Multi-modal ultrasonic fusion system and method based on dynamic conflict perception

Abstract

The invention discloses a multi-modal ultrasonic fusion system based on dynamic conflict perception and a method thereof, provides a method capable of evaluating and calibrating modal uncertainty aiming at dynamic characteristics of ultrasonic imaging and complex effects of the ultrasonic imaging in liver cancer, and solves prognosis reaction conflicts in combination with a calibration result. According to the method, efficient integration of cross-modal prognosis data is achieved, the accuracy of survival analysis is remarkably improved, individualized treatment selection of liver cancer patients is supported, and the method has important clinical application value.

Description

Technical Field

[0001] The present invention belongs to the technical field of medical ultrasound multimodal data processing, and in particular relates to a multimodal ultrasound fusion system and method based on dynamic conflict perception. Background Art

[0002] Despite the continuous advancement of treatment technology, the overall prognosis of liver cancer patients is still not ideal, with a five-year survival rate of only 14%. Among them, the survival rate of patients who undergo radical surgical resection is as high as 69% to 86%. However, only about 20% of patients meet the indications for surgical treatment when they are first diagnosed. Early detection and treatment of liver cancer is the key to improving the survival rate of liver cancer patients and is one of the important directions of current clinical research.

[0003] The choice of treatment for early-stage liver cancer is mainly based on the patient's liver function reserve, presence of large vessel invasion, and intra- and extra-hepatic metastasis. Surgical resection and radiofrequency ablation are the most common treatments, but due to the heterogeneity and complexity of liver tumors, the prognostic results of these two treatments are highly uncertain. Therefore, it is necessary to develop accurate survival analysis tools and personalized treatment selection methods to optimize patient treatment plans.

[0004] Survival analysis is a key step in personalized treatment of liver cancer patients. It can help clinicians predict the risk of disease recurrence or death by analyzing the patient's survival probability within a specific time period. The Cox proportional hazard model is currently the standard method for survival analysis. It can not only quantify the survival risk, but also evaluate the impact of multiple clinical variables (such as gender, age, and tumor size) on prognosis. However, traditional survival analysis models mainly rely on single-modal data and are difficult to fully capture the complex characteristics of liver cancer.

[0005] Multimodal ultrasound imaging technology combines B-ultrasound, ultrasound angiography, and clinical information to provide a comprehensive perspective for the diagnosis and treatment of liver tumors. B-ultrasound can visualize the size, shape, location, and morphological characteristics of the surrounding tissues of the tumor, and is the main tool for basic imaging analysis of liver cancer; ultrasound angiography can enhance the observation of tumor blood supply status by dynamically displaying tumor microcirculation characteristics in real time; clinical information such as pathological examination is used as a key indicator for evaluating pathological changes such as fibrosis or cirrhosis. By integrating morphological, functional, and clinical data, multimodal ultrasound improves the accuracy of survival analysis and supports personalized treatment selection.

[0006] Although multimodal ultrasound imaging has shown great potential in survival analysis and personalized treatment of liver cancer, its practical application still faces many challenges. First, the instability of modality quality is one of the key issues. In actual clinical environments, the quality and consistency of data may be affected by multiple factors such as imaging plane selection, equipment parameter adjustment, and imaging artifacts, making it difficult to meet the assumptions of traditional fusion methods on data stability and semantic alignment. Secondly, the inconsistency of modality information further exacerbates the difficulty of fusion. For example, B-ultrasound may show blurred tumor margins, suggesting a poor prognosis, while contrast-enhanced ultrasound may show a uniform high-enhancement pattern, pointing to a better prognosis, resulting in contradictory prognostic information from different modalities. In addition, most existing multimodal fusion methods rely on fixed or heuristic weight allocation, lack the ability to dynamically adjust data credibility, and are difficult to effectively deal with conflicts or uncertainties between modalities. These problems severely limit the application of multimodal ultrasound technology in the prognostic analysis of liver cancer. An innovative solution is urgently needed to improve the credibility and adaptability of fusion, effectively model modality-specific uncertainty, and dynamically adjust modal contributions, so as to extract stable and accurate prognostic information from multimodal ultrasound data, and provide solid technical support for personalized treatment options for liver cancer patients. Summary of the invention

[0007] In view of the shortcomings of existing multimodal fusion methods that rely on fixed or heuristic weight allocation, the present invention proposes a multimodal ultrasound fusion system and method based on dynamic conflict perception for reliable liver cancer survival analysis and treatment selection.

