Brain tumor curative effect analysis system

Abstract

The invention relates to the technical field of medical data analysis, in particular to a brain tumor curative effect analysis system which comprises a tumor data sensing layer for collecting multi-department diagnosis and treatment data, tumor image data and patient pathology monitoring data; the tumor feature processing center extracts data features, correlates data and core features before and after treatment through an attention mechanism, and generates a tumor complete-cycle unified feature map; the therapeutic effect dynamic analysis unit evaluates the therapeutic effect in stages, and outputs a therapeutic effect index and a recurrence risk value through a self-supervised model; the dynamic adaptation decision module is used for generating personalized treatment adjustment suggestions based on the dynamic change of the blood brain barrier in combination with the curative effect index, the recurrence risk value and the multi-omics characteristics of the patient; and the AI multi-department consultation unit automatically matches similar cases with field expert suggestions, and formulates a target diagnosis and treatment scheme based on a visual platform and multi-department doctor collaborative consultation in combination with personalized treatment adjustment suggestions. Therefore, the problems of lagging effect evaluation, insufficient diagnosis and treatment suggestions and the like in the prior art are solved.

Description

Technical Field

[0001] This invention relates to the field of medical data analysis technology, specifically a brain tumor treatment efficacy analysis system. Background Technology

[0002] As a crucial facility for ensuring the accuracy of brain tumor diagnosis and treatment and improving patient prognosis, the brain tumor efficacy analysis system, during its long-term application, is prone to problems such as large biases in efficacy assessment, insufficient adaptability of treatment plans, and delayed recurrence warnings due to factors such as strong tumor heterogeneity, fragmented diagnostic and treatment data, and significant individual patient differences. This leads to an increased risk of overtreatment or undertreatment, and even threatens patient life. With the advancement of medical technology towards precision and intelligence, higher demands are placed on efficacy analysis in terms of assessment accuracy, individual adaptability, timeliness of recurrence warnings, and efficiency of diagnostic and treatment collaboration. There is an urgent need for an adaptive analysis system capable of real-time sensing of multi-dimensional diagnostic and treatment data, dynamically optimizing the assessment model, and rapidly adapting to individual patient conditions to ensure efficient and accurate operation of efficacy analysis in diverse brain tumor scenarios.

[0003] However, traditional brain tumor efficacy analysis has inherent flaws: assessment strategies are rigid, relying solely on single imaging or pathological indicators without integrating multi-omics data and dynamic patient signs, making it difficult to adapt to the highly heterogeneous tumors and dynamic disease progression, resulting in low assessment accuracy and limited value in guiding treatment; detection equipment is poorly adaptable to dynamic changes in the tumor microenvironment, spatiotemporal heterogeneity of molecular expression, and complex pathological features, with sensitivity drift and data fusion errors significantly reducing analysis response speed; efficacy management models are crude, with passive assessments being highly lagging and proactive assessments lacking individual adaptability, leading to low efficiency in matching treatment plans and potential delays in optimal treatment; and there is no fault-tolerant design, where a single detection module failure or data loss can cause the analysis system to fail, resulting in a long annual adjustment time due to analysis errors, far exceeding the stringent requirements of precision medicine for the reliability of efficacy assessment. Overall, the technology faces multiple challenges, including rigid assessment strategies, lagging effect assessment, and insufficient treatment recommendations. Summary of the Invention

[0004] This application provides a brain tumor efficacy analysis system to address the problems of delayed efficacy evaluation and insufficient treatment recommendations in the prior art.

[0005] The first aspect of this application provides a brain tumor efficacy analysis system, comprising: a tumor data perception layer, a tumor feature processing center, an efficacy dynamic analysis unit, a dynamic adaptation decision module, and an AI multidisciplinary consultation unit; wherein, the tumor data perception layer is used to collect multidisciplinary diagnosis and treatment data, tumor multimodal imaging data, and patient pathological monitoring data; the tumor feature processing center is used to extract the core features of the multidisciplinary diagnosis and treatment data, tumor multimodal imaging data, and patient pathological monitoring data, and generate a unified feature map of the entire tumor cycle by associating pre- and post-treatment data and the core features through an attention mechanism; the efficacy dynamic analysis unit... The first component is used to evaluate treatment efficacy in stages based on the unified feature map of the entire tumor cycle. It outputs efficacy index and recurrence risk value by integrating a self-supervised model with a spatiotemporal pyramid Transformer architecture. The second component is used to generate personalized treatment parameter adjustment suggestions based on the dynamic changes of the blood-brain barrier in imaging examinations, combined with the efficacy index, recurrence risk value, and multi-omics characteristics of patients. The third component is used to automatically match similar cases with opinions from domain experts, conduct multi-disciplinary doctor collaborative consultations through a three-dimensional visualization platform, and formulate a target treatment plan based on the personalized treatment parameter adjustment suggestions.

[0006] Preferably, the tumor data sensing layer includes a data acquisition module, a multimodal image integration unit, and a pathological data monitoring unit. The data acquisition module is used to collect diagnostic and treatment records, medication regimens, and clinical examination data from multiple departments, including internal medicine, surgery, oncology, and radiology. The multimodal image integration unit is used to receive CT, MRI, and PET tumor multimodal image data and align and integrate them according to examination timestamps and scanning planes. The pathological data monitoring unit is used to continuously collect postoperative pathological slide analysis results, immunohistochemical indicators, and gene detection data from patients, establishing a time-seriesd pathological monitoring archive.

[0007] Preferably, the tumor feature processing center includes a core feature extraction unit, a cross-temporal data association module, and a feature map generation engine. The core feature extraction unit is used to extract treatment plan features from multidisciplinary diagnostic and treatment data, extract imaging features such as tumor size, enhancement degree, and edema extent from multimodal images, and extract Ki-67 index and MGMT methylation status pathological features from pathological data. The cross-temporal data association module is used to mine the temporal dependency relationship between treatment intervention and changes in tumor features by using a multi-head attention mechanism to target the core correlation points between pre-treatment baseline data and post-treatment follow-up data. The feature map generation engine is used to bind multi-dimensional features to individual patients, constructing a unified tumor feature map covering the entire lifecycle, including time and indicator dimensions.

[0008] Preferably, the efficacy dynamic analysis unit includes a phased efficacy assessment module, a self-supervised model training unit, and a risk and efficacy quantification engine. The phased efficacy assessment module divides the entire treatment cycle into initial treatment, intermediate treatment, late treatment, and follow-up periods, analyzing the adaptability of tumor characteristic changes and treatment response at each stage. The self-supervised model training unit performs self-supervised training based on unlabeled clinical data, capturing the spatiotemporal multi-scale changes in tumor characteristics and predicting tumor state evolution trends through a self-supervised model that integrates a spatiotemporal pyramid Transformer architecture. The risk and efficacy quantification engine generates an efficacy index based on the tumor state evolution trend, integrating tumor shrinkage rate and characteristic stability indicators, and calculates recurrence risk values ​​using historical recurrence data to form a quantitative result.

[0009] Preferably, the dynamic adaptation decision module includes a blood-brain barrier dynamic monitoring unit, a multi-dimensional data fusion module, and a treatment parameter optimization engine. The blood-brain barrier dynamic monitoring unit monitors the dynamic changes of the blood-brain barrier in real time during treatment by dynamically comparing the Ktrans value and vascular permeability parameters of enhanced MRI. The multi-dimensional data fusion module weightedly fuses blood-brain barrier change data, efficacy index, recurrence risk value, and the patient's multi-omics genetic characteristics and basic clinical information. The treatment parameter optimization engine generates personalized treatment parameter adjustment suggestions based on the fusion results, including drug dosage adjustment, radiotherapy target area correction, and treatment plan switching.

[0010] Preferably, the AI ​​multidisciplinary consultation unit includes a similar case matching module, an expert opinion integration unit, and a 3D visualization collaboration platform. The similar case matching module automatically retrieves historical similar cases and prognostic results based on tumor type, stage, treatment plan, and core feature maps using a cosine similarity algorithm. The expert opinion integration unit captures and integrates the diagnostic and treatment consensus, relevant guidelines, and previous consultation opinions of authoritative experts in the field. The 3D visualization collaboration platform uses WebGL to construct a 3D reconstruction model of the tumor and brain tissue, enabling doctors from multiple departments to annotate key areas online, exchange diagnostic and treatment opinions in real time, and simultaneously display personalized treatment parameter adjustment suggestions.

[0011] The second aspect of this application provides a method for analyzing the efficacy of brain tumor treatment, comprising: acquiring multidisciplinary diagnostic and treatment data, tumor multimodal imaging data, and patient pathological monitoring data; extracting core features from the multidisciplinary diagnostic and treatment data, tumor multimodal imaging data, and patient pathological monitoring data, and generating a unified feature map of the entire tumor cycle by associating pre- and post-treatment data and the core features through an attention mechanism; evaluating the treatment effect in stages based on the unified feature map of the entire tumor cycle, and outputting an efficacy index and recurrence risk value by using a self-supervised model that integrates a spatiotemporal pyramid Transformer architecture; generating personalized treatment parameter adjustment suggestions based on the dynamic changes of the blood-brain barrier observed in imaging examinations, combined with the efficacy index, recurrence risk value, and patient multi-omics characteristics; simultaneously, automatically matching similar cases and opinions from domain experts, conducting multidisciplinary physician collaborative consultations through a three-dimensional visualization platform, and formulating a target treatment plan based on the personalized treatment parameter adjustment suggestions.

