Osteosarcoma detection method and system based on multi-target collaborative recognition

Abstract

The invention discloses an osteosarcoma detection method and system based on multi-target collaborative recognition, and relates to the technical field of target recognition. The method comprises the following steps: firstly, acquiring multi-modal target point data of a target image area, performing feature coding processing to obtain a multi-target point feature vector, performing adaptability verification in a multi-target point combination process, and then performing abnormal coding vector identification on a gene target point vector in a corresponding target point combination after the adaptability verification is qualified, so as to obtain a multi-target point feature vector. And finally, carrying out gene target specific binding detection and non-linear monitoring on a detection signal amplification and response process to judge whether a detection result is output, so as to realize the purpose that in the multi-target identification process, the detection result is accurately detected, and the detection accuracy is improved. Through accurate cooperative processing of serum, image and gene cross-modal data, an effective data support is provided for accurate recognition of osteosarcoma mutation sites and an osteosarcoma detection process.

Description

Osteosarcoma Detection Method and System Based on Multi-Target Collaborative Recognition Technical Field

[0001] This invention relates to the field of target identification technology, and in particular to a method and system for detecting osteosarcoma based on multi-target collaborative identification. Background Technology

[0002] In the advancement of precision medicine in oncology, circulating tumor stem cells (CTSCs), as the core seed cells for tumor occurrence, development, metastasis, and recurrence, have become crucial for accurate detection in assessing the risk of tumor recurrence and metastasis, the efficacy of chemotherapy, and patient prognosis. CTSCs are closely related to tumor survival, proliferation, invasion, and metastasis. Detection can enable early warning of tumor metastasis and dynamic monitoring of treatment effects. However, CTSCs are extremely rare in blood and lack a single specific surface biomarker, posing a significant challenge to clinical detection. Currently, commonly used clinical methods for CTSC detection mainly include: biological function-based methods require the extraction of tumor tissue samples, making non-invasive liquid-based detection impossible; among biomarker-based methods, techniques such as immunofluorescence also rely on tissue samples; while flow cytometry can non-invasively detect CTSCs in blood, single-marker detection is prone to false positives, and multi-marker detection is prone to false negatives due to the scarcity of cells. Furthermore, these methods generally rely on expensive equipment and have cumbersome procedures, making it difficult to meet the clinical need for rapid and convenient detection.

[0003] To address the technical bottlenecks in clinical testing of CTSCs (tumor-specific cancer cells), building a precise and efficient testing system using cutting-edge technologies such as artificial intelligence and multimodal data fusion has become a research hotspot. These technologies can integrate multi-dimensional biological information, overcoming the limitations of single detection methods, and achieving precise target identification through feature extraction, association analysis, and model construction. For example, Chinese invention patent CN119889481B discloses an artificial intelligence-based method for detecting antitumor drug resistance. This method collects genomic data from tumor cells and analyzes gene changes; extracts tumor molecular markers and constructs an antitumor drug resistance knowledge graph; mines gene-protein characteristics and drug chemical properties, and constructs a drug resistance analysis model after fusion and dimensionality reduction; combines drug-target interaction relationships with gene-drug resistance association characteristics to evaluate the model's reliability, and finally outputs the antitumor drug resistance detection results for cancer patients.

[0004] In the field of osteosarcoma diagnosis and treatment, CTSCs are also closely related to tumor survival, proliferation, metastasis, and recurrence, making their detection value particularly prominent. However, the limitations of the aforementioned traditional detection methods are even more pronounced in this type of tumor. Electrochemical biosensors offer advantages such as high sensitivity, short detection time, and ease of operation, while DNA strand displacement reaction (SDR) leverages precise sequence orthogonality, mild experimental conditions, and programmability to achieve signal amplification and specific recognition through complementary base pairing. Combining these two technologies can transform the detection of osteosarcoma CTSCs into single-strand DNA detection, providing an innovative approach to solving the detection challenge. By screening and validating specific molecular markers for osteosarcoma CTSCs (CD133+ / CD49f- / CD146+), designing cascaded strand displacement reaction DNA probes, and assembling them onto corresponding antibodies to prepare multi-target probes, the detection specificity can be improved. Simultaneously, by optimizing signal amplification strategies using nanomaterials such as gold nanoparticles and graphene, an electrochemical-SDR sensor can be constructed, potentially achieving highly sensitive detection of rare osteosarcoma CTSCs in the blood, overcoming the shortcomings of traditional methods.

[0005] However, when faced with the complex environment of clinical blood samples, the assembly methods of multi-target probes lack optimization due to insufficient correlation between the specificity and signal amplification efficiency of multi-target combinations. This results in either probe spatial steric hindrance conflicts affecting recognition performance or failure to achieve synergistic amplification of multi-target signals, relying solely on the signal output of a single target and failing to compensate for signal attenuation caused by the scarcity of CTSCs. Furthermore, in existing multi-target detection schemes, the DNA strand displacement reaction and antibody conjugation process is not optimized for multi-target characteristics, easily leading to reduced antibody activity after probe modification and non-specific interactions between multi-target probes. This results in insufficient specific triggering efficiency of the cascade strand displacement reaction, further exacerbating the distortion of the detection signal.

[0006] Furthermore, dynamic monitoring of changes in CTSCs at different stages is necessary in clinical diagnosis and treatment. However, existing multi-target detection technologies lack dynamic adaptability in their biomarker combinations, failing to consider the impact of treatment interventions such as chemotherapy and surgery on the expression levels of biomarkers on the surface of CTSCs. Fixed multi-target combinations exhibit significantly reduced accuracy in scenarios where CTSC biomarker expression changes after treatment. Simultaneously, multi-target probes are prone to structural degradation or non-specific polymerization during storage at room temperature and in complex blood matrix environments, leading to fluctuations in multi-target recognition performance. The combined issues of unsystematic multi-target combination screening, unreasonable probe assembly and signal synergy design, and poor dynamic adaptability mean that existing osteosarcoma detection technologies cannot effectively meet the dynamic synergistic stability requirements of multi-target recognition, thus failing to provide accurate and reliable data support for evaluating chemotherapy efficacy and predicting postoperative recurrence. Summary of the Invention

[0007] To address the technical problem of low effectiveness of multi-target collaborative identification in osteosarcoma detection in existing technologies, this invention provides an osteosarcoma detection method and system based on multi-target collaborative identification. The technical solution is as follows: Firstly, a method for osteosarcoma detection based on multi-target collaborative identification is provided, comprising: Step 1, acquiring multimodal target data of the target image region and performing feature encoding processing to obtain multi-target feature vectors, while simultaneously verifying the compatibility of the multi-target combination process; Step 2, identifying abnormal encoded vectors of gene target vectors in the corresponding target combinations after successful compatibility verification to determine the presence of mutation site encoding, and quantifying the data matching degree based on the mutation site encoding determination result; Step 3, based on the quantified data matching degree result, performing gene target-specific binding detection, and simultaneously combining a multimodal feature extraction network to perform nonlinear monitoring of the detection signal amplification and response process, to determine whether to output the osteosarcoma detection result.