[0008] To achieve the above object, the technical solution adopted by the present invention is:

[0009] A multi-modal ultrasound fusion system based on dynamic conflict perception includes a B-mode ultrasound image acquisition module, a contrast-enhanced ultrasound image acquisition module, a clinical data acquisition module, and further includes:

[0010] A self-calibrated uncertainty estimation module is used to propose a cross-modal uncertainty calibration regularizer to align modal uncertainties with their prediction effects;

[0011] Dynamic conflict fusion module, which is used to propose a dynamic conflict-aware fusion mechanism to dynamically adjust the weights of different modalities according to cross-modal consistency and modal uncertainty;

[0012] Individual survival risk prediction module, used to predict survival risk based on model output and to perform risk stratification for liver cancer patients;

[0013] Personalized treatment plan recommendation module, used to recommend treatment plans for liver cancer patients.

[0014] A method for multi-modal ultrasound fusion based on dynamic conflict perception, firstly collects B-mode ultrasound images, contrast-enhanced ultrasound images and clinical data, including the following steps:

[0015] S1: Self-calibrated uncertainty estimation: Propose a cross-modal uncertainty calibration regularizer to align modal uncertainties with their prediction effects;

[0016] S2: Dynamic conflict fusion: A dynamic conflict-aware fusion mechanism is proposed to dynamically adjust the weights of different modalities according to cross-modal consistency and modality uncertainty;

[0017] S3: Based on the survival risk prediction output by the model, risk stratification and treatment plan recommendations are performed for liver cancer patients.

[0018] Furthermore, the method of self-calibration uncertainty estimation in step S1 is as follows:

[0019] In view of the characteristics of the Cox proportional hazard model in estimating relative risk, the partial likelihood loss function is used instead of the commonly used mean square error (MSE); this loss function quantifies the prediction accuracy by calculating the relative risk between the individual and the risk set; the formula is defined as follows:

[0020]

[0021] Where: i represents the i-th individual, j represents the j-th individual, m represents the mode, u i，m is the uncertainty constraint, E i，m is the prediction error, r i，m Predict the risk function for the i-th individual, r j，m The jth individual predicts the risk function, and the risk set R(·) refers to the risk at the time t i The individuals that still exist, δ i =1, it means that the event occurred in the ith individual. The ∝ in the formula represents direct proportion, -log(·) is a logarithmic function, and ∑ is the sum of all individuals that have experienced the event. Used to calculate the sum of partial likelihood losses of all individuals who have experienced an event, exp(r i，m ) and exp(r j，m ) represents the exponential form of the predicted risk scores of the i-th individual and the j-th individual under modality m; this alignment design can ensure that in the subsequent modality fusion process, the modality with larger errors is assigned higher uncertainty, thereby reducing its negative impact on the overall prediction result;

[0022] In batch data processing, the modal uncertainty u i，m The constraint is the same as the prediction error E i，mThey are proportional to each other and are transformed into discrete distributions U and E through normalization, which are defined as follows:

[0023]

[0024] Where: B is the batch size, M is the number of modalities, represents the uncertainty value of the last element in all samples and modes, Represents the prediction error of the last element in all samples and modes. The normalized uncertainty vector U and error vector E can eliminate the interference of the number of modes and batch size on uncertainty estimation, ensuring the consistency and stability of the distribution.

[0025] Furthermore, the method of dynamic conflict fusion in step S2 is as follows:

[0026] In order to quantify the conflict between risk predictions of different modalities, a multimodal conflict matrix S is defined, where each element in S represents the prognostic difference s between modality i and modality j. ij ,s ij =|r i -r j |, where: r i represents the predicted risk function of mode i, r j represents the predicted risk function of modality j, and the conflict matrix S quantifies the inconsistency across modalities by calculating the difference in prognostic results between modalities;

[0027] To adjust the difference between modes according to the confidence of the modal prediction, ij , further introducing the conflict adjustment factor d ij ; If the uncertainty u of mode i i Lower than the uncertainty u of mode j j , indicating that mode i is more reliable, in this case, the fusion is optimized by down-regulating the difference contribution from the less reliable mode j; on the contrary, if the uncertainty of mode j is lower, then retain s ij The complete contribution of , to maximize the consistency between the modes;