[0012] A third aspect of this application provides an electronic device, including: a memory, a processor, and a computer program stored in the memory and executable on the processor. The processor executes the program to implement a brain tumor efficacy analysis method as described in the above embodiments.

[0013] A fourth aspect of this application provides a computer-readable storage medium having a computer program stored thereon, which is executed by a processor to implement a brain tumor efficacy analysis method as described in the above embodiments.

[0014] A fifth aspect of this application provides a computer program product, including a computer program or instructions, for implementing a brain tumor efficacy analysis method as described in the above embodiments.

[0015] Therefore, this application has the following beneficial effects: This application's embodiments integrate multi-disciplinary, multimodal, and time-series diagnostic and treatment data through a tumor data perception layer, laying a comprehensive data foundation for efficacy analysis. The tumor feature processing center leverages attention mechanisms to uncover deep correlations between treatment and changes in tumor features, generating a unified feature map to enhance analytical relevance. The efficacy dynamic analysis unit achieves phased, precise assessment and recurrence warning through a spatiotemporal multi-scale self-supervised model, quantifying efficacy and risk. The dynamic adaptation decision module combines dynamic changes in the blood-brain barrier with multi-omics features to generate personalized treatment recommendations tailored to the individual. The AI ​​multi-disciplinary consultation unit promotes efficient multi-disciplinary collaboration through similar case matching, expert opinion integration, and 3D visualization, improving the accuracy, personalization, and collaborative efficiency of brain tumor diagnosis and treatment, effectively reducing the risk of overtreatment or undertreatment. Thus, it solves the problems of lagging efficacy evaluation and insufficient treatment recommendations in existing technologies.

[0016] Additional aspects and advantages of this application will be set forth in part in the description which follows, and in part will be obvious from the description, or may be learned by practice of this application. Attached Figure Description

[0017] The above and / or additional aspects and advantages of this application will become apparent and readily understood from the following description of the embodiments taken in conjunction with the accompanying drawings, wherein: Figure 1 This is a schematic diagram of the structure of a brain tumor treatment efficacy analysis system provided according to an embodiment of this application; Figure 2 This is a schematic diagram of a tumor data sensing layer according to an embodiment of this application; Figure 3 This is a schematic diagram of a tumor feature processing center according to an embodiment of this application; Figure 4 A schematic diagram of a dynamic efficacy analysis unit provided according to an embodiment of this application; Figure 5 This is a schematic diagram of a dynamic adaptation decision module provided according to an embodiment of this application; Figure 6 This is a schematic diagram of an AI multidisciplinary consultation unit provided according to an embodiment of this application; Figure 7 This is a schematic diagram of a brain tumor treatment efficacy analysis system provided according to an embodiment of this application; Figure 8 This is a flowchart of a method for analyzing the efficacy of brain tumor treatment according to an embodiment of this application; Figure 9 This is a schematic diagram of a brain tumor treatment efficacy analysis method according to an embodiment of this application; Figure 10 This is a schematic diagram of the structure of an electronic device provided according to an embodiment of this application. Detailed Implementation

[0018] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0019] The following description, with reference to the accompanying drawings, illustrates an embodiment of a brain tumor efficacy analysis system. Addressing the low response speed issue mentioned in the background section, this application provides a brain tumor efficacy analysis system. In this system, a tumor data perception layer integrates multidisciplinary, multimodal, and time-series diagnostic and treatment data, laying a comprehensive data foundation for efficacy analysis. A tumor feature processing center leverages an attention mechanism to uncover deep correlations between treatment and changes in tumor features, generating a unified feature map to enhance analytical relevance. A dynamic efficacy analysis unit utilizes a spatiotemporal multi-scale self-supervised model to achieve phased, precise assessment and recurrence warning, quantifying efficacy and risk. A dynamic adaptation decision module combines dynamic changes in the blood-brain barrier with multi-omics features to generate personalized treatment recommendations tailored to the individual. An AI multidisciplinary consultation unit promotes efficient multidisciplinary collaboration through similar case matching, expert opinion integration, and 3D visualization, improving the accuracy, personalization, and collaborative efficiency of brain tumor diagnosis and treatment, effectively reducing the risk of overtreatment or undertreatment. This solves the problems of delayed efficacy evaluation and insufficient treatment recommendations in existing technologies.

[0020] Figure 1 This is a schematic diagram of the structure of a brain tumor efficacy analysis system provided in an embodiment of this application.

[0021] This application provides a brain tumor treatment efficacy analysis system, the system 10 comprising: The system includes a tumor data perception layer (100), a tumor feature processing center (200), a dynamic efficacy analysis unit (300), a dynamic adaptation decision module (400), and an AI multidisciplinary consultation unit (500).

[0022] The system comprises the following components: a tumor data perception layer 100 for collecting multidisciplinary diagnostic and treatment data, multimodal tumor imaging data, and patient pathological monitoring data; a tumor feature processing center 200 for extracting core features from these data, and using an attention mechanism to correlate pre- and post-treatment data with core features to generate a unified feature map of the entire tumor lifecycle; a dynamic efficacy analysis unit 300 for evaluating treatment effectiveness in stages based on the unified feature map of the entire tumor lifecycle, and outputting efficacy index and recurrence risk value through a self-supervised model that integrates the spatiotemporal pyramid Transformer architecture; a dynamic adaptation decision module 400 for generating personalized treatment parameter adjustment suggestions based on dynamic changes in the blood-brain barrier observed in imaging examinations, combined with efficacy index, recurrence risk value, and multi-omics characteristics of patients; and an AI multidisciplinary consultation unit 500 for automatically matching similar cases with opinions from domain experts, conducting multidisciplinary collaborative consultations through a 3D visualization platform, and developing a target treatment plan based on personalized treatment parameter adjustment suggestions.

[0023] It is understood that in this embodiment, the tumor data perception layer integrates multidisciplinary, multimodal, and temporal diagnostic and treatment data to lay a comprehensive data foundation for efficacy analysis; the tumor feature processing center uses an attention mechanism to mine the deep correlation between treatment and changes in tumor features, generating a unified feature map to improve analytical relevance; the efficacy dynamic analysis unit achieves phased and accurate assessment and recurrence warning through a spatiotemporal multi-scale self-supervised model, quantifying efficacy and risk; the dynamic adaptation decision module combines dynamic changes in the blood-brain barrier and multi-omics features to generate personalized treatment suggestions tailored to the individual; the AI ​​multidisciplinary consultation unit promotes efficient multidisciplinary collaboration through similar case matching, expert opinion integration, and 3D visualization collaboration, improving the accuracy, personalization, and collaborative efficiency of brain tumor diagnosis and treatment, and effectively reducing the risk of overtreatment or undertreatment. Thus, it solves the problems of lagging effect evaluation and insufficient treatment suggestions in existing technologies.

[0024] In this embodiment of the application, the tumor data sensing layer 100 includes: Figure 2 As shown, the data acquisition module, multimodal image integration unit, and pathological data monitoring unit are included.

[0025] The data acquisition module is used to collect diagnosis and treatment records, medication plans and clinical examination data from multiple departments including internal medicine, surgery, oncology and radiology; the multimodal image integration unit is used to receive CT, MRI and PET tumor multimodal image data and align and integrate them according to examination timestamps and scanning layers; the pathology data monitoring unit is used to continuously collect postoperative pathological slide analysis results, immunohistochemical indicators and gene detection data of patients, and establish a time-seriesd pathology monitoring archive.

[0026] It is understood that the embodiments of this application comprehensively gather medical records, medication plans, and clinical examination data from internal medicine, surgery, oncology, and radiology through the data acquisition module, centrally integrating core medical information from multiple departments to avoid the limitations of the medical perspective and the omission of key information caused by data dispersion; the multimodal image integration unit accurately receives heterogeneous CT, MRI, and PET image data, and completes spatial-temporal dual-dimensional alignment according to the examination timestamp and scanning level, solving the problems of inconsistent image data formats and difficulties in correlation in traditional imaging, laying the foundation for subsequent cross-modal feature extraction; the pathological data monitoring unit continuously collects postoperative pathological slide analysis results, immunohistochemical indicators, and gene detection data, establishing a time-seriesd pathological monitoring archive to fully track the dynamic evolution of tumor molecular characteristics and pathological status, ensuring the integrity, consistency, and timeliness of the data, and significantly reducing the cost and error of manual integration of multi-source data, providing high-quality and comprehensive data support for subsequent tumor feature processing, dynamic analysis of efficacy, and personalized treatment decisions, and improving the efficiency and accuracy of medical data analysis.

