Postoperative intelligent monitoring system for gastrointestinal tumor patient

Abstract

The invention discloses a postoperative intelligent monitoring system for gastrointestinal tumor patients, particularly relates to the technical field of postoperative monitoring, and is used for solving the problem that a nonlinear mode in data cannot be effectively identified when facing high-dimensional and complex wound image data in the prior art. By deploying an ambient light sensor and an automatic exposure adjustment technology, it is ensured that a wound image can keep high quality under various illumination conditions, rich local and global features are extracted from the wound image, and a detailed data basis is provided for deep learning analysis; the chaos theory analysis module evaluates the complexity of the features by using a chaos theory model, and effectively screens out data which is high in complexity and needs to be deeply analyzed; the convolutional network classification module uses a convolutional neural network to accurately identify and classify different healing stages and infection signs of the wound; and the model calibration feedback module continuously calibrates and optimizes the model through historical healing data, so that the self-updating capability and the long-term operation accuracy of the system are guaranteed.

Description

Technical Field

[0001] The present invention relates to the technical field of postoperative monitoring, and more specifically, to an intelligent monitoring system for gastrointestinal tumor patients after surgery. Background Art

[0002] In the field of postoperative monitoring of gastrointestinal tumor patients, traditional wound monitoring methods mainly rely on regular visual inspections by medical staff and self-reporting by patients. This method relies on subjective judgment, is easily affected by human factors, and cannot achieve continuous monitoring. With the development of medical information technology, although digital image processing technology has been introduced to assist wound monitoring, existing systems usually lack efficient data processing and analysis capabilities, making it difficult to accurately evaluate the healing status and development trend of wounds.

[0003] Existing technologies for processing wound images: The acquisition and processing of wound images are not sufficient to provide continuous and accurate monitoring. When faced with high-dimensional and complex wound image data, traditional methods are usually unable to effectively identify nonlinear patterns in the data, resulting in inaccurate model predictions and low reliability.

[0004] In order to solve the above problems, a technical solution is now provided. Summary of the invention

[0005] In order to overcome the above-mentioned defects of the prior art, an embodiment of the present invention provides an intelligent monitoring system for gastrointestinal tumor patients after surgery to solve the problems raised in the above-mentioned background technology.

[0006] To achieve the above object, the present invention provides the following technical solutions:

[0007] An intelligent postoperative monitoring system for gastrointestinal tumor patients, comprising an ambient light adjustment module, a feature vector extraction module, a chaos theory analysis module, a convolutional network classification module, and a model calibration feedback module;

[0008] Ambient light adjustment module: deploys ambient light sensors and automatic exposure adjustment technology, and uses machine learning algorithms to adjust camera parameters in real time based on ambient light data;

[0009] Feature vector extraction module: extracts comprehensive local and global features from wound images through advanced image processing algorithms to generate a multi-dimensional feature vector containing texture, color and shape information;

[0010] Chaos theory analysis module: uses the chaos theory model to analyze the complexity of the extracted multi-dimensional feature vectors and determine whether the wound image data is suitable for the next step of deep learning analysis;

[0011] Convolutional network classification module: Use convolutional neural networks to extract features from wound images and classify wound status; train the model to identify various stages of wound healing and possible signs of infection;

[0012] Model calibration feedback module: Use historical healing data to calibrate the model to ensure the accuracy and reliability of the analysis results. Low confidence results will trigger manual review.

[0013] In a preferred embodiment, an ambient light sensor and automatic exposure adjustment technology are deployed, and a machine learning algorithm is used to adjust the camera parameters in real time according to the ambient light data, specifically:

[0014] Install multiple ambient light sensors on the image acquisition device to capture lighting conditions in all directions;

[0015] The sensor continuously collects light intensity data: pre-processes the light data;

[0016] Load the pre-trained machine learning model to parse the illumination data: Load the pre-trained machine learning model M light , M light The model formula is as follows: Among them, P exp is the recommended exposure parameter, f RNN represents a recurrent neural network function based on lighting data, is the light intensity value of the i-th sensor after preprocessing;

[0017] According to P exp , the system dynamically adjusts the camera’s exposure time and aperture size in real time.

