A lung cancer type prediction method and system based on fusion deep learning network

Abstract

The present application provides a lung cancer type prediction method and system based on a fusion deep learning network. It dynamically segments CT images of lung cancer patients, extracts tumor vascular distribution data and calculates three-dimensional density parameters and vascular rupture direction parameters, and simultaneously obtains a necrotic area volume ratio sequence; separates circulating tumor cells through spiral sorting of blood samples and fluid control, and generates stiffness change parameters and deformation recovery parameters derived from deformation trajectories; combines three-dimensional density, vascular rupture direction and stiffness parameters to extract the vascular rupture direction time series offset, and dynamically and synchronously matches the stiffness change response to generate a synchronization set; constructs a lag correlation between the vascular rupture direction offset and the deformation recovery period, and determines the weight matching through cross-modal analysis; combines the spatiotemporal characteristics of the necrotic area volume to output the lung cancer subtype classification results. The present application integrates a cross-modal correlation model of vascular dynamic offset and cell mechanical response to achieve accurate discrimination of lung cancer subtypes.

Description

Technical Field

[0001] The present application relates to the technical field of lung cancer type prediction, and in particular to a lung cancer type prediction method and system based on a fusion deep learning network. Background Art

[0002] Lung cancer is one of the malignant tumors with the highest morbidity and mortality rates worldwide. Its early diagnosis and accurate classification are crucial for patient treatment and prognosis. Traditional lung cancer classification relies primarily on pathological examinations, but this method is subject to significant subjectivity, time-consuming, and limited ability to identify early lesions. With the rapid development of medical imaging technology and artificial intelligence, deep learning-based lung cancer classification prediction methods have become a research hotspot. This type of technology can automatically extract features from medical images, assisting doctors in making rapid and accurate diagnoses, thereby meeting the clinical need for efficient and intelligent diagnostic tools.

[0003] Currently, deep learning-based lung cancer type prediction methods primarily employ single network models, such as convolutional neural networks, recurrent neural networks, or graph convolutional networks. These methods classify lung cancer types by extracting spatial or temporal features from medical images. For example, convolutional neural networks are widely used for lung cancer detection and classification in CT images, as they can capture local image features. Recurrent neural networks are used to process dynamic image data and extract time series information.

[0004] While existing methods have made some progress in lung cancer type prediction, the following issues remain: Single network models often only capture a specific set of features (such as spatial or temporal features), failing to fully reflect the multidimensional nature of lung cancer, which limits classification accuracy. Furthermore, simple model combination methods lack effective feature fusion mechanisms, easily introducing redundant information or overlooking key features, impacting model performance. Summary of the Invention

[0005] The present application provides a lung cancer type prediction method and system based on a fusion deep learning network to solve the problem of insufficient fusion of multimodal dynamic features in lung cancer classification in the prior art, resulting in low accuracy.

[0006] In a first aspect, the present application provides a method for predicting lung cancer types based on a fusion deep learning network, comprising:

[0007] Obtaining CT images and blood samples from lung cancer patients, dynamically segmenting the CT images to extract vascular distribution data of the tumor region, calculating three-dimensional density parameters and vascular rupture direction parameters based on the vascular distribution data, and obtaining a volume percentage sequence of the tumor necrotic region;

[0008] performing spiral sorting on the blood sample, separating circulating tumor cells through fluid control, collecting deformation trajectory data of the circulating tumor cells, and generating stiffness change parameters and deformation recovery parameters based on the deformation trajectory data;

[0009] jointly analyzing the three-dimensional density parameter, the blood vessel fracture direction parameter, and the stiffness change parameter, extracting a time series offset of the blood vessel fracture direction parameter, and synchronously matching the time series offset with a dynamic response of the stiffness change parameter to generate a synchronous parameter set;

[0010] Based on the synchronization parameter set, a hysteresis correlation relationship is established between the time series offset of the blood vessel rupture direction parameter and the recovery period of the deformation recovery parameter, so as to determine a weight matching result between the two through cross-modal analysis;

[0011] Outputting a lung cancer subtype classification result based on the weight matching result and the spatiotemporal distribution characteristics of the tumor necrosis area volume ratio.

[0012] Optionally, dynamically segmenting the CT image to extract vascular distribution data of the tumor area, calculating three-dimensional density parameters and vascular rupture direction parameters based on the vascular distribution data, and obtaining a volume percentage sequence of the tumor necrosis area includes:

[0013] Performing a multi-time point segmentation operation on the CT image to extract vascular distribution data of the tumor area at each time point, wherein the vascular distribution data includes the three-dimensional coordinates of vascular branch nodes and the connection status between adjacent nodes;

[0014] Calculating the density distribution of blood vessels in three-dimensional space based on the three-dimensional coordinates of the blood vessel branch nodes, and defining the number of blood vessel branch nodes per unit volume as a three-dimensional density parameter;

[0015] Count the fracture directions of all vascular branch nodes and generate the vascular fracture direction parameters according to the angular distribution of the fracture directions in the three-dimensional coordinate system;

[0016] The volume fraction series was generated by measuring the ratio of the volume of the tumor necrotic area to the total volume of the tumor area at each time point.

[0017] Optionally, the blood sample is subjected to spiral sorting processing to separate circulating tumor cells by fluid control, deformation trajectory data of the circulating tumor cells are collected, and stiffness change parameters and deformation recovery parameters are generated based on the deformation trajectory data, including:

[0018] injecting the blood sample into a spiral sorting device to separate circulating tumor cells from non-tumor cells by adjusting the direction of fluid dynamics;

[0019] Collecting movement trajectory data of the separated circulating tumor cells in the fluid environment, wherein the movement trajectory data includes the deformation amplitude and deformation direction of the cells at different time points;

[0020] Calculating the recovery speed after the deformation amplitude reaches a maximum value based on the curve of the deformation amplitude changing with time, and defining the reciprocal of the recovery speed as the deformation recovery parameter;

[0021] The frequency of changes in the deformation direction in the moving trajectory is counted, and the product of the frequency and the deformation amplitude is defined as a stiffness change parameter.

[0022] Optionally, the three-dimensional density parameter, the vascular rupture direction parameter, and the stiffness change parameter are jointly analyzed to extract a time series offset of the vascular rupture direction parameter, and the time series offset is synchronously matched with a dynamic response of the stiffness change parameter to generate a synchronous parameter set, including:

[0023] Obtaining direction angle values of the blood vessel rupture direction parameter at multiple time points to calculate angle differences between adjacent time points, and generating a time series offset according to the angle differences;

[0024] Obtaining parameter values of the stiffness variation parameters at the same time point, and extracting dynamic variation values of the parameter values over time;

[0025] The time series offset and the dynamic change value are aligned on a time axis, and the sum of the products of the two at the same time point is calculated to generate a synchronization parameter set.

[0026] Optionally, based on the synchronization parameter set, a hysteresis correlation relationship is established between the time series offset of the blood vessel rupture direction parameter and the recovery period of the deformation recovery parameter, so as to determine a weight matching result between the two through cross-modal analysis, including:

[0027] Counting the time delay between the time point at which the time series offset reaches a peak value and the time point at which the recovery period ends, to generate a lagged time series;

[0028] Calculating the association weight between the time series offset and the recovery period according to the length distribution of the lagged time series;

[0029] The association weight and the sum of the products in the synchronization parameter set are superimposed to generate a weight matching result.

[0030] Optionally, outputting a lung cancer subtype classification result based on the weight matching result and the spatiotemporal distribution characteristics of the tumor necrosis area volume ratio includes:

[0031] Extracting the distribution characteristics of the volume proportion values in the volume proportion sequence as they change with spatial positions, and associating the spatial positions with time points to generate spatiotemporal distribution characteristics;

[0032] The weight matching result is mapped and matched with the spatiotemporal distribution feature, and the corresponding lung cancer subtype classification result is output according to a preset threshold interval.

[0033] Optionally, mapping and matching the weight matching result with the spatiotemporal distribution feature, and outputting a corresponding lung cancer subtype classification result according to a preset threshold interval, includes:

[0034] Obtaining the parameter value of each time point in the weighted matching result and the spatial position ratio value of the corresponding time point in the spatiotemporal distribution feature;

[0035] Superimposing the parameter value and the proportion value at the same time point to generate a comprehensive parameter at each time point;

[0036] Count the maximum and minimum values of the comprehensive parameters at all time points to generate dynamic parameter intervals;

[0037] The dynamic parameter interval is compared with a preset threshold interval. If the dynamic parameter interval completely falls within any threshold interval, the lung cancer subtype classification result corresponding to the threshold interval is output.

[0038] In a second aspect, the present application provides a lung cancer type prediction system based on a fusion deep learning network, comprising:

[0039] an acquisition module, configured to acquire CT images and blood samples from lung cancer patients, dynamically segment the CT images to extract vascular distribution data of the tumor region, calculate three-dimensional density parameters and vascular rupture direction parameters based on the vascular distribution data, and obtain a volume percentage sequence of the tumor necrosis region;

[0040] a processing module, configured to perform spiral sorting processing on the blood sample, separate circulating tumor cells through fluid control, collect deformation trajectory data of the circulating tumor cells, and generate a stiffness change parameter and a deformation recovery parameter based on the deformation trajectory data;

[0041] an analysis module, configured to jointly analyze the three-dimensional density parameter, the vascular rupture direction parameter, and the stiffness change parameter, extract a time series offset of the vascular rupture direction parameter, and synchronously match the time series offset with a dynamic response of the stiffness change parameter to generate a synchronization parameter set;

[0042] an establishing module for establishing, based on the synchronization parameter set, a hysteresis correlation relationship between the time series offset of the blood vessel rupture direction parameter and the recovery period of the deformation recovery parameter, so as to determine a weight matching result between the two through cross-modal analysis;

[0043] An output module is used to output a lung cancer subtype classification result based on the weight matching result and the spatiotemporal distribution characteristics of the tumor necrosis area volume ratio.

