Network) and an LSTM (Long Short-Term Memory Network) is used to handle the joint prediction task of physiological characteristics, historical surgical data, medical image data, and the chemical and physical property data of the biomedical material to be implanted based on the cleaning process and normalization process.

[0034] Data fusion is carried out. The physiological characteristics, historical surgical data, medical image data, and the chemical and physical property data of the biomedical material to be implanted after cleaning and normalization are collected to generate a multi-dimensional data matrix H, where x11, x12,..., x35 are all the physiological characteristics, historical surgical data, and the chemical and physical property data of the biomedical material to be implanted after cleaning and normalization; where U1, U2,..., Un are the pixel color value matrices of the biomedical images after wavelet transform denoising;

[0035] The CNN (Convolutional Neural Network) is used to extract the spatial features of medical image data. The CNN gradually extracts the high-level features of the image through multiple convolutional layers, pooling layers, and fully connected layers, and finally outputs an image feature vector. The formula for the convolutional layer is:

[0036] ; Yi is the image feature vector extracted from the medical image Pi; among them is the kernel function of the convolutional layer, * is the convolution operator, is the bias coefficient.

[0037] The LSTM (Long Short-Term Memory) network is used to extract the temporal feature information from physiological features, historical surgical data, medical imaging data, and the chemical and physical property data of the biomedical material to be implanted, as well as the relationships between medical imaging data and physiological features, historical surgical data, medical imaging data, and the chemical and physical property data of the biomedical material to be implanted. The LSTM network includes an input gate, a forget gate, a memory gate, an output gate, and a candidate memory gate;

[0038] Among them, the input gate is responsible for receiving the multi-dimensional data matrix H and determining which information in the current input will be written into the memory cell through a convolution operation. The mathematical formula is:

[0039] ;

[0040] Among them, i(t) is the output of the input gate, indicating the influence of the current input on the memory cell, ranging from 0 to 1; among them, σ is the activation function; among them is the weight matrix of the input gate; is the bias term of the input gate;

[0041] Among them, the forget gate determines which information in the memory cell at the previous moment is forgotten. The mathematical formula is:

[0042] ;

[0043] Among them, σ is the activation function; among them is the output of the forget gate, indicating the proportion of the forgotten information, ranging from 0 to 1; among them is the weight matrix of the forget gate; is the bias term of the forget gate; among them is the hidden state at the previous moment, and the specific value is calculated by the output gate; among them is the combination of the multi-dimensional data matrix H input at the current moment t and the hidden state information at the previous moment t - 1;

[0044] The memory cell combines the outputs of the forget gate and the input gate to update the memory cell. The mathematical formula is:

[0045] ;

[0046] Among them, C(t) is the memory cell at the current moment; among them, C(t - 1) is the memory cell at the previous moment; among them, f(t) × C(t - 1) is the retained part of the memory cell at the previous moment; among them is the updated part of the current moment input; among them, tanh is the hyperbolic tangent function, and the output value ranges from -1 to 1. The formula is ; among them is the weight matrix of the memory cell; among them The bias term for memory cell update;

[0047] The output gate calculates the hidden state at the current time based on the memory cell and the input, and outputs the final result. The mathematical formula is:

[0048] ; where is the weight matrix of the output gate; where is the bias term of the output gate; where h(t) is the hidden state at the current time and also the final output of the LSTM neural network model; where o(t) is the output of the output gate, which determines which information in the memory cell C(t) is output.

[0049] As a preferred embodiment of the present invention, the image feature vector Yi extracted by the CNN is combined with h(t) output after the LSTM network is processed for joint prediction. These different types of data are fused through a fully connected layer to generate the final prediction result.

[0050] As a preferred embodiment of the present invention, the joint feature vector after being processed by the fully connected layer is numerically normalized through an activation function to obtain the prediction result. The prediction result can be a regression value Mark with a preset numerical range, and the numerical range is from 0 to 100. The higher the value, the better the prediction result of the implantation effect.

[0051] The comparison database is responsible for storing all the prediction results of the deep learning model, that is, the regression value Mark. For each implantation target, a data set including historical patient data, surgical data, implant material data, and the regression value Mark is generated. The data in this database is strictly screened and classified, which can provide a reference basis for the system to compare with the prediction results of the current patient.