[0018] In a preferred embodiment, comprehensive local and global features are extracted from the wound image by advanced image processing algorithms to generate a multidimensional feature vector containing texture, color and shape information, specifically:

[0019] Preprocessing wound images;

[0020] The local binary pattern algorithm is used for local texture extraction, and its formula is as follows: Among them, F LBP (x, y) represents the local binary pattern eigenvalue at the position (x, y), I p Represents the gray value of the pth pixel in the pixel neighborhood, I c represents the gray value of the central pixel, and P represents the number of pixels contained in the neighborhood;

[0021] S(z) is the sign function, defined as: Where z = I p -I c ;

[0022] The color distribution of the wound area is modeled using a color histogram, and the formula is as follows: Among them, F color (k) is the normalized color histogram value on color channel k, n k is the number of pixels on color channel k, and N is the total number of all pixels in the image;

[0023] Use Fourier transform to extract the shape features of the wound; perform discrete Fourier transform on the wound contour;

[0024] Through Fourier analysis, the global characteristics of the wound shape are extracted;

[0025] Form a multidimensional feature vector V feat , which is expressed as follows: V feat =[F LBP (x,y), F color (k), F shape (u)]; where V feat is a multidimensional feature vector.

[0026] In a preferred embodiment, the complexity of the extracted multi-dimensional feature vector is analyzed using a chaos theory model to determine whether the wound image data is suitable for the next step of deep learning analysis, specifically:

[0027] The extracted multidimensional feature vector V feat Standardize the process;

[0028] The standardized multidimensional feature vector V is transformed into std Mapping into high-dimensional phase space;

[0029] After reconstructing the phase space, the system is calculated by calculating the maximum Lyapunov exponent Λ max To measure the degree of chaos in the data;

[0030] If Λ max >γ, then the wound image data is judged to be suitable for the next step of deep learning analysis; if Λ max ≤γ, then the wound image data is judged to be unsuitable for the next step of deep learning analysis;

[0031] γ is the preset threshold.

[0032] In a preferred embodiment, the maximum Lyapunov exponent Λ max The calculation formula is: Among them, Λ max is the maximum Lyapunov exponent, X recon (t j ) is the time point t in the phase spacej A vector of time steps where r is the number of time steps used to compute the Lyapunov exponent and j is the index of the time step.

[0033] In a preferred embodiment, a convolutional neural network is used to extract features from wound images and classify wound states; the model is trained to identify various stages of wound healing and possible signs of infection, specifically:

[0034] The input wound image is normalized; the local features are extracted using a convolutional layer;

[0035] The convolution layer extracts local features through sliding windows to form multiple feature maps;

[0036] The pooling layer performs dimensionality reduction through the maximum pooling operation;

[0037] The fully connected layer integrates the output features of the pooling layer and completes the classification of wound status through weighted summation;

[0038] The cross entropy loss function is used to measure the accuracy of model prediction. The cross entropy loss function formula is as follows: Among them, L cross is the cross entropy loss value, is the true label of class v1, is the probability of class v1 predicted by the model, where V1 is the total number of classes.

[0039] In a preferred embodiment, the model is calibrated using historical healing data to ensure the accuracy and reliability of the analysis results. Low confidence results will trigger manual review, including:

[0040] Extract historical healing data of postoperative patients from the database, covering different healing stages and related characteristics;

[0041] Through real-time analysis, the deviation between model output and historical data is evaluated and the forecast error is calculated;

[0042] Based on the model's prediction results, calculate the confidence score to evaluate the model's credibility for the current data;

[0043] Automatically adjust model parameters to optimize prediction accuracy based on historical data and error analysis;

[0044] When the confidence level is lower than the set threshold, the system automatically triggers manual review.

[0045] The technical effects and advantages of the intelligent monitoring system for patients with gastrointestinal tumors after surgery provided by the present invention are as follows:

[0046] The ambient light adjustment module ensures that the wound image can maintain high quality under various lighting conditions by deploying ambient light sensors and automatic exposure adjustment technology, making the image data more suitable for subsequent processing. The feature vector extraction module can extract rich local and global features from the wound image, such as texture, color and shape information, providing a detailed data basis for deep learning analysis. The chaos theory analysis module uses chaos theory models to evaluate the complexity of these features, effectively screening out those data with high complexity and more in-depth analysis, thereby optimizing resource allocation and improving processing efficiency. The convolutional network classification module uses convolutional neural networks to accurately identify and classify different healing stages and signs of infection of wounds, greatly improving the accuracy and timeliness of diagnosis. Finally, the model calibration feedback module continuously calibrates and optimizes the model through historical healing data, ensuring the system's self-update ability and long-term operation accuracy. When the model confidence is low, manual review is automatically triggered to ensure the reliability of the monitoring results. The design of the entire system not only improves the monitoring quality of patients' postoperative recovery and reduces the misuse and overuse of medical resources, but also helps medical staff to adjust treatment plans in a timely manner, effectively improving patients' treatment effects and satisfaction. BRIEF DESCRIPTION OF THE DRAWINGS

[0047] Figure 1 The present invention is a schematic diagram of the structure of an intelligent monitoring system for gastrointestinal tumor patients after surgery. DETAILED DESCRIPTION

[0048] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0049] Example

[0050] Figure 1 A structural schematic diagram of a postoperative intelligent monitoring system for gastrointestinal tumor patients of the present invention is given, which includes an ambient light adjustment module, a feature vector extraction module, a chaos theory analysis module, a convolutional network classification module and a model calibration feedback module.