[0044] In a third aspect, an embodiment of the present application provides a computing device comprising a processing component and a storage component; the storage component stores one or more computer instructions; the one or more computer instructions are used to be called and executed by the processing component to implement a multi-dimensional data processing method for intelligent scoring of emergency patients as described in the first aspect above.

[0045] In a fourth aspect, an embodiment of the present application provides a computer storage medium storing a computer program. When the computer program is executed by a computer, it implements a multi-dimensional data processing method for intelligent scoring of emergency patients as described in the first aspect.

[0046] This application obtains CT images and blood samples from lung cancer patients, dynamically segments the CT images to extract vascular distribution data of the tumor area, calculates three-dimensional density parameters and vascular rupture direction parameters based on the vascular distribution data, and obtains a volume percentage sequence of the tumor necrosis area. It can integrate imaging genomics with microvascular spatial heterogeneity characteristics to achieve quantitative characterization of the tumor core necrosis volume and multi-dimensional analysis of the three-dimensional vascular network mechanical properties.

[0047] By performing spiral sorting on the blood sample, separating circulating tumor cells through fluid control, collecting deformation trajectory data of the circulating tumor cells, and generating stiffness change parameters and deformation recovery parameters based on the deformation trajectory data, it is possible to accurately capture the dynamic characteristics of the mechanical response of circulating tumor cells, eliminate interference from non-tumor cells, and establish a dynamic mapping between cell stiffness gradient and tumor invasion behavior.

[0048] By jointly analyzing the three-dimensional density parameters, the vascular rupture direction parameters and the stiffness change parameters, extracting the time series offset of the vascular rupture direction parameters, and synchronously matching the time series offset with the dynamic response of the stiffness change parameters, a synchronous parameter set is generated. This can achieve precise alignment of the time axis of the dynamic characteristics of the image and the cellular mechanical behavior, solve the problem of temporal faults in multimodal data and enhance the temporal and spatial correlation consistency of cross-scale features.

[0049] By establishing a hysteresis correlation between the time series offset of the vascular rupture direction parameter and the recovery period of the deformation recovery parameter based on the synchronization parameter set, and determining the weight matching result of the two through cross-modal analysis, the biomechanical coupling mechanism of the dynamic evolution of microvascular rupture direction and the mechanical recovery behavior of tumor cells can be revealed, providing a quantitative discrimination basis for cross-modal association for subtype classification.

[0050] By outputting the lung cancer subtype classification results based on the weight matching results and the spatiotemporal distribution characteristics of the volume proportion of the tumor necrosis area, it is possible to integrate the multidimensional characteristics of vascular network heterogeneity, cell mechanics dynamic response and spatiotemporal distribution of necrotic areas, and construct a lung cancer subtype identification model based on cross-modal deep correlation and spatiotemporal weight matching, significantly improving the pathological interpretability and predictive accuracy of the classification results.

[0051] Furthermore, by dynamically segmenting CT images at multiple time points and extracting the three-dimensional coordinates and connection status of vascular branch nodes, the spatiotemporal evolution characteristics of the vascular network in the tumor area can be accurately captured, providing high-precision spatial topological data for subsequent quantitative analysis; calculating the node density within a unit volume based on the three-dimensional coordinates of vascular branch nodes can effectively characterize the spatial heterogeneous distribution characteristics of blood vessels in the tumor microenvironment and reveal the association between angiogenesis and tumor aggressiveness; by statistically analyzing the three-dimensional angular distribution of vascular fracture directions, the mechanical properties of vascular structural damage can be quantified, providing important parameters for evaluating the mechanical properties of the tumor microenvironment; combined with time series analysis of the volume proportion of tumor necrosis areas, dynamic monitoring of the tumor necrosis process can be achieved, providing a quantitative discrimination basis for the identification of lung cancer subtypes that combines spatial distribution characteristics and temporal evolution laws, significantly improving the accuracy and reliability of diagnostic results.

[0052] These and other aspects of the present application will become more readily apparent from the description of the following embodiments. BRIEF DESCRIPTION OF THE DRAWINGS

[0053] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, a brief introduction will be given below to the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without any creative work.

[0054] Figure 1 A flowchart of a lung cancer type prediction method based on a fusion deep learning network provided by the present application is shown;

[0055] Figure 2 A scene diagram of a circulating tumor cell microfluidic analysis process provided by the present application is shown;

[0056] Figure 3 A flowchart of a dynamic imaging assessment process for lung adenocarcinoma provided by the present application is shown;

[0057] Figure 4 A scene diagram of a multimodal intelligent diagnosis process for lung adenocarcinoma provided by the present application is shown;

[0058] Figure 5 A schematic diagram of the structure of a lung cancer type prediction system based on a fusion deep learning network provided by the present application is shown. DETAILED DESCRIPTION

[0059] In order to enable those skilled in the art to better understand the solution of the present application, the technical solution in the embodiments of the present application will be clearly and completely described below in conjunction with the drawings in the embodiments of the present application.

[0060] In some of the processes described in the specification and claims of this application and the above-mentioned figures, multiple operations that appear in a specific order are included, but it should be clearly understood that these operations may not be executed in the order in which they appear in this document or may be executed in parallel. The serial numbers of the operations, such as 101, 102, etc., are only used to distinguish between different operations, and the serial numbers themselves do not represent any order of execution. In addition, these processes may include more or fewer operations, and these operations may be executed in sequence or in parallel. It should be noted that the descriptions of "first", "second", etc. in this document are used to distinguish different messages, devices, modules, etc., and do not represent a sequential order, nor do they limit "first" and "second" to being different types.

[0061] The researchers found that the existing lung cancer subtype classification method lacks cross-modal collaborative analysis of dynamic segmentation of CT images and mechanical response of blood samples, resulting in insufficient synchronization between vascular heterogeneity parameters and the dynamic behavior of circulating tumor cells, and the lagging correlation between parameters has not been effectively analyzed. Based on this, a lung cancer subtype classification method based on spatiotemporal dynamic fusion is provided. This method can achieve accurate modeling of subtype classification by matching the cross-modal weights of vascular rupture direction and cell mechanical parameters. The technical solution of this application can be applied to early screening of lung cancer, dynamic evaluation of tumor heterogeneity and prediction of targeted therapy response.

[0062] The entire R&D process embodies the fusion mechanism of synchronous modeling of multimodal dynamic parameters and spatiotemporal lag correlation analysis, aiming to overcome the limitations of existing solutions such as the separation of image and blood parameters, dynamic response mismatch, and ambiguous cross-modal weight distribution. Through the coordination of CT image dynamic segmentation and spiral sorting fluid control, the spatiotemporal matching problem of tumor vascular heterogeneity parameters and the mechanical behavior of circulating tumor cells is solved; based on the synchronous analysis of time series offset and dynamic response, the one-sided dependence of a single modality parameter on subtype classification is eliminated; combined with lag correlation modeling and weight matching optimization, a breakthrough is made in the traditional method's neglect of dynamic phase differences between parameters; and finally, through the joint decision-making of the spatiotemporal distribution of tumor necrosis volume and cross-modal weights, a systematic improvement in the robustness, dynamics, and clinical interpretability of lung cancer subtype classification is achieved.

[0063] The following will be combined with the drawings in the embodiments of this application to clearly and completely describe the technical solutions in the embodiments of this application. Obviously, the embodiments described are only part of the embodiments of this application, not all of the embodiments. Based on the embodiments in this application, all other embodiments obtained by those skilled in the art without making creative efforts are within the scope of protection of this application.

[0064] Figure 1 A flowchart of a lung cancer type prediction method based on a fusion deep learning network is provided for the embodiment of this application. Figure 1 As shown, the method includes:

[0065] 101. Obtain a CT image and a blood sample from a lung cancer patient, dynamically segment the CT image to extract vascular distribution data of the tumor region, calculate a three-dimensional density parameter and a vascular rupture direction parameter based on the vascular distribution data, and obtain a volume percentage sequence of the tumor necrosis region;

[0066] In this step, CT images refer to medical images acquired using computed tomography. Dynamic segmentation refers to the process of continuously segmenting images over time. Vascular distribution data refers to quantitative data describing the vascular network structure in the tumor region. 3D density parameters refer to the density characteristics of vascular distribution in 3D space. Vascular rupture direction parameters refer to the spatial directional characteristics of vascular ruptures. Volume fraction series refer to the volume ratio data of necrotic tumor regions that change over time.

[0067] In the embodiment of the present application, first, CT images of lung cancer patients are obtained from the hospital imaging system, and blood samples of the patients are collected for subsequent analysis. Secondly, a deep learning segmentation algorithm (such as 3D U-Net) is used to dynamically segment the CT images, identify the tumor area layer by layer, and extract vascular distribution data, including vascular density, direction, and branching characteristics. Then, based on the segmentation results, three-dimensional density parameters (such as the density change rate of the X / Y / Z axis) and vascular fracture direction parameters (such as the angular distribution of the fracture) are calculated. Finally, the volume percentage sequence of the tumor necrosis area is statistically analyzed by the voxel counting method, and the changes in the necrotic area at different time points are recorded to establish an imaging basis for subsequent multimodal analysis.