[0008] On the other hand, a multi-target collaborative identification-based osteosarcoma detection system is provided. This system includes: a feature vector acquisition and fit verification module, used to collect multimodal target data of the target image region and perform feature encoding processing to obtain multi-target feature vectors, while simultaneously verifying the fit of the multi-target combination process; a mutation site identification and data matching degree quantification module, used to identify abnormal encoded vectors of gene target vectors in the corresponding target combination after the fit verification is qualified, in order to determine whether there is a mutation site encoding, and simultaneously quantify the data matching degree based on the determination result of the mutation site encoding; and a nonlinear monitoring and detection result output determination module, used to perform gene target specific binding detection based on the quantification result of the data matching degree, and simultaneously combine a multimodal feature extraction network to perform nonlinear monitoring of the detection signal amplification and response process, in order to determine whether to output the osteosarcoma detection result.

[0009] The beneficial effects of the technical solution provided by the embodiments of the present invention include at least the following: 1. The present invention collects multimodal target data of the target image region and completes feature encoding, obtains multi-target feature vectors, and performs combination compatibility verification simultaneously. It systematically screens target combinations that match signal amplification efficiency, avoiding probe spatial steric hindrance conflicts from the source. For gene target vectors in qualified target combinations, it accurately identifies abnormal coding vectors to determine mutation sites, and simultaneously quantifies the data matching degree during sequence complementation, reducing data acquisition deviations caused by non-specific probe interactions. It combines sequence quantification results to perform gene target-specific binding detection, and uses a multimodal feature extraction network to achieve signal amplification and nonlinear monitoring, dynamically adapting to changes in the expression of CTSCs markers after chemotherapy, surgery, and other treatment interventions. This avoids the decrease in recognition accuracy caused by fixed combinations and enhances the stability of probes in complex blood environments and under normal temperature storage. This comprehensively improves the dynamic synergistic stability of multi-target recognition, providing accurate and reliable data support for clinical chemotherapy efficacy evaluation and postoperative recurrence warning.

[0010] 2. First, multi-target vectors are aligned in dimension and standardized to form a unified format of feature vectors to be verified. Then, a verification coefficient is calculated using a feature vector information entropy and norm fusion equation. The passability is determined by comparing the coefficients with reference values ​​for image, gene, and serum target vectors. If all values ​​meet the criteria, the final target feature vector is output. If only one type fails to meet the criteria, targeted optimization is performed. Otherwise, an alarm is triggered to check the data acquisition quality and parameter calibration. This approach ensures the reliability of target feature vectors from the source by quantifying the uniformity of feature distribution and the sufficiency of differentiated information. For multi-target combination compatibility verification, the final target feature vector is input into the mutual information calculation equation to calculate the mutual information values ​​for image-gene, image-serum, and gene-serum. Feature selection algorithms are used to optimize or the target feature vectors are manually reconstructed. If multiple combinations fail, related features are prioritized for selection and targeted adjustments. If the optimized combination meets the criteria, the target vector is deemed qualified and the gene target vector is obtained; otherwise, an adjustment to the target combination is prompted. The entire process, through standardized processing, classification verification, and hierarchical optimization, not only solves the problems of inconsistent data formats and uneven feature quality among multi-target points, but also ensures the collaborative adaptability of target point combinations through mutual information quantification, thereby improving the stability and accuracy of multi-target point identification.

[0011] 3. By first retrieving the gene coding vector benchmark library of the specified specific biomarker, the gene target vector in the target combination is compared with the corresponding coding vector in the benchmark library dimension by dimension, and the difference is calculated. If the difference is not less than the corresponding preset value, it is initially determined that the coding vector fragment contains a mutation site code, and then cross-validation of serum and imaging target vectors is initiated to confirm the result. If the difference is less than the corresponding preset value, it is initially determined that there is no mutation site code, and the preset personnel are prompted to further verify. This hierarchical judgment mechanism not only ensures the identification efficiency, but also avoids the risk of missed judgment through manual verification. In the cross-validation stage, the initially determined mutation site code is input into the pixel feature matrix of the lesion region to which the imaging target vector belongs, and two types of parameters, base matching ratio and overlapping area ratio, are extracted. The coding matching quantification index is obtained by arithmetic averaging. If the matching degree meets the standard, the final mutation site code is confirmed. If the standard is not met, it is determined that there is no clear mutation site code. Dual verification effectively improves the accuracy of mutation site code determination and avoids the bias of single-dimensional judgment. Subsequent statistical confirmation of the results and pre-detection of gene targets based on multimodal feature association not only narrowed the scope of serum target screening and improved detection efficiency, but also significantly reduced signal fluctuation interference caused by nonspecific hybridization by quantifying the degree of data matching between gene targets and the final mutation site encoding during sequence complementation.

[0012] 4. By acquiring complementary continuity values ​​reflecting sequence integrity and continuity, adaptive feedback for serum target pre-detection is achieved. This quantification method evaluates the degree of complementarity from two dimensions: core matching and process stability, effectively improving the accuracy of serum target screening and reducing interference from non-specific binding during data acquisition. In the signal amplification and nonlinear monitoring stage, adaptive feedback results are first acquired. The standardized vector of serum targets, the pixel feature matrix corresponding to image targets, and the gene sequence features encoded by the final mutation site are simultaneously imported into a multimodal feature extraction network for collaborative feature learning. After modal nonlinear mapping, a standardized high-dimensional vector is obtained. Then, the duration of cross-modal synergistic interaction is obtained through a multimodal fusion network. The signal is corrected by a nonlinear fitting algorithm, and a nonlinear monitoring index is calculated. When the nonlinear monitoring index is not lower than the corresponding preset value and the interaction response duration is within the corresponding allowable range, the osteosarcoma detection result can be output; otherwise, feedback is provided. The entire process provides dual assurance for the accuracy of data acquisition in the osteosarcoma detection process through multi-dimensional quantification of complementarity, cross-modal feature fusion, and dual signal verification. Attached Figure Description

[0013] To more clearly illustrate the technical solutions in the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0014] Figure 1 is a flowchart of the osteosarcoma detection method based on multi-target collaborative recognition provided in an embodiment of the present invention; Figure 2 is a flowchart of the adaptability verification of the multi-target combination process provided in an embodiment of the present invention; Figure 3 is a flowchart of the abnormal coding vector identification and cross-validation provided in an embodiment of the present invention; Figure 4 is a flowchart of the nonlinear monitoring of the detection signal amplification and response process provided in an embodiment of the present invention; Figure 5 is a schematic diagram of the structure of the osteosarcoma detection system based on multi-target collaborative recognition provided in an embodiment of the present invention. Detailed Implementation

[0015] The technical solution of the present invention will now be described with reference to the accompanying drawings.