[0051] Ambient light adjustment module: Deploys ambient light sensors and automatic exposure adjustment technology, and uses machine learning algorithms to adjust camera parameters in real time based on ambient light data.

[0052] Feature vector extraction module: Extracts comprehensive local and global features from wound images through advanced image processing algorithms to generate a multi-dimensional feature vector containing texture, color and shape information.

[0053] Chaos theory analysis module: Use the chaos theory model to analyze the complexity of the extracted multi-dimensional feature vectors and determine whether the wound image data is suitable for the next step of deep learning analysis.

[0054] Convolutional network classification module: Use convolutional neural networks to extract features from wound images and classify wound status; train the model to identify the various stages of wound healing and possible signs of infection.

[0055] Model calibration feedback module: Use historical healing data to calibrate the model to ensure the accuracy and reliability of the analysis results. Low confidence results will trigger manual review.

[0056] Deploy ambient light sensors and automatic exposure adjustment technology, and use machine learning algorithms to adjust camera parameters in real time based on ambient light data. Specifically:

[0057] Install multiple ambient light sensors on the image acquisition device to capture all lighting conditions:

[0058] Multiple ambient light sensors are installed, each capturing the lighting conditions of the surgical area and surrounding environment from a different angle. The position of the sensors is precisely designed to ensure that the coverage includes direct light sources, reflected light sources, and equipment shadow areas to fully capture various light information.

[0059] The sensor continuously collects light intensity data, providing real-time input to the algorithm:

[0060] The ambient light sensor continuously collects real-time light intensity data, which is quantified in terms of light intensity (unit: lux). To ensure the stability and accuracy of the data, the following formula is used to preprocess the light data: Among them, I adj is the light intensity value after preprocessing, I raw is the raw light intensity value detected by the ambient light sensor, A calib is the calibration factor (a fixed value adjusted according to environmental conditions to compensate for differences between sensors), n is the number of ambient light sensors, i is the number of ambient light sensors, is the raw light intensity value detected by the i-th ambient light sensor.

[0061] By normalizing and calibrating data from multiple sensors, we ensure that the data received by the system can reflect the actual ambient lighting conditions and avoid errors caused by misalignment of individual sensors.

[0062] Load a pre-trained machine learning model for parsing lighting data:

[0063] Load the pre-trained machine learning model M lightThe model is trained based on a large amount of lighting environment data and is used to analyze the impact of lighting changes on image quality. The model structure adopts the recurrent neural network (RNN) in deep learning. Its input is the light intensity data of each sensor, and its output is the corresponding exposure parameter adjustment value.

[0064] M light The model formula is as follows:

[0065] P exp : Recommended exposure parameters for adjusting the camera.

[0066] f RNN : Represents a recurrent neural network function based on lighting data.

[0067] The preprocessed light intensity value of the i-th sensor.

[0068] Through the application of this model, the system can intelligently predict the optimal exposure parameters based on the complex changes in ambient light, ensuring that the brightness and clarity of the image remain stable under various lighting conditions.

[0069] Dynamically adjust the camera's exposure and focus based on the algorithm analysis results:

[0070] According to the exposure parameter P output by the machine learning model exp , the system dynamically adjusts the exposure time and aperture size of the camera in real time, and the adjustment formula is:

[0071] t exp : Exposure time, which indicates how long the camera shutter is open.

[0072] A ap : Aperture size, indicating the opening width of the camera aperture.

[0073] k1 and k2: are the system preset calibration coefficients, which are debugged based on experimental data to ensure the stability of image brightness under different lighting conditions.

[0074] Through this step, the camera can automatically adapt to the current lighting conditions and adjust the shooting parameters in real time, thereby ensuring that the captured surgical wound images still have sufficient details and clarity when the light intensity changes.

[0075] Multi-point data is collected through ambient light sensors, and the light intensity is intelligently analyzed in combination with a recursive neural network model. The camera parameters are adjusted in real time through precise calculations. Through normalized preprocessing of light data and in-depth application of machine learning algorithms, the system can respond efficiently to complex changes in ambient light and avoid overexposure or underexposure problems in existing technologies.

