CN-122023987-A - Tumor identification method based on multi-mode information collaboration

CN122023987ACN 122023987 ACN122023987 ACN 122023987ACN-122023987-A

Abstract

The invention provides a tumor recognition method based on multi-mode information cooperation, and relates to the technical field of tumor recognition. The method comprises the steps of firstly constructing a multi-mode collaborative characterization unit, synchronously receiving medical images and electrical impedance time sequence data, cooperatively mining macro morphology association features and microscopic difference features of tumors by utilizing a multi-domain coupling characterization space technology, outputting multi-mode collaborative feature vectors, then constructing an electrical impedance time sequence analysis unit, extracting deep association features of electrical impedance signals through multi-order analysis of time sequence features, outputting deep electrical impedance feature vectors, then designing a dynamic feature adaptation fusion unit, carrying out self-adaptive fusion and dimensional recombination on the multi-mode collaborative feature vectors and the deep electrical impedance feature vectors based on a feature responsiveness adaptation technology, outputting stabilized fusion feature vectors, and finally constructing a feature association judgment engine, quantifying feature contribution degree based on a structural causal model and a multi-layer association verification technology, and outputting tumor diagnosis results. The problems of insufficient multi-mode feature cooperativity and interpretation deviation caused by black box mapping in the prior art are effectively solved, and the accuracy and logic robustness of tumor identification are improved.

Inventors

YAN ZIBAO
HOU JINXUAN
XIAO JIABO

Assignees

武汉中针智诊科技有限公司

Dates

Publication Date: 20260512
Application Date: 20260130

Claims (10)

1. A tumor identification method based on multi-mode information cooperation is characterized by comprising the following steps: S1, constructing a multi-mode collaborative characterization unit, synchronously receiving medical images and electrical impedance time sequence data, collaborative mining macro-morphology association features and micro-difference features of tumors based on a multi-domain coupling characterization space technology, and outputting multi-mode collaborative feature vectors ; S2, building an electrical impedance time sequence analysis unit, processing electrical impedance time sequence data, extracting deep association features of electrical impedance signals through a time sequence feature multi-stage analysis technology, and outputting electrical impedance deep feature vectors ; S3, designing a dynamic characteristic adaptation fusion unit, and adapting the multi-mode collaborative feature vector based on a characteristic responsiveness adaptation technology And electrical impedance deep feature vector Performing adaptive fusion and dimension recombination, and outputting stabilized fusion feature vectors ; And S4, constructing a feature association judging engine, quantifying feature contribution based on a multi-layer association checking technology, and outputting a tumor diagnosis result.
2. The tumor recognition method based on multi-modal information cooperation according to claim 1, wherein the step S1 specifically comprises the following steps: s11, acquiring a tumor tissue structure slice and an electrical impedance sequence at the same time through a synchronous trigger mechanism; S12, mapping the image pixel grid points to an image representation domain; s13, impedance time sequence signal Performing time sequence to space domain mapping to obtain a time sequence representation vector 。
3. The method for tumor recognition based on multi-modal information coordination according to claim 2, wherein the step S1 further comprises the steps of: S14, utilizing multi-domain coupling correlation function Integrating the double-domain information, and converting the feature extraction into a process of searching for a physical steady state in an energy field composed of images and electrical impedance data; s15, solving a stable representation state by adopting intra-domain evolution convergence, and enabling total energy of the system to tend to be minimum by iterative calculation to obtain a steady-state field ; S16, adopting a space frequency domain operator decomposition method to obtain a stable state field Extracting macro-associated features And microscopic difference features ; S17, fusing macroscopic association features by adopting covariance pair Ji Jiaquan fusion method And microscopic difference features Obtaining the multi-mode cooperative feature vector 。
4. The tumor recognition method based on multi-modal information coordination according to claim 3, wherein in step S16: Using low-pass operators For steady-state field Extracting macro-associated features ; Using high-pass residual operators For steady-state field Capturing microscopic differential features 。
5. The tumor recognition method based on multi-modal information cooperation according to claim 1, wherein the step S2 specifically comprises the following steps: S21, adopting a multistage physical filtering algorithm to perform filtering on original signals Purifying to obtain a purified signal ; S22, processing the purified signal by the same method as that of the step S13 Obtaining a purified time sequence representation vector ; S23, adopting a time sequence track analysis operator Extracting evolution rule to obtain a relation matrix 。
6. The method for tumor recognition based on multi-modal information coordination according to claim 5, wherein in step S21: The purification method comprises the following steps: s211, correcting the baseline drift to obtain corrected signals ; S212, utilizing frequency response function Filtering the power frequency interference to obtain a filtered signal ; S213 normalization processing of the filtered signals Outputting a purge signal 。
7. The method for tumor recognition based on multi-modal information coordination according to claim 5, wherein step S2 further comprises the steps of: s24, calculating the evolution intensity of the signal by adopting a multidimensional correlation integral method, and firstly obtaining a correlation integral function Through the associated integral function Calculating to obtain comprehensive association strength parameters for describing tissue electrophysiological complexity ; S25, constructing a time sequence evolution track structure characterization model, and constructing a nested simplex complex on a time sequence data point cloud Capturing topological features by calculating coherent groups, and calculating a current signal topological structure persistence point set And standard model Between (a) and (b) A distance; S26, fusing the comprehensive association strength parameters through a nonlinear projection fusion operator Feature point set with topology persistence Obtaining the deep characteristic vector of the electrical impedance 。
8. The tumor recognition method based on multi-modal information cooperation according to claim 1, wherein the step S3 specifically comprises the following steps: S31, based on a transducer architecture pair And (3) with Dynamic reorganization is carried out, and a matrix is obtained through transformation Weight parameters Then, the feature recombination is completed by using a cross attention mechanism to obtain an initial fusion feature vector ; S32, pair Proceeding with Normalization processing is carried out to obtain a final stabilized fusion feature vector 。
9. The tumor recognition method based on multi-modal information cooperation according to claim 1, wherein step S4 specifically comprises the following steps: S41, building a related network frame based on a structural causal model; First, define a set of input variables containing unobserved background factors ; Next, define a composition containing a synergistic component And impedance component Core variable set of (2) ; Finally, the cause and effect mapping rule is described directionally; S42, utilizing an adaptive mapping operator Fusion feature vectors will be stabilized Conversion to core variable nodes 。
10. The method for tumor recognition based on multi-modal information coordination according to claim 9, wherein step S4 further comprises the steps of: S43, performing multi-layer causal correlation verification on the core variable by adopting do-algorithm, and calculating a diagnosis predicted value of the dry prognosis And verifying causal link robustness between the input features and the diagnostic conclusions; s44, calculating the pure contribution degree of each characteristic component by using the contribution degree expectation operator Outputting final tumor diagnosis result based on high-contribution feature set 。

