Intraoperative brain glioma deep learning detection method and system based on hyperspectral imaging

Abstract

The present application relates to the technical field of intraoperative navigation, and particularly relates to an intraoperative brain glioma deep learning detection method and system based on hyperspectral imaging, the steps of the method comprising: analyzing the internal heterogeneity of brain glioma based on spectral difference, evaluating the coverage benefits of preoperative planning path and intraoperative detour path; analyzing the blood oxygen saturation distribution based on the absorption peak characteristics of hemoglobin in a specific wave band, evaluating the risk of local ischemia of the preoperative planning path and the intraoperative detour path; mapping the preoperative planning path and the intraoperative detour path to the brain function area topology network respectively, and evaluating the functional damage risk of the paths based on the topological connection relationship and the risk of local ischemia; analyzing the expected benefits and risk trends of the preoperative planning path and the intraoperative detour path, and generating surgical path decision suggestions. The present application quantifies the path coverage benefits and functional damage risks of the surgical path, effectively improving the accuracy and safety of the surgical path decision.

Description

Technical Field

[0001] This invention relates to the field of intraoperative navigation technology, and in particular to a method and system for intraoperative glioma detection based on hyperspectral imaging using deep learning. Background Technology

[0002] Gliomas are highly malignant and heterogeneous brain tumors. There are significant differences in the cellular and tissue structures of different regions of a glioma. Because gliomas often involve functional areas of the brain, it is difficult to clearly distinguish the boundary between the tumor tissue and the functional areas of the brain during surgery. Traditional image-guided surgery may result in residual malignant areas or unnecessary removal of normal tissue, increasing surgical risks and affecting prognosis.

[0003] Traditional preoperative planning mainly relies on techniques such as magnetic resonance imaging (MRI), functional magnetic resonance imaging (fMRI), and diffusion tensor imaging (DTI). Although these techniques can provide reference information for preoperative path planning, they still have shortcomings in terms of spatial resolution and real-time performance during surgery.

[0004] Hyperspectral imaging technology fuses spectral and image data, enabling it to acquire spatial information about tissues and analyze internal tissue properties by extracting the spectrum from each pixel in the image. It can also identify components invisible to the human eye, such as hemoglobin concentration and blood oxygen saturation, and non-contactly and non-invasively distinguish between different types of tissues, such as tumor tissue and non-tumor tissue.

[0005] Traditional imaging methods struggle to reveal the heterogeneity within tumors, making it impossible to assess the differentiated coverage benefits of the central, infiltrated, and peripheral areas of gliomas. Furthermore, existing technologies fail to couple the surgical path with the topological network of brain functional areas, neglecting the propagation effect of ischemic risk in topological connectivity. This results in insufficient accuracy in assessing the damage risk of the surgical path to brain functional areas, reducing the accuracy and safety of surgical path decisions. Summary of the Invention

[0006] To overcome the defects and shortcomings of existing technologies, this invention provides a deep learning detection method and system for intraoperative gliomas based on hyperspectral imaging. It dynamically corrects the local ischemia risk of the surgical path through the compensatory ability of collateral circulation, and combines the topological network of brain functional areas to quantify the propagation and amplification effect of local ischemia risk in topological connectivity, which significantly improves the accuracy and safety of the operation.

[0007] To achieve the above objectives, the present invention adopts the following technical solution:

[0008] In a first aspect, the present invention provides a deep learning-based method for intraoperative glioma detection based on hyperspectral imaging, comprising:

[0009] Acquire hyperspectral imaging data of the preoperative planned path and intraoperative detour path, topological network of brain functional areas, and preoperative vascular imaging data;

[0010] Based on spectral difference analysis of the internal heterogeneity of gliomas, the benefits of preoperative planning and intraoperative bypass routes in covering the central, infiltrative, and peripheral areas of gliomas were evaluated.

[0011] Based on the absorption peak characteristics of hemoglobin in a specific band, the distribution of blood oxygen saturation was analyzed to assess the local ischemic risk of the preoperative planned path and the intraoperative detour path.

[0012] The preoperative planned path and the intraoperative detour path were mapped to the topological network of brain functional areas, and the risk of functional impairment corresponding to the path was assessed based on the topological connectivity and the risk of local ischemia.

[0013] We conduct expected benefit and risk trend analysis on the preoperative planned path and the intraoperative bypass path, and generate surgical path decision recommendations.

[0014] Further, the coverage benefits are evaluated, including:

[0015] The hyperspectral imaging data was preprocessed and glioma characteristic bands were extracted. The preprocessing included illumination correction and noise removal.

[0016] Based on the characteristic bands of gliomas, gliomas are divided into subregions, including the central region, infiltrative region, and peripheral region of the glioma.

[0017] The spatial coverage of the preoperative planned path and the intraoperative bypass path in the central area, infiltrative area and peripheral area of ​​the glioma were calculated respectively.

[0018] The spectral heterogeneity weights of the central, infiltrative, and peripheral regions of gliomas are assessed based on differences in spectral characteristics. The spectral heterogeneity weights are the mean standard deviations of each characteristic band within the region.

[0019] The path coverage benefit coefficient of the preoperative planned path and the intraoperative detour path is obtained by weighting and summing the spatial coverage by spectral heterogeneity weights.

