Lung nodule benign and malignant fuzzy clustering method and system based on multi-dimensional feature modeling, electronic device and storage medium

Abstract

The application provides a lung nodule benign and malignant fuzzy clustering method and system based on multi-dimensional feature modeling, an electronic device and a storage medium, and relates to the technical field of multi-dimensional feature modeling. The method comprises the following steps: acquiring multi-view refractive index data and low-dose CT images of a lung nodule region; extracting local texture features of the lung nodule region from the low-dose CT images, and fusing the multi-view refractive index data and the local texture features to generate fusion feature information; based on the fusion feature information, performing multi-angle polarized light field scanning on the lung nodule region to generate time-series light field data; based on the time-series light field data, generating a pulse sequence by using a pulse neural network; inputting an entropy value result of the pulse sequence into a preset fuzzy membership function, performing fuzzy clustering analysis by using the fuzzy membership function, and outputting a probability distribution result of lung nodule benign and malignant classification. The application improves the accuracy of benign and malignant differentiation of micro-lung nodules under low-dose CT.

Description

Technical Field

[0001] This application relates to the field of multidimensional feature modeling technology, and in particular to a method, system, electronic device and storage medium for fuzzy clustering of benign and malignant lung nodules based on multidimensional feature modeling. Background Technology

[0002] In clinical lung CT screening, differentiating between benign and malignant pulmonary nodules smaller than 5 mm presents significant challenges. Due to the small size and indistinct features of the lesions, conventional imaging analysis methods are insufficient to obtain adequate diagnostic evidence, necessitating the development of novel detection technologies capable of accurately quantifying the tissue characteristics of nodules.

[0003] Current advanced solutions employ high-resolution micro-CT combined with digital image processing technology. They acquire the fine structure of nodules through sub-millimeter-level tomographic scanning and utilize morphological analysis and texture feature quantification algorithms to establish a machine learning-based classification model. This approach enhances feature extraction capabilities by improving spatial resolution.

[0004] This method relies on high-radiation-dose scanning, making it unsuitable for routine screening, and the processing is complex and time-consuming. Feature extraction is sensitive to equipment parameters, and data from different models is not comparable. Model training requires a large number of labeled samples, leading to unstable generalization performance in practical applications. Summary of the Invention

[0005] This application provides a fuzzy clustering method, system, electronic device, and storage medium for benign and malignant lung nodules based on multidimensional feature modeling, in order to solve the problem of low accuracy in distinguishing between benign and malignant small lung nodules under low-dose CT in the prior art.

[0006] To address the aforementioned technical problems, in a first aspect, this application provides a fuzzy clustering method for benign and malignant lung nodules based on multidimensional feature modeling, comprising:

[0007] Acquire multi-view refractive index data and low-dose CT images of pulmonary nodule regions;

[0008] Local texture features of the lung nodule region are extracted from the low-dose CT images, and the multi-view refractive index data and the local texture features are fused to generate fused feature information;

[0009] Based on the fused feature information, the lung nodule region is scanned by a multi-angle polarized light field to generate time-series light field data;

[0010] Based on the aforementioned temporal optical field data, a pulse sequence is generated using a spiking neural network.

[0011] The entropy value of the pulse sequence is input into a preset fuzzy membership function, and fuzzy clustering analysis is performed through the fuzzy membership function to output the probability distribution results of the benign and malignant classification of lung nodules.

[0012] Optionally, the step of inputting the entropy value of the pulse sequence into a preset fuzzy membership function, performing fuzzy clustering analysis through the fuzzy membership function, and outputting the probability distribution results of the benign and malignant classification of lung nodules includes:

[0013] The pulse sequence is divided into time-domain windows, and the pulse firing frequency distribution within each time-domain window is calculated;

[0014] Based on the pulse firing frequency distribution, the entropy value of each time domain window is calculated to form an entropy value sequence;

[0015] The entropy sequence is input into the fuzzy membership function, the cluster centers are initialized through the fuzzy membership function, and the cluster centers are iteratively optimized to output the probability distribution results of the benign and malignant classification of lung nodules.

[0016] Optionally, the step of inputting the entropy sequence into the fuzzy membership function, initializing cluster centers through the fuzzy membership function, iteratively optimizing based on the cluster centers, and outputting the probability distribution result of benign and malignant lung nodules includes:

[0017] The entropy sequence is divided into multiple feature segments, and the entropy fluctuation features of each feature segment are extracted.

[0018] Based on the entropy fluctuation characteristics, calculate the center offset parameter and distribution width parameter of the fuzzy membership function;

[0019] The cluster centers are initialized using the center offset parameter, and the cluster radius constraint is set based on the distribution width parameter.

[0020] Using the fused feature information as an input vector, the membership weight between the input vector and the cluster center is iteratively calculated under the constraint of the cluster radius;

[0021] When the change in membership weights in two consecutive iterations is less than the preset difference threshold, the iteration is terminated and the probability distribution result is output.

[0022] Optionally, calculating the center offset parameter and distribution width parameter of the fuzzy membership function based on the entropy fluctuation characteristics includes:

[0023] Extract extreme value features from the entropy fluctuation features, map the extreme value features to a preset feature space coordinate system, and generate a center offset reference vector;

[0024] Calculate the discrete feature index of the entropy fluctuation characteristic, and determine the distribution width scaling factor based on the ratio of the discrete feature index to the preset discrete threshold.

[0025] The center offset reference vector is multiplied by the anatomical position correction coefficient of the lung nodule to generate the center offset parameter of the fuzzy membership function;

[0026] The distribution width scaling factor is multiplied by the standard distribution cardinality to generate the distribution width parameter of the fuzzy membership function.

[0027] Optionally, generating a pulse sequence using a spiking neural network based on the temporal optical field data includes:

[0028] The refractive index change rate at continuous time points is calculated on the time-series optical field data to generate refractive index gradient data at all time points.

[0029] The refractive index gradient data at all time points are input into the synaptic transmission channel of the spiking neural network in chronological order;

[0030] When the magnitude of the refractive index gradient data at each time point exceeds a preset neuron activation threshold, a pulse firing event is triggered.

[0031] Multiple pulse emission events triggered within a predetermined time period are spatiotemporally integrated to output a pulse sequence with spatial coding characteristics.

[0032] Optionally, the step of fusing the multi-view refractive index data and the local texture features to generate fused feature information includes:

[0033] The multi-view refractive index data is separated into multiple single-view refractive index components according to different view directions;

[0034] The local texture features are subjected to structural feature separation to obtain the lobed contour component and the burr morphology component.

[0035] Establish the spatial positional mapping relationship between the single-viewpoint refractive index component and the lobed feature contour component;

[0036] Based on the spatial location mapping relationship, viewpoint aligned refractive index distribution data is generated;

[0037] The refractive index distribution data is adjusted by a dynamic weighting mechanism.

[0038] The adjusted refractive index distribution data is integrated with the burr morphology components to generate fused feature information.

[0039] Optionally, the step of performing multi-angle polarization light field scanning on the lung nodule region based on the fused feature information to generate time-series light field data includes:

[0040] Based on the fusion feature information, a scan path sequence containing N polarization angles is generated, where N is greater than or equal to 3;

[0041] According to the polarization angle sequence of the scanning path sequence, modulated beams with corresponding polarization directions are sequentially emitted into the lung nodule region;

[0042] The system receives the emitted light beam after it is refracted by lung nodule tissue in real time, and records the light field intensity distribution data at each polarization angle based on the emitted light beam.