[0052] The signal matching module converts the output result of the deep learning model into a signal suitable for performing operations, and selects different operations according to different prediction results.

[0053] If it is recognized that the regression value Mark is greater than the first preset threshold Mark1, a high matching feedback signal is generated: indicating that the predicted implantation effect is relatively ideal, and it is recommended to continue the current implantation plan;

[0054] If it is recognized that the regression value Mark is greater than the second preset threshold Mark2 and less than or equal to the first preset threshold Mark1, it indicates that there may be potential risks or complications in the predicted implantation effect. It is recommended to conduct further examinations or modify the treatment plan, and it is recommended to re-evaluate the current implantation plan;

[0055] If it is recognized that the regression value Mark is less than or equal to the second preset threshold Mark2, it indicates that the prediction result has a large difference from the historical data, there is a high risk, and a risk warning signal is generated to remind medical staff to take emergency intervention measures.

[0056] Compared with the prior art, the beneficial effects of the present invention are as follows:

[0057] 1. Through the training and prediction of the deep learning model, the present invention combines multiple data types and uses a hybrid model that combines a convolutional neural network and a long short-term memory network to jointly predict multi-modal data. This model structure can effectively fuse data from different sources, extract potential deep-level features, and improve the accuracy and robustness of implant effect prediction. Through the training and optimization of the model, the system can identify and predict potential surgical risks and complications, thereby providing a more scientific basis for doctors to make treatment decisions;

[0058] 2. The signal matching and feedback module of the present invention generates feedback signals for different prediction results by comparing with historical case data according to the prediction results of the deep learning model. These feedback signals include high matching feedback signals, re-evaluation plan signals, and risk warning signals. Through this mechanism, the system can dynamically adjust the treatment plan of the patient. For example, when the prediction result shows that the implant effect is ideal, the system generates a signal to maintain the status quo; if the prediction result shows potential risks, the system will send a signal to re-evaluate the treatment plan, and even send an emergency warning signal in case of serious risks, prompting medical staff to intervene;

[0059] 3. Through the intelligent prediction system of the present invention, medical staff can effectively adjust the treatment plan of the patient based on the accurate prediction and feedback signals provided by the system. This not only reduces the subjective judgment factors in traditional medical plans, but also greatly improves the utilization efficiency of medical resources. Doctors can make decisions at the appropriate time according to the prediction and feedback of the system, avoiding ineffective treatment and over-treatment. In addition, by comparing the historical case data in the database, the system can help medical staff refer to similar cases and further optimize the treatment plan. BRIEF DESCRIPTION OF THE DRAWINGS

[0060] For the convenience of those skilled in the art to understand, the present invention will be further described below with reference to the accompanying drawings:

[0061] Figure 1 is the system block diagram of the present invention; DETAILED DESCRIPTION OF THE INVENTION

[0062] The technical solution of the present invention will be clearly and completely described below in conjunction with the embodiments. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts shall fall within the protection scope of the present invention.

[0063] Please refer to Figure 1 As shown, a prediction system for the implantation effect of biomaterials based on deep learning includes a data acquisition module, a data preprocessing module, a deep learning and prediction module, a comparison database, and a signal matching and feedback module.

[0064] The data acquisition module is responsible for collecting biomedical data of the target patient for biomaterial implantation, including the physiological characteristics, historical surgical data, and medical imaging data of the target patient. Collect chemical and physical property data of the biomedical material to be implanted for data preparation for subsequent effect prediction modeling.

[0065] The specifically collected data includes:

[0066] Physiological characteristic data, including heart rate x11, blood pressure x12, body temperature x13, blood oxygen saturation x14, and respiratory rate x15.

[0067] Historical surgical data: surgical type identifier X21, surgical date X22, and postoperative complication identifier X23.

[0068] Among them, the surgical type identifier represents different preset surgical types through specific values, including orthopedic surgery, cardiovascular surgery, oncology surgery, neurosurgery, transplantation surgery, ophthalmic surgery, thoracic surgery, gastrointestinal surgery, urological surgery, and gynecological surgery;

[0069] Among them, the postoperative complication identifier represents different preset postoperative complication types through specific values, including postoperative infection, hematoma or bleeding, deep vein thrombosis, pulmonary complications, cardiovascular complications, renal failure, nerve injury, poor wound healing, nausea and vomiting, and drug allergic reactions.