[0020] Furthermore, the assessment of the local ischemia risk of the preoperative planned path and the intraoperative bypass path includes:

[0021] The absorption characteristics of oxyhemoglobin and deoxyhemoglobin in characteristic bands were extracted based on hyperspectral imaging data, and the tissue oxygen saturation at each pixel location was determined by optical density difference method.

[0022] Mark the pixel locations where tissue blood oxygen saturation is lower than a preset tissue blood oxygen saturation threshold and construct a tissue ischemia risk region;

[0023] The local ischemia risk coefficient is obtained by compensatory correction of the tissue ischemia risk coefficient in the tissue ischemia risk area and a local ischemia risk distribution map is constructed. The tissue ischemia risk coefficient represents the degree of deviation of tissue blood oxygen saturation from the preset tissue blood oxygen saturation threshold.

[0024] The preoperative planned path and the intraoperative detour path were mapped to the local ischemia risk distribution map, and the local ischemia risk coefficient at each pixel position of the path was extracted.

[0025] Furthermore, the method of obtaining the local ischemia risk coefficient by compensatory correction of the tissue ischemia risk coefficient within the tissue ischemia risk area includes:

[0026] Preoperative vascular imaging data were acquired and the collateral circulation compensation capacity was graded using the modified ASITN / SIR collateral circulation scoring system.

[0027] The sigmoid function is used to convert the lateral circulation compensation capacity into compensation redundancy weights.

[0028] The tissue ischemia risk coefficient within the tissue ischemia risk region is attenuated and corrected based on the compensatory redundancy weight, and the corrected tissue ischemia risk coefficient is used as the local ischemia risk coefficient.

[0029] Furthermore, the functional impairment risk corresponding to the assessment path includes:

[0030] The average Euclidean distance between functional nodes and paths in the topological network of brain functional areas is extracted, and the distance coupling coefficient between functional nodes and paths is determined using the distance decay function.

[0031] The local ischemia risk coefficient at each pixel position of the preoperative planned path and the intraoperative detour path is mapped to the functional nodes of the brain functional area topology network, and the product of the local ischemia risk coefficient and the distance coupling coefficient is used as the node functional damage risk coefficient.

[0032] The functional impairment risk coefficient of the path is obtained by using the betweenness centrality of each functional node in the brain functional area topology network as the node connection weight and by weighting and summing the node functional impairment risk coefficients of all nodes using the node connection weights.

[0033] Furthermore, the step of performing expected benefit and risk trend analysis on the preoperative planned path and intraoperative bypass path and generating surgical path decision recommendations includes:

[0034] Obtain the path coverage benefit coefficient and functional impairment risk coefficient corresponding to the preoperative planned path and the intraoperative detour path;

[0035] The difference between the path coverage benefit coefficient and the functional damage risk coefficient of the preoperative planned path is used as the benchmark net benefit coefficient, and the difference between the path coverage benefit coefficient and the functional damage risk coefficient of the intraoperative detour path is used as the detour net benefit coefficient.

[0036] When the detour net benefit coefficient is greater than the benchmark net benefit coefficient, the intraoperative detour path is taken as the preferred surgical path; when the detour net benefit coefficient is less than or equal to the benchmark net benefit coefficient, the preoperative planned path is taken as the preferred surgical path.

[0037] Secondly, the present invention provides an intraoperative glioma deep learning detection system based on hyperspectral imaging, comprising:

[0038] The data acquisition module is used to acquire hyperspectral imaging data of the preoperative planned path and the intraoperative detour path, the topological network of brain functional areas, and preoperative vascular imaging data.

[0039] The coverage benefit assessment module is used to analyze the internal heterogeneity of gliomas based on spectral differences and to evaluate the coverage benefit of preoperative planned paths and intraoperative bypass paths on the central, infiltrated, and peripheral areas of gliomas.

[0040] The ischemia risk assessment module is used to analyze blood oxygen saturation distribution based on the absorption peak characteristics of hemoglobin in a specific band, and to assess the local ischemia risk of the preoperative planned path and the intraoperative bypass path.

[0041] The damage risk assessment module is used to map the preoperative planned path and the intraoperative detour path to the topological network of brain functional areas and assess the functional damage risk corresponding to the path based on the topological connectivity and local ischemia risk.

[0042] The decision suggestion generation module is used to perform expected benefit and risk trend analysis on the preoperative planned path and the intraoperative detour path, and generate surgical path decision suggestions.

[0043] Thirdly, the present invention provides an electronic device, comprising: a processor and a memory, wherein the memory stores a computer program that can be called by the processor, and the processor executes a deep learning detection method for intraoperative glioma based on hyperspectral imaging by calling the computer program stored in the memory.

[0044] Fourthly, the present invention provides a computer-readable storage medium storing instructions that, when executed on a computer, cause the computer to perform a deep learning detection method for intraoperative gliomas based on hyperspectral imaging.