[0027] It should be noted that the continuous collection of postoperative pathological slide analysis results, immunohistochemical indicators, and gene testing data from patients will establish a time-seriesd pathological monitoring archive. The continuously collected postoperative pathological slide analysis results (including tumor cell morphology, differentiation degree, invasion range, and necrosis area ratio), immunohistochemical indicators (such as Ki-67 proliferation index, MGMT methylation status, and PD-L1 expression level), and gene testing data (including driver gene mutations, differentially expressed genes, and signaling pathway activity) will be categorized and organized according to "data type - core indicator - detection method." Furthermore, a precise time anchor will be attached to each data category, indicating the specific collection time (e.g., 1 month postoperatively). The system identifies follow-up points at 1 month, 3 months, and 6 months, and links them to corresponding treatment stages (such as after 2 cycles of chemotherapy or after radiotherapy) to ensure accurate matching of data with the treatment process. Subsequently, the categorized time-series data is structured and stored using standardized file templates. The templates include core fields such as unique patient identifier, data category, collection time, treatment stage, indicator value, reference range, and abnormality annotation. Finally, whenever a patient completes a new pathology-related test, the system automatically adds the new data to the corresponding time node according to the above rules, updating the file content in real time. This ultimately forms a time-series file that covers the entire postoperative follow-up period and clearly traces the dynamic evolution of tumor pathology and molecular characteristics.

[0028] For example, taking the postoperative efficacy monitoring of glioblastoma patients as an example, the multimodal image integration unit first receives the patient's head CT scan 1 week after surgery (to assess surgical residue and bleeding), enhanced MRI 1 month after surgery (to observe the extent of tumor enhancement and the degree of edema), and PET-CT 3 months after surgery (to monitor tumor metabolic activity). Then, it precisely aligns the images in two dimensions: "examination time stamp (1 week / 1 month / 3 months after surgery) + scanning plane (uniform slice thickness of 1mm for axial / coronal / sagittal planes)". This spatially registers the skull structure from CT, the soft tissue details from MRI, and the metabolic hotspots from PET, automatically marking the corresponding areas of the same anatomical location in different images (such as the primary tumor lesion and surrounding edema zone), ultimately generating an integrated "time-space-multimodal" image set. This integration method not only solves the problems of heterogeneous image formats and difficult correlation in traditional multimodal images, but also allows doctors to intuitively compare the dynamic changes in morphology, structure, and metabolism of tumors at different stages, providing intuitive and coherent image support for subsequent extraction of features such as tumor regression rate and edema regression trend, as well as accurate assessment of treatment effects.

[0029] In this embodiment of the application, the tumor feature processing center 200 includes: Figure 3 As shown, the core feature extraction unit, the cross-time series data association module, and the feature map generation engine are all included.

[0030] The core feature extraction unit is used to extract treatment plan features from multidisciplinary diagnosis and treatment data, extract imaging features such as tumor size, enhancement degree, and edema range from multimodal images, and extract Ki-67 index and MGMT methylation status pathological features from pathological data. The cross-temporal data association module is used to mine the temporal dependency relationship between treatment intervention and changes in tumor features by targeting the core correlation points between pre-treatment baseline data and post-treatment follow-up data through a multi-head attention mechanism. The feature map generation engine is used to bind multi-dimensional features with individual patients and construct a unified feature map of the entire tumor cycle that includes time and indicator dimensions.

[0031] Understandably, the embodiments of this application, through a core feature extraction unit, accurately extract key parameters of the treatment plan from multidisciplinary diagnostic and treatment data, capture core imaging features such as tumor size, enhancement degree, and edema range from multimodal images, and screen pathological features closely related to efficacy and recurrence, such as Ki-67 proliferation index and MGMT methylation status, from pathological data, transforming scattered multi-source data into standardized structured analysis elements; the cross-temporal data association module, with the help of a multi-head attention mechanism, focuses on the core correlation points between pre-treatment baseline data and post-operative follow-up data, deeply mining the temporal dependency relationship between treatment intervention and changes in tumor characteristics; the feature map generation engine binds these multi-dimensional features with the patient's unique identifier, constructing a unified feature map of the entire tumor cycle covering time and indicator dimensions, intuitively presenting the dynamic evolution of tumor characteristics with the treatment process, providing correlated and visualized core feature support for subsequent dynamic analysis of efficacy, accurate prediction of recurrence risk, and personalized treatment decisions, significantly improving the systematicness and accuracy of tumor diagnosis and treatment analysis.

[0032] It should be noted that the formula for the multi-head attention mechanism is:

[0033]

[0034]

[0035] Where MultiHead is the final output of the multi-head attention mechanism; Q is the query vector, which corresponds to postoperative tumor feature data; K is the key vector, which corresponds to pre-treatment baseline feature data; V is the value vector, which corresponds to treatment intervention parameter related information; and Concat is the feature concatenation operation. The output results for h attention heads; To output the projection matrix; This is the output of the first attention head; This is the output of the second attention head; This represents the output of the i-th attention head; Attention is the attention calculation function. Let be the query projection matrix of the i-th attention head; Let be the key projection matrix of the i-th attention head; The projection matrix is ​​the value of the i-th attention head; For query vector Q The result after projection; For the key vector K The result after projection; For the value vector V The result after projection; softmax is the normalized activation function; Multiply the transposes of Q and K; Let K be the dimension of the key vector. This is the scaling factor.

[0036] Formula for constructing a unified feature map of the entire tumor cycle:

[0037] in, The Pearson correlation coefficient measures the strength of the linear association between variables X and Y. Let X be the i-th sample value of variable X; Let X be the sample mean of variable X; Let Y be the i-th sample value of variable Y; Let Y be the sample mean of the variable Y; Let x be the square of the difference between the i-th sample value and the mean of variable X; Let be the square of the difference between the i-th sample value and the mean of variable Y; The number of samples; For sample index.

[0038] In this embodiment, the dynamic efficacy analysis unit 300 includes: Figure 4 As shown, the module includes a phased efficacy assessment module, a self-supervised model training unit, and a risk and efficacy quantification engine.

[0039] The phased efficacy assessment module divides the entire treatment cycle into the initial treatment period, the intermediate treatment period, the later treatment period, and the follow-up period, and analyzes the adaptability of tumor characteristic changes and treatment response at each stage. The self-supervised model training unit performs self-supervised training based on unlabeled clinical data. By integrating a self-supervised model with a spatiotemporal pyramid Transformer architecture, it captures the spatiotemporal multi-scale change patterns of tumor characteristics and predicts the evolution trend of tumor status. The risk and efficacy quantification engine is used to generate an efficacy index based on the tumor status evolution trend, integrate tumor shrinkage rate and characteristic stability indicators, and calculate the recurrence risk value by combining historical recurrence data to form a quantitative result.

[0040] It is understood that the embodiments of this application refine the entire treatment cycle into the initial, middle, late and follow-up periods through the phased efficacy evaluation module, accurately analyze the adaptability of tumor characteristic changes and treatment response at each stage, and conduct phased and refined evaluation of efficacy; the self-supervised model training unit relies on unlabeled clinical data and integrates a self-supervised model with a spatiotemporal pyramid Transformer architecture to effectively capture the spatiotemporal multi-scale change patterns of tumor characteristics. Even in scenarios lacking a large amount of labeled data, it can still accurately predict the evolution trend of tumor status, improving the foresight and generalization of prediction; the risk and efficacy quantification engine, based on the evolution trend of tumor status, integrates multiple indicators such as tumor shrinkage rate and feature stability to generate a quantitative efficacy index, and calculates the recurrence risk value by combining historical recurrence data, transforming abstract efficacy and risk into intuitive quantitative results. This not only conducts phased and precise evaluation of efficacy and prediction of tumor trends, but also provides a clear and scientific basis for subsequent treatment decisions through quantitative results, improving the accuracy and foresight of brain tumor efficacy analysis.

[0041] It should be noted that the formula for treatment suitability analysis is as follows:

[0042] in, The fit between tumor characteristic changes and treatment response in stage i (initial / intermediate / late stage / follow-up period); The total number of features participating in the evaluation; For feature index; This represents the actual change value of the j-th tumor feature in the i-th stage; This represents the expected treatment response value corresponding to the j-th feature in the i-th stage; Let be the historical standard deviation of the j-th feature; This represents the normal reference mean of the j-th tumor feature; It is an exponential function.

[0043] Spatiotemporal pyramid Transformer self-supervised model:

[0044]

[0045]

[0046] in, This represents the total loss value of the self-supervised model. These are the weighting coefficients; For spatial multi-scale feature reconstruction loss; Loss is for predicting time-series trends; This represents the number of levels in the spatial pyramid. Spatial scale index; Let the mean square error function be used. The true features of the m-th spatial scale; The m-th layer spatial features reconstructed for the model; For time steps; For time indexing; It is the mean absolute error function; The true temporal characteristics of the (t+1)th time node; The time series features at time point t+1 are predicted based on the features at time point t.

[0047] Therapeutic efficacy index fusion formula:

[0048] in, To quantify the efficacy index; These are the weighting coefficients; This represents the actual rate of tumor shrinkage. This represents the lowest tumor shrinkage rate in history. This represents the largest tumor shrinkage rate in history. As a characteristic stability indicator of tumors; This is the minimum reference value for characteristic stability; This is the maximum reference value for characteristic stability.