[0070] Medical imaging data, medical images Pi at the implantation position collected by medical imaging devices based on MRI, CT, and X-ray. Where i is the medical image identifier, i = 1, 2, 3,..., n; where n is the total number of medical images.

[0071] Chemical and physical property data of the biomedical material, including mechanical properties, chemical stability, and cell affinity data of the biomaterial. Specifically includes elastic modulus x31, tensile strength x32, half-maximal inhibitory concentration x33, degradation time x34, and cell attachment rate x35.

[0072] It should be noted that the half maximal inhibitory concentration (IC50): It is often used to represent the ability of a material to inhibit the generation of free radicals under specific conditions. The measurement method is to add antioxidants or materials with different concentrations to the free radical system and gradually increase the concentration until a 50% inhibitory effect is achieved. Its specific value is measured in the laboratory and reflects the antioxidant ability of the biological material in vivo.

[0073] The data preprocessing module is responsible for cleaning and normalizing the collected data, converting parameters with different dimensions into dimensionless parameters with similar numerical ranges, unifying them to the same scale, and providing high-quality input data for the subsequent deep learning model.

[0074] The specific process of data cleaning is as follows:

[0075] For numerical parameters: physiological characteristic data, remove the invalid data, duplicate data, and outliers, and remove the abnormal numerical values exceeding the preset threshold.

[0076] For the collected medical image data, use the wavelet transform denoising algorithm to suppress noise to improve the image quality. The wavelet transform denoising algorithm is specifically as follows:

[0077] Perform wavelet transform on the color value fi(x, y) of the pixels in the original image Pi:

[0078] Where fi(x, y) is the color value of the pixels in each medical image Pi before filtering, where (x, y) is the pixel coordinate; Wi(a, b) is the wavelet transform coefficient corresponding to the pixels in each medical image Pi after filtering; where a and b are the preset scale parameter and translation parameter respectively, and the preset (a, b) combination obtains different wavelet transform coefficients Wi(a, b) by controlling the different extraction precisions of high-frequency noise and low-frequency noise. Among them is the wavelet function; among them and are the scale parameters of the wavelet transform in the x-axis and y-axis directions respectively; among them and are the translation parameters of the wavelet transform in the x-axis and y-axis directions respectively;

[0079] It should be noted that the scale parameter controls the degree of dilation of the wavelet basis function in the image. Generally, a smaller scale value represents high-frequency information, while a larger scale value represents low-frequency information. When the scale parameter is small, the waveform of the wavelet basis function is relatively compact and can capture more detailed information. The translation parameter controls the position of the wavelet basis function in the image, and it determines the specific position of the center point of the wavelet basis function on the image. By changing the translation parameter, the wavelet basis function can be translated to different regions of the image for local analysis at different positions.

[0080] Perform hard thresholding on the obtained wavelet transform coefficients Wi(a, b) to obtain the wavelet transform coefficients after thresholding. ;

[0081] For the wavelet transform coefficients after thresholding , and then restore it to the denoised version of the original image through inverse wavelet transform to obtain the pixel color values of wavelet transform denoising. . The inverse wavelet transform formula is:

[0082] ; where is the inverse wavelet transform function.

[0083] Arrange the pixel color values of each biomedical image Pi after wavelet transform denoising according to the pixel coordinates (x, y), and compress them to the preset image resolution w×h to generate the pixel color value matrix Ui = U1, U2,..., Un of the biomedical image Pi.

[0084] The specific process of normalization is as follows:

[0085] For the chemical and physical property data of the biomedical material to be implanted and the physiological characteristic data after data cleaning, use the standardization formula to normalize the data so that its mean is 0 and the standard deviation is 1, and at the same time convert all numerical data into dimensionless parameters. The formula is as follows:

[0086] ; where X is the physiological characteristic data before normalization, including the heart rate x11, blood pressure x12, body temperature x13, blood oxygen saturation x14, and respiratory rate x15 after data cleaning; Z(X) is the physiological characteristic data after normalization. Among them and are the mean and standard deviation of the physiological characteristic data X respectively.

[0087] The deep learning and prediction module includes a CNN (Convolutional Neural Network) for medical image feature extraction and an LSTM (Long Short-Term Memory Network) for numerical data and associated feature extraction.