[0028]

[0029] Through this mechanism, the model can make weighted adjustments to the conflict matrix S based on uncertainty and construct the conflict adjustment matrix D, so that in, represents the reweighted conflict matrix S, ⊙, that is, the conflict matrix S and the conflict adjustment matrix D are multiplied element by element. This reweighting mechanism ensures that the more reliable modality has a greater influence in the fusion process, while the weight of the unreliable modality is appropriately reduced; at the same time, the overall conflict score of each modality is obtained by the following formula:

[0030]

[0031] Among them, s~ ij represents the weighted difference between modal i and j. It represents the final conflict score after mode i is adjusted. In order to dynamically adjust the contribution of each mode, the The reward and penalty functions of If the average conflict threshold θ is exceeded, the penalty function is applied; otherwise, the reward function is applied, and α controls the degree of the reward and punishment function; the reward and punishment function is defined as follows:

[0032]

[0033] Among them, e is a natural constant, approximately equal to 2.71828, represents the final conflict score after adjustment of modality i; through the above dynamic fusion mechanism, the present invention ensures that the more reliable modality occupies an important position in multimodal fusion, while the influence of the modality with high conflict or unreliable on the overall prediction is effectively suppressed; finally, the features x of each modality are i According to its weight w i Perform weighted aggregation to obtain the final multimodal fusion representation x f , for more accurate prognostic prediction;

[0034]

[0035] This method effectively improves the reliability and accuracy of multimodal data in liver cancer prognosis prediction through a dynamic fusion mechanism, and also provides a flexible method to deal with modal conflicts.

[0036] Compared with the prior art, the present invention has the following beneficial effects:

[0037] Compared with traditional fixed or heuristic methods, this method has the characteristics of strong dynamic adaptability and high robustness, and can effectively cope with the complexity of multimodal data in an open clinical environment, thereby achieving high accuracy and dynamics in liver cancer survival analysis. This method provides important support for the formulation of personalized treatment plans and helps promote the development of precision medicine for liver cancer. BRIEF DESCRIPTION OF THE DRAWINGS

[0038] Figure 1 It is a flow chart of the present invention. DETAILED DESCRIPTION

[0039] The present invention will be further described below in conjunction with the embodiments.

[0040] Example 1

[0041] A multi-modal ultrasound fusion system based on dynamic conflict perception includes a B-mode ultrasound image acquisition module, a contrast-enhanced ultrasound image acquisition module, a clinical data acquisition module, and further includes:

[0042] A self-calibrated uncertainty estimation module is used to propose a cross-modal uncertainty calibration regularizer to align modal uncertainties with their prediction effects;

[0043] Dynamic conflict fusion module, which is used to propose a dynamic conflict-aware fusion mechanism to dynamically adjust the weights of different modalities according to cross-modal consistency and modal uncertainty;

[0044] Individual survival risk prediction module, used to predict survival risk based on model output and to perform risk stratification for liver cancer patients;

[0045] Personalized treatment plan recommendation module, used to recommend treatment plans for liver cancer patients.

[0046] Example 2

[0047] A method for multi-modal ultrasound fusion based on dynamic conflict perception, firstly collects B-mode ultrasound images, contrast-enhanced ultrasound images and clinical data, including the following steps:

[0048] S1: Self-calibrated uncertainty estimation: A cross-modal uncertainty calibration regularizer is proposed to align modal uncertainty with its prediction effect; step S1 introduces a cross-modal uncertainty calibration regularizer, and this method uses normalized distribution matching to align the inherent uncertainty of each modality with the prediction error. This calibration ensures that the fusion process prioritizes the modality with reliable prediction, thereby significantly improving the credibility and accuracy of the overall prognosis.