[0076] Advanced image processing algorithms are used to extract comprehensive local and global features from wound images to generate a multidimensional feature vector containing texture, color and shape information, specifically:

[0077] The wound images are preprocessed to ensure the consistency and stability of the image data. The preprocessing process includes the following steps:

[0078] Use a denoising algorithm based on median filtering to reduce noise in the image caused by shooting conditions or hardware equipment. Use histogram equalization technology to optimize image contrast and improve the visibility of local features.

[0079] In the process of local feature extraction, the edge and local texture of the wound are analyzed to ensure that important details of the wound healing process are captured. The local binary pattern (LBP) algorithm is used for local texture extraction, and its formula is as follows: Among them, F LBP (x, y) represents the local binary pattern eigenvalue at the position (x, y), I p Represents the gray value of the pth pixel in the pixel neighborhood, I c represents the gray value of the central pixel, and P represents the number of pixels contained in the neighborhood.

[0080] S(z): is the sign function, defined as: Where z = I p -I c .

[0081] The key point of global feature extraction is to capture the color distribution and shape characteristics of the entire wound area. To achieve this goal, a combination of color histogram and Fourier shape analysis is used.

[0082] The color distribution of the wound area is modeled using a color histogram, and the formula is as follows: Among them, F color (k) is the normalized color histogram value on color channel k, n k is the number of pixels on color channel k, and N is the total number of all pixels in the image.

[0083] The shape features of the wound are extracted using Fourier transform. The wound contour is subjected to discrete Fourier transform (DFT), and the formula is as follows: Among them, F shape (u) is the shape feature in the Fourier frequency domain, f(x, y) represents the grayscale value at the position (x, y), W and H are the width and height of the wound image, respectively, and u and υ are the spatial frequency components in the horizontal direction and the vertical direction, respectively.

[0084] Through Fourier analysis, global features of wound shape, such as symmetry and edge curvature, are extracted.

[0085] Combining the above local and global features, a multidimensional feature vector V is formed feat , which is expressed as follows: V feat =[F LBP (x,y), F color (k), F shape (u)]; where V feat It is a multidimensional feature vector that contains comprehensive information of local texture, global color and shape features.

[0086] By combining local binary patterns, color histograms and Fourier transforms, features of wound images are extracted from local and global perspectives, ensuring that the generated multi-dimensional feature vector can fully describe the healing status of the wound. These features are extracted through complex image processing algorithms, which improves the accuracy and reliability of the system's assessment of wound status.

[0087] The complexity of the extracted multi-dimensional feature vectors is analyzed using the chaos theory model to determine whether the wound image data is suitable for the next step of deep learning analysis. Specifically:

[0088] The extracted multidimensional feature vector V feat Standardization is performed to ensure that features of different dimensions are compared on the same scale. The standardization formula is as follows: in, is the value of the qth feature dimension after standardization, is the value of the original qth feature dimension, μ q is the mean of the qth feature dimension, σ q is the standard deviation of the qth feature dimension.

[0089] Standardization can eliminate the dimensional differences between different feature dimensions and ensure that subsequent analysis is more accurate.

[0090] The standardized multidimensional feature vector V is transformed into std Mapped into a high-dimensional phase space in order to construct the phase space trajectory, the formula is as follows: recon (t)=[V std (t), V std (t+Δ), V std (t+2Δ), ..., V std (t+(d-1)Δ)].

[0091] X recon (t): The reconstructed phase space vector at time point t.

[0092] Δ is the time delay, and the optimal delay value is determined by the autocorrelation function or the mutual information method.

[0093] d: Embedding dimension, which represents the dimension of the phase space and is usually determined by the pseudo nearest neighbor method.

[0094] After reconstructing the phase space, the system is calculated by calculating the maximum Lyapunov exponent Λ max To measure the degree of chaos of the data, the formula is as follows:

[0095]

[0096] Λ max : Maximum Lyapunov exponent, used to evaluate the chaos of the system.

[0097] X recon (t j ): time point t in phase space j The vector on .

[0098] |·|: represents the Euclidean norm of a vector.

[0099] r: The number of time steps used to calculate the Lyapunov exponent, representing the total time step length.

[0100] j: represents the index of the time step, which is usually used to traverse the entire time series and is used to calculate the distance between adjacent vectors in phase space as part of the Lyapunov exponent calculation.

[0101] Maximum Lyapunov exponent Λ max A positive value of indicates that the system exhibits chaotic behavior and the data has complex dynamic characteristics.