[0088] Perform data fusion on all the data output by the data preprocessing module, and predict the effect of biomaterial implantation through a deep learning model. A hybrid model combining a convolutional neural network (CNN) and a long short-term memory network (LSTM) is used to process the joint prediction task of physiological features, historical surgical data, medical image data, and chemical and physical property data of the biomaterial to be implanted, which are based on cleaning and normalization processes.

[0089] Perform data fusion, collect physiological features, historical surgical data, medical image data, and chemical and physical property data of the biomaterial to be implanted after cleaning and normalization processes, and generate a multi-dimensional data matrix H. Among them, x11, x12,..., x35 are all physiological features, historical surgical data, and chemical and physical property data of the biomaterial to be implanted after cleaning and normalization processes; among them, U1, U2,..., Un are pixel color value matrices of biomedical images after wavelet transform denoising.

[0090] CNN uses a convolutional neural network to extract spatial features of medical image data. CNN gradually extracts high-level features of the image through multiple convolutional layers, pooling layers, and fully connected layers, and finally outputs an image feature vector. The formula for the convolutional layer is:

[0091] ; Yi is the image feature vector extracted from the medical image Pi; among them is the kernel function of the convolutional layer, * is the convolution operator, is the bias coefficient.

[0092] The LSTM long short-term memory network is used to extract temporal feature information from physiological features, historical surgical data, medical image data, and chemical and physical property data of the biomaterial to be implanted, as well as the relationship between medical image data and physiological features, historical surgical data, medical image data, and chemical and physical property data of the biomaterial to be implanted. The LSTM long short-term memory network includes an input gate, a forget gate, a memory gate, an output gate, and a candidate memory gate.

[0093] Among them, the input gate is responsible for receiving the multi-dimensional data matrix H and determining which information in the current input will be written into the memory unit through convolution operation. The mathematical formula is:

[0094] ;

[0095] Among them, i(t) is the output of the input gate, indicating the influence of the current input on the memory unit, ranging from 0 to 1; among them, σ is the activation function; among them is the weight matrix of the input gate; is the bias term of the input gate;

[0096] Among them, the forget gate determines which information in the memory cell at the previous moment is forgotten, and the mathematical formula is:

[0097] ;

[0098] where σ is the activation function; where is the output of the forget gate, representing the proportion of forgotten information, ranging from 0 to 1; where is the weight matrix of the forget gate; is the bias term of the forget gate; where is the hidden state at the previous moment, and the specific value is calculated by the output gate; where is the combination of the multi-dimensional data matrix H input at the current moment t and the hidden state information at the previous moment t-1;

[0099] The memory cell combines the outputs of the forget gate and the input gate to update the memory cell, and the mathematical formula is:

[0100] ;

[0101] where C(t) is the memory cell at the current moment; where C(t-1) is the memory cell at the previous moment; where f(t)×C(t-1) is the retained part of the memory cell at the previous moment; where is the updated part input at the current moment; where tanh is the hyperbolic tangent function, and the output value is between -1 and 1, and the formula is ; where is the weight matrix of the memory cell; where is the bias term for memory cell update;

[0102] The output gate calculates the hidden state at the current moment based on the memory cell and the input, and outputs the final result. The mathematical formula is:

[0103] ; where is the weight matrix of the output gate; where is the bias term of the output gate; where h(t) is the hidden state at the current moment and is also the final output of the LSTM neural network model; where o(t) is the output of the output gate, which determines which information in the memory cell C(t) is output.

[0104] Furthermore, the image feature vector Yi extracted by the CNN is combined with h(t) output after being processed by the LSTM network for joint prediction. These different types of data are fused through the fully connected layer to generate the final prediction result.

[0105] Further, the combined feature vector processed by the fully connected layer is numerically normalized through an activation function to obtain a prediction result. The prediction result can be a regression value Mark with a preset numerical range, and the numerical range is from 0 to 100. The higher the value, the better the prediction result of the implantation effect.

[0106] The comparison database is responsible for storing all the prediction results of the deep learning model, that is, the regression value Mark. For each implantation target, a data set including historical patient data, surgical data, implant material data, and the regression value Mark is generated. The data in this database has been strictly screened and classified, and can provide a reference basis for the system to compare with the prediction results of the current patient.