[0043] The light field intensity distribution data acquired under continuous polarization angles are arranged in chronological order according to the scanning time to form time-series light field data.

[0044] Secondly, this application provides a fuzzy clustering system for benign and malignant lung nodules based on multidimensional feature modeling, comprising:

[0045] The acquisition module is used to acquire multi-view refractive index data and low-dose CT images of the lung nodule region;

[0046] The extraction module is used to extract local texture features of the lung nodule region from the low-dose CT images and fuse the multi-view refractive index data and the local texture features to generate fused feature information.

[0047] The scanning module is used to perform multi-angle polarization light field scanning on the lung nodule region based on the fused feature information to generate time-series light field data;

[0048] The generation module is used to generate a pulse sequence based on the time-series optical field data using a spiking neural network;

[0049] The input module is used to input the entropy value of the pulse sequence into a preset fuzzy membership function, perform fuzzy clustering analysis through the fuzzy membership function, and output the probability distribution results of the benign and malignant classification of lung nodules.

[0050] Thirdly, this application provides an electronic device, comprising:

[0051] Memory, used to store computer programs;

[0052] A processor, configured to implement the steps of the fuzzy clustering method for benign and malignant lung nodules based on multidimensional feature modeling as described in the first aspect above when executing the computer program.

[0053] Fourthly, this application provides a computer-readable storage medium storing a computer program that, when executed by a processor, can implement the steps of the fuzzy clustering method for benign and malignant lung nodules based on multidimensional feature modeling as described in the first aspect above.

[0054] This application provides a fuzzy clustering method for benign and malignant lung nodules based on multidimensional feature modeling. The method includes: acquiring multi-view refractive index data and low-dose CT images of the lung nodule region; extracting local texture features of the lung nodule region from the low-dose CT images, and fusing the multi-view refractive index data and the local texture features to generate fused feature information; performing multi-angle polarization light field scanning on the lung nodule region based on the fused feature information to generate temporal light field data; generating a pulse sequence using a spiking neural network based on the temporal light field data; inputting the entropy value of the pulse sequence into a preset fuzzy membership function, performing fuzzy clustering analysis through the fuzzy membership function, and outputting the probability distribution results of benign and malignant lung nodule classification.

[0055] The technical solution provided in this application has the following beneficial effects:

[0056] This application achieves dual-modal data acquisition of optical properties and imaging features of nodular regions, providing a multi-source information foundation for subsequent analysis. A correlation model between imaging features and optical properties is established to enhance the characterization ability of nodular tissue heterogeneity. Microstructural changes in nodular tissue are acquired through dynamic optical detection, improving feature capture sensitivity. Continuous optical signals are converted into neural pulse codes to achieve efficient feature extraction of dynamic tissue characteristics. A diagnostic decision model based on probability distribution is established to improve the reliability of differentiating between benign and malignant small pulmonary nodules.

[0057] Furthermore, this application also divides the pulse sequence into time-domain windows, calculates the pulse firing frequency distribution within each window and converts it into an entropy value sequence, uses the entropy value sequence to dynamically construct a fuzzy membership function, initializes the cluster centers, and then iteratively optimizes to finally output the benign and malignant probability distribution.

[0058] Furthermore, this technical solution achieves quantitative evaluation of the dynamic characteristics of nodule tissue through temporal characteristic analysis of pulse sequences, and improves the accuracy and reliability of benign and malignant differentiation of small lung nodules by combining adaptive fuzzy clustering algorithm.

[0059] These or other aspects of this application will become more apparent in the following description of the embodiments. Attached Figure Description

[0060] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0061] Figure 1 A flowchart illustrating a fuzzy clustering method for benign and malignant lung nodules based on multidimensional feature modeling, provided for embodiments of this application;

[0062] Figure 2 A schematic diagram illustrating a specific implementation of a fuzzy clustering method for benign and malignant lung nodules based on multidimensional feature modeling, provided in this application embodiment;

[0063] Figure 3 This is a schematic diagram illustrating another specific implementation of a fuzzy clustering method for benign and malignant lung nodules based on multidimensional feature modeling, provided in an embodiment of this application.

[0064] Figure 4 This is a schematic diagram of the structure of a fuzzy clustering system for benign and malignant lung nodules based on multidimensional feature modeling, provided in an embodiment of this application. Detailed Implementation

[0065] In the field of differentiating between benign and malignant pulmonary nodules, existing methods based on high-resolution computed tomography (CT) can acquire fine nodule structures, but their high radiation levels limit their clinical application. Furthermore, the feature extraction process is sensitive to equipment parameters, leading to poor comparability of data from different CT models. More critically, these methods rely on static image analysis and cannot capture the dynamic optical characteristics of nodule tissue. They lack effective solutions to the feature ambiguity problem of small pulmonary nodules (<5mm), ultimately affecting the reliability of diagnosis.

[0066] To address the aforementioned issues, this application proposes a fuzzy clustering method for benign and malignant pulmonary nodules based on multidimensional feature modeling. Its innovation lies in combining the texture features of low-dose CT with the dynamic optical characteristics of multi-angle polarized light field scanning, and using a pulse neural network to quantitatively characterize tissue heterogeneity. Specifically, firstly, a multidimensional feature space is constructed by fusing CT image features and refractive index data. Then, the dynamic refractive characteristics of nodule tissue are obtained through time-series light field scanning, and the fuzzy clustering model is dynamically optimized using pulse sequence entropy values. This method overcomes the limitations of traditional static image analysis. Through multidimensional synergistic analysis of optical properties and image features, it improves the accuracy of benign and malignant identification of small pulmonary nodules while maintaining a low radiation dose, effectively solving the problems of strong equipment dependence and limited feature representation in existing technologies.

[0067] To enable those skilled in the art to better understand the present application, the present application will be further described in detail below with reference to the accompanying drawings and specific embodiments. Obviously, the described embodiments are merely some embodiments of the present application, and not all embodiments. Based on the embodiments in this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0068] The core of this application is to provide a fuzzy clustering method for benign and malignant lung nodules based on multidimensional feature modeling. A flowchart of one specific implementation is shown below. Figure 1 As shown, the method includes:

[0069] Step 101: Acquire multi-view refractive index data and low-dose CT images of the lung nodule region.

[0070] In step 101, multi-view refractive index data refers to a set of light refractive characteristic data of the lung nodule region obtained through optical scanning at different angles, including optical parameters from multiple viewing directions. Low-dose CT images refer to cross-sectional lung images obtained using computed tomography (CT) techniques that reduce radiation dose.

[0071] In this embodiment, an optical scanning device is first used to collect optical characteristics of the lung nodule region from different angles to obtain multiple sets of refractive index data. At the same time, a low-dose CT scanner is used to acquire lung image data. The two types of data are stored in the processing system to complete the raw data acquisition.

[0072] For example, in a case of lung nodule detection in a hospital, optical scans were performed on the nodule area in the upper lobe of the patient's right lung at three directions: 0 degrees, 45 degrees, and 90 degrees. The refractive index data were obtained as 1.35, 1.42, and 1.38, respectively. At the same time, low-dose CT image data were acquired and stored in the image processing system.

[0073] Step 102: Extract local texture features of the lung nodule region from the low-dose CT images, and fuse the multi-view refractive index data and the local texture features to generate fused feature information.