[0045] Compared with the prior art, the present invention has the following advantages and beneficial effects:

[0046] This invention improves the accuracy of surgical pathway coverage benefit assessment by introducing hyperspectral imaging and deep learning technologies to accurately extract spectral features of different subregions of gliomas and quantify the internal heterogeneity of gliomas. By combining the absorption peaks of hemoglobin in characteristic bands and the compensatory capacity of collateral circulation, it dynamically corrects local ischemia risk, achieving an ischemia risk assessment that more closely reflects the actual blood supply status. By coupling the pathway with the topological network of brain functional areas, it quantitatively reflects the propagation and amplification effect of ischemia risk in topological connections, thereby accurately assessing the risk of functional impairment. By constructing a comprehensive benefit and risk assessment system, it compares and analyzes preoperative planned pathways and intraoperative bypass pathways, generating optimal pathway decision suggestions, providing strong support for refined and individualized intraoperative pathway decision-making, and effectively improving the accuracy and safety of surgical pathway decisions. Attached Figure Description

[0047] Other features, objects, and advantages of the invention will become more apparent from the following detailed description of non-limiting embodiments with reference to the accompanying drawings:

[0048] Figure 1 This is a flowchart illustrating the intraoperative deep learning detection method for gliomas based on hyperspectral imaging provided in this embodiment of the invention.

[0049] Figure 2 This is a schematic diagram of the intraoperative glioma deep learning detection system based on hyperspectral imaging provided in an embodiment of the present invention;

[0050] Figure 3 This is a schematic diagram of the structure of the electronic device provided in an embodiment of the present invention. Detailed Implementation

[0051] The technical solution of the present invention will be described in detail below with reference to the accompanying drawings and specific embodiments. It should be understood that the embodiments of the present invention and the specific features in the embodiments are detailed descriptions of the technical solution of the present invention, rather than limitations thereof. In the absence of conflict, the embodiments of the present invention and the technical features in the embodiments can be combined with each other.

[0052] Please see Figure 1 , Figure 1 This is a schematic diagram of the overall process of the intraoperative deep learning detection method for gliomas based on hyperspectral imaging provided in this embodiment of the invention, which specifically includes the following steps:

[0053] S100. Obtain hyperspectral imaging data of the preoperative planned path and the intraoperative bypass path, the topological network of brain functional areas, and preoperative vascular imaging data. In the intraoperative process, hyperspectral imaging data of the exposed brain tissue areas are collected to obtain the real spectral features within the path coverage area. For the deep or occluded areas that have not yet been exposed, the spectral features are predicted using a deep learning model trained on preoperative imaging data, thereby achieving hyperspectral imaging data completion of the entire path domain. The deep learning model includes: (1) a spatial feature extraction network, which uses a three-dimensional convolutional neural network (3D-CNN) to process preoperative imaging data (such as MRI, DTI) and extract the anatomical and structural connectivity features of brain tissue; (2) a spectral feature extraction network, which uses a convolutional neural network (CNN) to process the hyperspectral imaging data collected from the exposed brain tissue areas and extract its spectral distribution features; (3) a spectral reconstruction network, which uses a fusion attention mechanism to jointly model spatial features and spectral features and uses the spectral reconstruction network to predict the hyperspectral data of the unexposed areas. During the training phase, paired training sample sets are first established, with each sample containing preoperative imaging data, hyperspectral imaging data of the exposed area, and real hyperspectral labels. Multimodal fusion features are obtained through spatial feature extraction networks and spectral feature extraction networks and input into the spectral reconstruction network. In the loss function design, mean squared error (MSE) loss is combined to ensure the accuracy of spectral intensity fitting. At the same time, spectral angle mapping loss (SAM) is introduced to maintain spectral morphology consistency, and adversarial loss (GAN) is further introduced to enhance the model's generative ability. Through iterative training and gradient backpropagation, the model parameters are continuously optimized, and a deep learning model that can accurately predict spectral distribution is finally obtained. In the prediction phase, preoperative imaging data is input into the spatial feature extraction network, and hyperspectral data of the exposed area during surgery is input into the spectral feature extraction network. The resulting fusion features are used by the spectral reconstruction network to generate spectral prior estimation data for the unexposed area. Deep learning models can complete the spectral data of unexposed tissue areas along the preoperative planned path and intraoperative bypass path, so that the entire path range has hyperspectral imaging data, providing complete data support for subsequent glioma heterogeneity analysis, ischemic risk assessment and functional impairment prediction.

[0054] A topological network of brain functional areas was constructed using preoperative functional magnetic resonance imaging (fMRI), diffusion tensor imaging (DTI), and other multimodal imaging data to reflect brain functional nodes and their connections. Preoperative vascular imaging data was obtained using digital subtraction angiography (DSA), magnetic resonance angiography (MRA), or computed tomography angiography (CTA) to analyze collateral circulation and blood supply distribution characteristics, thereby providing data support for subsequent ischemic risk correction and functional impairment assessment.

[0055] S200: Based on spectral difference analysis, the internal heterogeneity of gliomas is evaluated to assess the coverage benefits of preoperative planned paths and intraoperative bypass paths for the central, infiltrated, and peripheral areas of gliomas.