[0107] The signal matching module converts the output result of the deep learning model into a signal suitable for performing operations, and selects different operations according to different prediction results.

[0108] If it is recognized that the regression value Mark is greater than the first preset threshold Mark1, a high matching feedback signal is generated: indicating that the predicted implantation effect is relatively ideal, and it is recommended to continue the current implantation plan;

[0109] If it is recognized that the regression value Mark is greater than the second preset threshold Mark2 and less than or equal to the first preset threshold Mark1, it means that there may be potential risks or complications in the predicted implantation effect, and it is recommended to conduct further examinations or modify the treatment plan, and it is recommended to re-evaluate the current implantation plan;

[0110] If it is recognized that the regression value Mark is less than or equal to the second preset threshold Mark2, it means that the prediction result is too different from the historical data and there are high risks. A risk warning signal is generated to remind medical staff to take emergency intervention measures.

[0111] It should be understood that the terms "including" and "comprising" used in the specification and claims of this disclosure indicate the presence of the described features, wholes, steps, operations, elements, and / or components, but do not exclude the existence or addition of one or more other features, wholes, steps, operations, elements, components, and / or their combinations.

[0112] It should also be understood that the terms used in this disclosure specification are only for the purpose of describing specific embodiments and are not intended to limit this disclosure. As used in this disclosure specification and claims, unless the context clearly indicates otherwise, the singular forms "a", "an", and "the" are intended to include the plural forms. It should be further understood that the term "and / or" used in this disclosure specification and claims refers to any combination and all possible combinations of one or more of the associated listed items, and includes these combinations;

[0113] The preferred embodiments of the present invention disclosed above are only used to help illustrate the present invention. The preferred embodiments do not describe all the details in detail, nor do they limit the invention to the specific embodiments described. Obviously, many modifications and variations can be made according to the content of this specification. These embodiments are selected and specifically described in this specification in order to better explain the principles and practical applications of the present invention, so that those skilled in the art can well understand and utilize the present invention. The present invention is only limited by the claims and their full scope and equivalents.

Claims

1. A biomaterial implantation effect prediction system based on deep learning, comprising a data preprocessing module, a deep learning and prediction module and a signal matching module, characterized in that ； The data preprocessing module is responsible for cleaning and normalizing the collected data, converting parameters of different dimensions into dimensionless parameters with similar numerical ranges, unifying them to the same scale, and providing high-quality input data for subsequent deep learning models; The deep learning and prediction module includes CNN convolutional neural network for medical image feature extraction and LSTM long short-term memory network for numerical data and associated feature extraction; All data output by the data preprocessing module are fused, and the implantation effect of biomaterials is predicted through a deep learning model. A hybrid model combining a convolutional neural network and an LSTM long short-term memory network is used to process the joint prediction task based on the physiological characteristics of cleaning and normalization, historical surgical data, medical imaging data, and the chemical and physical properties of the biomedical materials to be implanted; The output results of CNN using convolutional neural network and LSTM long short-term memory network are fully connected and normalized to obtain the regression value representing the prediction result; The signal matching module converts the output results of the deep learning model into signals suitable for performing operations and selects different operations based on different prediction results.

2. A biomaterial implantation effect prediction system based on deep learning according to claim 1, characterized in that: It also includes a data acquisition module and a comparison database; The data acquisition module is responsible for collecting biomedical data of the target patient for biomaterial implantation, including physiological characteristics, historical surgical data and medical imaging data of the target patient; collecting chemical and physical property data of the biomedical material to be implanted, and preparing data for subsequent effect prediction modeling; the collected data specifically includes physiological characteristic data, historical surgical data, medical imaging data and chemical and physical property data of biomedical materials; The comparison database is responsible for storing all prediction results of the deep learning model, namely the regression value Mark. For each implant target, a data set containing historical patient data, surgical data, implant material data and regression value Mark is generated. The data in this database has been strictly screened and classified, which can provide a reference for the system for comparison with the prediction results of the current patient.