Description

Tumor identification method based on multi-mode information collaboration Technical Field The invention relates to the technical field of tumor recognition, in particular to a tumor recognition method based on multi-mode information collaboration. Background The clinical treatment effect of the tumor is improved, the life cycle of a patient is prolonged and is highly dependent on early detection and identification of the tumor, at present, clinical diagnosis mainly depends on medical imaging examination and electrophysiological detection, in the prior art, a single-mode-based identification scheme has obvious limitations, medical images can intuitively show macroscopic anatomical forms of the tumor, but have insufficient sensitivity to complex electrophysiological metabolism abnormality inside tumor tissues, and electrical impedance detection has extremely high sensitivity to functional changes of tissues but has weaker performance in spatial resolution and morphological positioning. In order to overcome the limitation of single mode, a multi-mode fusion scheme such as the Chinese patent CN115830017A has appeared in recent years, which discloses a tumor detection system based on image-text multi-mode fusion, and the detection accuracy is improved by fusing image feature vectors and text feature vectors of electronic medical records, however, the prior multi-mode scheme still has the following difficulties in practical application: The conventional weight distribution method easily causes that the model excessively depends on a certain mode to generate fusion weight unbalance so as to influence the generalization stability of the model in a multi-center and cross-equipment scene, the conventional AI identification model is mostly modeled based on statistical correlation, belongs to black box mapping, cannot distinguish pseudo-correlation interference caused by real pathological features and unobserved tissue background, causes poor interpretation of diagnosis suggestions and is difficult to meet the causal logic requirement of the medical field. In summary, in the prior art, when the medical image and the electrical impedance signal are processed and fused deeply, there are still problems of not deep information cooperation, low resolution precision, large fusion deviation, lack of logic interpretation and the like, and a tumor recognition method which has better causality and robustness and can realize multi-mode depth cooperation is urgently needed. Disclosure of Invention The invention mainly aims to provide a tumor recognition method based on multi-mode information cooperation, which solves the problems of low multi-mode fusion depth, strong pseudo-correlation interference and limited recognition accuracy. In order to solve the technical problems, the technical scheme adopted by the invention is that the tumor identification method based on multi-mode information cooperation comprises the following steps: S1, constructing a multi-mode collaborative characterization unit, synchronously receiving medical images and electrical impedance time sequence data, collaborative mining macro-morphology association features and micro-difference features of tumors based on a multi-domain coupling characterization space technology, and outputting multi-mode collaborative feature vectors 。 S2, constructing an electrical impedance time sequence analysis unit, independently processing electrical impedance time sequence data, extracting deep association characteristics of electrical impedance signals through a time sequence characteristic multi-stage analysis technology, and outputting electrical impedance deep characteristic vectors。 S3, designing a dynamic feature adaptation fusion unit, carrying out self-adaptive fusion and dimension recombination on the two-channel features output by S1 and S2 based on a feature responsiveness adaptation technology, and outputting a stabilized fusion feature vector。 And S4, constructing a feature association judging engine, quantifying feature contribution degree based on a multi-layer association checking technology, and outputting a high-precision tumor diagnosis result. In a preferred embodiment, the step S1 specifically includes the following steps: s11, synchronously starting the medical imaging equipment and the impedance acquisition device by utilizing hardware trigger pulse to acquire a tumor tissue structure slice and an electrical impedance sequence at the same time. S12, mapping the pixel grid points of the medical image to an image characterization domain, wherein a formula for defining a mapping relation is as follows: (1); Wherein, the For the intra-domain mapping operator, physical attribute conversion from the original pixel information to the characterization factors is realized; Characterizing intra-domain for images A characterization factor at the coordinates; Is the coordinates A standardized base value of the image; Outputting the characterization factor And the