[0056] The heterogeneity within gliomas manifests as differences in cell density, blood supply, and metabolic levels across different subregions (such as the central, infiltrative, and peripheral regions). These differences can be expressed as significantly different spectral characteristics in the characteristic bands of hyperspectral imaging. The coverage benefit of a surgical path for gliomas depends not only on whether the path passes through or resectes the tumor area, but also on the specific subregion it covers and the degree of heterogeneity within that subregion. For example, when the preoperatively planned path covers more infiltrative areas with higher spectral heterogeneity, it yields greater benefits in removing potential tumor residue and reducing the risk of recurrence. Conversely, if the intraoperative bypass path primarily covers the less heterogeneous peripheral region, the benefit in clearing the core tumor lesion is limited despite the increased path extension. Therefore, the coverage benefit of a surgical path is closely related to glioma heterogeneity; covering areas with high heterogeneity weights amplifies the benefit value of the path, while covering areas with low heterogeneity contributes relatively less. Assessing coverage benefit includes:

[0057] The hyperspectral imaging data was preprocessed and glioma characteristic bands were extracted. The preprocessing included illumination correction and noise removal.

[0058] Based on the characteristic bands of gliomas, gliomas are divided into subregions, including the central region, infiltrative region, and peripheral region of the glioma.

[0059] The spatial coverage of the preoperative planned path and the intraoperative bypass path in the central area, infiltrative area and peripheral area of ​​the glioma were calculated respectively.

[0060] The spectral heterogeneity weights of the central, infiltrative, and peripheral regions of gliomas are assessed based on differences in spectral characteristics. The spectral heterogeneity weights are the mean standard deviations of each characteristic band within the region. These weights reflect the degree of difference in spectral characteristics within different subregions of a glioma, essentially representing the complexity and heterogeneity of the tissue composition in that region. Higher weights indicate more significant differences in metabolism, blood supply, or cell structure in that region, potentially indicating more malignant cells or infiltration risks. Therefore, covering and resecting these regions during surgery is more meaningful. Introducing spectral heterogeneity weights into the assessment of pathway coverage benefits avoids the one-sidedness of simply measuring pathway value by spatial coverage rate. This allows coverage of highly heterogeneous infiltrative or central regions to contribute more to the overall benefit, while coverage of less heterogeneous peripheral regions has a smaller impact. This more accurately reflects the actual value of surgical pathways in resection effectiveness and tumor control, guiding surgeons to choose the surgical pathway that maximizes benefits.

[0061] The path coverage benefit coefficient of the preoperative planned path and the intraoperative detour path is obtained by weighting and summing the spatial coverage by spectral heterogeneity weights.

[0062] S300: Based on the absorption peak characteristics of hemoglobin in a specific band, analyze the distribution of blood oxygen saturation and assess the local ischemic risk of the preoperative planned path and the intraoperative bypass path.

[0063] Neurons in brain functional areas are extremely sensitive to ischemia. If the surgical path passes through or is adjacent to areas with a high risk of ischemia, insufficient local blood supply will amplify the functional vulnerability of that area, increasing the probability of postoperative functional impairment. Furthermore, ischemic risk not only directly affects local tissues but may also propagate through topological connections in the brain's functional network, leading to damage to distant nodes. Therefore, assessing local ischemic risk is crucial for evaluating the risk of subsequent functional impairment. Assessing the local ischemic risk of the preoperative planned path and the intraoperative bypass path includes:

[0064] The absorption characteristics of oxyhemoglobin and deoxyhemoglobin in characteristic bands are extracted based on hyperspectral imaging data, and the tissue oxygen saturation at each pixel location is determined by optical density difference method. The characteristic absorption peaks of oxyhemoglobin and deoxyhemoglobin are extracted based on hyperspectral imaging data, mainly relying on their different spectral characteristics in the visible and near-infrared bands. Oxyhemoglobin has significant absorption peaks at approximately 542 nm and 577 nm, while deoxyhemoglobin has more obvious absorption peaks at 555 nm and 760 nm. Hyperspectral imaging can obtain the reflectance spectrum of brain tissue in these specific bands pixel by pixel. Then, the reflectance is converted into optical density value using Lambert-Beer law. Furthermore, the concentration ratio of oxyhemoglobin to deoxyhemoglobin is calculated by optical density difference method, thereby obtaining the tissue oxygen saturation of each pixel and accurately characterizing the oxygenation status of local tissues.

[0065] Mark the pixel locations where tissue blood oxygen saturation is lower than a preset tissue blood oxygen saturation threshold and construct a tissue ischemia risk region;

[0066] By compensatoryly correcting the tissue ischemia risk coefficient within the tissue ischemia risk region, a local ischemia risk coefficient is obtained, and a local ischemia risk distribution map is constructed. The tissue ischemia risk coefficient represents the degree of deviation of tissue blood oxygen saturation from a preset tissue blood oxygen saturation threshold. The tissue ischemia risk coefficient can be:

[0067] ;

[0068] In the formula For the first The risk coefficient of tissue ischemia at each pixel location This is a preset tissue oxygen saturation threshold, with a default value of 60%. This value can be adjusted based on clinical literature or experimental calibration. For the first Tissue blood oxygen saturation at each pixel location, For the first The degree to which the tissue oxygen saturation at a given pixel location is lower than a preset tissue oxygen saturation threshold. Used to normalize the difference to a proportional value between 0 and 1. This is used to ensure that the tissue ischemia risk coefficient does not become negative;

[0069] The preoperative planned path and the intraoperative detour path were respectively mapped to the local ischemia risk distribution map, and the local ischemia risk coefficient at each pixel position of the path was extracted.