3. The biomaterial implantation effect prediction system based on deep learning according to claim 1, characterized in that: The specific process of data cleaning is as follows: For numerical parameters: physiological characteristic data, remove invalid data, duplicate data and abnormal values, and remove abnormal values ​​that exceed the preset threshold; For the collected medical image data, the wavelet transform denoising algorithm is used to suppress noise to improve the image quality. The specific wavelet transform denoising algorithm is as follows: Perform wavelet transform on the color value fi(x, y) of the pixel in the original image Pi: Where fi (x, y) is the color value of the pixel in each medical image Pi before filtering, where (x, y) is the pixel coordinate; Wi (a, b) is the wavelet transform coefficient corresponding to the pixel in each medical image Pi obtained after filtering; where a and b are preset scale parameters and translation parameters respectively, and the preset (a, b) combination is different wavelet transform coefficients Wi (a, b) obtained by controlling the different extraction accuracies of high-frequency noise and low-frequency noise; where is a wavelet function; and are the scale parameters of the wavelet transform in the x-axis and y-axis directions respectively; and are the translation parameters of the wavelet transform in the x-axis and y-axis directions respectively; Perform hard threshold processing on the obtained wavelet transform coefficients Wi (a, b) to obtain the wavelet transform coefficients after threshold processing ; The wavelet transform coefficients after threshold processing , and then restore the denoised version of the original image through inverse wavelet transform to obtain the pixel color value denoised by wavelet transform ; The inverse wavelet transform formula is: ;in is the inverse wavelet transform function; The pixel color values ​​of each biomedical image Pi are denoised by wavelet transform Arranged according to pixel coordinates (x, y) and compressed to a preset image resolution w×h, a pixel color value matrix Ui=U1, U2, ..., Un of the biomedical image Pi is generated.

4. The biomaterial implantation effect prediction system based on deep learning according to claim 1, characterized in that: The specific process of normalization is as follows: For the chemical and physical property data of the biomedical materials to be implanted and the physiological characteristic data after data cleaning, the data are normalized using a standardized formula to make the mean value 0 and the standard deviation 1, and all numerical data are converted into dimensionless parameters; the formula is as follows: ; X is the physiological characteristic data before normalization, including heart rate x11, blood pressure x12, body temperature x13, blood oxygen saturation x14 and respiratory rate x15 after data cleaning; Z (X) is the physiological characteristic data after normalization; in and are the mean and standard deviation of the physiological characteristic data X respectively.

5. The biomaterial implantation effect prediction system based on deep learning according to claim 1, characterized in that: The specific process of data fusion is as follows: Collect the physiological characteristics of the cleaning and normalization processing, historical surgical data, medical imaging data and the chemical and physical properties data of the biomedical materials to be implanted, and generate a multidimensional data matrix H, Among them, x11, x12, ..., x35 are all physiological characteristics, historical surgical data and chemical and physical property data of biomedical materials to be implanted after cleaning and normalization; among them, U1, U2, ..., Un are the pixel color value matrices of biomedical images after wavelet transform denoising.

6. The biomaterial implantation effect prediction system based on deep learning according to claim 1, characterized in that: CNN for medical image feature extraction uses convolutional neural network specifically as follows: CNN uses convolutional neural networks to extract spatial features of medical imaging data. CNN gradually extracts high-level features of the image through multiple convolutional layers, pooling layers, and fully connected layers, and finally outputs an image feature vector. The formula of the convolutional layer is: ; Yi is the image feature vector extracted from the medical image Pi; is the kernel function of the convolution layer, * is the convolution operator, is the bias coefficient.

