A method for monitoring intraoperative bleeding using an intelligent surgical robot

By employing a multi-level feature cross-recombination algorithm and an anomaly perturbation identification and analysis method, the real-time and accuracy issues of intraoperative bleeding monitoring were resolved. This enabled precise identification of occult bleeding and dynamic tracking of bleeding volume, thereby improving the real-time performance and interpretability of the monitoring.

CN123120280APending Publication Date: 2026-11-17山东衡昊信息技术有限公司
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202511206089.1
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2025-08-27
Publication Date
2026-11-17

AI Technical Summary

Technical Problem

Existing technologies lack real-time and quantitative capabilities in intraoperative bleeding monitoring, rely on physician experience, and cannot accurately identify occult bleeding or dynamically track the amount of bleeding. Furthermore, the combined monitoring capabilities of multimodal sensors are insufficient.

Method used

A multi-level feature cross-recombination algorithm is used to fuse multi-source data. Combined with anomaly perturbation identification and analysis, bleeding parameters are constructed and fed back to the surgeon through feature extraction and data augmentation.

Benefits of technology

It enables precise monitoring of intraoperative bleeding, possesses real-time and anti-interference capabilities, and can identify sudden bleeding events and provide interpretable monitoring data support.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN123120280A_ABST
    Figure CN123120280A_ABST
Patent Text Reader

Abstract

This invention relates to the field of intraoperative bleeding monitoring technology, and more particularly to an intelligent surgical robot method for intraoperative bleeding monitoring. The method includes: collecting and preprocessing multi-source raw data to obtain preprocessed multi-source data; performing data augmentation on the preprocessed multi-source data to obtain amplified and enhanced multi-source data; extracting features from the amplified and enhanced multi-source data to obtain temporal and spatial features, and performing normalization preprocessing; introducing a multi-level feature cross-recombination algorithm for fusion processing to obtain fused comprehensive data; and performing intelligent analysis on the fused comprehensive data using an anomaly perturbation identification and analysis method to obtain patient bleeding parameters, which are then fed back to the surgeon for intraoperative bleeding monitoring. This method solves the technical problem of inaccurate data processing leading to inaccurate bleeding monitoring in intraoperative bleeding monitoring.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] This invention relates to the field of intraoperative bleeding monitoring technology, and in particular to an intraoperative bleeding monitoring method using an intelligent surgical robot. Background Technology

[0002] In the context of current surgical assistance technologies, especially in rapidly developing fields such as minimally invasive surgery, robot-assisted surgery, and image-guided systems, intraoperative bleeding monitoring remains a technological bottleneck. Currently, in clinical practice, surgeons primarily rely on visual observation of changes in blood color, the rate of blood accumulation in the surgical field, and instrument feedback to determine the presence of abnormal bleeding. This method is heavily reliant on the surgeon's experience, highly subjective, and lacks real-time and quantitative capabilities. In cases of occult bleeding or unclear bleeding location, it can easily lead to delayed identification, increasing intraoperative risks.

[0003] While some advanced surgical robot systems incorporate technologies such as image enhancement, vascular structure recognition, and preoperative CT or MRI data fusion, they still cannot achieve real-time dynamic tracking and quantitative analysis of intraoperative blood loss. Furthermore, most lack the ability to combine multimodal sensors (such as temperature, pressure, optics, and images) for joint blood loss monitoring. Even studies attempting to identify blood areas within the surgical field using image processing algorithms (such as image segmentation and color matching) are limited to static images or short-term assessments, lacking modeling of blood flow behavior and failing to derive continuous blood loss variation curves.

[0004] Meanwhile, the existing technologies mentioned above also have technical problems such as inaccurate data processing in intraoperative bleeding monitoring, leading to inaccurate bleeding monitoring. Summary of the Invention

[0005] This invention provides a method for monitoring intraoperative bleeding using an intelligent surgical robot, in order to solve the technical problem of inaccurate data processing and consequently inaccurate bleeding monitoring in intraoperative bleeding monitoring.

[0006] The present invention provides a method for monitoring intraoperative bleeding using an intelligent surgical robot, which specifically includes the following technical solutions: A method for monitoring intraoperative bleeding using an intelligent surgical robot includes the following steps: S1. Collect and preprocess multi-source raw data to obtain preprocessed multi-source data. Perform data augmentation on the preprocessed multi-source data to obtain amplified and enhanced multi-source data. Extract features from the amplified and enhanced multi-source data to obtain temporal and spatial features. Perform normalization preprocessing and introduce a multi-level feature cross-recombination algorithm for fusion processing to obtain fused comprehensive data. S2. The integrated data after fusion is analyzed intelligently using the abnormal perturbation identification and analysis method to obtain the patient's bleeding parameters, and then fed back to the surgeon for intraoperative bleeding monitoring.