[0049] Furthermore, the method of self-calibration uncertainty estimation in step S1 is as follows:

[0050] In view of the characteristics of the Cox proportional hazard model in estimating relative risk, the present invention adopts the partial likelihood loss function instead of the commonly used mean square error (MSE); the loss function quantifies the prediction accuracy by calculating the relative risk between the individual and the risk set; the formula is defined as follows:

[0051]

[0052] Where: i represents the i-th individual, j represents the j-th individual, m represents the mode, u i，m is the uncertainty constraint, E i，m is the prediction error, r i，m Predict the risk function for the i-th individual, r j，m The jth individual predicts the risk function, and the risk set R(·) refers to the risk at the time t iThe individuals that still exist, δ i =1, it means that the event occurred in the ith individual. The ∝ in the formula represents direct proportion, -log(·) is a logarithmic function, and ∑ is the sum of all individuals that have experienced the event. Used to calculate the sum of partial likelihood losses of all individuals who have experienced an event, exp(r i，m ) and exp(r j，m ) represents the exponential form of the predicted risk scores of the i-th individual and the j-th individual under modality m; this alignment design can ensure that in the subsequent modality fusion process, the modality with larger errors is assigned higher uncertainty, thereby reducing its negative impact on the overall prediction result;

[0053] In batch data processing, the present invention converts the modal uncertainty u i，m The constraint is the same as the prediction error E i，m They are proportional to each other and are transformed into discrete distributions U and E through normalization, which are defined as follows:

[0054]

[0055] Where: B is the batch size, M is the number of modalities, represents the uncertainty value of the last element in all samples and modes, Represents the prediction error of the last element in all samples and modalities. The normalized uncertainty vector U and error vector E can eliminate the interference of the number of modalities and batch size on uncertainty estimation, ensuring the consistency and stability of the distribution;

[0056] S2: Dynamic conflict fusion: A dynamic conflict-aware fusion mechanism is proposed to dynamically adjust the weights of different modalities based on cross-modal consistency and modal uncertainty; in the fusion stage, the present invention dynamically adjusts the modal weights based on cross-modal consistency and modal uncertainty. When there is a conflict between multiple modalities, the degree of conflict is first quantified, and then the calibration uncertainty is used for adjustment, and finally the modal weights are re-weighted through a reward and punishment mechanism to ensure that more stable and reliable modal inputs are given priority. Compared with traditional fixed or heuristic methods, the present invention has the characteristics of strong dynamic adaptability and high robustness, and can effectively cope with the complexity of multimodal data in an open clinical environment, thereby achieving high precision and dynamics in liver cancer survival analysis. This method provides important support for the formulation of personalized treatment plans, and at the same time helps to promote the development of precision medicine for liver cancer;

[0057] Furthermore, the method of dynamic conflict fusion in step S2 is as follows:

[0058] In order to quantify the conflict between risk predictions of different modalities, the present invention defines a multimodal conflict matrix S, where each element in S represents the prognostic difference s between modality i and modality j.ij ,s ij =|r i -r j |, where: r i represents the predicted risk function of mode i, r j represents the predicted risk function of modality j, and the conflict matrix S quantifies the inconsistency across modalities by calculating the difference in prognostic results between modalities;

[0059] To adjust the difference between modes according to the confidence of the modal prediction, ij The present invention further introduces the conflict adjustment factor d ij ; If the uncertainty u of mode i i Lower than the uncertainty u of mode j j , indicating that mode i is more reliable, in this case, the fusion is optimized by down-regulating the difference contribution from the less reliable mode j; on the contrary, if the uncertainty of mode j is lower, then retain s ij The complete contribution of , to maximize the consistency between the modes;

[0060]

[0061] Through this mechanism, the model can make weighted adjustments to the conflict matrix S based on uncertainty and construct the conflict adjustment matrix D, so that in, represents the reweighted conflict matrix S, ⊙, that is, the conflict matrix S and the conflict adjustment matrix D are multiplied element by element. This reweighting mechanism ensures that the more reliable modality has a greater influence in the fusion process, while the weight of the unreliable modality is appropriately reduced; at the same time, the overall conflict score of each modality is obtained by the following formula:

[0062]

[0063] in, represents the weighted difference between modal i and j. It represents the final conflict score after mode i is adjusted. In order to dynamically adjust the contribution of each mode, the The reward and penalty functions of If the average conflict threshold θ is exceeded, the penalty function is applied; otherwise, the reward function is applied, and α controls the degree of the reward and punishment function; the reward and punishment function is defined as follows:

[0064]

[0065] Among them, e is a natural constant, approximately equal to 2.71828, represents the final conflict score after adjustment of modality i; through the above dynamic fusion mechanism, the present invention ensures that the more reliable modality occupies an important position in multimodal fusion, while the influence of the modality with high conflict or unreliable on the overall prediction is effectively suppressed; finally, the features x of each modality are i According to its weight w i Perform weighted aggregation to obtain the final multimodal fusion representation x f , for more accurate prognostic prediction;

[0066]

[0067] This method effectively improves the reliability and accuracy of multimodal data in liver cancer prognosis prediction through a dynamic fusion mechanism, and also provides a flexible method to deal with modal conflicts.