[0074] In step 102, local texture features refer to the surface morphological features of nodules extracted from CT images, including lobulated contour data and spiculated morphological data. Fusion feature information refers to the comprehensive feature dataset resulting from integrating optical property data and image feature data using a specific algorithm.

[0075] In this embodiment, the boundary of the nodule region is identified from low-dose CT images, surface concavity and convexity features are extracted as lobulation feature data, edge burr features are extracted as burr feature data, multi-view refractive index data is decomposed into independent components according to the viewing direction, a spatial correspondence between optical data and lobulation feature data is established, and optical data and burr feature data are fused through a weighted algorithm to generate a comprehensive dataset containing optical and image features.

[0076] For example, taking the aforementioned patient data as an example, the nodule lobulation feature parameter of 0.25 and the spiculation feature parameter of 0.12 mm were extracted from the CT images. The refractive index data from the three perspectives were then correlated with the lobulation feature. After assigning different weights according to the intensity of the lobulation feature, it was integrated with the spiculation feature to finally generate a fused feature value of 1.277.

[0077] Step 103: Based on the fused feature information, perform multi-angle polarization light field scanning on the lung nodule region to generate time-series light field data.

[0078] In step 103, the temporal optical field data refers to the set of optical intensity distribution data obtained at different polarization angles arranged in chronological order, which has both temporal and spatial dimensions.

[0079] In this embodiment of the application, the scanning angle sequence is determined based on the fusion feature information, the polarization light device is controlled to emit modulated light beams at different angles in sequence, the emitted light signal after passing through the nodule tissue is received in real time, the light intensity distribution at each angle is recorded, and the data is organized and stored as a three-dimensional data array in the order of acquisition time.

[0080] For example, based on the aforementioned fusion feature value of 1.277, the system determines the scanning angles as 0 degrees, 30 degrees, 60 degrees, 90 degrees and 120 degrees, and sequentially scans to obtain light intensity data of 85, 78, 82, 88 and 75 at each angle, which are stored as time-series data in chronological order.

[0081] Step 104: Based on the time-series optical field data, generate a pulse sequence using a spiking neural network.

[0082] In step 104, the pulse sequence refers to a set of discrete pulse signals converted from continuous optical signals by a neural network, which has spatiotemporal coding characteristics.

[0083] In this embodiment of the application, the rate of change of refractive index at adjacent time points is calculated from the time-series optical field data, and the rate of change data is input into a pulse neural network. When the change amplitude exceeds a set threshold, a pulse signal is triggered. Multiple pulse events are integrated within a fixed time window to generate a pulse sequence with spatial location encoding.

[0084] For example, the rate of change of adjacent points in the aforementioned time series data is calculated. When the rate of change exceeds 0.03, a pulse is triggered, and the pulse sequence 01101001 is recorded within 200 milliseconds, where 1 indicates that there is a pulse event and 0 indicates that there is no pulse event.

[0085] Step 105: Input the entropy value of the pulse sequence into a preset fuzzy membership function, perform fuzzy clustering analysis through the fuzzy membership function, and output the probability distribution results of the benign and malignant classification of lung nodules.

[0086] In step 105, the entropy value result refers to a quantitative indicator obtained by calculating the disorder of the pulse sequence. The probability distribution result refers to the assessment of the likelihood that the nodules belong to the benign or malignant categories obtained through fuzzy clustering analysis.

[0087] In this embodiment, the pulse sequence is divided into multiple time periods, the distribution frequency of pulse events in each time period is calculated, the entropy value is calculated based on the frequency distribution, the entropy value sequence is used to construct a fuzzy membership function, the cluster center position is initialized, the membership relationship between nodule features and cluster centers is determined through iterative calculation, and the benign or malignant probability is output when the change is less than a set threshold.

[0088] For example, the pulse sequence 01101001 is divided into four time periods, and the entropy values ​​of each period are calculated as 0.56, 0.69, 0.56, and 0.69. After constructing a fuzzy function and performing three iterations, the final output shows that the probability of the nodule being malignant is 72% and the probability of it being benign is 28%.

[0089] This method, by fusing optical properties and imaging features, combined with dynamic light field scanning and pulsed neural network processing, achieves accurate differentiation between benign and malignant small pulmonary nodules. It improves diagnostic reliability while maintaining a low radiation dose, providing an objective basis for clinical decision-making.

[0090] To address the issue of ambiguous features in differentiating between benign and malignant small pulmonary nodules and further improve diagnostic accuracy, in some embodiments, step 105 involves inputting the entropy value of the pulse sequence into a preset fuzzy membership function. Fuzzy clustering analysis is then performed through the fuzzy membership function to output the probability distribution results for classifying pulmonary nodules as benign or malignant. Figure 2 As shown, it includes:

[0091] Step 201: Divide the pulse sequence into time-domain windows and calculate the pulse firing frequency distribution within each time-domain window.

[0092] In step 201, time-domain window division refers to a processing method that divides continuous pulse signals into fixed time periods, with each window containing several pulse events. Pulse firing frequency distribution refers to the statistical result of the density of pulse signal occurrences within a single time-domain window.

[0093] In this embodiment, a suitable time period length is first determined, and the entire pulse sequence is divided into several equal-length time periods. The number of times the pulse signal appears in each time period is counted, and the pulse density distribution in that time period is calculated, providing basic data for subsequent entropy value calculation.

[0094] Step 202: Calculate the entropy value of each time-domain window based on the pulse firing frequency distribution to form an entropy value sequence.

[0095] In step 202, the entropy sequence refers to the ordered set of values ​​obtained by calculating the disorder of pulse distribution within each time window, used to quantify the spatiotemporal characteristics of the pulse signal. The entropy sequence is the entropy result.

[0096] In this embodiment of the application, based on the pulse density distribution data of each time-domain window, the entropy value of each window is calculated by applying the information entropy calculation formula, and the entropy values ​​of all windows are arranged in chronological order to form an entropy value sequence that reflects the dynamic characteristics of the entire pulse sequence.

[0097] Step 203: Input the entropy sequence into the fuzzy membership function, initialize the cluster centers through the fuzzy membership function, iteratively optimize based on the cluster centers, and output the probability distribution results of the benign and malignant classification of lung nodules.

[0098] In step 203, the fuzzy membership function refers to a mathematical function used to describe the degree of association between data points and cluster centers, and its parameters are determined by the entropy sequence characteristics.

[0099] In this embodiment, the center position and distribution range of the fuzzy function are first determined based on the statistical characteristics of the entropy sequence. The function is then used to initialize two cluster centers, representing benign and malignant features respectively. The fused feature data is then used as input, and iterative calculations are performed under the constraints of the fuzzy function. The membership relationship between the data points and the cluster centers is continuously adjusted. The calculation stops when the adjustment range is less than the set standard, and the final benign / malignant probability assessment is output.