[0007] Preferably, S1 specifically includes: The multi-level feature cross-recombination algorithm fuses the time-series and spatial features after normalization preprocessing, and generates fused comprehensive data by mapping the time-series and spatial features layer by layer through a mathematical-information mapping method.

[0008] Preferably, S1 specifically includes: In the implementation of the multi-level feature cross-recombination algorithm, in the first layer, a local cross-fusion initial transformation is performed, that is, at the feature level, temporal feature data and spatial feature data are combined in pairs, and the basic response relationship between temporal features and spatial features in terms of numerical amplitude and structural differences is quantified to construct the spatiotemporal modal primary cross-feature tensor.

[0009] Preferably, S1 specifically includes: In the implementation of the multi-level feature cross-recombination algorithm, in the second layer, nonlinear reconstruction and residual enhancement operations are performed. That is, based on the primary cross feature tensor generated in the previous layer, a global temporal convolutional cross signal is introduced. At the same time, its output domain is restricted by a nonlinear activation function, and the features of the first layer are retained to construct a deep residual path, thus obtaining the fused feature value.

[0010] Preferably, S1 specifically includes: In the implementation of the multi-level feature cross-recombination algorithm, the third layer performs aggregation and high-order nonlinear projection compression. The fusion feature values ​​generated in the second layer are amplified by a high power to increase the difference in the mode, and then the noise region is suppressed by the spatial exponential decay function. Finally, compression is achieved by a high-order root function to obtain the final fusion feature vector and output the fused comprehensive data.

[0011] Preferably, S2 specifically includes: In the implementation of the abnormal disturbance identification and analysis method, the attention weight of each fused comprehensive data is calculated, and the fused comprehensive data is weighted based on the attention weight to obtain the attention-adjusted weighted fused data.

[0012] Preferably, S2 specifically includes: In the implementation of the abnormal disturbance identification and analysis method, a local disturbance identification operator is introduced. The local disturbance identification operator identifies the change amplitude of local intervals in the weighted fusion data after attention adjustment, and introduces derivative constraints to determine whether the current physiological data shows a sudden change risk trend, thereby obtaining the disturbance response index.

[0013] Preferably, S2 specifically includes: In the implementation of the abnormal disturbance identification and analysis method, the disturbance response index is jointly mapped with the attention-adjusted weighted fusion data to construct the bleeding physical quantity estimation function, that is, the total bleeding volume estimation function per second during the operation, so as to obtain the bleeding volume per unit time at the current moment.

[0014] Preferably, S2 specifically includes: The bleeding rate is obtained by taking the first derivative of the bleeding volume per unit time at the current moment. For the bleeding location, the spatial features after normalization preprocessing are weighted and the spatial location corresponding to the spatial feature with the maximum weight is taken as the bleeding location. The bleeding parameters are obtained and fed back to the surgeon for intraoperative bleeding monitoring.

[0015] The beneficial effects of the technical solution of the present invention are: 1. For the amplified and enhanced multi-source data, spatial and temporal features are extracted using feature engineering methods. A multi-level feature cross-recombination algorithm is introduced. Through operations such as layer-by-layer nonlinear modeling, local-global modulation, residual enhancement, and energy compression, fusion features with spatial location memory, temporal state dependence, and response structure differences are constructed as the fused comprehensive data. This can more accurately express the complex correspondence between hemorrhage behavior and multimodal physiological changes, making the representation of hemorrhage status more physiologically reliable and spatiotemporally interpretable, thereby significantly improving monitoring accuracy and anti-interference ability.

[0016] 2. A local perturbation identification operator is introduced. This local perturbation identification operator does not rely on label classification or model discrimination, but is based on pure physical behavior. It captures the abnormal fluctuations of the attention-adjusted weighted fusion data in a short period of time, thereby identifying possible sudden bleeding events during surgery. It not only meets the real-time requirements of the surgical site, but also has good interpretability and can be used for retrospective analysis or medical behavior intervention reference. Attached Figure Description

[0017] Figure 1 This is a flowchart of an intraoperative bleeding monitoring method using an intelligent surgical robot according to the present invention. Detailed Implementation

[0018] To further illustrate the technical means and effects adopted by the present invention to achieve its intended purpose, the technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0019] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention pertains.

[0020] The following description, in conjunction with the accompanying drawings, details a specific scheme for an intraoperative bleeding monitoring method for an intelligent surgical robot provided by the present invention.