[0068] For example, at the hospital's tumor diagnosis and treatment center, a 62-year-old female patient was admitted due to a space-occupying lesion in the lungs. An enhanced CT scan showed irregular enhancement lesions in the left upper lobe of the lung. After obtaining her chest CT images and peripheral blood samples, the medical team used a dynamic segmentation algorithm to analyze the tumor area layer by layer and extracted the three-dimensional distribution data of the microvessels inside the lesion. Calculations showed that the three-dimensional density parameters of the blood vessels in the leading edge of the tumor showed a gradient-increasing feature, while the rupture direction parameters of multiple blood vessels were detected in the dorsal area, showing that the vascular network formed an acute angle intersection with the pleural plane. The synchronously generated tumor necrosis area volume ratio sequence showed that the necrotic volume in the central area of the lesion expanded in a step-by-step manner over time, which was highly correlated with the patient's progressive dyspnea symptoms. The three-dimensional reconstructed model clearly presents the spatial correspondence between the vascular rupture direction parameters and the boundary of the necrotic area.

[0069] 102. Perform spiral sorting on the blood sample to separate circulating tumor cells through fluid control, collect deformation trajectory data of the circulating tumor cells, and generate stiffness change parameters and deformation recovery parameters based on the deformation trajectory data;

[0070] In this step, spiral sorting refers to the technology used to separate cells using spiral microfluidic channels. Fluid control refers to the method of regulating cell movement through fluid dynamics parameters. Circulating tumor cells are tumor cells that have entered the peripheral blood circulation from the primary tumor. Deformation trajectory data refers to the time-series record of cell movement and deformation in fluid. The stiffness change parameter is a quantitative indicator of the change in cell stiffness in response to external force. The deformation recovery parameter refers to the dynamic characteristics of the cell's recovery to its original shape after deformation.

[0071] In the embodiment of this application, Figure 2As shown, a blood sample is first injected into a microfluidic chip, where circulating tumor cells are isolated through spiral sorting using a centrifugal force field and fluid control technology. Next, a high-speed microscopy system (1000 frames per second) is used to capture the movement of circulating tumor cells in the spiral channel, collecting deformation trajectory data of the cells under compression and rebound. Next, mechanical models (such as the Hertz contact model) are used to analyze these deformation trajectory data, calculate the change in elastic modulus of the cells under compression, and generate a stiffness change parameter that reflects the cell's hardness characteristics. Finally, the cell rebound curve is fitted to extract deformation recovery parameters (such as the time required to recover 80% of the deformation), fully characterizing the dynamic mechanical response characteristics of the cells.

[0072] For example, the patient's peripheral blood sample was processed by a spiral sorting device, and the blood cells were introduced into the spiral microfluidic channel through fluid control technology. Taking advantage of the inertia difference between tumor cells and blood cells, the system successfully separated five circulating tumor cells with cytokeratin markers. The high-speed microscopy imaging system recorded the deformation trajectory data of these cells as they passed through the narrow section of the spiral path: two of the cells showed continuous deformation under the flow field pressure, and the deformation recovery parameters were significantly lower than the threshold; the other three showed rapid elastic rebound characteristics. Further analysis showed that the fluctuation amplitude of the stiffness change parameters of abnormally deformed cells was more than twice the normal value, and the stiffness change parameter curve had a similar oscillation frequency to the blood vessel rupture direction parameters in CT images, suggesting a potential correlation between mechanical properties and the tumor microenvironment.

[0073] 103. Jointly analyze the three-dimensional density parameter, the blood vessel rupture direction parameter, and the stiffness change parameter to extract a time series offset of the blood vessel rupture direction parameter, and synchronously match the time series offset with a dynamic response of the stiffness change parameter to generate a synchronous parameter set;

[0074] In this step, joint analysis refers to the integration of multi-source data for comprehensive analysis. Time series offset refers to the phase difference between the time-varying parameters of the vascular rupture direction. Dynamic response refers to the real-time reaction of the stiffness parameter to external force stimulation. Synchronous matching refers to the process of aligning data from different time dimensions. The synchronized parameter set refers to the multimodal feature set after time alignment.

[0075] In this embodiment, the three-dimensional density parameters and vessel fracture direction parameters from step 101 are first spatiotemporally aligned with the stiffness change parameters from step 102 to construct a multimodal joint analysis matrix. Next, the time series offset of the vessel fracture direction parameters (e.g., the time delay of the vessel fracture event relative to the baseline) is detected using a dynamic time warping (DTW) algorithm. This offset is then matched to the dynamic response of the stiffness change parameters, and a phase synchronization algorithm is employed to eliminate timescale differences. This generates a synchronized parameter set containing vessel fracture time, stiffness response intensity, and delay time, establishing a bridge between imaging features and mechanical properties.

[0076] For example, the time series of vessel rupture direction parameters extracted from CT images (collected every four hours) were jointly analyzed with the stiffness change parameters of circulating tumor cells. Using a time phase alignment algorithm, it was found that when the vessel rupture direction parameters deflected southward, the stiffness change parameters of circulating tumor cells at the corresponding time node would synchronously rise in a pulsed manner. The synchronized parameter set showed that this dynamic response became particularly significant when the volume of the tumor necrosis area accounted for more than 40 percent. Of particular note, the deformation recovery parameters of circulating tumor cells corresponding to areas with higher three-dimensional vascular density parameters remained at a low level, revealing that the density of blood vessels within the tumor may affect the mechanical properties of tumor cells in the peripheral blood through a mechanical transduction mechanism.

[0077] 104. Based on the synchronization parameter set, establish a hysteresis correlation between the time series offset of the blood vessel rupture direction parameter and the recovery period of the deformation recovery parameter, so as to determine a weight matching result between the two through cross-modal analysis;

[0078] In this step, the hysteresis relationship refers to the time-delayed correspondence between different parameters. The recovery period refers to the time it takes for a deformation recovery parameter to complete a recovery process. Cross-modal analysis refers to an analytical method that integrates data from different detection modes. The weight matching result refers to the quantitative correspondence between the importance of different parameters.

[0079] In this example, the time series offset of the vessel rupture direction parameter and the recovery period of the deformation recovery parameter were first extracted from the synchronized parameter set, and the degree of hysteresis between the two was calculated through cross-correlation analysis. Secondly, a cross-modal association network was constructed, and a graph neural network was used to analyze the importance of different hysteresis patterns, determining the weighted matching results between the vessel rupture event and the cellular mechanical response. Next, backpropagation optimization was used to adjust the contribution weights of each feature, highlighting diagnostically significant hysteresis relationships (such as the sudden drop in stiffness 0.5-1 seconds after vessel rupture), providing a quantitative basis for subtype classification.

[0080] For example, the time series model constructed based on cross-modal analysis shows that there is a lagged correlation between the offset of the vascular rupture direction parameter and the recovery period of the deformation recovery parameter. When the vascular rupture direction parameter deflects southward three times in a row, the recovery period of the deformation recovery parameter is delayed by about six hours and shortened. Through deep learning network training, it is determined that the three-dimensional density parameter of the blood vessel occupies the main weight in the cross-modal association, while the time series offset of the vascular rupture direction parameter regulates the dynamic response of the stiffness change parameter. The weight matching results show that the correlation strength between the acute angle vascular rupture direction parameter on the dorsal side of the patient's tumor and the abnormal deformation recovery characteristics of circulating tumor cells reaches the clinical significance threshold, providing a key basis for subtype identification.

[0081] 105. Output a lung cancer subtype classification result based on the weight matching result and the spatiotemporal distribution characteristics of the tumor necrosis area volume ratio.

[0082] In this step, spatiotemporal distribution characteristics refer to the temporal and spatial variations of volume proportions. Lung cancer subtype classification results refer to the diagnosis of lung cancer types based on multi-parameter analysis.

[0083] In this embodiment, the weighted matching results from step 104 are first integrated with the spatiotemporal distribution characteristics of the tumor necrotic region volume fraction from step 101 to construct a three-dimensional classification feature space. Next, a random forest algorithm is used to train a classification model, jointly modeling the hysteresis pattern of vascular rupture, the characteristics of the cell mechanical response, and the distribution of necrotic regions. Next, feature importance analysis is performed to select the most discriminative combination of features (e.g., decreased stiffness after vascular rupture accompanied by central necrosis). Ultimately, a lung cancer subtype classification result, including specific types such as adenocarcinoma and squamous cell carcinoma, is output, providing an objective basis for clinical treatment decisions.

[0084] For example, by combining the weighted matching results with the spatiotemporal distribution of the tumor necrotic area volume, the system outputs a lung cancer subtype classification of "angioinvasive lung adenocarcinoma." This subtype is characterized by the coexistence of sharply angled distribution parameters of vascular rupture direction and a step-wise expansion of necrotic area volume. Furthermore, circulating tumor cells in the peripheral blood exhibit dual mechanical abnormalities: high-frequency oscillations in stiffness variation parameters and delayed deformation recovery. Comparison with historical case databases revealed that patients with this subtype are three times more sensitive to anti-angiogenic drugs than those with traditional adenocarcinomas. Based on this information, the clinical team developed a personalized regimen of bevacizumab combined with chemotherapy. A follow-up examination two weeks later showed reduced lesion enhancement and a reduction of over 50% in the number of circulating tumor cells, validating the subtype classification model's guiding value in precision medicine.