[0068] S3: Based on the survival risk prediction output by the model, risk stratification and treatment plan recommendations are performed for liver cancer patients; the method for recommending personalized treatment plans in step S3 is as follows:

[0069] This paper proposes a scientific and comprehensive personalized treatment plan recommendation method for hepatocellular carcinoma (HCC) patients based on a postoperative risk prediction model. This method combines multidimensional data analysis with risk stratification technology to provide patients with accurate treatment recommendations. The specific steps are as follows:

[0070] First, for new HCC cases, the postoperative risk prediction model h SR (t|x) and h RFA (t|x) is used to evaluate the postoperative risk of patients under surgical resection (SR) and ablation therapy (RFA). According to the prediction results, patients who underwent SR were divided into the original SR high-risk group and the original SR low-risk group, and patients who underwent RFA were divided into the original RFA high-risk group and the original RFA low-risk group. The original high-risk group represents patients with a poor prognosis, and the original low-risk group represents patients with a good prognosis. Then, the original SR group patients were input into the RFA risk prediction model λ RFA (t|x) were reclassified into new SR high-risk group and new SR low-risk group; at the same time, the original RFA group patients were entered into the SR risk prediction model λ SR(t|x), and re-divided into a new RFA high-risk group and a new RFA low-risk group. Through this two-way risk assessment method, the present invention can identify patients with potential improved prognosis for different treatment options. Focus on the following two types of non-overlapping patients: Non-overlapping patient type I: patients who originally belonged to the RFA high-risk group, but belong to the RFA low-risk group in the new risk grouping. These patients have a poor prognosis under RFA treatment, but if SR treatment is chosen, the prognosis may be significantly improved. Non-overlapping patient type II: patients who originally belonged to the SR high-risk group, but belong to the SR low-risk group in the new risk grouping. These patients have a poor prognosis under SR treatment, but if RFA treatment is chosen, the prognosis may be significantly improved. Through the accurate identification of non-overlapping patients, the present invention provides a clear decision-making basis for personalized treatment.

[0071] Finally, based on the risk grouping and non-overlapping characteristics of patients, the present invention proposes the following treatment recommendation principles: Type I patients: RFA is recommended as the preferred treatment option; Type II patients: SR is recommended as the preferred treatment option; For patients who show low risk for both SR and RFA: The decision for such patients can be based on a comprehensive consideration of multiple factors such as clinical preference, patient willingness or treatment cost, providing flexible personalized treatment options; For patients who show high risk for both SR and RFA: For such patients, it is necessary to make careful decisions based on existing prognostic models and other available treatment options to ensure the best treatment effect under controllable risks. Through the above steps, the present invention achieves a comprehensive assessment of the postoperative risks of HCC patients and the recommendation of accurate treatment options, providing patients with better personalized treatment options, and significantly improving the scientificity and effectiveness of clinical treatment.

[0072] In summary, the present invention aims at the shortcomings of existing multimodal fusion methods in clinical complexity and data diversity scenarios, and proposes a dynamic conflict-aware multimodal ultrasound fusion method. This method realizes the dynamic optimization and uncertainty calibration of modal weights through the two core steps of self-calibration uncertainty estimation and dynamic conflict-aware fusion, thereby improving the accuracy and reliability of multimodal data fusion. At the same time, based on the survival risk prediction results, the present invention further formulates personalized treatment plans to provide scientific and accurate treatment recommendations for patients with hepatocellular carcinoma. This not only significantly improves the clinical prognosis of patients, but also provides important technical support for promoting the development of precision medicine and personalized diagnosis and treatment of liver cancer.