[0021] See attached document Figure 1 The diagram illustrates a flowchart of an intraoperative bleeding monitoring method for an intelligent surgical robot according to an embodiment of the present invention. The method includes the following steps: S1. Collect and preprocess multi-source raw data to obtain preprocessed multi-source data. Perform data augmentation on the preprocessed multi-source data to obtain amplified and enhanced multi-source data. Extract features from the amplified and enhanced multi-source data to obtain temporal and spatial features. Perform normalization preprocessing and introduce a multi-level feature cross-recombination algorithm for fusion processing to obtain fused comprehensive data. Sensor data is acquired during surgery under conditions such as changes in blood flow and temperature fluctuations using sensors such as pressure sensors, optical sensors, and temperature sensors. Simultaneously, surgical video images are acquired in real time using endoscopes, ultrasound, or infrared imaging devices. The sensor data and surgical video images are used together as multi-source raw data, including pressure data, temperature data, physiological data, and image data. The multi-source raw data undergoes preprocessing such as cleaning, noise reduction, time synchronization, standardization, and normalization to obtain preprocessed multi-source data, including preprocessed sensor data and preprocessed image data. The preprocessing processes all employ techniques well-known to those skilled in the art and will not be elaborated upon here.

[0022] New data is generated from the preprocessed multi-source data using data augmentation techniques, resulting in expanded and enhanced multi-source data. The data augmentation techniques employed are well-known to those skilled in the art and will not be elaborated upon here. For example, more representative image data can be generated through existing image transformations to enhance the diversity and richness of the training data.

[0023] Feature extraction is performed on the expanded and enhanced multi-source data using existing feature engineering techniques to obtain temporal and spatial features. For example, feature extraction is performed on preprocessed sensor data using a bidirectional long short-term memory network to obtain temporal features, and on preprocessed image data using a deep convolutional neural network to obtain spatial features. A spatial coding layer is introduced between the convolutional layers of the deep convolutional neural network. By incorporating the spatial location information of the image, the spatial patterns of the bleeding area can be better identified, thus obtaining spatial features. The extracted temporal and spatial features are then subjected to normalization preprocessing. The normalization preprocessing process employs techniques well-known to those skilled in the art and will not be elaborated upon here.

[0024] After normalizing and preprocessing the temporal and spatial features, a multi-level feature cross-recombination algorithm is used for fusion processing to obtain the fused comprehensive data. This algorithm generates the final fused comprehensive data representing the bleeding state by mapping the temporal and spatial features layer by layer using a mathematical-information mapping method. The specific implementation process is as follows: Let the time series feature data be denoted as Spatial feature data is denoted as ,in, , These represent the total number of temporal feature data and spatial feature data, respectively. Indicates the first One time-series feature data; Indicates the first Spatial feature data; This indicates transpose.

[0025] In the first layer, a local cross-fusion initial transformation is performed, that is, temporal and spatial feature data are combined pairwise at the feature level, and the basic response relationship between temporal and spatial features in terms of numerical amplitude and structural differences is quantified. Based on existing nonlinear feature cross-modeling techniques, a spatiotemporal modal primary cross-feature tensor is constructed, and the specific mathematical expression is as follows: , in, It is the primary cross-feature tensor, representing the first... The time series feature data and the first The first-layer fusion response value generated by the cross-interaction of spatial feature data represents the response coupling strength; It is the first One time-series feature data; It is the first Spatial feature data; Yes and The strength of the linear coupling between them is compressed to control the impact of excessively large cross responses on the perturbation and to ensure the stability of the primary cross feature tensor. This represents a residual nonlinear enhancement factor between temporal and spatial feature data, effectively balancing and enhancing the differences between temporal and spatial features; In the second layer, nonlinear reconstruction and residual enhancement operations are performed. Specifically, based on the primary cross-feature tensor generated in the previous layer, a global temporal convolutional cross-signal is introduced. Simultaneously, its output domain is restricted using a nonlinear activation function, while retaining the features from the first layer to construct a deep residual path. The specific formula is as follows: , in, Is the current layer number The time series feature data and the first The fusion feature value corresponding to each spatial feature data, i.e. the constructed deep residual path, is fused with the primary cross feature tensor and the global temporal series before residual superposition. It is the first One time-series feature data; This is a residual weighting factor used to adjust the proportion of the primary cross-feature tensor obtained in the first layer in the residual path of the second layer. It is determined based on expert experience, and the reference value range is [value missing]. ; It is a global temporal convolutional cross signal, that is, for the current spatial feature data A nonlinear correlation metric between it and all time-series channels is introduced to emphasize the influence of global time-series features on local spatial features. It is the hyperbolic tangent function, which serves as a non-linear activation function to ensure that the numerical output is within a fixed range, thereby reducing the risk of overfitting and improving the information compression rate. It is a residual term, used to further ensure the non-destructive transmission of fused feature information.