[0085] In summary, steps 101 to 105 achieved an innovative intelligent prediction method for lung cancer subtypes based on multimodal data fusion. By integrating the vascular distribution characteristics of CT images with the mechanical properties of circulating tumor cells in blood samples, a cross-modal joint analysis framework was constructed, addressing the problem of incomplete tumor feature representation in traditional single-modality diagnosis. By extracting the three-dimensional density parameters and fracture direction parameters of tumor blood vessels through dynamic segmentation, and combining the stiffness changes and deformation recovery characteristics of circulating tumor cells, a synchronous matching mechanism between vascular dynamic changes and cellular mechanical responses was established, significantly improving the accuracy and reliability of lung cancer subtype classification and providing a multi-dimensional decision-making basis for precision medicine.

[0086] In order to quantify the spatiotemporal evolution characteristics of the vascular network in the tumor microenvironment through multi-time point dynamic segmentation and three-dimensional spatial analysis technology, the spatial density distribution is calculated based on the three-dimensional coordinates of vascular branch nodes to evaluate vascular heterogeneity, and the spatial pattern of vascular structural destruction is analyzed in combination with fracture direction parameters. The dynamic trend of tumor tissue degeneration changes is monitored using the volume percentage sequence of necrotic areas. In this way, a correlation model between vascular network morphology and tumor pathological progression is constructed, providing an accurate quantitative evaluation basis for dynamic monitoring of the tumor microenvironment and anti-angiogenic treatment.

[0087] In some embodiments, in step 101, dynamically segmenting the CT image to extract vascular distribution data of the tumor region, calculating three-dimensional density parameters and vascular rupture direction parameters based on the vascular distribution data, and obtaining a volume percentage sequence of the tumor necrosis region includes:

[0088] 201. Perform a multi-time point segmentation operation on the CT image to extract vascular distribution data of the tumor area at each time point, wherein the vascular distribution data includes the three-dimensional coordinates of vascular branch nodes and the connection status between adjacent nodes;

[0089] In step 201, the multi-time point segmentation operation refers to segmenting the CT image at different time points. A vascular branch node refers to a bifurcation connection point in the vascular network. A connection state refers to a description of the connectivity between vascular nodes.

[0090] In the embodiments of the present application, first, a time series registration algorithm is used to preprocess the CT images to ensure spatial alignment of images at different time points. Secondly, a three-dimensional vascular segmentation algorithm (such as VesselNet based on deep learning) is used at each time point to accurately segment the tumor region and extract the vascular network structure. Next, a skeletonization algorithm is used to convert the vascular network into a topological graph consisting of vascular branch nodes and connecting edges, recording the three-dimensional coordinates (such as the X / Y / Z axis position) and connection status (such as the connectivity between adjacent nodes) of each node. Finally, the vascular topology graphs at each time point are arranged along the time axis to construct a complete vascular evolution time series database.

[0091] 202. Calculate the density distribution of blood vessels in three-dimensional space based on the three-dimensional coordinates of the blood vessel branch nodes, and define the number of blood vessel branch nodes per unit volume as a three-dimensional density parameter;

[0092] In step 202, density distribution refers to the density characteristics of blood vessels in space. Unit volume refers to a set standard spatial measurement unit. The number of blood vessel branch nodes refers to the count of blood vessel bifurcation points within a specific spatial range.

[0093] In the embodiment of the present application, first, starting from the three-dimensional coordinates of the vascular branch nodes obtained in step 201, the tumor area is divided into a 1 mm³ cubic voxel grid. Secondly, the number of vascular nodes contained in each voxel is counted, and this value is defined as the three-dimensional density parameter of the voxel. Then, the density distribution is smoothed using a three-dimensional Gaussian filter to eliminate noise interference caused by small vascular fluctuations. Finally, the density parameters at each time point are integrated according to the spatial position to generate a density field sequence that can reflect the spatial distribution characteristics of the blood vessels.

[0094] 203. Count the fracture directions of all vascular branch nodes, and generate vascular fracture direction parameters according to the angular distribution of the fracture directions in the three-dimensional coordinate system;

[0095] In step 203, the fracture direction refers to the spatial orientation characteristics of the blood vessel fracture. The three-dimensional coordinate system refers to the spatial reference system composed of the X / Y / Z axes. The angular distribution refers to the orientation statistical characteristics of the fracture direction in three-dimensional space.

[0096] In the present embodiment, all fracture nodes (i.e., terminal nodes with only one connection point) are first identified in the vascular topology map, and their fracture directions (e.g., the vector from the last connection point to the fracture point) are extracted. Next, each fracture direction vector is projected into a three-dimensional coordinate system, and its angle with the X / Y / Z axes is calculated (e.g., in the range of 0-180 degrees). Spherical statistics are then used to analyze the angular distribution characteristics of the fracture directions, generating vascular fracture direction parameters that include the primary fracture direction (e.g., the 45° direction has the highest proportion) and the directional dispersion. Finally, the fracture direction parameters at each time point are arranged in a time series for subsequent dynamic analysis.

[0097] 204. By measuring the ratio of the volume of the tumor necrosis area to the total volume of the tumor area at each time point, a volume ratio sequence is generated.

[0098] In step 204, volume measurement refers to calculating the spatial proportion of the necrotic region of the tumor. The total volume refers to the complete spatial volume of the tumor region. The ratio calculation refers to the ratio of the volume of the necrotic region to the total volume of the tumor.

[0099] In this embodiment of the present application, based on the tumor region segmentation results from step 201, a threshold segmentation method (such as the Otsu algorithm) is first used to distinguish between necrotic tumor regions and active tissue regions. Next, the volume of the necrotic region is calculated using three-dimensional connected domain analysis and divided by the total tumor volume to obtain the volume fraction. Next, the volume fraction values at each time point are recorded in a time series to generate a volume fraction sequence reflecting the evolution of the necrotic region. Finally, the time series curve is smoothed using a moving average method to eliminate interference from short-term fluctuations on the analysis results.

[0100] The following is a specific example, Figure 3 As shown:

[0101] exist Figure 3 In a dynamic imaging assessment system for lung adenocarcinoma, multi-phase CT vascular analysis technology enables precise monitoring of tumor microenvironment evolution. For a patient with a mixed ground-glass nodule in the right lower lobe of the lung, the system first performed time-point segmentation on three-phase enhanced CT images (step 201). The system then extracted the three-dimensional coordinates and connectivity of vascular branch nodes in the tumor region during the arterial, venous, and delayed phases, revealing a characteristic vascular "pruning" phenomenon at the lesion's edge. Based on the spatial distribution of vascular nodes (step 202), the system calculated and showed that the three-dimensional density parameters in the lesion's core were significantly higher than those in the surrounding areas, exhibiting a typical "center-enriched" distribution pattern. By calculating the direction of vascular fractures (step 203), fracture direction parameters were generated, indicating that the superior pole of the lesion exhibited a clustering of angles between 90 and 120 degrees, indicating mechanical stress-induced vascular fractures in this region. The system also measured the proportion of necrotic area in each phase (step 204), generating a series of percentages showing the gradual expansion of necrotic volume over time. The direction of expansion closely matched the clustered areas of vascular fractures. When a new case's CT scan showed similar central enrichment density parameters and superior polar vascular disruption patterns, the system automatically predicted an invasive lung adenocarcinoma subtype, prioritizing EGFR gene testing. Ultimately, pathology confirmed the diagnosis as invasive adenocarcinoma with a micropapillary component. This case optimized the multi-phase vascular analysis algorithm, enhancing the ability to identify the invasive characteristics of lung adenocarcinoma early, and establishing an intelligent assessment system from dynamic vascular analysis to molecular subtype prediction.

[0102] In summary, steps 201 to 204 achieve dynamic quantification and three-dimensional modeling of tumor vascular distribution characteristics. Multi-time point segmentation technology extracts the three-dimensional coordinates and connectivity of vascular branch nodes, enabling precise quantification of the spatial distribution of the vascular network. Three-dimensional density parameters are calculated based on the number of vascular nodes per unit volume, and vascular fracture direction parameters are generated based on the angular distribution of fracture directions, constructing a dynamic assessment system for tumor vascular heterogeneity. By using a sequence of necrotic region volume fractions at consecutive time points, a vascular morphological indicator with spatiotemporal evolution characteristics is provided for lung cancer subtype classification, enhancing the ability to characterize tumor microenvironmental features.

[0103] In some embodiments, as described in step 102, the blood sample is subjected to spiral sorting processing, circulating tumor cells are separated by fluid control, deformation trajectory data of the circulating tumor cells are collected, and stiffness change parameters and deformation recovery parameters are generated based on the deformation trajectory data, including:

[0104] 301. Injecting the blood sample into a spiral separation device to separate circulating tumor cells from non-tumor cells by adjusting the direction of fluid dynamics;

[0105] In step 301, the spiral separation device refers to experimental equipment that uses a spiral microfluidic channel to separate cells. The hydrodynamic direction refers to the direction of the mechanical action exerted by the fluid in the spiral channel. Circulating tumor cells refer to tumor cells that have entered the peripheral blood from the primary tumor. Non-tumor cells refer to other cellular components in the blood, excluding tumor cells. Adjustment refers to the process of optimizing the separation effect by changing experimental parameters.