[0102] According to the maximum Lyapunov exponent Λ max The value of is used to evaluate the complexity of the multidimensional feature vector and make a decision on whether to proceed to the next step of analysis:

[0103] If Λ max >γ (where γ is the preset threshold), it indicates that the nonlinear characteristics of the data are complex enough and suitable for further deep learning analysis. At this time, the wound image data is determined to be suitable for the next step of deep learning analysis.

[0104] If Λ max ≤γ, the dynamic characteristics of the data are relatively simple and may have limited contribution to the deep learning model, requiring further adjustment or re-collection. In this case, the wound image data is determined to be unsuitable for the next step of deep learning analysis.

[0105] By performing complexity analysis on the multidimensional feature vectors of wound image data, the nonlinear and dynamic characteristics of the data can be accurately evaluated. Through phase space reconstruction and calculation of Lyapunov exponents, the system can identify image data with high complexity, thereby ensuring that these data can play the greatest role in the deep learning model. This process can effectively filter out image data that is not complex enough and avoid interference from invalid data, thereby improving the accuracy and efficiency of deep learning analysis. Secondly, the introduction of chaos theory models enables the system to flexibly adapt to different environments and individual differences among patients. For the wound healing conditions of different patients, the module can determine whether the data is suitable for further learning and prediction through complexity analysis. This feature provides strong support for personalized medicine. By improving the accuracy and effectiveness of data screening, not only the overall intelligence level of the system is improved, but also unnecessary data processing is greatly reduced, thereby improving the timeliness and accuracy of postoperative monitoring.

[0106] Convolutional neural networks are used to extract features from wound images and classify wound status; the model is trained to identify the various stages of wound healing and possible signs of infection, specifically:

[0107] The input wound images were standardized to ensure that the image data were processed within a uniform range to facilitate further analysis by the network.

[0108] The convolution layer is used to extract local features, such as edges, textures, and other information. The specific formula for the convolution operation is as follows:

[0109] R conv (p2, q2): Output feature map at position (p2, q2) after the convolution operation.

[0110] P filter (s1, t1): convolution kernel of size S1×T1, used to extract local features.

[0111] Q std (p2+s1, q2+t1): Normalized image pixel value, used as input to the convolution kernel.

[0112] The convolution layer extracts local features through a sliding window to form multiple feature maps for subsequent classification.

[0113] In order to reduce the dimension of the feature map and retain the most critical features, the pooling layer performs dimensionality reduction through the maximum pooling operation. The specific pooling formula is as follows:

[0114] R pool (p3, q3): Output value of the pooling layer at position (p3, q3).

[0115] S2×T2: Pooling window size, used to select local maximum values.

[0116] The pooling layer can effectively reduce the redundant information output by the convolutional layer while maintaining the representativeness of the features and improving the computational efficiency of the network.

[0117] The fully connected layer integrates the output features of the pooling layer and completes the classification of wound status through weighted summation. The specific formula is as follows:

[0118] R fc (p4): The output value of the fully connected layer at the p4th classification node.

[0119] The weight of the fully connected layer acts on the u1-th output value of the pooling layer.

[0120] R pool (u1): The u1-th output feature of the pooling layer.

[0121] The bias value of the fully connected layer adjusts the final output.

[0122] U1: represents the number of output features of the pooling layer, that is, the total number of features generated by the pooling layer after the dimensionality reduction operation.

[0123] The cross entropy loss function is used to measure the accuracy of model prediction. The specific cross entropy loss function formula is as follows:

[0124] L cross : Cross entropy loss value, which measures the difference between the model prediction and the true label.

[0125] The true label of class v1, represented as a one-hot encoding.

[0126] The probability of class v1 predicted by the model.

[0127] V1 is the total number of categories.

[0128] p1, p2, p3, and p4 represent the pixel position in the image or the pixel coordinates in the feature map, respectively, and are used to identify the positions at different processing stages. q1, q2, and q3 represent another dimension in the pixel coordinates, which are used to represent the positions in the image and feature map.

[0129] s1 and s2 are used to represent the dimensions of the convolution kernel or pooling window, and are used to identify the starting position or size of the sliding window. t1 and t2 are used to represent the other dimension of the convolution kernel or pooling window.

[0130] u1 is used as the index in the fully connected layer, identifying the position in the feature map output by the pooling layer. v1 is used to represent the class index of the classification, indicating the category number in the model prediction or the true label.

[0131] S1 represents the number of rows (or height) of the convolution kernel, which is used to describe the vertical dimension of the convolution kernel when it slides on the image. S2 represents the number of rows (or height) of the pooling window, which is used to select the vertical dimension of the local maximum value on the feature map during the pooling operation.