[0026] In the third layer, aggregation and high-order nonlinear projection compression are performed. This process amplifies important mode differences in the fused feature values ​​generated in the second layer using high powers, suppresses noise regions using a spatial exponential decay function, and then compresses them using a high-order root function to obtain the final fused feature vector. Based on existing nonlinear energy compression fusion techniques, the formula is as follows: , in, It is the third layer output feature value, that is, the final fused feature vector, which is used to characterize the fused feature after aggregating the influence of all spatial features in each temporal dimension; This is the spatial exponential decay factor, used to control the degree of penalty imposed on spatial feature data. It is determined based on expert experience, and the reference value range is as follows: ; It is a high-order feature amplification term that increases the strong response index of the fusion feature value generated in the second layer, thereby increasing its weight in the overall prediction contribution. It is a spatial feature penalty factor that exponentially compresses features in spatial features that are too high in energy or in abnormal regions.

[0027] Finally, the output fused and integrated data is represented as follows: , It is the total number of integrated data after merging.

[0028] S2. The integrated data after fusion is analyzed intelligently using the abnormal perturbation identification and analysis method to obtain the patient's bleeding parameters, and then fed back to the surgeon for intraoperative bleeding monitoring.

[0029] Intelligent analysis is performed on the fused comprehensive data using an abnormal perturbation identification and analysis method. This method introduces a local perturbation identification operator to determine the mutation risk trend of the current physiological data, and further obtains the patient's bleeding parameters. The specific implementation process is as follows: For the fused aggregated data, the attention weights for each piece of aggregated data are calculated using existing attention mechanisms. And based on attention weights The fused data is then weighted to obtain attention-adjusted weighted fused data. .

[0030] Furthermore, to enable abnormal monitoring of sudden intraoperative bleeding (such as vascular rupture or tissue damage), a local perturbation identification operator is introduced. The mathematical objective of this operator is to identify the magnitude of change in local intervals within the attention-adjusted weighted fusion data and introduce derivative constraints to determine whether the current physiological data exhibits a trend of sudden change risk. The implementation formula for the local perturbation identification operator is expressed as follows: , in, It is the first Attention-adjusted weighted fusion data at time [time] The disturbance response index, i.e. the intensity of abnormal volatility, indicates that the current channel characteristics have a sudden unstable trend, suggesting that intraoperative bleeding events may occur. This is the width of the time window, representing the length of the sliding window. It is determined based on the specific scenario and will not be elaborated here. It is a time window index; At any moment The Attention-adjusted weighted fusion data; It is the first Attention-adjusted weighted fusion data at time [time] The average value of the sliding window; This is the derivative weighting coefficient, used to control the influence of the rate of change (time derivative) on the overall disturbance response index. It is set based on expert experience, and the reference value range is [value missing]. ; At any moment The Attention-adjusted weighted fusion data; At any moment The Attention-adjusted weighted fusion data; It is the time interval between adjacent time points, a preset value based on specific needs and expert experience.

[0031] Furthermore, determine the bleeding parameters; and define the disturbance response index. With attention-adjusted weighted fusion data By performing a joint mapping, a function for estimating the physical quantity of bleeding is constructed, namely, a function for estimating the total intraoperative blood loss per second. The formula is as follows: , in, This represents the amount of bleeding per unit of time at the current moment; This is the conversion factor for converting features to physical units, determined based on expert experience, with a reference range of values. The unit is ml / characteristic unit; It is a mapping function used to compress the perturbation response exponentially to Range, select function; To monitor changes in bleeding trends, such as a sudden increase, the amount of bleeding per unit time at the current moment is measured. Find the first derivative to obtain the bleeding rate.

[0032] For the bleeding location, the spatial features after normalization preprocessing in step S1 are weighted using the existing attention mechanism. When the weight reaches its maximum value, the spatial location corresponding to the spatial feature is used as the coordinate location of the suspected bleeding point.

[0033] The above process yields bleeding parameters such as bleeding volume, bleeding rate, and bleeding location, which are then fed back to the surgeon to provide data support and enable intraoperative bleeding monitoring.

[0034] In summary, a method for monitoring intraoperative bleeding using an intelligent surgical robot has been developed.

[0035] The order of the embodiments is for illustrative purposes only and does not represent the superiority or inferiority of the embodiments. The processes depicted in the drawings do not necessarily require a specific or sequential order to achieve the desired result. In some embodiments, multitasking and parallel processing are possible or may be advantageous.