[0070] Collateral circulation compensatory capacity reflects whether tissues can maintain perfusion and oxygenation through bypass blood flow when the main supplying artery is damaged or local perfusion decreases. Better collateral circulation means a smaller immediate risk to tissue function from the same degree of local tissue oxygen saturation decline (greater buffering effect). Conversely, poor collateral circulation makes irreversible damage more likely with the same degree of tissue oxygen saturation decline. Therefore, when quantifying tissue ischemia risk, it is necessary to perform attenuation correction on the tissue ischemia risk according to collateral compensatory capacity. This is achieved by compensatoryly correcting the tissue ischemia risk coefficient within the tissue ischemia risk area to obtain the local ischemia risk coefficient, which includes:

[0071] Preoperative vascular imaging data was obtained and the collateral circulation compensation capacity was graded using the modified ASITN / SIR collateral circulation scoring system, which included grades 1, 2 and 3.

[0072] The sigmoid function is used to convert the compensatory capacity of lateral circulation into compensatory redundancy weights. ,in, , For the first The lateral cyclic compensation capability of each pixel position is mapped into compensation redundancy weights by the Sigmoid function to the discrete hierarchical smoothness.

[0073] The tissue ischemia risk coefficient within the tissue ischemia risk region is attenuated and corrected based on the compensatory redundancy weight, and the corrected tissue ischemia risk coefficient is used as the local ischemia risk coefficient. ,in, , This indicates that the stronger the collateral circulation compensation capacity, the more linearly the risk of local ischemia is reduced.

[0074] S400: Map the preoperative planned path and the intraoperative detour path to the topological network of brain functional areas respectively, and assess the risk of functional impairment corresponding to the path based on the topological connectivity and local ischemia risk.

[0075] The topological connectivity between the surgical pathway and its functional nodes reflects the potential impact of the pathway on different brain functional areas, while the local ischemia risk coefficient along the surgical pathway reflects the physiological tendency for functional damage to tissues along that pathway. When the local ischemia risk of the pathway is high and it is close to a highly important functional node, the risk coefficient of node functional damage increases. For example, if the surgical bypass path passes near the motor cortex and the local blood oxygen saturation in that area is significantly reduced (high local ischemia risk), the functional damage risk coefficient of the corresponding motor functional node will increase significantly. If the pathway passes through the same area but collateral circulation is good or blood oxygen saturation is normal (low local ischemia risk), the functional damage risk coefficient is low. This demonstrates the relationship between local ischemia risk and topological location in determining the potential functional damage of the pathway. Assessing the functional damage risk corresponding to the pathway includes:

[0076] The average Euclidean distance between functional nodes and paths in the topological network of brain functional areas is extracted, and the distance coupling coefficient between functional nodes and paths is determined by the distance decay function, which can be either Gaussian distance decay function or exponential decay function.

[0077] The local ischemia risk coefficient at each pixel position of the preoperative planned path and the intraoperative detour path is mapped to the functional nodes of the brain functional area topology network, and the product of the local ischemia risk coefficient and the distance coupling coefficient is used as the node functional damage risk coefficient.

[0078] The betweenness centrality of each functional node in the brain functional area topology network is used as the node connection weight. The functional impairment risk coefficient of the path is obtained by weighted summation of the node functional impairment risk coefficients of all nodes through the node connection weights. Nodes with high betweenness centrality are located on the key pathways of the brain functional network and play a pivotal role in information flow or functional transmission. Therefore, their local ischemia risk will not only affect their own node function, but may also transmit the ischemia risk to other connected nodes through the network topology, leading to more widespread functional impairment.

[0079] S500 performs expected benefit and risk trend analysis on the preoperative planned path and the intraoperative detour path and generates surgical path decision suggestions;

[0080] By integrating the aforementioned pathway coverage benefits and functional impairment risks into actionable surgical decision-making criteria, this study quantifies the comprehensive effects of different surgical pathways in maximizing tumor coverage while minimizing brain functional impairment, and identifies the surgical pathway with the superior overall net benefit. This provides objective and quantifiable decision-making recommendations for surgery. Furthermore, it conducts expected benefit and risk trend analysis on preoperative planned pathways and intraoperative bypass pathways, generating surgical pathway decision recommendations, including:

[0081] Obtain the path coverage benefit coefficient and functional impairment risk coefficient corresponding to the preoperative planned path and the intraoperative detour path;

[0082] The difference between the path coverage benefit coefficient and the functional damage risk coefficient of the preoperative planned path is used as the benchmark net benefit coefficient, and the difference between the path coverage benefit coefficient and the functional damage risk coefficient of the intraoperative detour path is used as the detour net benefit coefficient.

[0083] When the detour net benefit coefficient is greater than the benchmark net benefit coefficient, the intraoperative detour path is taken as the preferred surgical path; when the detour net benefit coefficient is less than or equal to the benchmark net benefit coefficient, the preoperative planned path is taken as the preferred surgical path.

[0084] Please see Figure 2 , Figure 2 This is a schematic diagram of the structure of the intraoperative glioma deep learning detection system based on hyperspectral imaging provided in an embodiment of the present invention, including:

[0085] Data acquisition module 210 is used to acquire hyperspectral imaging data of the preoperative planned path and intraoperative bypass path, brain functional area topology network and preoperative vascular imaging data.