[0016] In embodiments of the present invention, words such as "exemplarily," "for example," etc., are used to indicate that something is an example, illustration, or description. Any embodiment or design described as "exemplary" in the present invention should not be construed as being more preferred or advantageous than other embodiments or designs. Specifically, the use of the word "exemplary" is intended to present the concept in a concrete manner. Furthermore, in embodiments of the present invention, the meaning expressed by "and / or" can be both, or either one.

[0017] To make the technical problems, technical solutions and advantages of the present invention clearer, a detailed description will be given below in conjunction with the accompanying drawings and specific embodiments.

[0018] This invention provides a method for detecting osteosarcoma based on multi-target collaborative recognition. As shown in Figure 1, the flowchart of the method includes the following steps: Step 1: Collect multimodal target data of the target image region and perform feature encoding processing to obtain multi-target feature vectors. Simultaneously, verify the adaptability of the multi-target combination process, achieving standardized encoding of multimodal target data and preliminary screening of suitable target combinations. The target image region refers to the tissue region to be detected, located by medical imaging (such as CT, MRI, PET-CT), where osteosarcoma lesions are suspected or potentially present. The multimodal target data includes... Image target data, gene target data, and serum target data are included. Image target data typically includes morphological parameters (size, boundary sharpness, density / signal uniformity), texture features (gray-level co-occurrence matrix, entropy value), and blood perfusion parameters (enhancement peak, time to peak). Gene target data typically includes promoter / enhancer regulatory region sequences, coding region sequences, and gene expression levels. Serum target data typically includes serum microRNA (such as miR-21, miR-143) expression levels and osteosarcoma DNA (ctDNA) related sequence information. Target feature vectors include image target feature vectors, gene target feature vectors, and serum target feature vectors.

[0019] Step two involves identifying anomalous coding vectors in the gene target vectors of the corresponding target combinations after the fit verification is qualified, in order to determine whether there are mutation site codes. At the same time, the degree of data matching is quantified based on the determination results of mutation site codes, realizing the accurate identification of mutation site codes and the quantitative characterization of data matching degree. Mutation site codes refer to the coding information corresponding to specific sequence regions with mutations such as base substitution, insertion, and deletion in gene fragments (such as ctDNA and target gene transcripts) related to the occurrence and development of osteosarcoma in serum target data.

[0020] Step 3: Based on the quantitative results of data matching, gene target-specific binding detection is performed. Simultaneously, a multimodal feature extraction network is used to monitor the nonlinearity of the detection signal amplification and response process to determine whether to output osteosarcoma detection results. This achieves the verification of specific binding of osteosarcoma targets and the accurate output of detection results. Nonlinear monitoring is used to accurately capture unexpected fluctuations in the signal amplification and response process (such as interference signals generated by nonspecific binding) during serum target-specific binding detection, eliminate false positive or false negative interference, and ensure the authenticity and stability of the detection signal.

[0021] It is important to further understand that the target imaging regions, imaging target data, gene target data, and serum target data involved in this invention, such as morphological parameters, coding region sequences, and serum microRNA expression levels, all originate from the output results of existing third-party clinical testing platforms (such as reports issued by gene sequencing institutions, detection data from serum testing equipment, and scanning results from medical imaging equipment). This invention does not directly acquire, store, or process any genetic materials such as DNA, RNA, or cells, nor does it involve the collection, preservation, or utilization of genetic resources. It does not conduct any biological experimental operations such as nucleic acid extraction, sequencing, hybridization, amplification, or verification; it only performs subsequent processing on the digital detection data already generated by third-party testing platforms. Its technical essence lies in achieving cross-modal data fusion and feature mining through standardized coding, fit verification, mutation site association quantification, and nonlinear signal monitoring of multimodal medical data. It is a purely computer data processing technology, and its core lies in improving the accuracy of detection results through integrated analysis at the data level.

[0022] In this embodiment, multimodal target data integration and standardized coding are employed to transform heterogeneous data into feature vectors of a unified dimension. This avoids the information limitations of a single data dimension, enables collaborative representation of multi-source data, and improves the comprehensiveness of target information mining. Pre-processing for multi-target combination compatibility verification selectively filters effective target combinations, eliminates redundant and invalid data, optimizes data processing efficiency, and avoids unnecessary computational waste. Quantitative processing of data matching transforms base pairing relationships into quantifiable analytical indicators, making the analysis of mutation-related data more objective and accurate, and reducing human judgment errors. Combined with nonlinear monitoring, real-time identification of signal response data effectively filters out interference data such as nonspecific bindings, ensuring the authenticity and reliability of core detection data. This provides high-quality data support for subsequent result determination and comprehensively improves the accuracy of osteosarcoma detection data processing.

[0023] Furthermore, the specific methods for obtaining multi-target feature vectors are as follows: The multi-class discriminant function is a decision function used to distinguish between three types of target data: images, genes, and serum. The core is to set classification rules based on data type features (such as the pixel matrix structure of images, sequencing text information of genes, and numerical detection results of serum). The collected multimodal target data is classified and processed using a multi-class discriminant function to ensure that different types of data are transformed into standardized feature vectors that are uniformly adapted. Specifically, for image target data, the target image region located by medical imaging is used as the basis to extract the length and width pixel parameters of the corresponding image. Combined with the gray value distribution, edge texture features and spatial topological relationship of the lesion region, the region contour is optimized and noise interference is removed through morphological opening and closing operations. Then, the lesion region is resampled and the coordinates are calibrated to extract the gray intensity value, gradient magnitude and direction information of each pixel. The pixels are arranged in a regular order according to their spatial position to generate a two-dimensional pixel matrix containing pixel coordinates, gray intensity and gradient information. Subsequently, the two-dimensional pixel matrix is ​​subjected to feature extraction and dimension transformation through convolution operation. The matrix is ​​traversed by sliding convolution kernel to realize local feature aggregation and global information integration, mapping the high-dimensional pixel data into a standardized vector of a specified dimension, and finally generating a structured image target vector.

[0024] For gene target data, binary encoding is used to uniquely identify different gene states using binary vectors, transforming non-numerical gene information into computer-recognizable discrete numerical vectors, thus forming standardized gene target vectors. For serum target data, detection results such as serum microRNA expression levels and ctDNA sequence-related indicators are collected. Considering the dimensional differences and numerical dispersion of different serum indicators, the Z-score standardization method is used to eliminate the influence of dimensions. By calculating the deviation of each indicator from the mean and converting it into a standard score, it is uniformly mapped to the same numerical range, generating comparable standardized serum target vectors. The generated multi-target vectors are summarized, and feature vector transformation and qualification verification are performed to determine whether to output the target feature vectors corresponding to the multimodal target data. Multi-target vectors include imaging target vectors, gene target vectors, and serum target vectors.