[0049] Formula for calculating recurrence risk value:

[0050] in, The posterior probability of recurrence when the patient possesses characteristic F; The conditional probability of the characteristic combination F occurring in historical relapse cases; The prior probability of historical recurrence of brain tumors; Let F be the marginal probability of the characteristic combination F occurring in all cases; This is a recurrence event; This refers to the combination of features corresponding to the evolution trend of tumor status.

[0051] For example, taking postoperative treatment of glioblastoma patients as an example, the phased efficacy assessment module divides the entire treatment cycle into the initial treatment period (1-3 months postoperatively), the intermediate period (4-6 months), the late period (7-9 months), and the follow-up period (1 year or more): The initial period focuses on changes in tumor morphology and edema. Analysis showed that the patient's tumor volume shrank from 4.2 cm to 2.8 cm, the edema area decreased by 40%, and the cell-killing effect of chemotherapy drugs achieved a match of 0.85 (high match); the intermediate period focuses on assessing tumor metabolic and proliferative characteristics, through P... ET-CT revealed a 35% decrease in tumor metabolic activity and a reduction in the Ki-67 index from 32% to 18%, with a fit of 0.78 (medium-high fit) to the expected response of synergistic inhibition of tumor proliferation by radiotherapy and chemotherapy. Later monitoring of tumor characteristic stability showed that the tumor size remained stable at approximately 2.5 cm with no significant increase in enhancement, achieving a fit of 0.91 (high fit) to maintenance therapy. Follow-up monitoring of recurrence-related characteristics showed no tumor progression and stable MGMT methylation status at one year, with a fit of 0.83 to the expected long-term prognosis. This module, through precise matching of tumor characteristic changes and treatment mechanisms at different stages, clearly determines the treatment response at each stage, providing a refined assessment basis for whether to adjust chemotherapy dosage in the mid-term and whether to continue maintenance therapy in the later stages.

[0052] In this embodiment of the application, the dynamic adaptation decision module 400 includes, as follows: Figure 5 As shown, the blood-brain barrier dynamic monitoring unit, multi-dimensional data fusion module, and treatment parameter optimization engine are all included.

[0053] Among them, the blood-brain barrier dynamic monitoring unit is used to monitor the dynamic changes of the blood-brain barrier in real time during treatment by dynamically comparing the Ktrans value and vascular permeability parameters of enhanced MRI; the multi-dimensional data fusion module is used to weightedly fuse blood-brain barrier change data, efficacy index, recurrence risk value with the patient's multi-omics characteristics and clinical basic information; the treatment parameter optimization engine is used to generate personalized treatment parameter adjustment suggestions based on the fusion results, such as drug dosage adjustment, radiotherapy target area correction, and treatment plan switching.

[0054] It is understood that the embodiments of this application, through the blood-brain barrier dynamic monitoring unit, utilize the Ktrans value and vascular permeability parameters of dynamic contrast-enhanced MRI to capture the dynamic changes of the blood-brain barrier in real time during treatment, breaking the limitations of static assessment; the multi-dimensional data fusion module weightedly integrates blood-brain barrier change data with efficacy index, recurrence risk value, patient gene multi-omics characteristics, and clinical basic information, efficiently integrating and mining the value of multi-source heterogeneous data; the treatment parameter optimization engine, based on the fusion results, accurately generates personalized suggestions such as drug dosage adjustment, radiotherapy target area correction, and treatment plan switching, ensuring that treatment decisions are adapted to the dynamic state of the blood-brain barrier and the individual characteristics of the patient, avoiding poor efficacy due to insufficient drug penetration or inappropriate treatment parameters, and improving the targeting, safety, and effectiveness of brain tumor treatment.

[0055] It should be noted that by using dynamic contrast-enhanced MRI to measure Ktrans values ​​and vascular permeability parameters, the dynamic changes of the blood-brain barrier during treatment can be monitored in real time. First, a small molecule contrast agent (such as gadopentetate dimeglumine) is injected intravenously into the patient. Then, the device performs continuous dynamic scanning of the tumor and surrounding brain tissue at high temporal resolution (once every 10-30 seconds) for 3-5 minutes to generate a time-signal intensity curve (TIC). The curve data is then analyzed using a pharmacokinetic model (such as a standard two-compartment model) to calculate key parameters and correlate them with the state of the blood-brain barrier. The Ktrans value (volume transport constant) is a core indicator, representing the rate (unit: min⁻¹) of contrast agent transport from the vascular lumen across the blood-brain barrier into the interstitial space of the brain tissue. Normally, the blood-brain barrier is highly intact, resulting in extremely low Ktrans values ​​(usually <0.05 min⁻¹). During treatment, an increase in Ktrans (e.g., to 0.12 min⁻¹) indicates damage to the barrier structure and increased openness, allowing the drug to penetrate the tumor tissue more easily. A persistently low and unchanged Ktrans value indicates that the barrier is not effectively open, potentially indicating insufficient drug penetration. Simultaneously calculated vascular permeability parameters (such as vascular surface permeability PS and vascular volume Vp) can supplement the assessment: PS reflects the permeability of the vascular wall per unit area, while Vp reflects changes in vascular bed volume. Combining these two values ​​can distinguish whether an increase in Ktrans is due to barrier opening or angiogenesis. By acquiring the temporal changes of parameters through baseline scanning before treatment and repeated scanning during treatment (such as every 2 chemotherapy cycles), the dynamic process of the blood-brain barrier from "intact → partially open → repair" or "continuous closure" can be captured in real time. This breaks through the limitation of traditional single static imaging that can only assess the barrier state at a certain point in time, and provides direct functional evidence for judging the drug penetration effect and adjusting the treatment strategy.

[0056] Pharmacokinetic model:

[0057] in, The permeability coefficient of the blood-brain barrier; The initial drug concentration for the arterial input function; Let τ be the drug concentration in the tumor tissue at time τ; The rate of change of drug concentration in tumor tissue over time; This is the rate constant for drug reflux from tumor tissue into blood vessels; τ represents the drug reflux attenuation term; t is the monitoring time point; τ is the integral variable; It is an integral infinitesimal element.

[0058] Multi-dimensional data weighted fusion formula:

[0059] in, This is a comprehensive decision value derived from the fusion of multi-dimensional data. The total number of data dimensions to be merged; Indexing data dimensions; The weight coefficients for the m-th dimension data; This represents the original data value for the m-th dimension; The historical minimum value of the data in the m-th dimension; The m-th dimension represents the historical maximum value of the data. This is the result of normalization processing for the data in the m-th dimension.

[0060] For example, taking postoperative radiotherapy and chemotherapy in glioblastoma patients as an example, the treatment parameter optimization engine first integrates multi-dimensional data: dynamic monitoring of the blood-brain barrier shows that the Ktrans value increased from 0.04 min⁻¹ to 0.11 min⁻¹ (increased barrier opening), the efficacy index was 68 (moderate efficacy), the recurrence risk value was 0.32 (low recurrence risk), and the patient was MGMT methylation positive with normal liver and kidney function. Based on the fusion results, the engine generates personalized adjustment suggestions: optimizing the temozolomide dose from 75 mg / m² to 90 mg / m² (adapting to barrier opening and improved drug penetration), revising the radiotherapy target area from the initial 3.2 cm × 3.0 cm to 2.8 cm × 2.6 cm (matching the tumor shrinkage trend), and recommending to maintain the combined radiotherapy and chemotherapy regimen (low recurrence risk, no need to switch), which adapts to the patient's real-time physiological status and treatment response, while avoiding insufficient or excessive treatment, precisely improving treatment efficacy and safety.

[0061] In this embodiment of the application, the AI ​​multidisciplinary consultation unit 500 includes, as follows: Figure 6 As shown, the similar case matching module, expert opinion integration unit, and three-dimensional visualization collaboration platform are included.

[0062] The similar case matching module automatically retrieves historical similar cases and prognostic results based on tumor type, stage, treatment plan, and core feature map using a cosine similarity algorithm. The expert opinion integration unit is used to capture and structure the diagnostic and treatment consensus, relevant guidelines and suggestions, and previous consultation opinions of authoritative experts in the field. The three-dimensional visualization collaboration platform uses WebGL to construct a three-dimensional reconstruction model of tumors and brain tissue, enabling doctors from multiple departments to annotate key areas online, exchange diagnostic and treatment opinions in real time, and simultaneously display personalized treatment parameter adjustment suggestions.

[0063] It is understood that the embodiments of this application use a similar case matching module to automatically retrieve historical similar cases and prognostic results for reference based on core information such as tumor type and stage and a cosine similarity algorithm; the expert opinion integration unit captures authoritative treatment consensus, guidelines and previous consultation opinions and integrates them in a structured manner to form standardized evidence; the three-dimensional visualization collaboration platform uses WebGL to construct a three-dimensional reconstruction model of tumor and brain tissue, enabling doctors from multiple departments to annotate key areas online, communicate in real time, and simultaneously display personalized treatment parameter suggestions, which not only provides doctors with case references, authoritative evidence and visualization tools, reducing the bias of diagnosis and treatment decisions, but also improves the efficiency of consultation and the accuracy of collaboration.