[0106] In an embodiment of the present application, first, the collected blood sample is injected into the inlet of the spiral sorting device through a precision syringe pump, and the initial flow rate is set to 1 ml / min. Secondly, the direction of the fluid dynamics in the device is adjusted by the microfluidic control system so that the fluid generates a stable laminar flow and centrifugal force field in the spiral channel. Then, by utilizing the difference in mass and deformation ability between tumor cells and blood cells, circulating tumor cells are moved along the outside of the spiral channel due to their greater inertia, while non-tumor cells are thrown to the inside due to their strong deformation ability. Finally, a sorting valve is set at the end of the channel to collect high-purity circulating tumor cells at the outer outlet to complete the cell separation process.

[0107] 302. Collecting movement trajectory data of the separated circulating tumor cells in the fluid environment, wherein the movement trajectory data includes deformation amplitude and deformation direction of the cells at different time points;

[0108] In step 302, movement trajectory data refers to the spatiotemporal recording of cell movement in a fluid. Deformation amplitude refers to the degree of cell deformation under the action of an external force. Deformation direction refers to the primary spatial orientation of cell deformation. Acquisition refers to the process of acquiring experimental data using detection equipment.

[0109] In the examples of the present application, first, the isolated circulating tumor cells are suspended in a buffer solution and re-injected into the observation channel at a constant flow rate. Secondly, a high-speed microscopic imaging system (e.g., 1000 frames per second) is used to continuously capture the cell movement process and obtain a time-series image of the cell contour. Next, the cell contour at each time point is extracted using an edge detection algorithm, and the deformation amplitude (e.g., the change in the ratio of the cell's major axis to the minor axis) and deformation direction (e.g., the angle between the main axis of deformation and the flow direction) are calculated. Finally, the measurement results at each time point are integrated into a movement trajectory data table containing timestamps, deformation amplitudes, and deformation directions.

[0110] 303. Calculate the recovery speed after the deformation amplitude reaches a maximum value based on the curve of the deformation amplitude changing with time, and define the reciprocal of the recovery speed as a deformation recovery parameter;

[0111] In step 303, the change curve refers to a graph showing the deformation amplitude over time. The recovery rate refers to the speed at which the cell recovers from its maximum deformation state. The reciprocal refers to the reciprocal operation in mathematics. The deformation recovery parameter refers to a characteristic parameter that quantifies the cell's elastic recovery ability.

[0112] In the embodiment of the present application, first, a curve of the change of deformation amplitude over time is extracted from the movement trajectory data, and the time point when the deformation reaches the maximum value is determined by the peak detection algorithm. Secondly, the deformation recovery data within the 100ms time window after the maximum value is selected, and the slope of the recovery curve is fitted by linear regression. Then, the inverse of the slope of the recovery curve (that is, the time required to recover the unit deformation amount) is defined as the deformation recovery parameter. The smaller the parameter value, the stronger the elastic recovery ability of the cell. Finally, the average value of multiple deformation-recovery cycles of the same batch of cells is taken to obtain a representative deformation recovery parameter.

[0113] 304. Count the frequency of changes in the deformation direction in the moving trajectory, and define the product of the frequency and the deformation amplitude as a stiffness change parameter.

[0114] In step 304, the frequency of change refers to how frequently the deformation direction changes during the trajectory. The product refers to the result of a multiplication operation in mathematics. The stiffness change parameter refers to a quantitative indicator reflecting the dynamic changes in cell stiffness. Statistics refers to a method for summarizing and analyzing data.

[0115] In the embodiment of the present application, first, the deformation direction change in the movement trajectory data is analyzed, and the frequency of change of the deformation direction in the 0.1-10 Hz frequency band is calculated by Fourier transform. Secondly, the frequency value of each frequency band is multiplied by the deformation amplitude at the corresponding moment to obtain the frequency-amplitude product spectrum. Then, the frequency band corresponding to the maximum value in the product spectrum is selected as the characteristic frequency band, and the product value of the frequency band is normalized and defined as the stiffness change parameter, which reflects the dynamic response characteristics of the cell stiffness to the external fluid force. Finally, the stiffness change parameter is correlated with the deformation recovery parameter for analysis to establish a complete cell mechanical property description system.

[0116] Here's a specific example:

[0117] In the intelligent lung cancer liquid biopsy analysis system, microfluidic chip technology enables precise assessment of the mechanical properties of circulating tumor cells. For peripheral blood samples from patients suspected of small cell lung cancer, the system first injects the blood into a spiral sorting device (step 301). By optimizing the fluid dynamics, it successfully isolates a population of circulating tumor cells with characteristic rapid motion. A high-speed microscopic imaging system captures the trajectory of these cells within the spiral channel (step 302), recording the "slingshot" deformation pattern and unique deformation direction exhibited by the cells as they pass through narrow areas. The system analyzes the time-varying deformation amplitude curve (step 303) and calculates that small cell lung cancer cells exhibit significantly rapid deformation recovery parameters, consistent with the high elasticity of these tumor cell membranes. Further, the frequency of deformation direction changes is calculated (step 304), generating a pattern map showing the characteristic high-frequency stiffness variation parameters of small cell lung cancer. When new samples exhibit similar rapid recovery parameters and high-frequency stiffness variations, the system automatically associates them with the small cell lung cancer prediction model, guiding clinical neuroendocrine marker testing. Ultimately, pathological confirmation confirms small cell lung cancer with a characteristic "oatmeal-like" morphology. This case optimized the fluid parameter settings for spiral sorting, enhanced the specific identification capability of circulating tumor cells of small cell lung cancer, and formed an intelligent diagnostic closed loop from cell mechanical property analysis to pathological subtype determination.

[0118] In summary, steps 301 to 304 achieve precise detection and parameterized characterization of the mechanical properties of circulating tumor cells. Through the fluid dynamics of the spiral sorting device, efficient separation and purification of circulating tumor cells is achieved. High-speed trajectory acquisition technology captures the dynamic process of cell deformation. The deformation recovery parameter is defined by the inverse of the deformation recovery rate, and the stiffness change parameter is defined by combining the product of the frequency and amplitude of the deformation direction change. This establishes a quantitative indicator system for cell mechanical properties. This technology overcomes the static limitations of traditional cell detection methods and provides characteristic parameters reflecting the dynamic mechanical behavior of cells for lung cancer subtype classification.

[0119] In some embodiments, step 103 includes jointly analyzing the three-dimensional density parameter, the vascular rupture direction parameter, and the stiffness change parameter to extract a time series offset of the vascular rupture direction parameter, and synchronously matching the time series offset with the dynamic response of the stiffness change parameter to generate a synchronized parameter set, including:

[0120] 401. Obtaining direction angle values of the blood vessel rupture direction parameter at multiple time points to calculate angle differences between adjacent time points, and generating a time series offset based on the angle differences;

[0121] In step 401, the directional angle value refers to the specific angular value of the vascular rupture direction in three-dimensional space. The angle difference refers to the change in directional angle between adjacent time points. The time series offset refers to the phase difference of the vascular rupture direction over time. Calculation refers to the process of obtaining characteristic parameters through mathematical operations.

[0122] In this embodiment, first, measurement data of the vessel fracture direction parameter at consecutive time points is extracted from the vascular imaging analysis system, obtaining the recorded directional angle value (e.g., 45°, 60°, etc.) at each time point. Next, a differential calculation method is used to subtract the angle value at the previous time point from the angle value at the next time point to obtain the angular difference between adjacent time points (e.g., 60° - 45° = 15°). These angular differences are then arranged in chronological order to form a time series offset that reflects the changing trend of the vessel fracture direction. Finally, the offset sequence is smoothed using a sliding average method to eliminate random fluctuations caused by measurement noise.

[0123] 402. Obtain parameter values of the stiffness variation parameters at the same time point, and extract dynamic variation values of the parameter values over time;

[0124] In step 402, the parameter value refers to the specific value of the stiffness change parameter at a specific time point. The dynamic change value refers to the fluctuation characteristics of the parameter over time. Extraction refers to the analytical method of obtaining specific information from the data. The same time point refers to the time point when the vascular rupture direction and the stiffness change parameter are synchronized.

[0125] In the examples of this application, first, the stiffness change parameter values at the same time point as the vascular rupture direction parameters were obtained from the cell mechanics test data. Second, a time series analysis method was used to calculate the difference in stiffness change parameters between adjacent time points to obtain a dynamic change value reflecting the parameter fluctuation. Next, a normalization process was used to convert the dynamic change values at different time points to the same dimension to facilitate subsequent comparative analysis. Finally, a curve chart showing the dynamic change value over time was drawn to intuitively demonstrate the variation pattern of the stiffness parameter.

[0126] 403. Align the time series offset and the dynamic change value on a time axis, calculate the sum of the products of the two at the same time point, and generate a synchronization parameter set.

[0127] In step 403, time axis alignment refers to synchronizing the time dimensions of different data. The sum of products refers to the cumulative result of multiplying the parameters at corresponding time points. The synchronization parameter set refers to a feature set that reflects the temporal synchronization of multiple parameters. Generation refers to the process of obtaining the final result through calculation.

[0128] In the embodiment of the present application, first, the time series offset generated in step 401 and the dynamic change value obtained in step 402 are matched and aligned at the same time point to ensure that the time axes of the two are completely synchronized. Secondly, the corresponding offset and dynamic change value are multiplied at each time point to obtain the coordinated change amount at that time point. Then, the product results of all time points are accumulated to generate a synchronization parameter set that reflects the degree of coordination between the change in the direction of vascular rupture and the change in cell stiffness. Finally, the values in the synchronization parameter set are converted to the range of 0-1 through normalization processing to facilitate subsequent weight matching analysis.