[0132] T1 represents the number of columns (or width) of the convolution kernel, which is used to describe the horizontal dimension of the convolution kernel when it slides on the image. T2 represents the number of columns (or width) of the pooling window, which is used to select the horizontal dimension of the local maximum value on the feature map in the pooling operation.

[0133] By extracting local features of wound images, such as edges and textures, through the convolution layer, the system can accurately identify the key features of the wound, which provides richer image information for subsequent classification. Secondly, the pooling layer effectively reduces redundant information through dimensionality reduction operations, retains the most representative features, improves the computational efficiency of the model and reduces the risk of overfitting. With the help of the fully connected layer, the convolutional network efficiently classifies the wound healing stage and potential signs of infection. Through the cross-entropy loss function and gradient optimization algorithm, the system can gradually improve the accuracy and robustness of classification. The beneficial effect of this module is that it significantly improves the intelligence level of postoperative wound monitoring, can automatically and accurately evaluate the healing status of the wound, and help medical staff obtain the healing information of the patient's wound in a timely manner, thereby optimizing the postoperative treatment plan and reducing the risk of complications.

[0134] The model is calibrated using historical healing data to ensure the accuracy and reliability of the analysis results. Low confidence results will trigger manual review, including:

[0135] The historical healing data of postoperative patients were extracted from the database, covering different healing stages and related characteristics:

[0136] The historical data of postoperative healing of previous patients are extracted from the system database, including specific physiological characteristics of each healing stage (such as wound healing speed, signs of infection, etc.), as well as corresponding image features. These data cover the conditions and healing progress of different patients, ensuring that the model calibration has a broad reference basis. The historical data is used to help the current model accurately compare the wound images of new patients.

[0137] Through real-time analysis, the deviation between model output and historical data is evaluated and the prediction error is calculated:

[0138] The output of the model is analyzed in real time and compared with historical healing data to evaluate whether the model's predictions are consistent with past experience data. Deviations are calculated by analyzing the difference between the predicted results (such as wound healing stages) and the actual results in the historical healing data. The deviation of the model output is quantified as an error percentage. For example, if historical data indicates that a certain type of wound should have a specific healing pattern in a specific time period, and the model predicts a difference from this pattern, the system calculates this difference and marks it as a deviation.

[0139] Based on the model's prediction results, calculate the confidence score to evaluate the model's credibility for the current data:

[0140] Based on the current prediction results of the model, the system calculates a confidence score to evaluate the reliability of the model's prediction. Confidence reflects the model's confidence in the current judgment. Specifically, the confidence assessment is based on the probability distribution of the model. For example, if the model's judgment results for the healing stage are relatively concentrated (high confidence), the system considers the model prediction reliable; conversely, if the results are too dispersed, the confidence is low.

[0141] The confidence calculation uses a logical rule analysis of comprehensive feature weights. For example, the system assigns weights to each feature based on the color and shape changes of the wound and the standard progression in historical data, and gives a confidence score based on how well these features match. If multiple key features do not match historical data, the confidence is reduced.

[0142] Automatically adjust model parameters to optimize prediction accuracy based on historical data and error analysis:

[0143] When a deviation or low confidence is detected, the model will optimize itself by dynamically adjusting its internal parameters. The adjustment process relies on feedback from historical data, and the system automatically fine-tunes the model's learning rate, weights, and bias parameters to reduce the error with historical data. For example, the system may increase the weight of specific features, such as wound color or healing speed, to correct the model's prediction results. This process enables the model to be more adaptable to the specific situation of the current patient.

[0144] By analyzing the sources of error, the system gradually optimizes the model’s prediction accuracy to ensure that it is closer to the actual situation in historical data. For example, if historical data shows that the wound color changes greatly when infection occurs, the model may increase the weight of this feature to more sensitively capture color changes.

[0145] When the confidence level is lower than the set threshold, the system automatically triggers a manual review, prompting medical staff to re-evaluate:

[0146] When the confidence score is lower than the threshold set by the system, the system will automatically trigger the manual review mechanism. At this time, the system will notify medical personnel to manually verify the model's prediction results to ensure that the analysis of wound status is not biased due to the model's misjudgment. For example, when the system finds that the model's prediction is seriously inconsistent with historical data and the confidence score is low, it will prompt the doctor to re-examine the patient's wound status. In this way, the system can ensure the reliability of each model analysis and reduce treatment delays caused by automatic analysis errors.

[0147] By combining historical healing data with real-time analysis, a dynamic model calibration feedback mechanism is created to ensure high accuracy and reliability of model predictions. The technical details are disclosed through a series of detailed steps, including the use of historical data, the automated process of model calibration, and the trigger conditions for manual review, ensuring that the module technology is fully disclosed and avoids the defects in existing technologies.