[0064] It should be noted that the formula for the cosine similarity algorithm is as follows:

[0065] in, Cosine similarity between the case to be matched and historical cases; The feature vector of the case to be matched; The feature vector of historical cases; For vectors and The dot product; For vectors The L2 norm; For vectors The L2 norm; This represents the total number of feature dimensions. Indexed by feature dimensions; For the first case to be matched Standardized values ​​of 3D features; For the first historical case Standardized values ​​of 3D features; For vectors and Summation of the products of each dimension; For vectors The square root of the sum of squares of all dimensions; For vectors The square root of the sum of squares of all dimensions.

[0066] The expert opinion integration unit leverages natural language processing technology to accurately capture authoritative diagnostic and treatment information from multiple channels. This includes authoritative guidelines related to central nervous system tumors, professional consensus on neuro-oncology, consensus from top domestic neurosurgeons, oncologists, and radiotherapy experts on relevant tumors, and multidisciplinary consultation records of similar tumor cases from hospitals in recent years, covering various diagnostic and treatment opinions, meeting minutes, and explanations of treatment plan adjustments. Subsequently, entity recognition technology is used to extract core entities such as tumor type, gene status, treatment plan, dosage range, and contraindications. Using keyword clustering and semantic association analysis, this unstructured information is standardized and structured: categorized into five core modules: initial treatment plan recommendation, dosage adjustment principles, blood-brain barrier dynamic adaptation suggestions, adverse reaction management, and post-relapse salvage plans. Duplicate statements are removed, evidence levels are labeled, and patient-specific dimensions are associated through feature mapping. Ultimately, a structured opinion package is formed for the target patient, clearly presenting the corresponding recommended treatment plan, dosage adjustment consensus, and relevant contraindications with precise matching information.

[0067] Using WebGL to construct a 3D reconstruction model of tumors and brain tissue, the system first receives multimodal image data. Image preprocessing techniques are used to eliminate scanning noise and standardize the data format. Then, intelligent segmentation algorithms are employed to accurately identify and extract key structures such as tumor tissue, normal brain tissue, cerebral blood vessels, and nerve fiber bundles. Simultaneously, feature labeling of different tissues is performed based on anatomical principles. Subsequently, WebGL performs volumetric data reconstruction and rendering on the segmented 2D image sequence, transforming the planar images into a 3D model. By setting different transparency, color mapping, and lighting effects, the spatial relationship between the tumor and surrounding normal brain tissue and blood vessels is clearly distinguished, accurately restoring the size, shape, boundary features, and infiltration range of the tumor. This makes the originally abstract image information intuitive and perceptible, allowing doctors to rotate, scale, and translate the model by dragging and dropping with a mouse. They can also hide or show specific tissue structures individually, allowing for in-depth observation of the tumor's proximity to important nerves and blood vessels. The platform also supports online synchronous access to the model by doctors from multiple departments, enabling real-time sharing of operational perspectives. This provides a visual foundation for subsequent online annotation of key areas and exchange of diagnostic opinions, reducing the difficulty for doctors in interpreting image data.

[0068] For example, taking a patient with a stage IV glioblastoma in the right frontal lobe, positive for MGMT methylation, an initial tumor size of 4.3 cm, and a blood-brain barrier Ktrans value of 0.05 min⁻¹, who is scheduled for temozolomide combined with radiotherapy, as an example, the similar case matching module first converts the patient's core features, such as tumor type, stage, MGMT methylation status, initial tumor size, blood-brain barrier parameters, and proposed treatment plan, into a standardized feature vector. Then, it uses the cosine similarity algorithm to search the hospital's 5-year historical case database (containing 2000+ glioblastoma cases) and calculates the match between the case to be matched and historical cases. Based on the similarity of feature vectors, three matched cases with a similarity ≥0.92 were ultimately selected. All three cases were MGMT methylation positive, had initial tumor sizes of 4.0-4.5 cm, similar basic blood-brain barrier permeability, and all received the "temozolomide (75 mg / m²) + local radiotherapy" regimen. Two of these cases achieved a tumor shrinkage rate of 58%-62% and an efficacy index ≥85 after 6 months of treatment. In one case, the efficacy significantly improved after adjusting the dose to 90 mg / m² following a blood-brain barrier Ktrans value increase to 0.10 min⁻¹ after 3 months of treatment. All three cases achieved recurrence-free survival for more than 18 months. The module simultaneously compiled the complete treatment history of the three cases (including dose adjustment points, adverse reaction management, and follow-up data), marked them with an "A-level match" rating, and associated them with the characteristics of the patients to be matched. This provided physicians with valuable historical data for confirming the feasibility of the initial treatment plan, predicting efficacy, and planning dose adjustments after dynamic monitoring of the blood-brain barrier, significantly reducing the uncertainty in treatment plan development.

[0069] This application proposes a brain tumor treatment efficacy analysis system. Through a tumor data perception layer, it integrates multidisciplinary, multimodal, and time-series diagnostic and treatment data, laying a comprehensive data foundation for efficacy analysis. The tumor feature processing center utilizes an attention mechanism to uncover the deep correlation between treatment and changes in tumor features, generating a unified feature map to enhance analytical relevance. The efficacy dynamic analysis unit achieves phased, precise assessment and recurrence warning through a spatiotemporal multi-scale self-supervised model, quantifying efficacy and risk. The dynamic adaptation decision module combines dynamic changes in the blood-brain barrier with multi-omics features to generate personalized treatment recommendations tailored to the individual. The AI ​​multidisciplinary consultation unit promotes efficient multidisciplinary collaboration through similar case matching, expert opinion integration, and 3D visualization, improving the accuracy, personalization, and collaborative efficiency of brain tumor diagnosis and treatment, effectively reducing the risk of overtreatment or undertreatment. This solves the problems of lagging efficacy evaluation and insufficient treatment recommendations in existing technologies.

[0070] The following will illustrate a brain tumor treatment efficacy analysis system through a specific embodiment, such as... Figure 7 As shown, it includes: Using a neurosurgery center of a top-tier hospital as an example, this project develops a practical brain tumor efficacy analysis system to address the efficacy evaluation needs of patients with common brain tumors such as glioblastoma and meningioma. The system's hardware deployment employs an "edge computing + cloud collaboration" architecture. The edge computing layer is deployed in the hospital's data center, configured with eight high-performance servers (each equipped with an Intel Xeon Gold 6330 processor, 256GB of RAM, and 4TB of SSD storage) for local data acquisition and real-time processing. The cloud layer utilizes Alibaba Cloud ECS instances, configured with elastic computing nodes and a distributed database for model training, case storage, and multi-center data collaboration. The software is built on a Linux operating system, using a hybrid Python and Java architecture. The front-end uses the Vue.js framework for the interactive interface, while the back-end uses the Spring Boot framework to build a service cluster. Data transmission adopts the HL7FHIR standard protocol to ensure compatibility and security for data exchange across multiple departments. The implementation of the tumor data perception layer focuses on the seamless integration of multi-source data: the data acquisition module achieves real-time connection with the hospital's HIS, LIS, and PACS systems through interface development, automatically synchronizing internal medicine medication records, surgical plans, oncology radiotherapy and chemotherapy plans, and radiology examination reports at 2:00 AM daily, with MD5 encryption algorithm ensuring data integrity during synchronization; the multimodal imaging integration unit uses GESignaArchitect 3.0T MRI, Siemens Somatom Force CT, and Philips Ingenuity TFPET equipment to collect patients' preoperative baseline, data every 2 weeks during treatment, and postoperative data. The follow-up imaging data were sorted by examination timestamp using ITK-SNAP software, and a voxel-based registration algorithm was used to align the scanning planes of different modalities of images. For example, the T1 enhanced sequence of MRI and the plain scan sequence of CT were spatially calibrated, with the error controlled within 0.5 mm. The pathology data monitoring unit was connected to the Leica Aperio AT2 digital slide scanner in the pathology department to collect HE staining images, Ki-67 immunohistochemistry results and IDH1 gene detection data of postoperative pathology slides in real time. A time-series archive was established using a structured data entry tool, and at least 5 pathology monitoring data were recorded for each patient according to the timeline of "pre-treatment - treatment - post-treatment".