[0129] Here's a specific example:

[0130] In the multimodal dynamic monitoring system for lung adenocarcinoma, spatiotemporal feature fusion technology enables precise assessment of tumor vascular evolution. For a patient with a mixed ground-glass nodule in the left upper lobe, the system first acquired time series data on vessel fracture direction parameters from three-phase enhanced CT images (step 401). The system calculated the offset in directional angles from the arterial to venous phases, revealing a characteristic 45-degree directional abrupt shift at the leading edge of the lesion. Dynamic monitoring of circulating tumor cell stiffness parameters was then simultaneously analyzed (step 402), extracting the stiffness parameter fluctuation characteristics aligned with the CT scan time points, revealing a significant decrease in stiffness during the venous phase. Through precise timeline alignment (step 403), the system synchronously correlated the vessel fracture direction offset with the cell stiffness change values, generating a synchronized parameter set demonstrating a strong correlation between the two during the venous phase. When a new case exhibited a similar synchronous pattern of directional abrupt shift and stiffness decrease, the system automatically predicted a high-risk lung adenocarcinoma subtype with micropapillary components, guiding clinical targeted treatment decisions. Postoperative pathology confirmed the presence of characteristic vascular co-infiltration in the predicted region. This case optimized the multimodal time alignment algorithm, enhanced the early warning capability of the aggressive biological behavior of lung adenocarcinoma, and formed an intelligent evaluation system from dynamic analysis of imaging features to prediction of treatment response.

[0131] In summary, steps 401 to 403 achieve spatiotemporal synchronization and correlation analysis of cross-modal feature parameters. By extracting the time series offset of the vascular fracture direction parameter and aligning it with the dynamic response of the stiffness change parameter, a synchronization matching mechanism for multimodal data is established. By calculating the sum of the products of the time series offset and the dynamic change value to generate a set of synchronization parameters, dynamic correlation modeling of vascular morphological changes and cellular mechanical properties is achieved, significantly improving the spatiotemporal consistency of cross-modal feature fusion and providing a precise data foundation for subsequent hysteresis relationship analysis.

[0132] In some embodiments, in step 104, establishing a hysteresis correlation between the time series offset of the vascular rupture direction parameter and the recovery period of the deformation recovery parameter based on the synchronization parameter set, so as to determine a weighted matching result between the two through cross-modal analysis, includes:

[0133] 501. Count the time delay between the time point when the time series offset reaches a peak value and the time point when the recovery period ends, to generate a lag time series.

[0134] In step 501, the lag time series refers to the time difference between the time series offset and the recovery period. The peak time point refers to the time when the time series offset reaches its maximum. The recovery period end time point refers to the time point when the deformation recovery process is completed. The time delay refers to the time interval between two events. Statistics refers to the process of summarizing and analyzing data.

[0135] In the present embodiment, first, peak detection is performed on the time series offsets generated in step 401, and the peak time point corresponding to each offset peak is determined using a sliding window extreme value analysis method. Secondly, the end time point of each recovery cycle is extracted from the deformation recovery parameters in step 303. Next, the time difference between each offset peak and the end point of the most recent recovery cycle is calculated to generate a lagged time series reflecting the delayed relationship between the vascular rupture event and the cellular mechanical response. Finally, a box plot analysis is performed to eliminate abnormal lag values and retain biologically meaningful delay data.

[0136] 502. Calculate the association weight between the time series offset and the recovery period based on the length distribution of the lagged time series;

[0137] In step 502, length distribution refers to the statistical characteristics of the duration of the lagged time series. Correlation weight refers to a quantitative indicator reflecting the correlation between the time series offset and the recovery period. Calculation refers to the process of obtaining characteristic parameters through mathematical operations. The recovery period refers to the length of time it takes for a deformation recovery parameter to complete a recovery process.

[0138] In the embodiment of the present application, first, the lag time series are grouped and counted according to the delay time, and a histogram of the lag time length distribution is drawn. Secondly, the kernel density estimation method is used to calculate the probability density of different lag times, and the inverse of the probability density is used as the associated weight of the lag time (for example, the shorter the delay, the higher the weight). Then, the weight value is normalized to the range of 0-1 by the sigmoid function to ensure that the weight difference between long delay and short delay conforms to biological laws. Finally, a mapping relationship table between lag time and weight is established, and a corresponding weight value is assigned to each observed lag time.

[0139] 503. Superimpose the association weight and the sum of the products in the synchronization parameter set to generate a weight matching result.

[0140] In step 503, superposition refers to the mathematical accumulation of different parameters. The weighted matching result is a comprehensive parameter reflecting the degree of match between the vascular rupture direction and the cell's mechanical properties. Generation refers to the process of obtaining the final result through calculation. The sum of products refers to the cumulative result of multiplying the parameters at corresponding time points in the synchronization parameter set.

[0141] In the present embodiment, first, the sum of products at each time point is extracted from the synchronization parameter set in step 403. Next, a corresponding association weight is assigned to each sum of products based on the mapping relationship table established in step 502. Next, the sum of products is multiplied by the association weight at each time point to implement a weighted superposition calculation. Finally, the weighted results for all time points are cumulatively summed to generate a weighted matching result that comprehensively considers time delay and synchronization strength. This result quantifies the overall correlation between vascular rupture events and cellular mechanical responses.

[0142] Here's a specific example:

[0143] In the multimodal dynamic assessment system for squamous cell lung carcinoma, spatiotemporal feature fusion technology enables precise quantification of tumor vascular-mechanical coupling characteristics. Based on continuous monitoring data from an irregular mass in the right middle lobe of a patient, the system first calculates the time delay between the peak of the time series offset in the direction of vascular rupture and the end of the mechanical recovery cycle of circulating tumor cells (step 501). This generates a time series showing a characteristic 0.8-second lag in the superior region of the lesion. By analyzing the distribution characteristics of the lag time series (step 502), the correlation weight between vascular directional abrupt changes and cellular mechanical recovery is calculated, revealing a high-weighted coupling pattern in the tumor core. The system then superimposes this correlation weight with the previously generated set of vascular-mechanical synchronization parameters (step 503), ultimately generating a quantitative map showing significant weighted matching results in the leading edge of the lesion. When a new case exhibits similar lag characteristics in the superior region and high-weighted coupling in the core, the system automatically predicts a squamous cell lung carcinoma subtype with a propensity for keratin bead formation, guiding clinical PD-L1 expression testing. Pathology confirms the presence of characteristic dyskeratotic cell clusters in the predicted region. This case optimized the spatiotemporal coupling weight algorithm, improved the recognition accuracy of squamous cell carcinoma-specific vascular-mechanical characteristics, and formed an intelligent decision-making system from multimodal dynamic analysis to immunotherapy response prediction.

[0144] In summary, steps 501 to 503 implement a model of the hysteresis-related association between vascular dynamic changes and cellular mechanical recovery. By statistically analyzing the time delay between the peak offset of the time series and the end of the recovery cycle, a hysteresis-related association model between vascular rupture direction and cellular deformation recovery was established. Association weights were calculated based on the length distribution of the lagged time series and optimized overlaid with the synchronization parameter set to generate weighted results that reflect the dynamic matching relationship between cross-modal features. This technique overcomes the limitations of traditional static association analysis and enables quantitative characterization of the dynamic interaction mechanism between tumor vascular abnormalities and cellular mechanical responses.

[0145] In some embodiments, in step 105, outputting a lung cancer subtype classification result based on the weight matching result and the spatiotemporal distribution characteristics of the tumor necrosis area volume ratio includes:

[0146] 601. Extracting distribution characteristics of the volume proportion values in the volume proportion sequence as they change with spatial positions, and associating spatial positions with time points to generate spatiotemporal distribution characteristics;

[0147] In step 601, the proportion value refers to the ratio of the volume of the necrotic area to the total volume of the tumor. Spatial location refers to the specific location coordinates in a three-dimensional coordinate system. Spatiotemporal distribution characteristics refer to the changing patterns of volume proportion across time and space. Correlation refers to establishing corresponding relationships between data of different dimensions. Extraction refers to analytical methods for obtaining specific information from data.

[0148] In the embodiment of the present application, first, a volume percentage sequence is extracted from the tumor image analysis system to obtain the percentage value of the tumor necrosis area at different spatial positions (such as front / back / left / right / up / down) at each time point. Secondly, a spatial interpolation algorithm is used to convert the discrete spatial position percentage values into a continuous three-dimensional distribution map. Then, the three-dimensional distribution map of each time point is arranged in chronological order, and the spatial position is associated with the time point through spatiotemporal coding technology to generate a spatiotemporal distribution feature that can simultaneously reflect the spatial distribution and temporal evolution of the necrotic area. Finally, principal component analysis is performed on the spatiotemporal distribution features to extract the most representative spatiotemporal change pattern.

[0149] 602. Map and match the weight matching result with the spatiotemporal distribution feature, and output a corresponding lung cancer subtype classification result according to a preset threshold interval.

[0150] In step 602, mapping and matching refers to the process of correlating different feature parameters. The preset threshold range refers to the pre-set range of classification criteria. The lung cancer subtype classification result refers to the lung cancer type diagnosis conclusion derived based on multi-parameter analysis. Output refers to the operation of displaying or returning the final result.