[0025] The process of feature vector transformation and qualification verification includes: aligning and standardizing the generated multi-target vectors according to a preset standard format (i.e., vector structure with unified dimensions, consistent data types, and standardized numerical ranges) to form a unified format feature vector to be verified. This feature vector represents the target feature vector to be verified. The verification coefficient of each feature vector is calculated using the information entropy and norm fusion equation. Specifically, the norm fusion equation is a joint operation equation that directly integrates the calculation results of information entropy and L2 norm. Its core function is to integrate the representational advantages of the two types of indicators. Information entropy quantifies the dispersion and effective redundancy ratio of biological target association information in the feature vector to be verified, intuitively reflecting the uniformity of the distribution of features in different dimensions within the vector. A higher value indicates more comprehensive target differentiation information contained in the feature vector, focusing on measuring the degree of disorder within the vector. The norm (preferably L2 norm) focuses on characterizing the spatial amplitude of each feature vector to be verified. By calculating the square root of the sum of squares of each element in the vector, the overall strength of the feature vector is intuitively reflected. During the fusion process, the original calculation results of information entropy and norm are first normalized to the [0,1] interval to eliminate the difference in their dimensions. Then, information fusion is achieved through direct summation. Finally, the verification coefficients of each feature vector to be verified are output, providing an objective quantitative basis for the subsequent adaptation verification of multi-target combinations.

[0026] Reference values ​​for the verification coefficients of each feature vector to be verified were retrieved: 0.35 for image target vectors, 0.42 for gene target vectors, and 0.38 for serum target vectors. These values ​​were set through statistical analysis and validation experiments using multimodal target data from historical clinical osteosarcoma samples. If the verification coefficients of all target feature vectors to be verified are not less than their corresponding reference values, the feature vector transformation is considered successful, and the corresponding feature vector to be verified is used as the final target feature vector. If only the verification coefficient of an image target feature vector to be verified is less than its corresponding reference value, the convolution parameters are adjusted based on the ratio of the difference between the verification coefficient and the corresponding reference value. The convolution kernel size and pooling stride are adjusted using the gradient descent algorithm, and the feature encoding process and corresponding verification coefficients are recalculated.

[0027] If only the check coefficient of the gene target feature vector to be checked is less than the corresponding reference value, then the gene sequence data of the specified specific markers (such as the coding region sequence and promoter region sequence of TP53 and RB1 genes) are re-screened and the dimension corresponding to the binary code is updated, and the corresponding check coefficient is recalculated. If only the check coefficient of the serum target feature vector to be checked is less than the corresponding reference value, then the update feedback is based on the ratio of the difference between the check coefficient and the corresponding reference value (i.e., the ratio of the difference between the corresponding reference value and the check coefficient corresponding to the serum target feature vector to the corresponding reference value), and the Z-score standard is re-executed through an iterative update algorithm. The data is processed and the corresponding verification coefficients are recalculated. Except for the above situations, all other cases are judged as multi-target vector collaborative transformation errors. The essence is that the collaborative imbalance of multimodal data in feature transformation occurs: the encoding logic of image, gene, and serum data has not formed a consistent association. This may be due to the failure to effectively eliminate cross-modal data heterogeneity, mismatch of feature dimensions, or the accumulation of deviations in the data acquisition / processing links, which leads to the inability of various target features to achieve accurate complementarity and collaborative representation, violating the collaborative requirements of multi-target vector transformation. Therefore, it is judged as a collaborative transformation error. At this time, an abnormal alarm feedback should be issued to prompt the multimodal target data acquisition quality check, such as checking whether the image scanning equipment is compliant.

[0028] In this embodiment, standardized integration of multimodal target data is achieved through unified format standardization and information entropy-norm fusion verification, eliminating heterogeneous data differences and improving the consistency and comparability of feature vectors. Reference values ​​are set based on clinical sample statistics to ensure the scientific validity and clinical applicability of the verification standards. Targeted non-compliance handling mechanisms—namely, adjusting convolution parameters for image vectors, re-screening sequences and updating coding dimensions for gene vectors, and iteratively optimizing and standardizing serum vectors—can accurately correct single-modal data transformation deviations, effectively improving feature vector quality. Simultaneously, a collaborative transformation error alarm feedback mechanism can promptly identify data acquisition and function parameter issues, avoiding invalid data transfer. The overall process, through a closed-loop design of standardized processing, quantitative verification, and precise correction, significantly reduces redundant interference and transformation errors in feature vectors, ensuring the integrity, validity, and reliability of target feature information, laying a data foundation for subsequent multi-target combination suitability verification and precise osteosarcoma detection.

[0029] Figure 2 shows the flowchart of the adaptability verification process for multi-target combination. The specific process of adaptability verification is as follows: The final target feature vectors, namely the influence target feature vector X, the gene target feature vector Y, and the serum target feature vector Z, are used as inputs to the mutual information calculation equation. The correlation quantification of the three sets of target combinations (i.e., image-gene target group, image-serum target group, and gene-serum target group) is performed sequentially. The calculation formula is as follows: ,in, This represents the joint probability distribution of the target feature vectors A and B. , Let A and B represent the marginal probability distributions of target feature vectors A and B, respectively. The calculation first estimates the joint and marginal probability distributions of each vector group using kernel density estimation, then substitutes these values ​​into the formula for calculation: For the image-gene combination, input X and Y to obtain I(X,Y); for the image-serum combination, input X and Z to obtain I(X,Z); for the gene-serum combination, input Y and Z to obtain I(Y,Z). The output consists of three types of mutual information values: image-gene, image-serum, and gene-serum mutual information values.