[0071] The core of the tumor feature processing center lies in the accuracy of feature extraction and atlas construction. This unit is deployed on an edge computing server, employing a distributed computing architecture to improve processing efficiency. The core feature extraction unit achieves multi-dimensional feature extraction through a customized algorithm package: extracting treatment plan features (such as temozolomide dosage and total radiotherapy dose) and clinical indicator features (such as white blood cell count and liver and kidney function indicators in blood routine tests) from multi-departmental diagnosis and treatment data; extracting 12 imaging features from multimodal images using the U-Net++ segmentation model, including tumor size (expressed as maximum diameter and volume), enhancement degree (calculated by the signal intensity ratio of enhanced scans), and edema extent (segmented using high-signal regions of FLAIR sequences), among which the accuracy of tumor volume measurement was manually verified by three radiologists, with a consistency rate of over 92%; extracting six core pathological features from pathological data, including Ki-67 index (percentage of positive cells), MGMT methylation status (methylation rate ≥10% is positive), and IDH1 mutation type, with data accuracy ensured through double-blind review by pathologists. The cross-temporal data association module uses the TensorFlow framework to build a multi-head attention mechanism model, setting up 8 attention heads to focus on different feature dimensions. For example, the first attention head focuses on the correlation between changes in tumor volume and drug dosage, and the second attention head focuses on the correlation between changes in enhancement intensity and radiotherapy target area. The temporal dependency between treatment intervention and changes in tumor features is mined by calculating self-attention weights. For example, the correlation between the decrease in Ki-67 index and tumor volume reduction 2 weeks after temozolomide administration is identified. The feature map generation engine uses a graph neural network (GNN) to construct a unified feature map of the entire tumor lifecycle. With the patient ID as the root node, it derives time-dimensional branches (preoperative, treatment stage 1, treatment stage 2, treatment stage 3, follow-up) and indicator-dimensional branches (diagnostic features, imaging features, pathological features). Each branch node contains specific feature values ​​and confidence levels. The map is stored in JSON format and supports real-time updates and retrospective queries. For example, the map of a glioblastoma patient can clearly show the dynamic changes from a preoperative tumor volume of 32 cm³ and positive MGMT methylation to a tumor volume shrinking to 28 cm³ and a Ki-67 index decreasing from 60% to 45% after 2 weeks of radiotherapy.

[0072] The efficacy dynamic analysis unit adopts a "local model inference + cloud model update" model. The self-supervised model is pre-trained and deployed on an edge server, and incremental training is performed daily based on new data. The phased efficacy assessment module divides the entire treatment cycle into four phases: the initial treatment phase (weeks 1-2) focuses on treatment tolerance assessment, with a focus on analyzing changes in the extent of edema in imaging and the occurrence of adverse reactions in clinical practice; the mid-treatment phase (weeks 3-8) focuses on efficacy response assessment, with a core focus on monitoring the rate of tumor volume reduction and the degree of improvement in pathological indicators; the late treatment phase (weeks 9-12) focuses on efficacy stabilization assessment, with a focus on determining whether tumor characteristics have entered a plateau phase; and the follow-up phase (every 3 months after the end of treatment) focuses on recurrence monitoring, continuously tracking changes in small lesions. The self-supervised model training unit collected unlabeled data (including multimodal images, medical records, and pathological results) from 8,200 brain tumor patients over the past 5 years at the hospital. The model was trained using a spatiotemporal pyramid Transformer architecture, which consists of three layers: the bottom layer processes high-resolution image features at a single time point, the middle layer integrates feature changes at three adjacent time points, and the top layer integrates the feature evolution patterns throughout the entire cycle. The Transformer module has a 12-layer encoder and uses a multi-head self-attention mechanism to capture long-distance dependencies between features. After 80 rounds of training, the model achieved an accuracy of 89% in predicting treatment efficacy and an AUC of 0.91 in predicting recurrence risk. The risk and efficacy quantification engine constructs a quantitative system based on the model output results: The efficacy index adopts a percentage system, which is calculated by integrating the tumor shrinkage rate (weight 0.4), edema regression rate (weight 0.2), pathological indicator improvement degree (weight 0.3), and clinical tolerability (weight 0.1). ≥80 points is excellent, 60-79 points is good, and <60 points is poor. The recurrence risk value adopts a risk scoring method, which is based on the accumulation of 6 indicators, including tumor type (3 points for glioblastoma, 1 point for meningioma), efficacy index (3 points for <60 points, 2 points for 60-79 points, and 1 point for ≥80 points), and Ki-67 index (3 points for ≥50%, 2 points for 20%-49%, and 1 point for <20%). 0-3 points is low risk, 4-6 points is medium risk, and 7-12 points is high risk. The system automatically generates a daily efficacy analysis report. For example, a patient's mid-treatment report shows an efficacy index of 75 (good) and a relapse risk value of 4 (medium risk), suggesting that the treatment plan needs to be adjusted to improve efficacy.

[0073] The implementation of the dynamic adaptation decision-making module is closely integrated with clinical treatment needs, focusing on achieving personalized and precise adjustments to treatment parameters. The blood-brain barrier dynamic monitoring unit employs dynamic contrast-enhanced MRI (DCE-MRI) technology, using Gd-DTPA as the contrast agent. After intravenous injection at 0.1 mmol / kg body weight, continuous scanning is performed using a fast spin-echo sequence with scanning parameters set to TR=500ms, TE=15ms, and slice thickness of 3mm. Images are acquired every 30 seconds, for a total of 20 time points. The time-signal intensity curve of the tumor region is extracted using the Otsu threshold segmentation algorithm. Vascular permeability parameters such as Ktrans (volume transfer constant) and Ve (extracellular space volume fraction) are calculated using the Tofts dual-compartment model. An increase of more than 20% in the Ktrans value compared to the previous examination is considered an increase in blood-brain barrier permeability. The multi-dimensional data fusion module uses a weighted fusion algorithm to integrate blood-brain barrier change data (weight 0.3), efficacy index (weight 0.25), recurrence risk value (weight 0.25) with patient multi-omics characteristics (weight 0.1) and clinical basic information (weight 0.1). The multi-omics characteristics include sequencing results of 10 tumor-related genes such as IDH1 and TP53, and the clinical basic information covers indicators such as age, gender, and underlying diseases. The weights are jointly determined by experts from neurosurgery, oncology, and radiology through hierarchical analysis. The treatment parameter optimization engine generates three types of personalized suggestions based on the fusion results: drug dosage adjustment uses a linear regression model, calculating the adjustment range based on the efficacy index and blood-brain barrier permeability. For example, when the efficacy index is <60 and the Ktrans value increases, the temozolomide dose can be increased from 150 mg / m² to 200 mg / m²; radiotherapy target area correction uses image registration and segmentation technology, automatically adjusting the target area range according to changes in tumor volume. For example, when the tumor volume shrinks by more than 20%, the target area shrinks synchronously and avoids normal brain tissue; treatment plan switching is based on a decision tree model. When the recurrence risk value is ≥7 and the efficacy does not improve after drug adjustment, it is recommended to switch from drug therapy alone to a combined "drug + stereotactic radiotherapy" regimen. After the suggestions are generated, they must be reviewed by the attending physician. Once approved, they are synchronized to the HIS system to guide clinical treatment. For example, after review, a patient with intermediate-risk recurrence may have their temozolomide dose adjusted from 150 mg / m² to 180 mg / m², and the radiotherapy target area reduced by 3 mm, based on the system's suggestion.

[0074] The AI-powered multidisciplinary consultation unit focuses on breaking down departmental barriers to achieve efficient collaborative diagnosis and treatment. This unit supports both web and mobile access, meeting the needs of consultations across multiple scenarios. The similar case matching module constructs a database containing 12,000 historical cases. Each case includes information such as tumor type, stage, treatment plan, feature map, and prognosis. A cosine similarity algorithm is used for matching: using eight core indicators, including the current patient's tumor type, MGMT methylation status, and efficacy index, as search vectors, the similarity to historical cases is calculated. The top 10 cases with a similarity ≥ 0.85 are selected, simultaneously displaying the treatment process, efficacy changes, and prognosis. For example, for one patient with MGMT methylation-positive glioblastoma, eight similar cases were matched, six of whom received the "temozolomide + radiotherapy" regimen, achieving a 2-year survival rate of 65%. The expert opinion integration unit uses web crawling technology to crawl authoritative materials such as the NCCN guidelines, and employs natural language processing (NLP) technology for structured extraction to establish a knowledge base containing 320 consensus statements on diagnosis and treatment. Simultaneously, it connects to the hospital's expert database, integrating past consultation opinions from 15 chief physicians in neurosurgery, oncology, radiology, and other departments to form a structured expert suggestion database. The 3D visualization collaboration platform is developed based on WebGL technology and uses the VTK library to achieve 3D reconstruction of tumors and brain tissue. The reconstructed data comes from aligned multimodal images, with a resolution of 0.3mm × 0.3mm × 0.5mm. During consultations, doctors from multiple departments log in to the platform with their accounts, and can annotate tumor boundaries, edema areas, and key brain functional areas on the 3D model. Real-time voice communication and text messaging are supported, and the platform simultaneously displays the patient's feature atlas, efficacy analysis results, and personalized treatment suggestions. For example, in a consultation for a patient with a recurrent meningioma, the radiologist marked the positional relationship between the tumor and the optic nerve, the neurosurgeon proposed a surgical resection plan, and the oncologist, based on the system's recommendations, proposed a postoperative radiotherapy plan. Ultimately, a target treatment plan of "surgical resection + postoperative local radiotherapy" was formed. The entire consultation process took approximately 40 minutes, representing a 60% improvement in efficiency compared to traditional in-person consultations. In the year since its implementation, the system has served over 1200 patients, achieving a 90% accuracy rate in efficacy assessment for glioblastoma patients, an average recurrence risk warning time of 3.2 months, and a 55% improvement in multidisciplinary consultation efficiency, fully validating the system's practicality and reliability.