7. The biomaterial implantation effect prediction system based on deep learning according to claim 1, characterized in that: For numerical data and related features, the LSTM long short-term memory network is extracted as follows: The LSTM long short-term memory network is used to extract temporal feature information from physiological characteristics, historical surgical data, medical imaging data, and chemical and physical property data of biomedical materials to be implanted, as well as the connection between medical imaging data and physiological characteristics, historical surgical data, medical imaging data, and chemical and physical property data of biomedical materials to be implanted; the LSTM long short-term memory network includes an input gate, a forget gate, a memory gate, an output gate, and a candidate memory gate; Among them, the input gate is responsible for receiving the multidimensional data matrix H, and determines which information in the current input will be written into the memory unit through convolution operation. The mathematical formula is: ； Where i(t) is the output of the input gate, which indicates the influence of the current input on the memory unit, ranging from 0 to 1; where σ is the activation function; where is the weight matrix of the input gate; is the bias term of the input gate; Among them, the forget gate determines which information in the memory unit of the previous moment is forgotten. The mathematical formula is: ； Where σ is the activation function; is the output of the forget gate, indicating the proportion of information forgotten, ranging from 0 to 1; is the weight matrix of the forget gate; is the bias term of the forget gate; is the hidden state of the previous moment, and the specific value is calculated by the output gate; The multidimensional data matrix H input at the current time t and the hidden state information at the previous time t-1 The combination of The memory unit combines the output of the forget gate and the input gate to update the memory unit. The mathematical formula is: ； Where C(t) is the memory unit at the current moment; Where C(t-1) is the memory unit at the previous moment; Where f(t)×C(t-1) is the retained part of the memory unit at the previous moment; Where is the updated part of the current input; tanh is the hyperbolic tangent function, the output value is between -1 and 1, and the formula is ;in is the weight matrix of the memory unit; Bias term for memory unit update; The output gate calculates the hidden state at the current moment based on the memory unit and input, and outputs the final result. The mathematical formula is: ;in is the weight matrix of the output gate; is the bias term of the output gate; h(t) is the hidden state at the current moment, which is also the final output of the LSTM neural network model; o(t) is the output of the output gate, which determines which information in the memory unit C(t) is output.

8. A biomaterial implantation effect prediction system based on deep learning according to claim 6 or 7, characterized in that: for: Combine the image feature vector Yi extracted by CNN with h(t) output by LSTM network for joint prediction; fuse these different types of data through the fully connected layer to generate the final prediction result; The joint feature vector processed by the fully connected layer is numerically normalized through an activation function to obtain a prediction result; the prediction result can be a regression value Mark with a preset numerical range, ranging from 0 to 100, and the higher the value, the better the prediction result of the implantation effect; The comparison database is responsible for storing all prediction results of the deep learning model, namely the regression value Mark. For each implant target, a data set containing historical patient data, surgical data, implant material data and regression value Mark is generated. The data in this database has been strictly screened and classified, which can provide a reference for the system for comparison with the prediction results of the current patient.

9. The biomaterial implantation effect prediction system based on deep learning according to claim 1, characterized in that: The specific process of selecting different operations according to different prediction results is as follows: If the identified regression value Mark is greater than the first preset threshold Mark1, a high matching feedback signal is generated: indicating that the predicted implantation effect is relatively ideal, and it is recommended to continue the current implantation plan; If the identified regression value Mark is greater than the second preset threshold Mark2 and less than or equal to the first preset threshold Mark1, it indicates that the predicted implantation effect may have potential risks or complications, and further examination or modification of the treatment plan is recommended, and re-evaluation of the current implantation plan is recommended; If the regression value Mark is identified to be less than or equal to the second preset threshold Mark2, it means that the prediction result is too different from the historical data and there is a high risk. A risk warning signal is generated to remind medical staff to take emergency intervention measures.

10. The biomaterial implantation effect prediction system based on deep learning according to claim 2, characterized in that: The collected data specifically include: Physiological characteristic data, including heart rate x11, blood pressure x12, body temperature x13, blood oxygen saturation x14 and respiratory rate x15; Historical surgery data: surgery type code symbol X21, surgery date X22, postoperative complication code symbol X23; The surgical type code represents different preset surgical types through specific numerical values, including orthopedic surgery, cardiovascular surgery, tumor surgery, neurosurgery, transplant surgery, ophthalmic surgery, thoracic surgery, gastrointestinal surgery, urological surgery and gynecological surgery; Among them, the postoperative complication code represents different preset postoperative complication types through specific numerical values, including postoperative infection, hematoma or bleeding, deep vein thrombosis, pulmonary complications, cardiovascular complications, renal failure, nerve damage, poor wound healing, nausea and vomiting, and drug allergic reaction; Medical imaging data, medical images Pi at the implantation location acquired by medical imaging equipment based on MRI, CT and X-ray; wherein i is the medical image number, i=1,2,3,...,n; wherein n is the total number of medical images; Chemical and physical property data of biomedical materials, including mechanical properties, chemical stability and cell affinity data of biomaterials; specifically including elastic modulus x31, tensile strength x32, half-maximum inhibitory concentration x33, degradation time x34 and cell attachment rate x35.

Images