[0030] If all three mutual information values ​​are not less than the preset fit verification value (usually set to 0.7), the multi-target group is deemed fit, and the serum target vector in the corresponding target combination after fit verification is obtained. If any group's mutual information value is less than the preset fit verification value, the multi-target group is deemed unfit. If two or more groups' mutual information values ​​are less than the preset fit verification value, the target feature vectors in the unfit target groups are first filtered using a feature filtering algorithm to retain the correlation feature dimension with the target category up to a preset percentage (e.g., the top 40%), which is then used as a new target feature vector. Targeted optimization is then performed on the selected target feature vectors for each group. If only one group's mutual information value is less than the preset fit verification value, the mutual information value of the corresponding group is used as the basis for the optimization. The deviation is directly optimized in a targeted manner. Specifically, if the mutual information value deviation of any group is not greater than the preset allowable mutual information value deviation (usually set to 0.08), then based on the feature selection algorithm, the mutual information value between the target feature vector of the corresponding group and the target category is calculated (using the aforementioned formula for calculating mutual information value, replacing the target feature vector of another type with the target category label). After sorting from largest to smallest, the target feature vector with the largest mutual information value is selected as the new target feature subset, which replaces the target feature vector in the original group, thereby achieving accurate selection and updating of target features for each group. The mutual information value deviation represents the difference between the preset fit verification value and the obtained mutual information value of the corresponding target group. If the mutual information value deviation of any group is greater than the preset allowable mutual information value deviation, then the preset personnel are prompted to reconstruct the target feature vector of the corresponding group based on the mutual information value deviation.

[0031] The aforementioned fit verification value of 0.7, preset percentage of 40%, and allowable mutual information value deviation of 0.08 are all based on statistical analysis and repeated verification results of historical multimodal target data in the historical multi-target combination process. These values ​​were determined after quantitative analysis of the correlation between mutual information value and detection accuracy, and after multi-center sample verification and parameter optimization. After targeted optimization, the three types of mutual information values ​​are re-acquired. If all three mutual information values ​​are not less than the preset fit verification value, the multi-target combination is deemed fit and the serum target vector in the corresponding target combination after fit verification is obtained. Otherwise, a target combination adjustment prompt is sent. The preset personnel need to adjust the combination type based on the mutual information value not meeting the standard: if the mutual information value of a combination is insufficient, the feature screening algorithm is re-executed to expand the selection range of the target feature subset for that combination (e.g., taking the top 2-3 high mutual information value features), or new data for that category of targets is added and re-encoded to generate feature vectors.

[0032] In this embodiment, the formula for calculating the mutual information value is based on information theory to accurately measure the statistical dependence and information sharing degree of two types of target vectors, avoiding the limitations of linear correlation analysis. It can capture the complex nonlinear correlation between multimodal data and fit the complex action mechanism of multi-target data. The kernel density estimation method can efficiently estimate the probability distribution without pre-setting the data distribution type, and is suitable for the unknown distribution characteristics of target feature vectors. By quantifying the mutual information values ​​of the three sets of combinations, the synergistic correlation strength of image, gene, and serum data can be objectively reflected. The higher the value, the stronger the complementarity of the combined information, providing a scientific quantitative basis for subsequent fit verification, effectively screening out target combinations with less information redundancy and better synergy, and improving the data utilization and result accuracy of subsequent detection.

[0033] Subsequently, by setting tiered verification standards and differentiated optimization strategies, multi-target combinations with excellent information synergy are precisely selected, effectively eliminating redundant features and low-correlation data, thereby improving the targeting and information utilization rate of target combinations. Gradient optimization mechanisms (such as updating target feature subsets and reconstructing feature vectors) can precisely correct based on the degree of mutual information value deviation, avoiding information loss due to over-optimization and quickly compensating for insufficient data correlation. The overall process, through a closed-loop design of quantitative evaluation and precise optimization, strengthens the collaborative characterization capability of multimodal data, providing high-quality data support for subsequent mutation site identification and serum target sequence analysis.

[0034] Figure 3 shows the flowchart of abnormal coding vector identification and cross-validation. The specific process of abnormal coding vector identification is as follows: A pre-set serum coding vector benchmark library for the specified specific biomarker (i.e., osteosarcoma-specific biomarker) is retrieved. The serum target vector in the target combination is aligned sequentially with the coding vector of the corresponding biomarker in the benchmark library along the same dimensions. The Euclidean distance method is used to calculate the coding vector difference: first, the difference between the corresponding dimension values ​​of the two vectors is calculated and squared; then, the squared differences of all dimensions are summed; finally, the square root is taken to obtain the coding vector difference. This value directly reflects the degree of deviation between the serum target vector and the standard vector. The smaller the difference, the closer the serum index is to the characteristic pattern of the corresponding biomarker, providing a quantitative basis for subsequent mutation site identification. The pre-set serum coding vector benchmark library is a database established based on serum sample data of historically diagnosed osteosarcoma patients after standardized coding and validation. It contains the standard coding vectors corresponding to each osteosarcoma-specific biomarker.

[0035] If the coding vector difference is not less than the preset coding vector difference, the coding vector fragment of the corresponding marker is initially determined to have a mutation site coding, and cross-validation of serum target vector and imaging target vector is performed to finally determine the mutation site coding. The preset coding vector difference is represented by the sum and average of the historical coding vector differences in the historical lesion area mutation site identification process. If the coding vector difference is less than the preset coding vector difference, the coding vector fragment of the corresponding marker is initially determined to not have a mutation site coding, and the preset personnel are prompted to further verify, for example, review the serum sample testing process.

[0036] The specific steps of cross-validation are as follows: The mutation site codes obtained from the preliminary determination are input into the pixel feature matrix of the lesion region to which the image target vector belongs. Mutation site codes are compared to obtain the base matching ratio used to evaluate the consistency between the preliminary mutation site codes and the image mutation site codes, i.e., the ratio of the number of matching bases between the preliminary mutation site codes and the image mutation site codes to the total number of bases in the preliminary mutation site codes; and the overlapping region ratio used to evaluate the homologous segments between the preliminary mutation site codes and the image mutation site codes, i.e., the ratio of the overlap length of the homologous segments to the total length of the image mutation site codes. The pixel feature matrix contains the image mutation site codes of the lesion region. The pixel feature matrix is ​​based on medical image localization and combined with multi-dimensional image data such as the spatial location, morphological contour, gray value distribution, density gradient, edge texture and metabolic activity of the lesion. After feature coding mapping and dimension normalization, a structured matrix containing image-level mutation site association coding information and spatial structural features is generated.

[0037] The obtained base matching ratio and overlapping region ratio are arithmetically averaged to avoid single-dimensional evaluation bias, resulting in a coding matching degree. This matching degree can comprehensively and objectively measure the reliability of mutation site coding. If the coding matching degree is not less than the preset coding matching degree, the mutation site coding obtained in the preliminary judgment is confirmed as the final mutation site coding; otherwise, it is determined that there is no clear mutation site coding. The preset coding matching degree is represented by the sum and average of the historical coding matching degrees in the historical lesion region mutation site verification process. The results of cross-validation are statistically analyzed, and serum target pre-detection based on multimodal feature association is performed to narrow the range of serum target screening. In the pre-detection process, the data matching degree between serum targets and final mutation site coding is quantified to reduce signal fluctuation interference caused by non-specific hybridization.