[0151] In the embodiment of the present application, first, the weight matching result obtained in step 503 is aligned with the spatiotemporal distribution feature generated in step 601 to ensure that the time points of the two are completely corresponding. Secondly, the weight matching result of each time point is matched with the spatiotemporal distribution feature at the corresponding moment through a feature mapping algorithm to establish a parameter-feature mapping relationship table. Then, based on the threshold interval set by clinical experience (such as 0.3-0.5 for adenocarcinoma and 0.6-0.8 for squamous cell carcinoma), the mapping matching results are classified into the corresponding interval. Finally, the majority voting method is used to count the number of matches in each interval, and the lung cancer subtype classification result corresponding to the interval with the highest frequency is output.

[0152] Here's a specific example:

[0153] In the intelligent lung adenocarcinoma classification system, multimodal spatiotemporal feature fusion technology enables accurate assessment of tumor heterogeneity. For a patient with a mixed-density nodule in the left upper lobe, the system first extracted the spatial distribution features of the necrotic volume fraction sequence from three-phase enhanced CT scans (step 601). This then established a "fan-shaped" expansion pattern that indicated the temporal progression of the posterior basal region of the lesion, generating a distribution feature with temporal and spatial evolution patterns. By three-dimensionally mapping the previously obtained vascular-mechanical weight matching results with this spatiotemporal feature (step 602), the system discovered that the central region of the lesion exhibited a synergistic pattern of high-weight coupling and rapid spatial expansion. Based on a preset classification threshold, the system automatically outputted a signature spectrum consistent with micropapillary lung adenocarcinoma. When a new case showed a similar central synergistic pattern on CT scan, the system triggered a prediction of the probability of an EGFR-sensitive mutation, guiding the clinical decision to prioritize targeted therapy. Postoperative pathology confirmed the diagnosis of invasive adenocarcinoma with a micropapillary architecture. This case optimized the spatiotemporal synergistic analysis algorithm, enhancing the ability to early identify high-risk subtypes and forming an intelligent decision-making closed loop from image heterogeneity analysis to personalized treatment.

[0154] In summary, steps 601 to 602 achieve intelligent classification of lung cancer subtypes based on spatiotemporal distribution characteristics. By extracting the spatiotemporal evolution of the volume fraction of necrotic regions in tumors and mapping it with the weighted matching results, a multi-dimensional feature fusion classification decision model is constructed. By integrating multiple sources of features, such as vascular dynamics, cellular mechanical responses, and necrotic region distribution, lung cancer subtype classification transitions from single morphological indicators to spatiotemporal dynamic features, significantly improving the biological plausibility and clinical applicability of the classification results.

[0155] In some embodiments, as described in step 602, mapping and matching the weight matching result with the spatiotemporal distribution feature, and outputting the corresponding lung cancer subtype classification result according to a preset threshold interval, includes:

[0156] 701. Obtain a parameter value for each time point in the weighted matching result, and a spatial position ratio value of the corresponding time point in the spatiotemporal distribution feature;

[0157] In step 701, the parameter value refers to the specific value of the weighted matching result at a specific time point. The spatial position percentage value refers to the volume percentage value of a specific spatial position. Acquisition refers to the process of reading the required information from the data. The corresponding time point refers to the moment when two parameters have the same time stamp.

[0158] In the embodiment of the present application, first, the parameter value recorded at each time point is extracted from the weight matching result of step 503, and the spatial position proportion value of the same time point is obtained from the spatiotemporal distribution characteristics of step 601. Secondly, timestamp matching is used to ensure that the parameter value and the proportion value are from the same observation time. Then, the data is standardized to eliminate the dimensional differences between different parameters. Finally, a corresponding relationship table of time-parameter-proportion value is established to provide structured data for subsequent comprehensive analysis.

[0159] 702. Superimpose the parameter value and the proportion value at the same time point to generate a comprehensive parameter at each time point;

[0160] In step 702, the comprehensive parameter refers to a composite parameter that integrates weight matching and spatial proportion. Superposition refers to the mathematical combination of different parameters. The same time point refers to data points with the same time stamp. Generation refers to the process of obtaining new parameters through calculation.

[0161] In the embodiment of the present application, first, the parameter values and proportion values obtained at the same time point in step 701 are weighted, and different weights are assigned to the two types of parameters based on their clinical importance (e.g., a parameter value weight of 0.6 and a proportion weight of 0.4). Secondly, a linear weighting method is used to superimpose the two types of parameters according to their weights to generate a comprehensive parameter that reflects the comprehensive characteristics of the time point. Next, the comprehensive parameters of adjacent time points are smoothed using a sliding average method to eliminate random fluctuation interference. Finally, a curve chart of the comprehensive parameter changes over time is drawn to intuitively display the characteristic evolution trend.

[0162] 703. Count the maximum and minimum values of the comprehensive parameters at all time points and generate a dynamic parameter interval;

[0163] In step 703, the dynamic parameter interval refers to the range within which the comprehensive parameter changes over time. The maximum value refers to the highest value in a set of data. The minimum value refers to the lowest value in a set of data. Statistics refers to the method of summarizing and analyzing data.

[0164] In this embodiment, first, an extreme value detection algorithm is used to identify the maximum and minimum values of the comprehensive parameters generated at all time points in step 702. Next, the difference between the maximum and minimum values is calculated, and this difference is divided into a number of intervals (e.g., five intervals). Next, the distribution frequency of the comprehensive parameter in each interval is statistically analyzed to determine the primary range of variation of the parameter. Finally, a dynamic parameter interval is generated, encompassing the complete range of fluctuation of the parameter, using the maximum and minimum values as boundaries, to provide a quantitative basis for subtype discrimination.

[0165] 704. Compare the dynamic parameter interval with a preset threshold interval. If the dynamic parameter interval completely falls within any threshold interval, output the lung cancer subtype classification result corresponding to the threshold interval.

[0166] In step 704, range comparison refers to comparing the inclusion relationship between two numerical intervals. The preset threshold interval refers to the pre-set classification judgment standard range. Complete inclusion refers to the complete inclusion of one interval by another interval. The lung cancer subtype classification result refers to the lung cancer type diagnosis conclusion derived based on the parameter interval matching. Output refers to the operation of displaying or returning the final result.

[0167] In this embodiment of the present application, the dynamic parameter interval determined in step 703 is first compared with the threshold interval preset by clinical experts, and the overlap ratio is calculated using an interval inclusion algorithm. Secondly, if the dynamic parameter interval completely falls within a certain threshold interval (e.g., the interval of 0.3-0.5), it is determined to be the subtype corresponding to that interval (e.g., adenocarcinoma). Next, when the dynamic parameter interval spans multiple threshold intervals, the maximum overlap principle is used to determine the most likely subtype category. Finally, a lung cancer subtype classification result is output, including the specific subtype name (e.g., lung adenocarcinoma stage T2) and confidence level, to provide decision support for clinical treatment.

[0168] The following is a specific example, Figure 4 As shown,

[0169] exist Figure 4In a multimodal intelligent diagnosis system for lung adenocarcinoma, dynamic parameter fusion technology enables accurate classification of tumor subtypes. For a patient with a mixed ground-glass nodule in the right middle lobe of the lung, the system first obtains weighted matching results for vascular-mechanical coupling from three-phase enhanced CT scans (step 701). It also extracts the spatial fraction of necrotic volume at corresponding time points. It finds that the leading edge of the lesion exhibits a characteristic combination of high-weighted parameters and rapid expansion during the arterial phase. The coupling parameters and spatial fraction at the same time point are dynamically superimposed (step 702) to generate a comprehensive parameter sequence that peaks during the venous phase. By analyzing the extreme value distribution of this sequence (step 703), the system establishes a dynamic parameter interval reflecting the overall heterogeneity of the lesion. This interval is compared with pre-set subtype classification criteria (step 704). The range falls completely within the characteristic threshold range for micropapillary lung adenocarcinoma. The system automatically outputs the classification result and indicates the possibility of an EGFR-sensitive mutation. When a new case exhibits similar venous phase peak characteristics, the system recommends targeted therapy, achieving significant therapeutic results. This case optimized the dynamic interval comparison algorithm, improved the specificity of identifying special subtypes of lung adenocarcinoma, and formed an intelligent decision-making closed loop from multi-parameter dynamic analysis to precise treatment.

[0170] In summary, steps 701 to 704 implement intelligent matching decisions between dynamic parameter intervals and preset thresholds. By superimposing weighted matching parameters and spatiotemporal proportions to generate comprehensive parameters and statistically analyzing their dynamic variation intervals, a threshold-based classification decision mechanism is constructed. By intelligently matching dynamic parameter intervals with preset threshold ranges, automated determination of lung cancer subtypes is achieved, addressing the subjectivity inherent in traditional manual interpretation. This technology significantly improves the stability and interpretability of the classification system, providing objective and reliable decision support for clinical diagnosis.