[0100] Here is a specific example:

[0101] In the same lung nodule detection case, based on the previously obtained pulse sequence 01101001, the sequence was first divided into four time periods at 50-millisecond intervals. The number of pulse occurrences in each time period was counted as 1, 2, 1, and 2, respectively, and the pulse firing frequencies for each time period were calculated to be 0.25, 0.5, 0.25, and 0.5, respectively. The frequency calculation was obtained by dividing the number of pulses in the time period by the total possible number of 4. Then, the entropy formula was used. Calculate the entropy value for each time period, where This represents the entropy value in the i-th time period. This represents the pulse frequency in the i-th time interval, and the first time interval is calculated as follows: The second time period is Similarly, the entropy values ​​for the third and fourth time periods were obtained as 0.56 and 0.69, respectively, forming an entropy value sequence of 0.56, 0.69, 0.56, and 0.69. This sequence was input into a preset fuzzy membership function. The function automatically adjusted the center position parameter to 1.15 and the distribution range parameter to 0.8 based on the entropy value fluctuation characteristics, initializing the two cluster centers to 1.21 and 1.09, respectively, and setting the cluster radius to 0.8 × 0.5 = 0.4. The previously generated fusion feature value of 1.277 was used as input data, and iterative calculations were performed under the radius constraint. The first iteration yielded a malignant membership degree of 0.68 and a benign membership degree of 0.32. The second iteration yielded 0.71 and 0.29, respectively. The change of 0.03 was greater than the threshold of 0.01, so the iteration continued. The third iteration yielded 0.72 and 0.28, respectively. The change of 0.01 reached the stopping criterion, and the final output showed a malignant probability of 72% and a benign probability of 28% for the nodule.

[0102] In this embodiment, the method achieves quantitative evaluation of the dynamic characteristics of lung nodule tissue through the analysis of the time-domain characteristics of pulse signals. Combined with an adaptive fuzzy clustering algorithm, it improves the accuracy and reliability of differentiating between benign and malignant small lung nodules, providing an objective basis for clinical diagnosis.

[0103] To address the issue of unreasonable fuzzy clustering parameter settings in the classification of benign and malignant lung nodules and further improve diagnostic accuracy, in some embodiments, step 203 involves inputting the entropy value sequence into the fuzzy membership function, initializing cluster centers through the fuzzy membership function, iteratively optimizing based on the cluster centers, and outputting the probability distribution results for the classification of benign and malignant lung nodules, including:

[0104] Step 301: Divide the entropy sequence into multiple feature segments and extract the entropy fluctuation features of each feature segment.

[0105] In step 301, the feature segment is an analysis unit composed of multiple adjacent "entropy values ​​of time-domain windows" according to specific rules. The difference is that the entropy value of the time-domain window is the basic calculation unit, while the feature segment is a higher-level analysis structure formed by secondary grouping of multiple window entropy values, which is used to capture the entropy value fluctuation pattern on a larger time scale.

[0106] In this embodiment of the application, a suitable segment length is first determined, the entire entropy value sequence is divided into several continuous segments, the entropy value in each segment is statistically analyzed, and the characteristic quantity that can reflect its change range is calculated, so as to provide a basis for subsequent parameter calculation.

[0107] Step 302: Calculate the center offset parameter and distribution width parameter of the fuzzy membership function based on the entropy fluctuation characteristics.

[0108] In step 302, the center offset parameter represents the adjustment amount of the center position of the fuzzy membership function relative to the standard position. Its physical meaning is to reflect the overall offset trend of the lung nodule entropy value fluctuation, used to correct systematic biases caused by nodule location characteristics. The distribution width parameter represents the distribution range control amount of the fuzzy membership function. Its physical meaning is to quantify the dispersion of the lung nodule entropy value fluctuation, used to constrain the cluster radius to suppress noise interference. This parameter is directly related to the degree of heterogeneity of the nodule tissue.

[0109] In this embodiment of the application, the magnitude of adjustment required for the center position is calculated based on the entropy fluctuation characteristics of each feature segment through a preset conversion rule. At the same time, the scaling ratio of the function distribution range is determined based on the dispersion of the fluctuation characteristics, and finally two key parameters are output.

[0110] Step 303: Initialize the cluster centers using the center offset parameter, and set cluster radius constraints based on the distribution width parameter.

[0111] In step 303, the cluster radius constraint is a threshold that limits the maximum distance between a data point and the cluster center, and is used to control the range of data points during the iteration process.

[0112] In this embodiment, the position of the initial cluster center is adjusted using the center offset parameter to better match the actual data distribution. At the same time, a reasonable search radius is set according to the distribution width parameter to avoid interference from outliers during the iteration process.

[0113] Step 304: Using the fused feature information as an input vector, iteratively calculate the membership weight between the input vector and the cluster center under the cluster radius constraint.

[0114] In step 304, the membership weight refers to the probability value of a data point belonging to a certain cluster center, reflecting the degree of similarity between the two.

[0115] In this embodiment of the application, the fused feature information is converted into vector form, and the distance between it and each cluster center is calculated within a set radius. The distance is then converted into a weight value in probability form to complete one iteration calculation.

[0116] Step 305: When the change in membership weights in two consecutive iterations is less than the preset difference threshold, terminate the iteration and output the probability distribution result.

[0117] In step 305, the difference threshold refers to the minimum change criterion for determining whether the iterative process has converged.

[0118] In this embodiment of the application, the weight changes generated by two adjacent iterations are compared. When the weight changes of all data points are less than the set standard, the result is considered to be stable, the calculation process is terminated, and the final classification probability is output.

[0119] Here is a specific example:

[0120] Based on the aforementioned lung nodule detection case, the obtained entropy value sequence 0.56, 0.69, 0.56, 0.69 was divided into two feature segments. Each segment contained two adjacent entropy values. The range between the first segment (0.56 and 0.69) and the second segment (0.56 and 0.69) was calculated to be 0.13. The average of the ranges of the two segments, 0.13, was taken as the entropy fluctuation feature. According to the system's preset conversion rules, when the fluctuation feature was in the range of 0.1 to 0.2, the center offset parameter was calculated as 1.1 plus the fluctuation feature multiplied by 0.5, resulting in 1.1 + 0.13 × 0.5 = 1.165. The distribution width parameter was calculated as 0.7 plus the fluctuation feature multiplied by 0.8, resulting in 0.7 + 0.13 × 0.8 = 0.804. The center offset parameter of 1.165 was used to initialize the two clusters. The values ​​of the cluster centers were set to 1.165 x 1.05 = 1.223 and 1.165 x 0.95 = 1.107, respectively. Simultaneously, based on the distribution width parameter of 0.804, the cluster radius constraint was set to 0.804 x 0.5 = 0.402. The light field and texture coupling feature value of 1.277 from the fused feature information was used as the input vector. Under the constraint that the radius does not exceed 0.402, the distance between the input vector and the two cluster centers was calculated. The first iteration yielded membership weights of 0.68 and 0.32, respectively. The second iteration yielded weights of 0.71 and 0.29. The weight change between the two iterations was 0.03, which is greater than the preset difference threshold of 0.01, so the iteration continued. The third iteration yielded weights of 0.72 and 0.28, with a change of 0.01, reaching the termination condition. The final output showed that the malignancy probability of the nodule was 72%, and the benign probability was 28%.

[0121] In this embodiment, the method achieves accurate modeling of lung nodule features by dynamically adjusting fuzzy clustering parameters, making the classification results more consistent with actual clinical situations and improving the reliability of benign and malignant differentiation of small lung nodules.

[0122] To further improve the accuracy of fuzzy membership function parameter calculation, in some embodiments, step 302: calculating the center offset parameter and distribution width parameter of the fuzzy membership function based on the entropy fluctuation characteristics, includes:

[0123] Step 401: Extract extreme value features from the entropy fluctuation features, map the extreme value features to a preset feature space coordinate system, and generate a center offset reference vector.