[0038] The quantification of data matching degree involves: monitoring the data matching process between serum target sites and the final mutation site encoding during sequence complementation; calculating the ratio of the number of successfully matched nucleotide sites to the total number of serum target sites, denoted as the site matching percentage, to quantify the fit between serum target sites and mutation site encoding, intuitively reflecting the accuracy of effective data matching; calculating the ratio of the number of failed-matching nucleotide site fragments to the total number of serum target sequence sites, denoted as the mismatch fragment percentage, to quantify the concentration of matching interruptions, highlighting continuity defects and data biases in the sequence complementation process; and normalizing the difference between the obtained site matching percentage and mismatch fragment percentage, i.e., the difference between the site matching percentage and the mismatch fragment percentage, to obtain a complementation continuity value, which comprehensively reflects the matching quality and continuity. The closer the value is to 1, the better the matching, providing an intuitive and accurate quantitative basis for fitness judgment and effectively supporting the fitness feedback of serum target pre-detection.

[0039] In this embodiment, during the abnormal coding vector identification stage, a historically constructed serum coding vector benchmark library provides a scientific reference for calculating the degree of difference. Dimensional comparison using Euclidean distance accurately quantifies the deviation between serum target vectors and standard vectors, effectively screening out potential abnormal codes and laying a solid data foundation for mutation site identification, thus avoiding missed detection of key variant information. The criterion for determining the degree of difference in coding vectors is derived from the historical data mean, ensuring the objectivity and clinical suitability of the initial judgment. Furthermore, the serum sample testing process review prompts further reduce the risk of data errors.

[0040] In the cross-validation phase, serum target data from multiple clinical samples were iteratively grouped and validated against corresponding mutation site coding sequences. The complementary continuity values ​​were repeatedly calculated using the aforementioned quantification methods to further verify the stability and universality of the data matching quantification logic. The overall approach improved the quantification accuracy of serum target and mutation site coding matching, reduced the risk of misjudgment of suitability due to data bias, provided scientific and stable quantitative support for suitability screening in serum target pre-detection, and laid a high-quality data foundation for subsequent multimodal feature fusion.

[0041] As shown in Figure 4, the nonlinear monitoring flowchart illustrates the specific process as follows: The adaptation feedback results of serum target pre-detection are obtained, indicating whether the serum target pre-detection is successful (complementary continuity values ​​meet the expected serum target pre-detection requirements) or unsuccessful (complementary continuity values ​​do not meet the expected serum target pre-detection requirements). The serum target vector from the adaptation feedback results is then input into a multimodal feature extraction network. Simultaneously, the pixel feature matrix of the image target vector and the gene sequence features encoded by the final mutation site are imported. The multimodal feature extraction network first performs a modal-wise nonlinear mapping on the three types of data: the serum target processing branch transforms the vector into a quantized feature vector through a fully connected layer, preserving the association information between serum indicators and mutation sites; the image branch uses a convolutional neural network to extract spatial structural features from the pixel feature matrix, capturing the morphological association of the lesion region; and the gene branch analyzes gene sequence features through a recurrent neural network. During the mapping process, a nonlinear transformation is introduced through an activation function to eliminate the interference of data heterogeneity between modalities. At the same time, the three types of features are uniformly mapped to a high-dimensional vector space of the same dimension through standardization processing. Finally, the standardized high-dimensional vectors corresponding to serum target quantitative features, image spatial structure features and gene sequence functional features are output.

[0042] Based on the interaction response process of three types of standardized high-dimensional vectors in the multimodal fusion network, the synergistic effect duration for quantifying the complementary enhancement effect of cross-modal features is obtained. This duration is the time interval from the vector input to the fusion layer until the cross-modal features achieve stable complementary enhancement and output synergistic feature vectors. The ratio of the obtained synergistic effect duration to the preset synergistic effect duration is combined with a nonlinear fitting algorithm (such as Gaussian fitting) for signal correction to eliminate random interference and data fluctuations in modal interactions, resulting in a nonlinear monitoring index that reflects the stability of the detection signal and the correlation of features. If the obtained nonlinear monitoring index is lower than the preset nonlinear monitoring index, it is determined that there is fluctuation interference in the detection signal. In this case, the osteosarcoma detection result is not output, and feedback is sent after re-monitoring with multimodal feature mapping parameters. The preset nonlinear monitoring index is represented by the summation and averaging of historical nonlinear monitoring indices in the historical feature vector fusion process.

[0043] If the acquired nonlinear monitoring index is not lower than the preset nonlinear monitoring index, it is determined that the detected signal has no abnormal interference. At the same time, the interaction response duration of the three types of standardized high-dimensional vectors is monitored. The interaction response duration represents the activation duration of the signals between the corresponding three types of standardized high-dimensional vectors during the signal correction process. If the interaction response duration is within the preset allowable range, the osteosarcoma detection result is output. Otherwise, the osteosarcoma detection result is not output, and feedback on re-monitoring after multimodal feature mapping parameters is sent to prompt the preset personnel to optimize the activation function parameters of each modality feature extraction branch. For example, the ReLU6 function is used to limit the output range of the serum branch, and the GELU function is used to improve the gradient flow efficiency of the imaging branch. The preset allowable range of interaction response duration is set based on the statistical distribution law of historical cross-modal feature interaction data and is verified by multiple sets of samples.

[0044] In this embodiment, a multimodal feature extraction network is used to design dedicated processing branches for three types of heterogeneous data: serum, imaging, and genetic data. Combined with nonlinear transformation and standardization, this effectively eliminates data heterogeneity between modalities while fully preserving core features, laying a solid foundation for high-quality cross-modal fusion. By quantifying the duration of synergistic effects and using a Gaussian fitting algorithm for signal correction, random interference and data fluctuations in modal interactions are precisely filtered out. Based on historical data and established judgment criteria, the nonlinear monitoring index objectively reflects the stability and feature correlation of the detection signal, promptly identifying fluctuation interference and triggering parameter adjustments. The overall process employs a closed-loop design of precise feature extraction and dual-standard monitoring to achieve efficient signal amplification and precise interference removal, synergistically ensuring high accuracy and stability in osteosarcoma detection.

[0045] This invention provides an osteosarcoma detection system based on multi-target collaborative recognition. As shown in Figure 5, the system includes: a feature vector acquisition and fit verification module, used to collect multimodal target data from the target image region and perform feature encoding processing to obtain multi-target feature vectors, while simultaneously verifying the fit of the multi-target combination process; a mutation site identification and data matching degree quantification module, used to identify abnormal encoded vectors in the serum target vectors of the corresponding target combinations after the fit verification is qualified, to determine whether mutation site encoding exists, and simultaneously quantifying the data matching degree based on the mutation site encoding determination result; and a nonlinear monitoring and detection result output determination module, used to perform serum target specific binding detection based on the quantification result of the data matching degree, and simultaneously combine a multimodal feature extraction network to perform nonlinear monitoring of the detection signal amplification and response process, to determine whether to output the osteosarcoma detection result.