[0086] The coverage benefit assessment module 220 is connected to the data acquisition module 210 and is used to analyze the internal heterogeneity of glioma based on spectral differences and to assess the coverage benefit of the preoperative planned path and the intraoperative bypass path on the central area, infiltrative area and peripheral area of ​​glioma.

[0087] The ischemia risk assessment module 230 is connected to the coverage benefit assessment module 220, which is used to analyze the blood oxygen saturation distribution based on the absorption peak characteristics of hemoglobin in a specific band, and to assess the local ischemia risk of the preoperative planned path and the intraoperative bypass path.

[0088] The damage risk assessment module 240 is connected to the ischemia risk assessment module 230, which is used to map the preoperative planned path and the intraoperative detour path to the brain functional area topology network respectively, and assess the functional damage risk corresponding to the path based on the topological connectivity and local ischemia risk.

[0089] The decision suggestion generation module 250 is connected to the damage risk assessment module 240. It is used to perform expected benefit and risk trend analysis on the preoperative planned path and the intraoperative bypass path and generate surgical path decision suggestions.

[0090] In this embodiment of the invention, the coverage benefit assessment module 220 is used to analyze the internal heterogeneity of gliomas based on spectral difference analysis, and to assess the coverage benefit of preoperative planned paths and intraoperative bypass paths on the central area, infiltrative area, and peripheral area of ​​the glioma, including:

[0091] The hyperspectral imaging data was preprocessed and glioma characteristic bands were extracted. The preprocessing included illumination correction and noise removal.

[0092] Based on the characteristic bands of gliomas, gliomas are divided into subregions, including the central region, infiltrative region, and peripheral region of the glioma.

[0093] The spatial coverage of the preoperative planned path and the intraoperative bypass path in the central area, infiltrative area and peripheral area of ​​the glioma were calculated respectively.

[0094] The spectral heterogeneity weights of the central, infiltrative, and peripheral regions of gliomas are assessed based on differences in spectral characteristics. The spectral heterogeneity weights are the mean standard deviations of each characteristic band within the region.

[0095] The path coverage benefit coefficient of the preoperative planned path and the intraoperative detour path is obtained by weighting and summing the spatial coverage by spectral heterogeneity weights.

[0096] In this embodiment of the invention, the ischemia risk assessment module 230 is used to analyze blood oxygen saturation distribution based on the absorption peak characteristics of hemoglobin in a specific wavelength band, and to assess the local ischemia risk of the preoperative planned path and the intraoperative bypass path, including:

[0097] The absorption characteristics of oxyhemoglobin and deoxyhemoglobin in characteristic bands were extracted based on hyperspectral imaging data, and the tissue oxygen saturation at each pixel location was determined by optical density difference method.

[0098] Mark the pixel locations where tissue blood oxygen saturation is lower than a preset tissue blood oxygen saturation threshold and construct a tissue ischemia risk region;

[0099] The local ischemia risk coefficient is obtained by compensatory correction of the tissue ischemia risk coefficient in the tissue ischemia risk area and a local ischemia risk distribution map is constructed. The tissue ischemia risk coefficient represents the degree of deviation of tissue blood oxygen saturation from the preset tissue blood oxygen saturation threshold.

[0100] The preoperative planned path and the intraoperative detour path were mapped to the local ischemia risk distribution map, and the local ischemia risk coefficient at each pixel position of the path was extracted.

[0101] In this embodiment of the invention, the injury risk assessment module 240 is used to map the preoperative planned path and the intraoperative detour path to the topological network of brain functional areas, and to assess the functional impairment risk corresponding to the path based on the topological connectivity and the risk of local ischemia, including:

[0102] The average Euclidean distance between functional nodes and paths in the topological network of brain functional areas is extracted, and the distance coupling coefficient between functional nodes and paths is determined using the distance decay function.

[0103] The local ischemia risk coefficient at each pixel position of the preoperative planned path and the intraoperative detour path is mapped to the functional nodes of the brain functional area topology network, and the product of the local ischemia risk coefficient and the distance coupling coefficient is used as the node functional damage risk coefficient.

[0104] The functional impairment risk coefficient of the path is obtained by using the betweenness centrality of each functional node in the brain functional area topology network as the node connection weight and by weighting and summing the node functional impairment risk coefficients of all nodes using the node connection weights.

[0105] In this embodiment of the invention, the decision suggestion generation module 250 is used to perform expected benefit and risk trend analysis on the preoperative planned path and the intraoperative detour path and generate surgical path decision suggestions, including:

[0106] Obtain the path coverage benefit coefficient and functional impairment risk coefficient corresponding to the preoperative planned path and the intraoperative detour path;

[0107] The difference between the path coverage benefit coefficient and the functional damage risk coefficient of the preoperative planned path is used as the benchmark net benefit coefficient, and the difference between the path coverage benefit coefficient and the functional damage risk coefficient of the intraoperative detour path is used as the detour net benefit coefficient.

[0108] When the detour net benefit coefficient is greater than the benchmark net benefit coefficient, the intraoperative detour path is taken as the preferred surgical path; when the detour net benefit coefficient is less than or equal to the benchmark net benefit coefficient, the preoperative planned path is taken as the preferred surgical path.