[0124] In step 401, the extreme value feature refers to the difference between the maximum and minimum values ​​in the entropy fluctuation feature. The center offset reference vector refers to the basic adjustment amount obtained after transforming the extreme value feature to a specific coordinate system.

[0125] In this embodiment of the application, the maximum and minimum values ​​in the entropy fluctuation feature are first identified, and the difference between the two is calculated as the extreme value feature. Then, the extreme value is mapped to the feature space coordinate system through a preset linear transformation formula to generate the basic vector for subsequent parameter calculation.

[0126] Step 402: Calculate the discrete feature index of the entropy fluctuation characteristic, and determine the distribution width scaling factor based on the ratio of the discrete feature index to the preset discrete threshold.

[0127] In step 402, the discrete characteristic index of entropy fluctuation refers to a statistical measure used to quantify the degree of dispersion of the entropy value sequence. This index is obtained by calculating the average of the absolute values ​​of the differences between adjacent entropy values. Its physical meaning is to characterize the drastic change in entropy value of lung nodules over different time periods; the larger the value, the more obvious the heterogeneity of the nodule tissue. The distribution width scaling factor refers to the adjustment coefficient of the function distribution range determined according to the degree of dispersion.

[0128] In this embodiment of the application, the average value of the change amplitude of adjacent values ​​of entropy fluctuation characteristics is calculated as a discrete index. This index is compared with the standard threshold preset by the system, and the final scaling factor is determined according to the proportional relationship.

[0129] Step 403: Multiply the center offset reference vector with the lung nodule anatomical position correction coefficient to generate the center offset parameter of the fuzzy membership function.

[0130] In step 403, the anatomical location correction coefficient for lung nodules is derived from a pre-established weighted mapping table of lung lobe regions. Different weight values ​​are assigned based on the location of the lung nodules in the upper / middle / lower lobes of the lungs in CT images. The physical meaning of this value is to reflect the anatomical statistical differences in the probability of malignancy in different lung lobe regions (e.g., if the malignancy rate of the upper lobe is higher, the correction coefficient is >1).

[0131] In this embodiment of the application, a preset correction coefficient table is queried according to the lung lobe region where the nodule is located to obtain the corresponding position weight, and the center offset reference vector is multiplied by the weight to obtain the final center offset parameter.

[0132] Step 404: Multiply the distribution width scaling factor by the standard distribution cardinality to generate the distribution width parameter of the fuzzy membership function.

[0133] In step 404, the standard distribution base is a constant pre-calibrated through statistical analysis of the entropy distribution of historically diagnosed pulmonary nodule cases. Its physical meaning is a benchmark value (unit consistent with the entropy dimension) characterizing the ideal distribution width of typical benign and malignant nodules under noise interference.

[0134] In this embodiment of the application, the calculated distribution width scaling factor is multiplied by the system's preset standard base to generate a distribution width parameter that conforms to the characteristics of the current nodule.

[0135] Here is a specific example:

[0136] In the aforementioned lung nodule detection case, based on the obtained entropy fluctuation feature of 0.13, this feature is first extracted as an extreme value feature. It is then mapped to the feature space coordinate system using the conversion formula y = 0.8x + 0.1, where y is the center offset reference vector value and x represents the extreme value feature 0.13. The calculated center offset reference vector value is 0.8 × 0.13 + 0.1 = 0.204. Next, the discrete feature index of the entropy sequence 0.56, 0.69, 0.56, and 0.69 is calculated. The average of the absolute values ​​of the differences between adjacent entropy values ​​is used as the index. The calculation process is |0.69 - 0.56| + |0.56 - 0.69| + |0.69 - 0.56| divided by 3 equals 0.13. The system's preset discrete threshold is 0. 0.15, the ratio is calculated to be 0.13 / 0.15≈0.867. According to the system rules, when the ratio is in the range of 0.8 to 1.0, the distribution width scaling factor is equal to the ratio value multiplied by 0.9, resulting in 0.867×0.9≈0.78. Since the nodule is located in the upper lobe of the right lung, the correction coefficient for the upper lobe region is obtained by querying the pre-stored anatomical position correction coefficient table. The center offset reference vector 0.204 is multiplied by 1.15 to obtain the center offset parameter 0.2346. The system's preset standard distribution base is 0.35. The distribution width scaling factor 0.78 is multiplied by 0.35 to obtain the distribution width parameter 0.273. The final generated fuzzy membership function has a center offset of 0.2346 and a distribution width of 0.273.

[0137] In this embodiment of the application, the method achieves adaptive calculation of fuzzy function parameters by quantitatively analyzing the entropy fluctuation characteristics and combining them with anatomical location information, making the cluster analysis more consistent with the actual clinical data characteristics and improving the accuracy of the judgment of benign and malignant lung nodules.

[0138] To further improve the accuracy of dynamic feature extraction of lung nodules, in some embodiments, step 104 involves generating a pulse sequence using a spiking neural network based on the temporal light field data, such as... Figure 3 As shown, it includes:

[0139] Step 501: Calculate the rate of change of refractive index at continuous time points on the time-series optical field data to generate refractive index gradient data at all time points.

[0140] In step 501, consecutive time points refer to multiple sampling moments arranged in a fixed time interval. For example, when performing polarized light field scanning on a lung nodule region, light field intensity data at 20 consecutive time points within 200 milliseconds are collected at 10-millisecond sampling intervals, forming a complete observation period with a time series of t1, t2, t3, ..., t20. Refractive index gradient data refers to data obtained by calculating the magnitude of refractive index changes between adjacent time points, reflecting the dynamic changes in the optical properties of the nodule tissue.

[0141] In this embodiment, the refractive index values ​​at each time point in the time-series optical field data are first determined, and then the change amplitude between adjacent time points is calculated sequentially to obtain a series of gradient data reflecting the rate of change of refractive index, providing a basis for subsequent pulse triggering.

[0142] Step 502: Input the refractive index gradient data at all time points into the synaptic transmission channel of the spiking neural network in chronological order.

[0143] In step 502, the synaptic transmission channel refers to the information transmission path that simulates the connection of biological neurons in a spiking neural network.

[0144] In this embodiment, the calculated refractive index gradient data are sequentially fed into the input channel of the neural network in chronological order, with each time point corresponding to a specific channel, thus maintaining the temporal continuity of the data.

[0145] Step 503: When the magnitude of the refractive index gradient data at each time point exceeds the preset neuron activation threshold, a pulse firing event is triggered.

[0146] In step 503, the neuron activation threshold refers to the minimum input intensity value required to trigger the pulse signal. A pulse firing event refers to the discrete electrical pulse signal generated by the simulated biological neurons in the spiking neural network when the input signal intensity exceeds a set threshold. Its physical meaning is to mark the moment when the optical properties of the nodule tissue change; each pulse event corresponds to a characteristic mutation at a specific time point.

[0147] In this embodiment, the refractive index gradient data at each time point is compared with a preset threshold. When the data value exceeds the threshold, a pulse signal is generated at the corresponding time point; otherwise, no signal is generated.

[0148] Step 504: Spatiotemporally integrate multiple pulse emission events triggered within a predetermined time period to output a pulse sequence with spatial coding characteristics.

[0149] In step 504, spatial coding characteristics refer to the position information carried in the pulse signal.