[0046] In this embodiment, the feature vector acquisition and fit verification module ensures the integrity and synergy of target feature vectors through multimodal target data acquisition, encoding, and combination fit verification. The mutation site identification and data matching degree quantification module accurately identifies abnormal coding and mutation sites of serum targets, quantifies the data acquisition quality during sequence complementation, effectively narrows the detection range, reduces non-specific interference, and improves detection specificity. The nonlinear monitoring and detection result output judgment module combines multimodal feature extraction network to amplify signals, eliminates interference through nonlinear monitoring, ensures signal stability, and guarantees reliable detection results. The overall modules have clear division of labor and closed-loop linkage, realizing precise control of the entire process from data acquisition and mutation identification to result output, improving the effectiveness of osteosarcoma detection.

[0047] The above embodiments can be implemented, in whole or in part, by software, hardware (such as circuits), firmware, or any other combination thereof. When implemented using software, the above embodiments can be implemented, in whole or in part, as a computer program product. A computer program product includes one or more computer instructions or computer programs. When the computer instructions or computer programs are loaded or executed on a computer, all or part of the flow or function according to the embodiments of the present invention is generated. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device. Computer instructions can be stored in a computer-readable storage medium or transmitted from one computer-readable storage medium to another. For example, computer instructions can be transmitted from one website, computer, server, or data center to another website, computer, server, or data center via wired (e.g., infrared, wireless, microwave, etc.) means. A computer-readable storage medium can be any available medium that a computer can access or a data storage device such as a server or data center that includes one or more sets of available media. Available media can be magnetic media (e.g., floppy disks, hard disks, magnetic tapes), optical media (e.g., DVDs), or semiconductor media. Semiconductor media can be solid-state drives.

[0048] It should be understood that the term "and / or" in this article is merely a description of the relationship between related objects, indicating that three relationships can exist. For example, A and / or B can represent: A existing alone, A and B existing simultaneously, or B existing alone. A and B can be singular or plural. Additionally, the character " / " in this article generally indicates an "or" relationship between the preceding and following related objects, but it can also represent an "and / or" relationship. Please refer to the context for a more accurate understanding.

[0049] In various embodiments of the present invention, the order of the above-mentioned process numbers does not imply the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of the present invention.

[0050] Those skilled in the art will 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, or a combination of computer software and electronic hardware. 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.

[0051] The above are merely specific embodiments of the present invention, but the scope of protection of the present invention is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the technical scope disclosed in the present invention should be included within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be determined by the scope of the claims.

Claims

1. A method for detecting osteosarcoma based on multi-target collaborative recognition, characterized in that, The method includes: Step 1, collecting multimodal target data of the target image region and performing feature encoding processing to obtain multi-target feature vectors, while simultaneously verifying the compatibility of the multi-target combination process; Step 2, identifying abnormal encoded vectors for gene target vectors in the corresponding target combinations after the compatibility verification is qualified, in order to determine whether there are mutation site codes, and simultaneously quantifying the degree of data matching based on the determination result of mutation site codes; Step 3, performing gene target specific binding detection based on the quantification result of the degree of data matching, and simultaneously combining a multimodal feature extraction network to perform nonlinear monitoring of the detection signal amplification and response process, in order to determine whether to output the osteosarcoma detection result.

2. The osteosarcoma detection method based on multi-target collaborative recognition as described in claim 1, characterized in that, The specific method for obtaining the multi-target feature vectors is as follows: The acquired multimodal target data is classified using a multi-classification discriminant function to generate target feature vectors of corresponding types. The multimodal target data includes image target data, gene target data, and serum target data. The target feature vectors include image target feature vectors, gene target feature vectors, and serum target feature vectors. The data classification process specifically involves: if it is image target data, then based on the two-dimensional pixel matrix determined by the width and height pixel values ​​of the corresponding image in the target image region, the visual features in the two-dimensional pixel matrix are transformed into a specified dimension through convolution operations. The system generates image target vectors by standardizing the vectors of the target data. If the data is gene target data, the gene state in the gene sequencing results of the specified specific marker is converted into a discrete numerical vector based on binary encoding to generate gene target vectors. If the data is serum target data, the detection results of serum indicators are converted into standardized numerical vectors based on the Z-score standardization method to generate serum target vectors. The generated multi-target vectors are summarized, and feature vector conversion and qualification verification are performed to determine whether to output the target feature vectors of the corresponding type of multimodal target data. The multi-target vectors include image target vectors, gene target vectors, and serum target vectors.

3. The osteosarcoma detection method based on multi-target collaborative recognition as described in claim 2, characterized in that, The feature vector transformation and qualification verification specifically includes: aligning and standardizing the generated multi-target vectors according to a preset standard format to form a unified format feature vector to be verified; obtaining the verification coefficients of the feature vectors to be verified through the information entropy and norm fusion equation of the feature vectors, and simultaneously retrieving the reference values ​​of the verification coefficients corresponding to each feature vector to be verified; if the verification coefficients corresponding to the target feature vectors to be verified are all not less than the corresponding reference values, it indicates that the feature vector transformation is qualified, and the corresponding feature vectors to be verified are used as the final target feature vectors; if only the verification coefficients corresponding to the image target feature vectors to be verified are less than the corresponding reference values, then the convolution parameters are adjusted and fed back based on the difference ratio between the verification coefficient and the corresponding reference value. The gradient descent algorithm adjusts the convolution kernel size and pooling stride, and re-encodes the features and recalculates the corresponding verification coefficients. If only the verification coefficient of the gene target feature vector to be verified is less than the corresponding reference value, the gene sequence data of the specified specific marker is re-screened and the dimension of the binary encoding is updated, and the corresponding verification coefficient is recalculated. If only the verification coefficient of the serum target feature vector to be verified is less than the corresponding reference value, the update feedback is based on the difference ratio between the verification coefficient and the corresponding reference value. The Z-score standardization process is re-executed through an iterative update algorithm, and the corresponding verification coefficient is recalculated. Except for the above cases, all other cases are judged as multi-target vector collaborative transformation errors, and an anomaly alarm is issued.