[0148] The model mentioned in the calibration of the model using historical healing data refers to a convolutional neural network (CNN) or other machine learning model used to analyze postoperative wound healing in patients with gastrointestinal tumors. The main task of this model is to analyze and classify postoperative wound images to determine the patient's wound healing stage or whether there is a risk of infection.

[0149] The above formulas are all dimensionless and numerical calculations. The formula is a formula for the most recent real situation obtained by collecting a large amount of data and performing software simulation. The preset parameters and thresholds in the formula are set by technicians in this field according to actual conditions.

[0150] The above embodiments can be implemented in whole or in part by software, hardware, firmware or any other combination. When implemented using software, the above embodiments can be implemented in whole or in part in the form of a computer program product. The computer program product includes one or more computer instructions or computer programs. When the computer instructions or computer programs are loaded or executed on a computer, the process or function described in the embodiment of the present application is generated in whole or in part. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable devices. The computer instructions can be stored in a computer-readable storage medium, or transmitted from one computer-readable storage medium to another computer-readable storage medium. For example, the computer instructions can be transmitted from one website site, computer, server or data center to another website site, computer, server or data center by wired (e.g., infrared, wireless, microwave, etc.). The computer-readable storage medium can be any available medium that a computer can access or a data storage device such as a server or data center that contains one or more available media sets. The available medium can be a magnetic medium (e.g., a floppy disk, a hard disk, a tape), an optical medium (e.g., a DVD), or a semiconductor medium. The semiconductor medium can be a solid-state hard disk.

[0151] Those of ordinary skill in the art will appreciate that the modules and algorithm steps of each example described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are performed in hardware or software depends on the specific application and design constraints of the technical solution. Professional and technical personnel can use different methods to implement the described functions for each specific application, but such implementation should not be considered to be beyond the scope of this application.

[0152] Those skilled in the art can clearly understand that, for the convenience and brevity of description, the specific working processes of the systems, devices and modules described above can refer to the corresponding processes in the aforementioned method embodiments and will not be repeated here.

[0153] In the several embodiments provided in the present application, it should be understood that the disclosed systems, devices and methods can be implemented in other ways. For example, the device embodiments described above are only schematic. For example, the division of the modules is only a logical function division. There may be other division methods in actual implementation, such as multiple modules or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed can be through some interfaces, indirect coupling or communication connection of devices or modules, which can be electrical, mechanical or other forms.

[0154] The modules described as separate components may or may not be physically separated, and the components shown as modules may or may not be physical modules, and may be located in one place or distributed on multiple network modules. Some or all of the modules may be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0155] In addition, each functional module in each embodiment of the present application may be integrated into one processing module, or each module may exist physically separately, or two or more modules may be integrated into one module.

[0156] If the functions are implemented in the form of software function modules and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present application can be essentially or partly embodied in the form of a software product that contributes to the prior art. The computer software product is stored in a storage medium and includes several instructions for a computer device (which can be a personal computer, a server, or a network device, etc.) to perform all or part of the steps of the methods described in each embodiment of the present application. The aforementioned storage medium includes: various media that can store program codes, such as a USB flash drive, a mobile hard disk, a read-only memory (ROM), a random access memory (RAM), a magnetic disk or an optical disk.

[0157] The above is only a specific implementation of the present application, but the protection scope of the present application is not limited thereto. Any person skilled in the art who is familiar with the present technical field can easily think of changes or substitutions within the technical scope disclosed in the present application, which should be included in the protection scope of the present application. Therefore, the protection scope of the present application should be based on the protection scope of the claims.

[0158] Finally: The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention should be included in the protection scope of the present invention.

Claims

1. An intelligent monitoring system for patients with gastrointestinal tumors after surgery, characterized in that: It includes an ambient light adjustment module, a feature vector extraction module, a chaos theory analysis module, a convolutional network classification module, and a model calibration feedback module; Ambient light adjustment module: deploys ambient light sensors and automatic exposure adjustment technology, and uses machine learning algorithms to adjust camera parameters in real time based on ambient light data; Feature vector extraction module: extracts comprehensive local and global features from wound images through advanced image processing algorithms to generate a multi-dimensional feature vector containing texture, color and shape information; Chaos theory analysis module: uses the chaos theory model to analyze the complexity of the extracted multi-dimensional feature vectors and determine whether the wound image data is suitable for the next step of deep learning analysis; Convolutional network classification module: Use convolutional neural networks to extract features from wound images and classify wound status; train the model to identify various stages of wound healing and possible signs of infection; Model calibration feedback module: Use historical healing data to calibrate the model to ensure the accuracy and reliability of the analysis results. Low confidence results will trigger manual review.