[0171] Figure 5 The present invention provides a schematic diagram of a lung cancer type prediction system based on a fusion deep learning network. Figure 5 As shown, the system includes:

[0172] an acquisition module 21 for acquiring CT images and blood samples from lung cancer patients, dynamically segmenting the CT images to extract vascular distribution data of the tumor region, calculating three-dimensional density parameters and vascular rupture direction parameters based on the vascular distribution data, and obtaining a volume fraction sequence of the tumor necrosis region;

[0173] a processing module 22 for performing spiral sorting on the blood sample, separating circulating tumor cells through fluid control, collecting deformation trajectory data of the circulating tumor cells, and generating stiffness change parameters and deformation recovery parameters based on the deformation trajectory data;

[0174] an analysis module 23 for jointly analyzing the three-dimensional density parameter, the vessel rupture direction parameter, and the stiffness change parameter, extracting a time series offset of the vessel rupture direction parameter, and synchronously matching the time series offset with a dynamic response of the stiffness change parameter to generate a synchronous parameter set;

[0175] an establishing module 24 for establishing, based on the synchronization parameter set, a hysteresis correlation relationship between the time series offset of the blood vessel rupture direction parameter and the recovery period of the deformation recovery parameter, so as to determine a weight matching result between the two through cross-modal analysis;

[0176] The output module 25 is configured to output a lung cancer subtype classification result based on the weight matching result and the spatiotemporal distribution characteristics of the tumor necrosis area volume ratio.

[0177] Figure 5 The lung cancer type prediction system based on the fusion deep learning network can be performed Figure 1 The implementation principles and technical effects of the lung cancer type prediction method based on a fused deep learning network described in the illustrated embodiment are not further elaborated. The specific manner in which each module and unit performs operations in the lung cancer type prediction system based on a fused deep learning network described in the above embodiment has been described in detail in the relevant embodiments of the method and will not be further elaborated here.

[0178] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present application, rather than to limit them. Although the present application has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present application.

Claims

1. A lung cancer type prediction method based on a fusion deep learning network, characterized in that: include: Obtaining a CT image and a blood sample from a lung cancer patient, dynamically segmenting the CT image to extract vascular distribution data of the tumor region, calculating a three-dimensional density parameter and a vascular rupture direction parameter based on the vascular distribution data, and obtaining a volume fraction sequence of the tumor necrosis region, the volume fraction sequence being generated by measuring the ratio of the volume of the tumor necrosis region to the total volume of the tumor region at each time point; performing spiral sorting on the blood sample, separating circulating tumor cells through fluid control, collecting deformation trajectory data of the circulating tumor cells, and generating stiffness change parameters and deformation recovery parameters based on the deformation trajectory data; jointly analyzing the three-dimensional density parameter, the blood vessel fracture direction parameter, and the stiffness change parameter, extracting a time series offset of the blood vessel fracture direction parameter, and synchronously matching the time series offset with a dynamic response of the stiffness change parameter to generate a synchronous parameter set; Based on the synchronization parameter set, a hysteresis correlation relationship is established between the time series offset of the blood vessel rupture direction parameter and the recovery period of the deformation recovery parameter, so as to determine a weight matching result between the time series offset and the recovery period through cross-modal analysis; Outputting a lung cancer subtype classification result according to the weight matching result and the spatiotemporal distribution characteristics of the tumor necrosis area volume ratio; The three-dimensional density parameter, the vascular rupture direction parameter, and the stiffness change parameter are jointly analyzed to extract the time series offset of the vascular rupture direction parameter, and the time series offset is synchronously matched with the dynamic response of the stiffness change parameter to generate a synchronous parameter set, including: Obtain the directional angle values of the vascular rupture direction parameter at multiple time points to calculate the angle difference between adjacent time points, and generate a time series offset based on the angle difference; obtain the parameter value of the stiffness change parameter at the same time point, and extract the dynamic change value of the parameter value over time; align the time series offset with the dynamic change value on the time axis, calculate the sum of the products of the two at the same time point, and generate a synchronization parameter set.

2. The method according to claim 1, characterized in that Dynamically segmenting the CT image to extract vascular distribution data of the tumor area, calculating three-dimensional density parameters and vascular rupture direction parameters based on the vascular distribution data, and obtaining a volume percentage sequence of the tumor necrosis area, including: Performing a multi-time point segmentation operation on the CT image to extract vascular distribution data of the tumor area at each time point, wherein the vascular distribution data includes the three-dimensional coordinates of vascular branch nodes and the connection status between adjacent nodes; Based on the three-dimensional coordinates of the blood vessel branch nodes, calculating the density distribution of the blood vessels in the three-dimensional space, and defining the number of blood vessel branch nodes in a unit volume as a three-dimensional density parameter; Count the fracture directions of all vascular branch nodes and generate the vascular fracture direction parameters according to the angular distribution of the fracture directions in the three-dimensional coordinate system; The volume fraction series was generated by measuring the ratio of the volume of the tumor necrotic area to the total volume of the tumor area at each time point.

3. The method according to claim 1, characterized in that The blood sample is subjected to spiral sorting processing, circulating tumor cells are separated by fluid control, deformation trajectory data of the circulating tumor cells are collected, and stiffness change parameters and deformation recovery parameters are generated according to the deformation trajectory data, including: injecting the blood sample into a spiral sorting device to separate circulating tumor cells from non-tumor cells by adjusting the direction of fluid dynamics; Collecting movement trajectory data of the separated circulating tumor cells in the fluid environment, wherein the movement trajectory data includes the deformation amplitude and deformation direction of the cells at different time points; Calculating the recovery speed after the deformation amplitude reaches a maximum value based on the curve of the deformation amplitude changing with time, and defining the reciprocal of the recovery speed as the deformation recovery parameter; The frequency of changes in the deformation direction in the moving trajectory is counted, and the product of the frequency and the deformation amplitude is defined as a stiffness change parameter.

4. The method according to claim 1, wherein Based on the synchronization parameter set, a hysteresis correlation relationship is established between the time series offset of the blood vessel rupture direction parameter and the recovery period of the deformation recovery parameter, so as to determine a weight matching result between the time series offset and the recovery period through cross-modal analysis, including: Counting the time delay between the time point when the time series offset reaches a peak value and the end time point of the recovery period to generate a lagged time series; Calculating the association weight between the time series offset and the recovery period according to the length distribution of the lagged time series; The association weight and the sum of the products in the synchronization parameter set are superimposed to generate a weight matching result.

5. The method according to claim 1, wherein Outputting lung cancer subtype classification results based on the weight matching results and the spatiotemporal distribution characteristics of the tumor necrosis area volume ratio includes: Extracting the distribution characteristics of the volume proportion values in the volume proportion sequence as they change with spatial positions, and associating the spatial positions with time points to generate spatiotemporal distribution characteristics; The weight matching result is mapped and matched with the spatiotemporal distribution feature, and the corresponding lung cancer subtype classification result is output according to a preset threshold interval.

6. The method according to claim 5, characterized in that Mapping and matching the weight matching result with the spatiotemporal distribution feature, and outputting the corresponding lung cancer subtype classification result according to a preset threshold interval, including: Obtaining the parameter value of each time point in the weighted matching result and the spatial position ratio value of the corresponding time point in the spatiotemporal distribution feature; Superimposing the parameter value and the proportion value at the same time point to generate a comprehensive parameter at each time point; Count the maximum and minimum values of the comprehensive parameters at all time points to generate dynamic parameter intervals; The dynamic parameter interval is compared with a preset threshold interval. If the dynamic parameter interval completely falls within any threshold interval, the lung cancer subtype classification result corresponding to the threshold interval is output.

7. A lung cancer type prediction system based on a fusion deep learning network, characterized in that: include: an acquisition module, configured to acquire CT images and blood samples from lung cancer patients, dynamically segment the CT images to extract vascular distribution data of the tumor region, calculate three-dimensional density parameters and vascular rupture direction parameters based on the vascular distribution data, and acquire a volume fraction sequence of the tumor necrosis region, the volume fraction sequence being generated by measuring the ratio of the volume of the tumor necrosis region to the total volume of the tumor region at each time point; a processing module, configured to perform spiral sorting processing on the blood sample, separate circulating tumor cells through fluid control, collect deformation trajectory data of the circulating tumor cells, and generate a stiffness change parameter and a deformation recovery parameter based on the deformation trajectory data; an analysis module, configured to jointly analyze the three-dimensional density parameter, the vascular rupture direction parameter, and the stiffness change parameter, extract a time series offset of the vascular rupture direction parameter, and synchronously match the time series offset with a dynamic response of the stiffness change parameter to generate a synchronization parameter set; an establishing module for establishing, based on the synchronization parameter set, a hysteresis correlation relationship between a time series offset of the blood vessel rupture direction parameter and a recovery period of the deformation recovery parameter, so as to determine a weighted matching result between the time series offset and the recovery period through cross-modal analysis; an output module, configured to output a lung cancer subtype classification result based on the weight matching result and the spatiotemporal distribution characteristics of the tumor necrosis area volume ratio; The three-dimensional density parameter, the vascular rupture direction parameter, and the stiffness change parameter are jointly analyzed to extract the time series offset of the vascular rupture direction parameter, and the time series offset is synchronously matched with the dynamic response of the stiffness change parameter to generate a synchronous parameter set, including: Obtain the directional angle values of the vascular rupture direction parameter at multiple time points to calculate the angle difference between adjacent time points, and generate a time series offset based on the angle difference; obtain the parameter value of the stiffness change parameter at the same time point, and extract the dynamic change value of the parameter value over time; align the time series offset with the dynamic change value on the time axis, calculate the sum of the products of the two at the same time point, and generate a synchronization parameter set.

8. A computing device, characterized in that The method comprises a processing component and a storage component; the storage component stores one or more computer instructions; the one or more computer instructions are used to be called and executed by the processing component to implement a lung cancer type prediction method based on a fusion deep learning network as described in any one of claims 1 to 6.

9. A computer storage medium, characterized in that A computer program is stored, and when the computer program is executed by a computer, the method for predicting the type of lung cancer based on a fusion deep learning network according to any one of claims 1 to 6 is implemented.

Images