4. The osteosarcoma detection method based on multi-target collaborative recognition as described in claim 3, characterized in that, The specific process for verifying the compatibility of multi-target combination is as follows: The final target feature vector is used as input to the mutual information calculation equation. The mutual information values ​​of image-gene, image-serum, and gene-serum are calculated sequentially using the mutual information calculation equation. If all three mutual information values ​​are not less than a pre-set compatibility verification value, the multi-target combination is deemed compatible, and the gene target vector in the corresponding target combination is obtained. If any group's mutual information value is less than the pre-set compatibility verification value, the multi-target combination is deemed uncompatible. If two or more groups' mutual information values ​​are less than the pre-set compatibility verification value, the target feature vectors in the uncompatible target groups are first filtered using a feature filtering algorithm to retain the correlation feature dimension with the target category up to a pre-set percentage of the mutual information value, and this is used as a new target feature vector. Then, targeted optimization is performed on each group's filtered target feature vectors. If only one group has a mutual information value less than the pre-set compatibility verification value... If the mutual information value of a group is less than the preset fit verification value, targeted optimization is performed directly based on the corresponding mutual information value deviation. Specifically: if the mutual information value deviation of any group is not greater than the preset allowable mutual information value deviation, then based on the feature selection algorithm, the mutual information value between the target feature vector of the corresponding group and the target category is calculated, sorted from largest to smallest, and the target feature vector with the largest mutual information value is selected as the new target feature subset to update the target feature vector of the corresponding group; if the mutual information value deviation of any group is greater than the preset allowable mutual information value deviation, then the preset personnel are prompted to reconstruct the target feature vector of the corresponding group based on the mutual information value deviation; after targeted optimization, the three types of mutual information values ​​are re-acquired. If all three types of mutual information values ​​are not less than the preset fit verification value, then the multi-target group is deemed to be fit and qualified, and the gene target vector in the corresponding target combination after the fit verification is qualified is obtained; otherwise, a target combination adjustment prompt is sent.

5. The osteosarcoma detection method based on multi-target collaborative recognition as described in claim 4, characterized in that, The specific process for identifying abnormal coding vectors is as follows: A pre-defined gene coding vector baseline library for a specified specific biomarker is retrieved. The gene target vectors in the target combination are compared dimension-by-dimensionally with the coding vectors of the corresponding biomarkers in the baseline library, and the coding vector difference is calculated. If the coding vector difference is not less than a pre-defined coding vector difference, the coding vector fragment of the corresponding biomarker is preliminarily determined to contain a mutation site, and cross-validation of serum target vectors and imaging target vectors is performed. If the coding vector difference is less than a pre-defined coding vector difference, the coding vector fragment of the corresponding biomarker is preliminarily determined to not contain a mutation site, and a pre-defined personnel is prompted for further verification.

6. The osteosarcoma detection method based on multi-target collaborative recognition as described in claim 5, characterized in that, The specific steps of the cross-validation are as follows: input the mutation site code obtained from the preliminary determination into the pixel feature matrix of the lesion region to which the image target vector belongs, perform mutation site code comparison, and obtain code comparison parameters; the code comparison parameters include the base matching ratio and the overlapping region ratio; the pixel feature matrix contains the image mutation site code of the lesion region; perform arithmetic mean processing on the obtained base matching ratio and overlapping region ratio, and the arithmetic mean processing result is the code matching degree. If the coding matching degree is not less than the preset coding matching degree, the mutation site code obtained in the preliminary judgment will be confirmed as the final mutation site code; otherwise, it will be judged as having no clear mutation site code. The results of statistical cross-validation were analyzed, and gene target pre-detection based on multimodal feature association was performed. During the pre-detection process, the degree of data matching between gene target and final mutation site encoding was quantified.

7. The osteosarcoma detection method based on multi-target collaborative recognition as described in claim 6, characterized in that, The quantification of data matching degree specifically involves: monitoring the data matching process between the gene target and the final mutation site encoding during sequence complementation; calculating the ratio of the number of successfully matched nucleotide sites to the total number of gene target sites, denoted as the site matching percentage; calculating the ratio of the number of unmatched nucleotide site fragments to the total number of gene target sequence sites, denoted as the mismatch fragment percentage; normalizing the difference between the obtained site matching percentage and mismatch fragment percentage to obtain a complementarity continuity value, and providing adaptation feedback for gene target pre-detection.

8. The osteosarcoma detection method based on multi-target collaborative recognition as described in claim 7, characterized in that, The nonlinear monitoring of the detection signal amplification and response process specifically involves: obtaining the adaptation feedback results of gene target pre-detection, inputting the gene target vectors therein into a multimodal feature extraction network, simultaneously importing the pixel feature matrix of the image target vectors and the gene sequence features encoded by the final mutation sites, performing a modal-wise nonlinear mapping to obtain standardized high-dimensional vectors corresponding to serum target quantification features, image spatial structure features, and gene sequence functional features; and obtaining the synergistic effect duration for quantifying the cross-modal feature complementary enhancement effect based on the interactive response process of the three types of standardized high-dimensional vectors in the multimodal fusion network. The ratio of the obtained synergistic effect duration to the preset synergistic effect duration is used to perform signal correction in conjunction with a nonlinear fitting algorithm to obtain a nonlinear monitoring index that reflects the stability and feature correlation of the detection signal. This index is then compared with a preset nonlinear monitoring index to determine whether to output the bone and flesh detection result.

9. The osteosarcoma detection method based on multi-target collaborative recognition as described in claim 8, characterized in that, The determination of whether to output the osteosarcoma detection result is as follows: if the obtained nonlinear monitoring index is lower than the preset nonlinear monitoring index, it is determined that the detection signal has fluctuation interference, the osteosarcoma detection result is not output, and feedback of re-monitoring after sending multimodal feature mapping parameters is sent; if the obtained nonlinear monitoring index is not lower than the preset nonlinear monitoring index, it is determined that the detection signal has no abnormal interference, and the interaction response duration of the three types of standardized high-dimensional vectors is monitored at the same time. The interaction response duration represents the activation duration of the signals between the corresponding three types of standardized high-dimensional vectors during the signal correction process. If the interaction response time is within the preset allowable range, the osteosarcoma detection result will be output; otherwise, the osteosarcoma detection result will not be output, and feedback on re-monitoring after sending multimodal feature mapping parameters will be sent.

10. An osteosarcoma detection system based on multi-target collaborative recognition, employing the osteosarcoma detection method based on multi-target collaborative recognition as described in any one of claims 1-9, characterized in that, include: The feature vector acquisition and fit verification module is used to collect multimodal target data of the target image region and perform feature encoding processing to obtain multi-target feature vectors, while verifying the fit of the multi-target combination process; the mutation site identification and data matching degree quantification module is used to identify abnormal encoded vectors of gene target vectors in the corresponding target combination after the fit verification is qualified, so as to determine whether there is a mutation site encoding, and quantify the data matching degree based on the determination result of mutation site encoding. The nonlinear monitoring and detection result output judgment module is used to perform gene target specific binding detection based on the quantification results of the data matching degree. Simultaneously, it combines a multimodal feature extraction network to perform nonlinear monitoring of the detection signal amplification and response process to determine whether to output the osteosarcoma detection result.

Images