[0150] In this embodiment of the application, all triggered pulse signals are collected within a fixed time range, and integrated and arranged according to the time sequence of signal generation and the corresponding spatial location information to form a complete pulse sequence with spatiotemporal characteristics.

[0151] Here is a specific example:

[0152] In the aforementioned lung nodule detection case, based on the obtained temporal light field data (including light intensity data of 85, 78, 82, 88, and 75 collected at five polarization angles: 0°, 30°, 60°, 90°, and 120°), the refractive index change rate at adjacent time points was first calculated. Using the formula (current value - previous value) / previous value, four change rate values ​​were obtained: the change rate of 30° relative to 0° was (78-85) / 85≈-0.082, the change rate of 60° relative to 30° was (82-78) / 78≈0.051, the change rate of 90° relative to 60° was (88-82) / 82≈0.073, and the change rate of 120° relative to 90° was (75-88) / 88≈-0.148. These change rate data were then input into a spiking neural network in chronological order, and the neurons were set to activate... The threshold for activity is 0.05. When the absolute value of the rate of change exceeds this threshold, a pulse event is triggered. The absolute values ​​of the second, third, and fourth rates of change (0.082, 0.051, and 0.073, respectively) all exceed the threshold, thus generating a pulse signal 1 at the corresponding time point. The first rate of change does not reach the threshold, generating a signal 0, forming the original pulse sequence 0110. Within a 200-millisecond time window, based on the spatial distribution of the nodule region (upper edge, anterior edge, and lower edge), the pulse sequence 0110 is encoded as upper edge region 01 (indicating no pulse and pulse at the first two time points), anterior edge region 1 (pulse at the third time point), and lower edge region 0 (no pulse at the fourth time point). The final output is a complete pulse code 0110 with clear spatial location information, where pulse 1 indicates a change in the optical properties of the tissue in the region, and pulse 0 indicates no change.

[0153] In this embodiment, the method uses a spiking neural network to convert continuous optical signals into discrete pulse codes, effectively extracting the dynamic features of lung nodule tissue and providing a reliable data foundation for subsequent benign and malignant analysis.

[0154] To address the issue of insufficient fusion of multimodal medical data and further improve the comprehensiveness of lung nodule feature representation, in some embodiments, step 102: fusing the multi-view refractive index data and the local texture features to generate fused feature information includes:

[0155] Step 601: Separate the multi-view refractive index data into multiple single-view refractive index components according to different viewing directions.

[0156] In step 601, the single-viewpoint refractive index component refers to an independent refractive index data unit obtained from a specific scanning angle.

[0157] In the embodiments of this application, each scanning angle of the light field imaging is first determined, and the acquired multi-view refractive index data is divided into several independent data units according to the original acquisition angle. Each unit corresponds to optical characteristic data of a specific viewing direction.

[0158] Step 602: Perform structural feature separation on the local texture features to obtain the lobed contour component and the burr morphology component.

[0159] In step 602, the lobulation feature contour component refers to the characteristic parameter of the uneven surface of the nodule. The spiculation feature morphology component refers to the characteristic parameter of the spiculated protrusions at the edge of the nodule.

[0160] In this embodiment, edge detection and morphological analysis are performed on the nodule region in the CT image, the undulation features of the surface contour are extracted as the lobulation feature component, and the length and density features of the edge burrs are extracted as the burr feature component.

[0161] Step 603: Establish the spatial position mapping relationship between the single-viewpoint refractive index component and the lobed profile component.

[0162] In step 603, the spatial location mapping relationship refers to the coordinate correspondence rule between optical data and image features in three-dimensional space.

[0163] In the embodiments of this application, a coordinate transformation relationship is established between the refractive index data of each viewpoint and the lobed features of the nodule surface to ensure that the optical property data can accurately correspond to the specific spatial location of the image features.

[0164] Step 604: Based on the spatial location mapping relationship, generate viewpoint-aligned refractive index distribution data.

[0165] In step 604, the viewpoint-aligned refractive index distribution data refers to the standardized optical property dataset after spatial correction.

[0166] In this embodiment of the application, the refractive index data from various perspectives are subjected to coordinate transformation and normalization based on the established spatial mapping relationship to generate a refractive index dataset with uniform position.

[0167] Step 605: Adjust the refractive index distribution data using a dynamic weighting mechanism.

[0168] In step 605, the dynamic weighting mechanism refers to the data fusion weighting rules that are automatically adjusted based on the intensity of the foliation features.

[0169] In this embodiment, the intensity distribution characteristics of the lobed feature contour component are analyzed, and corresponding fusion weight values ​​are assigned to the refractive index data from different viewpoints to perform weighted processing on the refractive index distribution data.

[0170] Step 606: Integrate the adjusted refractive index distribution data with the burr morphology component to generate fused feature information.

[0171] In step 606, feature integration refers to the process of comprehensively calculating the weighted optical data and burr features.

[0172] In this embodiment of the application, the weighted refractive index data and the burr morphology component are calculated using a specific algorithm to generate a comprehensive feature value that includes both optical properties and morphological features.

[0173] Here is a specific example:

[0174] In the aforementioned lung nodule detection case, the refractive index data (1.35, 1.42, and 1.38) acquired from three scanning angles (0 degrees, 45 degrees, and 90 degrees) were first stored as independent data units. The surface features of the nodules were analyzed from the patient's CT images, and the contour undulation parameter (0.25, lobulation sign) and edge spiculation length (0.12 mm, spiculation sign) were measured. A correspondence between the refractive index data and the lobulation features was established, and the refractive index data was mapped to the lobulation feature space using a coordinate transformation formula. At 0 degrees: , 45 degrees: , 90 degrees: Based on the intensity distribution with a foliation parameter of 0.25, the system automatically assigns weights to each viewing angle as follows: 1.2 for 0-degree viewing angle, 0.9 for 45-degree viewing angle, and 1.0 for 90-degree viewing angle. The converted refractive index data is then weighted and calculated. , , Finally, the weighted refractive index data are averaged ( The fusion value is obtained by integrating the burr characteristic parameters with the mean refractive index, using the formula: fusion value equals the mean refractive index multiplied by (1 plus the burr length coefficient), where the burr length coefficient is... The final fusion feature value is calculated. .

[0175] In this embodiment, the method achieves an organic combination of optical properties and image morphological features through precise spatial mapping and dynamic weighted fusion of multiple features, thereby improving the comprehensiveness and accuracy of lung nodule feature characterization and laying a reliable data foundation for subsequent benign and malignant differentiation.

[0176] To further improve the systematicness and completeness of optical feature acquisition for lung nodules, in some embodiments, step 103: based on the fused feature information, performing multi-angle polarization light field scanning on the lung nodule region to generate time-series light field data includes:

[0177] Step 701: Based on the fusion feature information, generate a scan path sequence containing N polarization angles, where N is greater than or equal to 3.

[0178] In step 701, the scan path sequence refers to a set of polarization angles arranged according to specific rules, which is used to guide the order and range of the light field scan.

[0179] In this embodiment, the correlation characteristics between light field and texture features in the fused feature information are first analyzed, and the number of polarization angles to be scanned and the specific angle values ​​are determined according to the feature intensity distribution, so as to generate the optimal scanning path scheme.

[0180] Step 702: According to the polarization angle sequence of the scanning path sequence, modulated beams with corresponding polarization directions are sequentially emitted into the lung nodule region.