[0036] The various embodiments in this specification are described in a progressive manner. The same or similar parts between the various embodiments can be referred to each other. Each embodiment focuses on describing the differences from other embodiments.

[0037] The above embodiments are only used to illustrate the technical solutions of the present invention, and are not intended to limit it. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention, and should all be included within the protection scope of the present invention.

Claims

1. A method for monitoring intraoperative bleeding using an intelligent surgical robot, characterized in that, Includes the following steps: S1. Collect and preprocess multi-source raw data to obtain preprocessed multi-source data. Perform data augmentation on the preprocessed multi-source data to obtain amplified and enhanced multi-source data. Extract features from the amplified and enhanced multi-source data to obtain temporal and spatial features. Perform normalization preprocessing and introduce a multi-level feature cross-recombination algorithm for fusion processing to obtain fused comprehensive data. S2. The integrated data after fusion is analyzed intelligently using the abnormal perturbation identification and analysis method to obtain the patient's bleeding parameters, and then fed back to the surgeon for intraoperative bleeding monitoring.

2. The method for monitoring intraoperative bleeding using an intelligent surgical robot according to claim 1, characterized in that, S1 specifically includes: The multi-level feature cross-recombination algorithm fuses the time-series and spatial features after normalization preprocessing, and generates fused comprehensive data by mapping the time-series and spatial features layer by layer through a mathematical-information mapping method.

3. The method for monitoring intraoperative bleeding using an intelligent surgical robot according to claim 2, characterized in that, S1 specifically includes: In the implementation of the multi-level feature cross-recombination algorithm, in the first layer, a local cross-fusion initial transformation is performed, that is, at the feature level, temporal feature data and spatial feature data are combined in pairs, and the basic response relationship between temporal features and spatial features in terms of numerical amplitude and structural differences is quantified to construct the spatiotemporal modal primary cross-feature tensor.

4. The method for monitoring intraoperative bleeding in an intelligent surgical robot according to claim 3, characterized in that, S1 specifically includes: In the implementation of the multi-level feature cross-recombination algorithm, in the second layer, nonlinear reconstruction and residual enhancement operations are performed. That is, based on the primary cross feature tensor generated in the previous layer, a global temporal convolutional cross signal is introduced. At the same time, its output domain is restricted by a nonlinear activation function, and the features of the first layer are retained to construct a deep residual path, thus obtaining the fused feature value.

5. The method for monitoring intraoperative bleeding in an intelligent surgical robot according to claim 4, characterized in that, S1 specifically includes: In the implementation of the multi-level feature cross-recombination algorithm, the third layer performs aggregation and high-order nonlinear projection compression. The fusion feature values ​​generated in the second layer are amplified by a high power to increase the difference in the mode, and then the noise region is suppressed by the spatial exponential decay function. Finally, compression is achieved by a high-order root function to obtain the final fusion feature vector and output the fused comprehensive data.

6. The method for monitoring intraoperative bleeding in an intelligent surgical robot according to claim 1, characterized in that, S2 specifically includes: In the implementation of the abnormal disturbance identification and analysis method, the attention weight of each fused comprehensive data is calculated, and the fused comprehensive data is weighted based on the attention weight to obtain the attention-adjusted weighted fused data.

7. The method for monitoring intraoperative bleeding in an intelligent surgical robot according to claim 6, characterized in that, S2 specifically includes: In the implementation of the abnormal disturbance identification and analysis method, a local disturbance identification operator is introduced. The local disturbance identification operator identifies the change amplitude of local intervals in the weighted fusion data after attention adjustment, and introduces derivative constraints to determine whether the current physiological data shows a sudden change risk trend, thereby obtaining the disturbance response index.

8. The method for monitoring intraoperative bleeding in an intelligent surgical robot according to claim 7, characterized in that, S2 specifically includes: In the implementation of the abnormal disturbance identification and analysis method, the disturbance response index is jointly mapped with the attention-adjusted weighted fusion data to construct the bleeding physical quantity estimation function, that is, the total bleeding volume estimation function per second during the operation, so as to obtain the bleeding volume per unit time at the current moment.

9. The method for monitoring intraoperative bleeding in an intelligent surgical robot according to claim 8, characterized in that, S2 specifically includes: The bleeding rate is obtained by taking the first derivative of the bleeding volume per unit time at the current moment. For the bleeding location, the spatial features after normalization preprocessing are weighted and the spatial location corresponding to the spatial feature with the maximum weight is taken as the bleeding location. The bleeding parameters are obtained and fed back to the surgeon for intraoperative bleeding monitoring.