[0109] The parameters and steps for implementing the corresponding functions of each unit module in the intraoperative glioma deep learning detection system based on hyperspectral imaging of the present invention can be referred to the parameters and steps in the embodiments of the intraoperative glioma deep learning detection method based on hyperspectral imaging described above, and will not be repeated here.

[0110] Please refer to Figure 3 The present invention also provides an electronic device 300, including a memory 310, a processor 320, and a communication bus 330; the memory 310 and the processor 320 are connected via the communication bus 330. The memory 310 stores a method for intraoperative glioma detection based on hyperspectral imaging, which can be loaded and executed by the processor 320 as provided in the above embodiments.

[0111] The memory 310 can be used to store instructions, programs, code, code sets, or instruction sets. The memory 310 may include a program storage area and a data storage area. The program storage area may store instructions for implementing an operating system, instructions for at least one function, and instructions for implementing the intraoperative glioma deep learning detection method based on hyperspectral imaging provided in the above embodiments, etc. The data storage area may store data involved in the intraoperative glioma deep learning detection method based on hyperspectral imaging provided in the above embodiments, etc.

[0112] Processor 320 may include one or more processing cores. Processor 320 executes instructions, programs, code sets, or instruction sets stored in memory 310, and calls data stored in memory 310 to perform various functions and process data according to the present invention. Processor 320 may be at least one of the following: Application Specific Integrated Circuit (ASIC), Digital Signal Processor (DSP), Digital Signal Processing Device (DSPD), Programmable Logic Device (PLD), Field Programmable Gate Array (FPGA), Central Processing Unit (CPU), Controller, Microcontroller, and Microprocessor. It is understood that, for different devices, the electronic devices used to implement the functions of processor 320 may also be other types, and the embodiments of the present invention do not specifically limit this.

[0113] The communication bus 330 may include a path for transmitting information between the aforementioned components. The communication bus 330 may be a PCI (Peripheral Component Interconnect) bus or an EISA (Extended Industry Standard Architecture) bus, etc. The communication bus 330 can be divided into an address bus, a data bus, a control bus, etc. For ease of representation, Figure 3 The symbol is represented by a single double arrow, but this does not mean that there is only one bus or one type of bus.

[0114] This invention provides a computer-readable storage medium storing a computer program that can be loaded by a processor and executed as described in the above embodiments for the intraoperative glioma deep learning detection method based on hyperspectral imaging.

[0115] In this embodiment of the invention, the computer-readable storage medium can be a tangible device that holds and stores instructions used by an instruction execution device. The computer-readable storage medium can be, but is not limited to, an electrical storage device, a magnetic storage device, an optical storage device, an electromagnetic storage device, a semiconductor storage device, or any combination thereof. Specifically, the computer-readable storage medium can be a portable computer disk, a hard disk, a USB flash drive, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), lectern random access memory (SRAM), portable compact disc read-only memory (CD-ROM), digital multifunction disc (DVD), memory stick, floppy disk, optical disk, magnetic disk, mechanical encoding device, or any combination thereof.

[0116] The terms “comprising,” “including,” or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article, or apparatus.

[0117] The above description is merely a preferred embodiment of the present invention and an explanation of the technical principles employed. Those skilled in the art should understand that the scope of this application is not limited to the technical solutions formed by specific combinations of the above-described technical features, but should also cover other technical solutions formed by arbitrary combinations of the above-described technical features or their equivalents without departing from the foregoing concept. For example, technical solutions formed by substituting the above-described features with (but not limited to) technical features with similar functions claimed in this invention.

Claims

1. A deep learning-based method for intraoperative glioma detection using hyperspectral imaging, characterized in that, include: Acquire hyperspectral imaging data of the preoperative planned path and intraoperative detour path, topological network of brain functional areas, and preoperative vascular imaging data; Based on spectral difference analysis of the internal heterogeneity of gliomas, the benefits of preoperative planning and intraoperative bypass routes in covering the central, infiltrative, and peripheral areas of gliomas were evaluated. Based on the absorption peak characteristics of hemoglobin in a specific band, the distribution of blood oxygen saturation was analyzed to assess the local ischemic risk of the preoperative planned path and the intraoperative detour path. The preoperative planned path and the intraoperative detour path were mapped to the topological network of brain functional areas, and the risk of functional impairment corresponding to the path was assessed based on the topological connectivity and the risk of local ischemia. Expected benefit and risk trend analysis is performed on the preoperative planned path and the intraoperative detour path, and surgical path decision recommendations are generated. The method of acquiring hyperspectral imaging data includes: acquiring hyperspectral imaging data of the exposed brain tissue region during surgery to obtain the real spectral features within the path coverage area; and predicting the spectral features of the deep or obscured areas that have not yet been exposed using a deep learning model trained based on preoperative imaging data, thereby completing the hyperspectral imaging data of the entire path. The assessment of the local ischemia risk of the preoperative planned path and the intraoperative bypass path includes: The absorption characteristics of oxyhemoglobin and deoxyhemoglobin in characteristic bands were extracted based on hyperspectral imaging data, and the tissue oxygen saturation at each pixel location was determined by optical density difference method. Mark the pixel locations where tissue blood oxygen saturation is lower than a preset tissue blood oxygen saturation threshold and construct a tissue ischemia risk region; The local ischemia risk coefficient is obtained by compensatory correction of the tissue ischemia risk coefficient in the tissue ischemia risk area and a local ischemia risk distribution map is constructed. The tissue ischemia risk coefficient represents the degree of deviation of tissue blood oxygen saturation from the preset tissue blood oxygen saturation threshold. The preoperative planned path and the intraoperative detour path were respectively mapped to the local ischemia risk distribution map, and the local ischemia risk coefficient at each pixel position of the path was extracted. The functional impairment risks corresponding to the assessment path include: The average Euclidean distance between functional nodes and paths in the topological network of brain functional areas is extracted, and the distance coupling coefficient between functional nodes and paths is determined using the distance decay function. The local ischemia risk coefficient at each pixel position of the preoperative planned path and the intraoperative detour path is mapped to the functional nodes of the brain functional area topology network, and the product of the local ischemia risk coefficient and the distance coupling coefficient is used as the node functional damage risk coefficient. The functional impairment risk coefficient of the path is obtained by using the betweenness centrality of each functional node in the brain functional area topology network as the node connection weight and by weighting and summing the node functional impairment risk coefficients of all nodes using the node connection weights.