[0075] In summary, this application's embodiments achieve real-time acquisition and secure integration of multi-source data through an edge computing and cloud-based collaborative architecture. The multimodal imaging and pathological data processing module accurately aligns time-series data and extracts core features, while the feature processing center constructs a unified tumor feature map for the entire treatment cycle, significantly improving data integration and utilization efficiency. The efficacy dynamic analysis unit accurately quantifies treatment effects and provides early warnings of recurrence risks using a self-supervised model. The dynamic adaptation decision module generates personalized treatment parameter adjustment suggestions, and the AI ​​multidisciplinary consultation unit breaks down departmental barriers to achieve efficient collaborative diagnosis and treatment, effectively optimizing the treatment process. This enables full-cycle dynamic monitoring, precise efficacy evaluation, and multidisciplinary personalized treatment for brain tumor patients, improving the accuracy of efficacy evaluation and consultation efficiency, ensuring targeted treatment and prognostic quality, and reducing recurrence risk and treatment time.

[0076] Next, referring to the accompanying drawings, a method for analyzing the efficacy of brain tumor treatment according to an embodiment of this application is described.

[0077] like Figure 8 As shown, this method for analyzing the efficacy of brain tumor treatment includes the following steps: In step S101, multidisciplinary diagnosis and treatment data, tumor multimodal imaging data, and patient pathological monitoring data are acquired.

[0078] Among them, tumor multimodal imaging data refers to multi-dimensional imaging information collected by different imaging devices such as MRI, CT, and PET throughout the entire tumor diagnosis and treatment cycle, which is then integrated after registration and alignment and used for tumor feature extraction, efficacy evaluation, and diagnosis and treatment decision support.

[0079] It is understood that the embodiments of this application, by acquiring multimodal tumor imaging data, can integrate complementary information from different imaging devices such as MRI, CT, and PET, and perform consistency calibration on the spatial and temporal aspects of multimodal images. This not only provides comprehensive and three-dimensional data support for extracting core tumor features, accurately capturing subtle changes in key indicators such as tumor size, enhancement degree, edema extent, and metabolic activity throughout the entire period of preoperative baseline, dynamic monitoring during treatment, and postoperative follow-up, but also provides objective quantitative evidence for phased efficacy evaluation and early warning of recurrence risk. Simultaneously, through visualization, it provides intuitive reference for personalized treatment parameter optimization and multidisciplinary collaborative consultation, effectively reducing the limitations and errors of single image evaluation, improving the scientific rigor and accuracy of brain tumor diagnosis and treatment decisions, optimizing treatment plans, reducing recurrence risk, and improving patient prognosis.

[0080] In step S102, the core features of multidisciplinary diagnosis and treatment data, tumor multimodal imaging data and patient pathological monitoring data are extracted. The data and core features before and after treatment are associated through the attention mechanism to generate a unified feature map of the entire tumor cycle.

[0081] Among them, the unified feature map of the entire tumor cycle takes the patient ID as the root node and derives branches from the time dimension from preoperative to follow-up period and the dimensions of diagnosis, treatment, imaging and pathology indicators. It is a structured feature integration carrier that integrates the full cycle feature values ​​and confidence levels and supports real-time updates and retrospective queries, providing core data support for tumor efficacy evaluation and diagnosis and treatment decisions.

[0082] It is understood that the embodiments of this application integrate the core features of multidisciplinary diagnosis and treatment, multimodal tumor imaging, and pathological monitoring. By using the attention mechanism to associate the temporal dependencies between pre- and post-treatment data and various features, the full-cycle feature values ​​and confidence levels are clearly presented in a structured branch. This supports real-time updates and retrospective queries, allowing doctors to intuitively grasp the dynamic changes in key information such as tumor size, pathological indicators, and treatment plans. It also provides data support for phased efficacy quantitative assessment, recurrence risk warning, and personalized treatment parameter optimization, reducing the subjectivity and error of diagnosis and treatment decisions, improving the scientific nature, accuracy, and efficiency of brain tumor diagnosis and treatment, and helping to improve patient treatment prognosis.

[0083] It should be noted that the extraction of core features from multidisciplinary diagnostic and treatment data, tumor multimodal imaging data, and patient pathological monitoring data involves several steps. For structured diagnostic and treatment data, core features are extracted through standardized medical data mapping, redundant information cleaning, and clinical significance screening. For tumor multimodal imaging data, radiomics analysis tools are used to automatically / semi-automatically segment lesion areas. After quantifying and extracting first-order statistical features, shape features, texture features, and functional parameters, redundant information is removed using feature selection algorithms such as correlation analysis and LASSO regression, retaining key features with discriminative power. For patient pathological monitoring data, a digital pathological image analysis system is used to extract visual features such as cell morphology and tissue structure. Simultaneously, standardized results from laboratory test data and structured information from pathology reports are integrated to form a core feature set that combines clinical relevance and data representativeness.

[0084] In step S103, the treatment effect is evaluated in stages based on the unified feature map of the entire tumor cycle. By integrating the self-supervised model of the spatiotemporal pyramid Transformer architecture, the efficacy index and recurrence risk value are output.

[0085] Among them, the self-supervised model that integrates the spatiotemporal pyramid Transformer architecture is a model that combines the multi-scale spatiotemporal feature extraction capability of the spatiotemporal pyramid with the attention mechanism of the Transformer. It learns effective feature representations by generating supervision signals through the data's own structure without the need for manual data labeling.

[0086] It is understood that the embodiments of this application accurately learn effective representations through multi-scale extraction capabilities and attention mechanisms, output efficacy indices and recurrence risk values ​​in stages, reduce annotation costs and errors by eliminating the need for manual data annotation, dynamically capture the evolution of tumors throughout the entire cycle to improve the accuracy of treatment effect evaluation, provide early warning of recurrence risk to provide data support for timely clinical adjustments to the treatment plan, and provide more timely and reliable decision-making basis for precision diagnosis and treatment of tumors.

[0087] For example, taking a patient diagnosed with glioblastoma and positive for MGMT methylation as an example, a self-supervised model integrating the spatiotemporal pyramid Transformer architecture is first trained on thousands of unlabeled clinical data of brain tumors after desensitization in the hospital. Without the need for manual annotation of each case, the model automatically mines potential patterns in the data through self-supervised learning. The model constructs input features based on the patient's multimodal imaging data (CT, MRI, PET), pathological features (Ki-67 index), blood-brain barrier dynamic parameters (Ktrans value), and other multi-source information from the pre-treatment baseline and early / mid-treatment stages, as well as the treatment process. Using the multi-scale feature extraction module of the spatiotemporal pyramid Transformer, it captures features from different levels, including microscopic texture changes, mesoscopic morphological evolution, and macroscopic volume increases and decreases in the tumor. By capturing spatiotemporal correlation patterns and strengthening the temporal dependence between treatment intervention and changes in tumor characteristics through self-attention mechanisms, the model accurately predicted that the patient's tumor would enter a stable shrinkage phase in the later stages of treatment, and that the stability of characteristics would significantly improve after 6 months of treatment, with the risk of recurrence in a low-risk range. It also identified the key node that "the blood-brain barrier opening will reach its peak after 3 months of treatment." Based on these predictions, clinicians developed targeted treatment plans in advance, appropriately adjusting drug dosages at the peak node to optimize penetration. Ultimately, after 8 months of treatment, the patient's tumor shrinkage rate reached 65%, and the efficacy index rose to 88, far exceeding initial expectations. The model not only significantly improved the accuracy and timeliness of efficacy prediction but also provided a scientific basis for the dynamic adjustment of treatment plans, effectively avoiding the problems of blind medication or insufficient dosage.

[0088] In step S104, based on the dynamic changes of the blood-brain barrier observed in imaging examinations, combined with the efficacy index, recurrence risk value, and multi-omics characteristics of patients, personalized treatment parameter adjustment suggestions are generated. At the same time, similar cases and opinions from domain experts are automatically matched, and multi-disciplinary physicians conduct collaborative consultations through a three-dimensional visualization platform. Combined with personalized treatment parameter adjustment suggestions, a target treatment plan is formulated.

[0089] Among them, the multi-omics characteristics of patients are a comprehensive set of characteristics that integrate multiple molecular levels such as genomics, transcriptomics, proteomics, and metabolomics, reflecting the expression, variation and regulation of biomolecules in tumor tissues and the body of brain tumor patients, providing core molecular-level evidence for personalized diagnosis and treatment.

[0090] It is understood that the embodiments of this application, by integrating multi-level molecular information such as core genomic biomarkers, transcriptomic gene expression levels, proteomics expression of proliferation and differentiation proteins, and metabolomics changes in tumor microenvironment metabolites, provide key individual difference support for multi-dimensional data fusion of dynamic changes in the blood-brain barrier, efficacy index, and recurrence risk value. This allows for the prediction of patient sensitivity to relevant drugs through genomic characteristics, and the precise generation of personalized recommendations for drug selection and dosage gradient adjustment based on dynamic blood-brain barrier parameters, avoiding insufficient efficacy or toxic side effects caused by indiscriminate drug use. Furthermore, it can supplement similar cases through matching. By fully utilizing sub-level feature dimensions, the accuracy of case matching is significantly improved, eliminating pseudo-similar cases that are only similar in clinical features, and providing doctors with more valuable historical clinical experience. At the same time, it provides decision-making basis at the molecular mechanism level for multidisciplinary expert consultations, provides clear direction for discussions on targeted therapy and other options, and assists multidisciplinary doctors in forming a consensus on diagnosis and treatment based on clinical features, imaging dynamics, and molecular mechanisms. This ensures that the developed targeted treatment plan is not only adapted to the patient's real-time treatment response and blood-brain barrier status, but also conforms to their intrinsic molecular biological characteristics, thereby improving the accuracy, safety, and long-term efficacy of treatment, and reducing the rate of ineffective treatment and the risk of relapse.