[0181] In step 702, the modulated beam refers to an optical signal that has been adjusted through a specific polarization direction.

[0182] In this embodiment, the optical device is controlled to generate a modulated beam with a corresponding polarization direction according to the angular sequence of the preset scanning path, so as to accurately irradiate the target area of ​​the lung nodule and ensure that the beam parameters at each angle are consistent with the scanning requirements.

[0183] Step 703: Receive the emitted light beam after it is refracted by the lung nodule tissue in real time, and record the light field intensity distribution data at each polarization angle based on the emitted light beam.

[0184] In step 703, the lung nodule region refers to the anatomical area to be detected identified through medical imaging (such as CT), while "lung nodule tissue" specifically refers to the biological tissue entity with pathological characteristics within this region. The two have a spatial inclusion relationship—the "lung nodule region," as the spatial carrier of the detection operation, contains the specific "lung nodule tissue." The changes in the refractive characteristics of the modulated beam are mainly caused by the microstructure of the "lung nodule tissue," but the object of optical detection is the macroscopically defined "lung nodule region." The light field intensity distribution data refers to the set of light intensity measurements recorded at different spatial locations of the emitted beam.

[0185] In this embodiment, the emitted light beam passing through the nodule tissue is detected in real time, and the light intensity values ​​are measured simultaneously at multiple spatial sampling points to form a spatial distribution dataset reflecting the optical characteristics of the nodule.

[0186] Step 704: Arrange the light field intensity distribution data obtained under continuous polarization angles in the order of scanning time to form time-series light field data.

[0187] In step 704, the continuous polarization angle represents the optical field scanning process performed in a preset angular interval sequence. For example, the lung nodule region is sequentially scanned with polarized optical fields at four fixed angular intervals of 0 degrees, 45 degrees, 90 degrees, and 135 degrees. Each angular interval is 45 degrees and the scanning process is uninterrupted, forming a complete continuous angular scanning sequence from 0° to 45° to 90° to 135°.

[0188] In this embodiment, light intensity data collected at different polarization angles are arranged and stored in chronological order according to the actual scanning time to construct a three-dimensional data matrix with time stamps.

[0189] Here is a specific example:

[0190] In the aforementioned lung nodule detection case, based on the obtained fusion feature value of 1.277, the system first analyzed the feature intensity range of this value and determined that five polarization angles (0 degrees, 30 degrees, 60 degrees, 90 degrees, and 120 degrees) were needed for scanning. Following this angle sequence, the optical device sequentially emitted modulated beams with corresponding polarization directions into the nodule region, with the emission power adjusted to 12.77 milliwatts (base power 10 milliwatts multiplied by 1.277) according to the fusion feature value. At the 0-degree angle, the light intensity measurements of the received outgoing beam at nine sampling points in the nodule region were 85, 83, 86, 84, 87, 82, 85, 84, and 86, respectively. Similarly, at the 30-degree angle, the data for nine points were 78, 76, 79, 77, 80, and 75. The data are collected at 0 degrees, 78, 77, 79; at 60 degrees, 82, 80, 83, 81, 84, 79, 82, 81, 83; at 90 degrees, 88, 86, 89, 87, 90, 85, 88, 87, 89; and at 120 degrees, 75, 73, 76, 74, 77, 72, 75, 74, 76. These data are arranged and stored in chronological order of actual scanning time, with the 0-degree angle data collected first, followed by the 30-degree, 60-degree, and 90-degree data, and finally the 120-degree angle data, forming a time-series light field dataset containing 45 light intensity measurements. The scanning time interval for each angle is 50 milliseconds, and the entire scanning process takes 200 milliseconds. This time-series data will be used for subsequent refractive index gradient calculations and pulse sequence generation.

[0191] In this embodiment, the method achieves comprehensive acquisition of the optical properties of lung nodules through feature-guided multi-angle systematic scanning, providing a high-quality time-series data foundation for subsequent dynamic feature analysis and improving the reliability of small lung nodule detection.

[0192] Figure 4This application provides a schematic diagram of a specific implementation of a fuzzy clustering system for benign and malignant lung nodules based on multidimensional feature modeling, as shown in the embodiments of this application. Figure 4 The system may include:

[0193] The acquisition module 41 is used to acquire multi-view refractive index data and low-dose CT images of the lung nodule region.

[0194] Extraction module 42 is used to extract local texture features of the lung nodule region from the low-dose CT image and fuse the multi-view refractive index data and the local texture features to generate fused feature information.

[0195] The scanning module 43 is used to perform multi-angle polarization light field scanning on the lung nodule region based on the fused feature information to generate time-series light field data.

[0196] The generation module 44 is used to generate a pulse sequence based on the time-series optical field data using a spiking neural network.

[0197] The input module 45 is used to input the entropy value result of the pulse sequence into a preset fuzzy membership function, perform fuzzy clustering analysis through the fuzzy membership function, and output the probability distribution result of the benign and malignant classification of lung nodules.

[0198] The fuzzy clustering system for benign and malignant lung nodules based on multidimensional feature modeling in this application embodiment is used to implement the aforementioned fuzzy clustering method for benign and malignant lung nodules based on multidimensional feature modeling. Therefore, the specific implementation of the fuzzy clustering system for benign and malignant lung nodules based on multidimensional feature modeling can be found in the embodiment section of the fuzzy clustering method for benign and malignant lung nodules based on multidimensional feature modeling above. The specific implementation can be referred to the description of the corresponding embodiments, and will not be repeated here.

[0199] This application also provides an electronic device, comprising: a memory for storing a computer program; and a processor for executing the computer program to implement the steps of the above-described fuzzy clustering method for benign and malignant lung nodules based on multidimensional feature modeling.

[0200] This application also provides a computer-readable storage medium storing a computer program, which, when executed by a processor, implements the steps of any of the above-described methods for fuzzy clustering of benign and malignant lung nodules based on multidimensional feature modeling.

[0201] In one exemplary embodiment, the aforementioned computer-readable storage medium may include, but is not limited to, various media capable of storing computer programs, such as USB flash drives, read-only memory, random access memory, portable hard drives, magnetic disks, or optical disks.

[0202] Embodiments of the present invention also provide a computer program product, which includes a computer program that, when executed by a processor, implements the steps in any of the embodiments of the above-described fuzzy clustering method for benign and malignant lung nodules based on multidimensional feature modeling.

[0203] Those skilled in the art will further recognize that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, computer software, or a combination of both. To clearly illustrate the interchangeability of hardware and software, the components and steps of the various examples have been generally described in terms of functionality in the foregoing description. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementations should not be considered beyond the scope of this invention.

[0204] The foregoing has provided a detailed description of a fuzzy clustering method, system, electronic device, and storage medium for benign and malignant lung nodules based on multidimensional feature modeling, as provided in this application. Specific examples have been used to illustrate the principles and implementation methods of this application. The descriptions of the embodiments above are merely for the purpose of helping to understand the method and its core ideas. It should be noted that those skilled in the art can make various improvements and modifications to this application without departing from its principles, and these improvements and modifications also fall within the protection scope of this application.