2. The intraoperative deep learning detection method for gliomas based on hyperspectral imaging according to claim 1, characterized in that, The assessment of the coverage benefits includes: The hyperspectral imaging data was preprocessed and glioma characteristic bands were extracted. The preprocessing included illumination correction and noise removal. Based on the characteristic bands of gliomas, gliomas are divided into subregions, including the central region, infiltrative region, and peripheral region of the glioma. The spatial coverage of the preoperative planned path and the intraoperative bypass path in the central area, infiltrative area and peripheral area of ​​the glioma were calculated respectively. The spectral heterogeneity weights of the central, infiltrative, and peripheral regions of gliomas are assessed based on differences in spectral characteristics. The spectral heterogeneity weights are the mean standard deviations of each characteristic band within the region. The path coverage benefit coefficient of the preoperative planned path and the intraoperative detour path is obtained by weighting and summing the spatial coverage by spectral heterogeneity weights.

3. The intraoperative deep learning detection method for gliomas based on hyperspectral imaging according to claim 1, characterized in that, The method of obtaining a local ischemia risk coefficient by compensatory correction of the tissue ischemia risk coefficient within the tissue ischemia risk area includes: Preoperative vascular imaging data were acquired and the collateral circulation compensation capacity was graded using the modified ASITN / SIR collateral circulation scoring system. The sigmoid function is used to convert the lateral circulation compensation capacity into compensation redundancy weights. The tissue ischemia risk coefficient within the tissue ischemia risk region is attenuated and corrected based on the compensatory redundancy weight, and the corrected tissue ischemia risk coefficient is used as the local ischemia risk coefficient.

4. The intraoperative deep learning detection method for gliomas based on hyperspectral imaging according to claim 1, characterized in that, The process of analyzing the expected benefits and risks of preoperative planned routes and intraoperative bypass routes, and generating surgical route decision recommendations, includes: Obtain the path coverage benefit coefficient and functional impairment risk coefficient corresponding to the preoperative planned path and the intraoperative detour path; The difference between the path coverage benefit coefficient and the functional damage risk coefficient of the preoperative planned path is used as the benchmark net benefit coefficient, and the difference between the path coverage benefit coefficient and the functional damage risk coefficient of the intraoperative detour path is used as the detour net benefit coefficient. When the detour net benefit coefficient is greater than the benchmark net benefit coefficient, the intraoperative detour path is taken as the preferred surgical path; when the detour net benefit coefficient is less than or equal to the benchmark net benefit coefficient, the preoperative planned path is taken as the preferred surgical path.

5. A deep learning detection system for intraoperative gliomas based on hyperspectral imaging, used to implement the deep learning detection method for intraoperative gliomas based on hyperspectral imaging as described in any one of claims 1-4, characterized in that, The system includes: The data acquisition module is used to acquire hyperspectral imaging data of the preoperative planned path and the intraoperative detour path, the topological network of brain functional areas, and preoperative vascular imaging data. The coverage benefit assessment module is used to analyze the internal heterogeneity of gliomas based on spectral differences and to evaluate the coverage benefit of preoperative planned paths and intraoperative bypass paths on the central, infiltrated, and peripheral areas of gliomas. The ischemia risk assessment module is used to analyze blood oxygen saturation distribution based on the absorption peak characteristics of hemoglobin in a specific band, and to assess the local ischemia risk of the preoperative planned path and the intraoperative bypass path. The damage risk assessment module is used to map the preoperative planned path and the intraoperative detour path to the topological network of brain functional areas and assess the functional damage risk corresponding to the path based on the topological connectivity and local ischemia risk. The decision suggestion generation module is used to perform expected benefit and risk trend analysis on the preoperative planned path and the intraoperative detour path, and generate surgical path decision suggestions.

6. An electronic device, comprising: A processor and a memory, wherein the memory stores a computer program that can be called by the processor; characterized in that the processor executes the intraoperative glioma deep learning detection method based on hyperspectral imaging as described in any one of claims 1-4 by calling the computer program stored in the memory.

7. A computer-readable storage medium, characterized in that, The device stores instructions that, when executed on a computer, cause the computer to perform the intraoperative glioma deep learning detection method based on hyperspectral imaging as described in any one of claims 1-4.

Images