2. According to claim 1, a postoperative intelligent monitoring system for patients with gastrointestinal tumors is characterized in that: Deploy ambient light sensors and automatic exposure adjustment technology, and use machine learning algorithms to adjust camera parameters in real time based on ambient light data. Specifically: Install multiple ambient light sensors on the image acquisition device to capture lighting conditions in all directions; The sensor continuously collects light intensity data: pre-processes the light data; Load the pre-trained machine learning model to parse the illumination data: Load the pre-trained machine learning model M light , M light The model formula is as follows: Among them, P exp is the recommended exposure parameter, f RNN represents a recurrent neural network function based on lighting data, is the light intensity value of the i-th sensor after preprocessing; According to P exp , the system dynamically adjusts the camera’s exposure time and aperture size in real time.

3. The intelligent monitoring system for gastrointestinal tumor patients after surgery according to claim 2, characterized in that: Advanced image processing algorithms are used to extract comprehensive local and global features from wound images to generate a multidimensional feature vector containing texture, color and shape information, specifically: Preprocessing wound images; The local binary pattern algorithm is used for local texture extraction, and its formula is as follows: Among them, F LBP (x, y) represents the local binary pattern eigenvalue at the position (x, y), I p Represents the gray value of the pth pixel in the pixel neighborhood, I c represents the gray value of the central pixel, P represents the number of pixels contained in the neighborhood; S(z) is the sign function, defined as: Where z = I p -I c ; The color distribution of the wound area is modeled using a color histogram, and the formula is as follows: Among them, F color (k) is the normalized color histogram value on color channel k, n k is the number of pixels on color channel k, and N is the total number of all pixels in the image; Use Fourier transform to extract the shape features of the wound; perform discrete Fourier transform on the wound contour; Through Fourier analysis, the global characteristics of the wound shape are extracted; Form a multidimensional feature vector V feat , which is expressed as follows: V feat =[F LBP (x,y), F color (k), F shape (u)]; where V feat is a multidimensional feature vector.

4. The intelligent monitoring system for gastrointestinal tumor patients after surgery according to claim 3, characterized in that: The complexity of the extracted multi-dimensional feature vectors is analyzed using the chaos theory model to determine whether the wound image data is suitable for the next step of deep learning analysis. Specifically: The extracted multidimensional feature vector V feat Standardize the process; The standardized multidimensional feature vector V is transformed into std Mapping into high-dimensional phase space; After reconstructing the phase space, the system is calculated by calculating the maximum Lyapunov exponent Λ max To measure the degree of chaos in the data; If Λ max >γ, then the wound image data is judged to be suitable for the next step of deep learning analysis; if Λ max ≤γ, then the wound image data is judged to be unsuitable for the next step of deep learning analysis; γ is the preset threshold.

5. The intelligent monitoring system for gastrointestinal tumor patients after surgery according to claim 4, characterized in that: Maximum Lyapunov exponent Λ max The calculation formula is: Among them, Λ max is the maximum Lyapunov exponent, X recon (t j ) is the time point t in the phase space j A vector of time steps where r is the number of time steps used to compute the Lyapunov exponent and j is the index of the time step.

6. The intelligent monitoring system for gastrointestinal tumor patients after surgery according to claim 5, characterized in that: Convolutional neural networks are used to extract features from wound images and classify wound status; the model is trained to identify the various stages of wound healing and possible signs of infection, specifically: The input wound image is normalized; the local features are extracted using a convolutional layer; The convolution layer extracts local features through sliding windows to form multiple feature maps; The pooling layer performs dimensionality reduction through the maximum pooling operation; The fully connected layer integrates the output features of the pooling layer and completes the classification of wound status through weighted summation; The cross entropy loss function is used to measure the accuracy of model prediction. The cross entropy loss function formula is as follows: Among them, L cross is the cross entropy loss value, is the true label of class v1, is the probability of class v1 predicted by the model, where V1 is the total number of classes.

7. The intelligent monitoring system for gastrointestinal tumor patients after surgery according to claim 6, characterized in that: The model is calibrated using historical healing data to ensure the accuracy and reliability of the analysis results. Low confidence results will trigger manual review, including: Extract historical healing data of postoperative patients from the database, covering different healing stages and related characteristics; Through real-time analysis, the deviation between model output and historical data is evaluated and the forecast error is calculated; Based on the model's prediction results, calculate the confidence score to evaluate the model's credibility for the current data; Automatically adjust model parameters to optimize prediction accuracy based on historical data and error analysis; When the confidence level is lower than the set threshold, the system automatically triggers manual review.

Images