[0073] Example 3

[0074] The professional medical team of the Ultrasound Diagnosis Department of a tertiary hospital in Nanjing has been engaged in liver cancer diagnosis and liver cancer ablation treatment guided by ultrasound angiography for a long time. They have rich experience in ultrasound image analysis and clinical data processing, and provide solid technical support and data guarantee for the research of this invention. In previous studies, the team has retrospectively collected case data of more than 400 patients with primary single hepatocellular carcinoma, including more than 250 patients who underwent surgical resection (SR) and more than 150 patients who underwent thermal ablation (RFA). During the implementation of this invention, it is planned to further collect preoperative imaging data and clinical information of liver cancer patients who underwent surgical resection or thermal ablation in a tertiary hospital in Nanjing to support the continued advancement of the research.

[0075] In terms of data collection, the present invention covers preoperative clinical data, liver ultrasound angiography, postoperative follow-up, and inclusion and exclusion criteria. Preoperative clinical data include the patient's gender, age, concomitant diseases (such as hepatitis B, hepatitis C, fatty liver), liver cirrhosis, biochemical indicators (alpha-fetoprotein, alanine aminotransferase, total bilirubin, albumin, platelet count) and tumor characteristics (lesion size, location, whether it is located around blood vessels or at the edge of the liver). These clinical information provides a comprehensive reference for the stratification of liver cancer patients and the formulation of treatment plans.

[0076] The liver ultrasound contrast examination was performed by an ultrasound physician with more than five years of experience, using the second-generation ultrasound contrast agent SonoVue, and contrast agent-specific imaging in low mechanical index imaging mode. During the contrast imaging process, the patient injected 2.4 ml of contrast agent through the elbow vein, followed by 5 ml of saline flushing, while collecting contrast images and recording the spatial distribution and flow path of the contrast agent in the lesion and surrounding tissues until the contrast agent was completely cleared. All images were stored in DICOM or AVI format to provide standardized imaging resources for subsequent data analysis.

[0077] Postoperative follow-up involves systematic observation of patients who received SR or RFA for two years. The follow-up plan includes regular evaluations at 1, 3, 6, 9, and 12 months after surgery, as well as follow-up examinations every 3-6 months thereafter. During follow-up, enhanced CT or contrast-enhanced MRI images are used as the gold standard for postoperative evaluation, with a focus on observing tumor progression characteristics (such as local tumor progression, vascular invasion, distant intrahepatic recurrence, and extrahepatic metastases). The time when postoperative tumor progression characteristics first appear is defined as progression-free survival (PFS); if there is no tumor progression, PFS is defined as the time point of death due to any cause or the last follow-up.

[0078] In terms of patient selection, the present invention clarifies the inclusion and exclusion criteria. The inclusion criteria include patients with primary single hepatocellular carcinoma confirmed by histology or cytology, with a maximum lesion diameter of ≤5.0cm, good liver function (Child-Pugh A grade), no serious dysfunction of important organs such as heart, lung, and kidney, and the ability to follow the study plan and follow-up procedures. The exclusion criteria cover patients who have received other treatments, those who cannot observe lesions or have poor image quality under CEUS, and those who cannot complete the study due to drug abuse or other factors.

[0079] In summary, the present invention has carried out detailed design in each link of data collection, processing and follow-up, and combined with strict inclusion and exclusion conditions, to ensure the scientificity and reliability of the research data. This implementation method provides a comprehensive and standardized operating framework for subsequent research, and lays a solid foundation for the clinical application of accurate diagnosis and treatment of liver cancer.

[0080] The above is only a preferred embodiment of the present invention. It should be pointed out that for ordinary technicians in this technical field, several improvements and modifications can be made without departing from the principle of the present invention. These improvements and modifications should also be regarded as the scope of protection of the present invention.

Claims

1. A multi-modal ultrasound fusion system based on dynamic conflict perception, comprising a B-mode ultrasound image acquisition module, a contrast-enhanced ultrasound image acquisition module, and a clinical data acquisition module, characterized in that: Also includes: A self-calibrated uncertainty estimation module is used to propose a cross-modal uncertainty calibration regularizer to align modal uncertainties with their prediction effects; Dynamic conflict fusion module, which is used to propose a dynamic conflict-aware fusion mechanism to dynamically adjust the weights of different modalities according to cross-modal consistency and modal uncertainty; Individual survival risk prediction module, used to predict survival risk based on model output and to perform risk stratification for liver cancer patients; Personalized treatment plan recommendation module, used to recommend treatment plans for liver cancer patients.