Claims

1. A fuzzy clustering method for benign and malignant lung nodules based on multidimensional feature modeling, characterized in that, include: Acquire multi-view refractive index data and low-dose CT images of pulmonary nodule regions; Local texture features of the lung nodule region are extracted from the low-dose CT images, and the multi-view refractive index data and the local texture features are fused to generate fused feature information; Based on the fused feature information, the lung nodule region is scanned by a multi-angle polarized light field to generate time-series light field data; Based on the aforementioned temporal optical field data, a pulse sequence is generated using a spiking neural network. The entropy value of the pulse sequence is input into a preset fuzzy membership function, and fuzzy clustering analysis is performed through the fuzzy membership function to output the probability distribution results of the benign and malignant classification of lung nodules. The step of inputting the entropy value of the pulse sequence into a preset fuzzy membership function, performing fuzzy clustering analysis through the fuzzy membership function, and outputting the probability distribution results of the benign and malignant classification of lung nodules includes: The pulse sequence is divided into time-domain windows, and the pulse firing frequency distribution within each time-domain window is calculated; Based on the pulse firing frequency distribution, the entropy value of each time domain window is calculated to form an entropy value sequence; The entropy sequence is input into the fuzzy membership function, the cluster centers are initialized through the fuzzy membership function, and the cluster centers are iteratively optimized to output the probability distribution results of the benign and malignant classification of lung nodules. The step of generating a pulse sequence using a spiking neural network based on the temporal optical field data includes: The refractive index change rate at continuous time points is calculated on the time-series optical field data to generate refractive index gradient data at all time points. The refractive index gradient data at all time points are input into the synaptic transmission channel of the spiking neural network in chronological order; When the magnitude of the refractive index gradient data at each time point exceeds a preset neuron activation threshold, a pulse firing event is triggered. Multiple pulse emission events triggered within a predetermined time period are spatiotemporally integrated to output a pulse sequence with spatial coding characteristics; The process of fusing the multi-view refractive index data and the local texture features to generate fused feature information includes: The multi-view refractive index data is separated into multiple single-view refractive index components according to different view directions; The local texture features are subjected to structural feature separation to obtain the lobed contour component and the burr morphology component. Establish the spatial positional mapping relationship between the single-viewpoint refractive index component and the lobed profile component; Based on the spatial location mapping relationship, viewpoint aligned refractive index distribution data is generated; The refractive index distribution data is adjusted by a dynamic weighting mechanism. The adjusted refractive index distribution data is integrated with the burr morphology components to generate fused feature information.

2. The method according to claim 1, characterized in that, The process of inputting the entropy sequence into the fuzzy membership function, initializing cluster centers through the fuzzy membership function, iteratively optimizing based on the cluster centers, and outputting the probability distribution results of benign and malignant lung nodules includes: The entropy sequence is divided into multiple feature segments, and the entropy fluctuation features of each feature segment are extracted. Based on the entropy fluctuation characteristics, calculate the center offset parameter and distribution width parameter of the fuzzy membership function; The cluster centers are initialized using the center offset parameter, and the cluster radius constraint is set based on the distribution width parameter. Using the fused feature information as an input vector, the membership weight between the input vector and the cluster center is iteratively calculated under the constraint of the cluster radius; When the change in membership weights in two consecutive iterations is less than the preset difference threshold, the iteration is terminated and the probability distribution result is output.

3. The method according to claim 2, characterized in that, The step of calculating the center offset parameter and distribution width parameter of the fuzzy membership function based on the entropy fluctuation characteristics includes: Extract extreme value features from the entropy fluctuation features, map the extreme value features to a preset feature space coordinate system, and generate a center offset reference vector; Calculate the discrete feature index of the entropy fluctuation characteristic, and determine the distribution width scaling factor based on the ratio of the discrete feature index to the preset discrete threshold. The center offset reference vector is multiplied by the anatomical position correction coefficient of the lung nodule to generate the center offset parameter of the fuzzy membership function; The distribution width scaling factor is multiplied by the standard distribution cardinality to generate the distribution width parameter of the fuzzy membership function.

4. The method according to claim 1, characterized in that, The step of performing multi-angle polarization light field scanning on the lung nodule region based on the fused feature information to generate time-series light field data includes: Based on the fusion feature information, a scan path sequence containing N polarization angles is generated, where N is greater than or equal to 3; According to the polarization angle sequence of the scanning path sequence, modulated beams with corresponding polarization directions are sequentially emitted into the lung nodule region; The system receives the emitted light beam after it is refracted by lung nodule tissue in real time, and records the light field intensity distribution data at each polarization angle based on the emitted light beam. The light field intensity distribution data acquired under continuous polarization angles are arranged in chronological order according to the scanning time to form time-series light field data.

5. A fuzzy clustering system for benign and malignant lung nodules based on multidimensional feature modeling, characterized in that, include: The acquisition module is used to acquire multi-view refractive index data and low-dose CT images of the lung nodule region; The extraction module is used to extract local texture features of the lung nodule region from the low-dose CT images and fuse the multi-view refractive index data and the local texture features to generate fused feature information. The scanning module is used to perform multi-angle polarization light field scanning on the lung nodule region based on the fused feature information to generate time-series light field data; The generation module is used to generate a pulse sequence based on the time-series optical field data using a spiking neural network; The input module is used to input the entropy value of the pulse sequence into a preset fuzzy membership function, perform fuzzy clustering analysis through the fuzzy membership function, and output the probability distribution results of the benign and malignant classification of lung nodules; The step of inputting the entropy value of the pulse sequence into a preset fuzzy membership function, performing fuzzy clustering analysis through the fuzzy membership function, and outputting the probability distribution results of the benign and malignant classification of lung nodules includes: The pulse sequence is divided into time-domain windows, and the pulse firing frequency distribution within each time-domain window is calculated; Based on the pulse firing frequency distribution, the entropy value of each time domain window is calculated to form an entropy value sequence; The entropy sequence is input into the fuzzy membership function, the cluster centers are initialized through the fuzzy membership function, and the cluster centers are iteratively optimized to output the probability distribution results of the benign and malignant classification of lung nodules. The step of generating a pulse sequence using a spiking neural network based on the temporal optical field data includes: The refractive index change rate at continuous time points is calculated on the time-series optical field data to generate refractive index gradient data at all time points. The refractive index gradient data at all time points are input into the synaptic transmission channel of the spiking neural network in chronological order; When the magnitude of the refractive index gradient data at each time point exceeds a preset neuron activation threshold, a pulse firing event is triggered. Multiple pulse emission events triggered within a predetermined time period are spatiotemporally integrated to output a pulse sequence with spatial coding characteristics; The process of fusing the multi-view refractive index data and the local texture features to generate fused feature information includes: The multi-view refractive index data is separated into multiple single-view refractive index components according to different view directions; The local texture features are subjected to structural feature separation to obtain the lobed contour component and the burr morphology component. Establish the spatial positional mapping relationship between the single-viewpoint refractive index component and the lobed profile component; Based on the spatial location mapping relationship, viewpoint aligned refractive index distribution data is generated; The refractive index distribution data is adjusted by a dynamic weighting mechanism. The adjusted refractive index distribution data is integrated with the burr morphology components to generate fused feature information.

6. An electronic device, characterized in that, include: Memory, used to store computer programs; A processor, configured to implement the steps of the fuzzy clustering method for benign and malignant lung nodules based on multidimensional feature modeling as described in any one of claims 1 to 4 when executing the computer program.

7. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a computer program that, when executed by a processor, enables the implementation of the fuzzy clustering method for benign and malignant lung nodules based on multidimensional feature modeling as described in any one of claims 1 to 4.

Images