2. According to the method of a system for multi-modal ultrasound fusion based on dynamic conflict perception in claim 1, firstly, B-mode ultrasound images, contrast-enhanced ultrasound images and clinical data are collected, characterized in that: The following steps are involved: S1: Self-calibrated uncertainty estimation: Propose a cross-modal uncertainty calibration regularizer to align modal uncertainties with their prediction effects; S2: Dynamic conflict fusion: A dynamic conflict-aware fusion mechanism is proposed to dynamically adjust the weights of different modalities according to cross-modal consistency and modal uncertainty.

3. The multimodal ultrasound fusion method based on dynamic conflict perception according to claim 2, characterized in that: The method of self-calibration uncertainty estimation in step S1 is as follows: In view of the characteristics of the Cox proportional hazard model in estimating relative risk, the partial likelihood loss function is used instead of the commonly used mean square error MSE; the loss function quantifies the prediction accuracy by calculating the relative risk between the individual and the risk set; the formula is defined as follows: Where: i represents the i-th individual, j represents the j-th individual, m represents the mode, u i,m is the uncertainty constraint, E i,m is the prediction error, r i,m Predict the risk function for the i-th individual, r j,m Predict the risk function for the jth individual. The risk set R(·) refers to the risk at the time t when the event occurs. i The individuals that still exist, δ i =1, it means that the event occurred in the ith individual. The ∝ in the formula represents direct proportion, -log(·) is a logarithmic function, and ∑ is the sum of all individuals that have experienced the event. Used to calculate the sum of partial likelihood losses of all individuals who have experienced an event, exp(r i,m ) and exp(r j,m ) represents the exponential form of the predicted risk score of the i-th individual and the j-th individual under mode m; In batch data processing, the modal uncertainty u i,m The constraint is the same as the prediction error E i,m They are proportional to each other and are transformed into discrete distributions U and E through normalization, which are defined as follows: Where: B is the batch size, M is the number of modalities, represents the uncertainty value of the last element in all samples and modes, Represents the prediction error of the last element in all samples and modes. The normalized uncertainty vector U and error vector E can eliminate the interference of the number of modes and batch size on uncertainty estimation, ensuring the consistency and stability of the distribution.

4. The multimodal ultrasound fusion method based on dynamic conflict perception according to claim 3, characterized in that: The method of dynamic conflict fusion in step S2 is as follows: In order to quantify the conflict between risk predictions of different modalities, a multimodal conflict matrix S is defined, where each element in S represents the prognostic difference s between modality i and modality j. ij ,s ij =|r i -r j |, where: r i represents the predicted risk function of mode i, r j represents the predicted risk function of modality j, and the conflict matrix S quantifies the inconsistency across modalities by calculating the difference in prognostic results between modalities; To adjust the difference between modes according to the confidence of the modal prediction, ij , further introducing the conflict adjustment factor d ij ; If the uncertainty u of mode i i Below the uncertainty u of mode j j , indicating that mode i is more reliable, in this case, the fusion is optimized by down-regulating the difference contribution from the less reliable mode j; on the contrary, if the uncertainty of mode j is lower, then retain s ij The complete contribution of , to maximize the consistency between the modes; Through this mechanism, the model can make weighted adjustments to the conflict matrix S based on uncertainty and construct the conflict adjustment matrix D, so that in, represents the reweighted conflict matrix S, ⊙, that is, the conflict matrix S and the conflict adjustment matrix D are multiplied element by element, and the overall conflict score of each mode is obtained by the following formula: in, represents the weighted difference between modal i and j. represents the final conflict score after mode i is adjusted; in order to dynamically adjust the contribution of each mode, the The reward and penalty functions of If the average conflict threshold θ is exceeded, the penalty function is applied; otherwise, the reward function is applied, and α controls the degree of the reward and punishment function; the reward and punishment function is defined as follows: Among them, e is a natural constant, represents the final conflict score after mode i is adjusted; finally, each mode feature x i According to its weight w i Perform weighted aggregation to obtain the final multimodal fusion representation x f , for more accurate prognostic prediction;

Images