[0091] It should be noted that dynamic changes in the blood-brain barrier refer to the real-time dynamic changes in the permeability and structural integrity of the blood-brain barrier (a natural barrier composed of blood vessel walls and glial cells) under the intervention of drugs, radiotherapy, etc., which are reflected by parameters such as the Ktrans value. These changes directly affect the penetration effect of drugs into tumor tissue.

[0092] According to the embodiments of this application, a method for analyzing the efficacy of brain tumor treatment is proposed. This method integrates multidisciplinary, multimodal, and time-series diagnostic and treatment data through a tumor data perception layer, laying a comprehensive data foundation for efficacy analysis. The tumor feature processing center utilizes an attention mechanism to uncover the deep correlation between treatment and changes in tumor features, generating a unified feature map to enhance analytical relevance. The efficacy dynamic analysis unit achieves phased, precise assessment and recurrence warning through a spatiotemporal multi-scale self-supervised model, quantifying efficacy and risk. The dynamic adaptation decision module combines dynamic changes in the blood-brain barrier with multi-omics features to generate personalized treatment recommendations tailored to the individual. The AI ​​multidisciplinary consultation unit promotes efficient multidisciplinary collaboration through similar case matching, expert opinion integration, and 3D visualization, improving the accuracy, personalization, and collaborative efficiency of brain tumor diagnosis and treatment, and effectively reducing the risk of overtreatment or undertreatment. This solves the problems of lagging efficacy evaluation and insufficient treatment recommendations in existing technologies.

[0093] The following will illustrate a method for analyzing the efficacy of treatment for brain tumors through a specific embodiment, such as... Figure 9 As shown, it includes: This study used patients with gliomas treated at the neuro-oncology center of a tertiary hospital as research subjects. Targeting glioblastoma (GBM), a common and highly recurrent type of brain tumor, a method for analyzing the efficacy of treatment for brain tumors was designed and implemented. A total of 86 newly diagnosed GBM patients were included, and all patients signed informed consent forms. During the data acquisition phase, a multi-system integration approach was used to construct a data collection system, achieving comprehensive coverage and standardized collection of multi-departmental diagnostic and treatment data, multimodal tumor imaging data, and patient pathological monitoring data. Multidisciplinary diagnostic and treatment data are automatically extracted through interfaces of the hospital's electronic medical record (EMR), laboratory information system (LIS), and inpatient management system (HIS). Specifically, this includes surgical records from neurosurgery (including tumor resection range, operation time, and intraoperative blood loss), chemotherapy regimens from oncology (such as temozolomide dosage, cycle, and adverse reaction grade), radiotherapy plans from radiotherapy (target dose distribution, number of irradiations, and normal tissue dose), symptomatic treatment records from neurology (such as intracranial pressure control drug usage), and functional score data from rehabilitation (Karnofsky Performance Status Score, KPS). At the same time, basic patient information (age, gender, and history of underlying diseases) and laboratory test data (complete blood count, liver and kidney function, and tumor markers such as glial fibrillary acidic protein GFAP) are also extracted. Tumor multimodal imaging data were acquired through a Picture Archiving System (PACS) and standardized scans were performed using a 3.0T magnetic resonance imaging (MRI) device. Multi-sequence images were acquired 1 week before treatment, every 3 weeks during treatment, and every month after treatment, including T1-weighted imaging (T1WI), T2-weighted imaging (T2WI), fluid attenuation inversion recovery sequence (FLAIR), dynamic contrast-enhanced MRI (DCE-MRI), and diffusion-weighted imaging (DWI). Pre-treatment PET-CT images were also added for tumor metabolic assessment. All imaging data were exported in DICOM standard format and the resolution was uniformly adjusted to 512×512 pixels. Patient pathological monitoring data encompasses the pathological diagnostic results of preoperative biopsy, intraoperative frozen sections, and postoperative paraffin sections, including tumor pathological grade (WHO IV), histological type (classical, mesenchymal, etc.), and immunohistochemical indicators (Ki-67 proliferation index, MGMT promoter methylation status, IDH1 / 2 mutation status). Simultaneously, cerebrospinal fluid circulating tumor cell (CTC) detection data and circulating tumor DNA (ctDNA) methylation sequencing data are collected every 3 months after treatment to achieve dynamic monitoring at both the pathological and molecular levels. During data collection, data anonymization (removal of patient names, ID numbers, and other identifying information) is employed, and a unique patient ID is generated using MD5 encryption to ensure data compliance. A data quality control mechanism is also established, with two attending physicians verifying the accuracy of the diagnostic data, two radiologists double-blindly reviewing the completeness of the imaging data, and one chief pathologist reviewing the consistency of the pathological data. Unqualified data is fed back to the corresponding department for re-collection through the system, and the data pass rate must reach over 98%.

[0094] Entering the core feature extraction and unified feature map generation stage, a multimodal feature fusion strategy combined with an attention mechanism is adopted to achieve data association and map construction. Differentiated feature extraction schemes are designed for the characteristics of different data types: In multidisciplinary diagnostic and treatment data, discrete data (such as pathological grading and adverse reaction grading) are converted into numerical features using one-hot encoding; continuous data (such as KPS scores and GFAP concentrations) are standardized using Z-scores to eliminate the influence of dimensions; time-series data (such as changes in drug dosage and dynamic values ​​of blood routine tests) are extracted using a sliding window method to extract trend features (such as mean, variance, and rate of change). Finally, 32 core diagnostic and treatment features are selected, including MGMT methylation status, Ki-67 index, KPS score, temozolomide cumulative dose, and GFAP dynamic rate of change. For tumor multimodal imaging data, features are extracted using a combination of deep learning and traditional radiomics. First, the 3DU-Net model is used to automatically segment the tumor region in the images, and two senior radiologists analyze the segmentation results. After correction to ensure a Dice similarity coefficient ≥ 0.92, based on the segmented tumor region, 18 morphological features such as tumor enhancement degree and enhancement uniformity were extracted from T1WI enhanced sequences; 12 texture features such as edema volume and tumor-edema ratio were extracted from T2WI and FLAIR sequences; 8 perfusion parameters such as blood volume (BV) and blood flow (BF) were extracted from DCE-MRI; and 4 diffusion features such as the mean and minimum of apparent diffusion coefficient (ADC) were extracted from DWI, resulting in a total of 44 core imaging features. In the patient pathological monitoring data, in addition to immunohistochemical indicators being directly used as features, ctDNA data were used to obtain molecular features such as mutation frequency and copy number variation of 20 tumor-related genes through targeted sequencing; and 6 phenotypic features such as cell count and survival rate were extracted from CTC data, resulting in a total of 30 core pathological molecular features. After feature extraction, an attention mechanism module was constructed to achieve data association. This module includes two sub-modules: temporal attention and modal attention. The temporal attention sub-module calculates the weight coefficients of features at each time point before and after treatment, highlighting the contribution of key efficacy-related features such as changes in tumor volume, increases in ADC value, and decreases in Ki-67 index. The modal attention sub-module learns the correlation strength between different modal features, strengthening the correlation between cross-modal features such as image perfusion parameters and pathological proliferation index, and the dosage of diagnostic and therapeutic drugs and molecular mutation status. Based on the associated features output by the attention mechanism, feature splicing and dimensionality reduction (reducing 106 features to 20 principal components through principal component analysis) are used to generate a unified feature map of the entire tumor cycle. This map uses the time axis (pre-treatment, induction therapy, consolidation therapy, and follow-up) as the horizontal axis and the 20 principal component features as the vertical axis, visually displaying the dynamic changes of tumor features at each stage in the form of a heatmap.

[0095] Based on a unified feature map of the entire tumor lifecycle, a phased treatment efficacy evaluation was conducted, constructing a self-supervised model integrating a spatiotemporal pyramid Transformer architecture to output efficacy indices and recurrence risk values. First, the phased evaluation criteria were clearly defined: during the induction treatment phase (weeks 1-6), the focus was on evaluating tumor volume reduction rate, edema resolution, and patient tolerability; during the consolidation treatment phase (weeks 7-24), the focus was on evaluating tumor stability, molecular marker seroconversion rate, and long-term adverse reactions; during the follow-up phase (weeks 25-52), the focus was on evaluating recurrence warning indicators and neurological function recovery. In the model construction process, the spatiotemporal pyramid Transformer architecture was divided into a spatial pyramid module and a temporal pyramid module: the spatial pyramid module used three-scale convolutional kernels (3×3, 5×5, 7×7) to extract spatial features from the feature map at multiple scales, capturing spatial information at different levels, such as overall tumor morphological changes, local lesion progression, and microenvironmental alterations; the temporal pyramid module used three time windows (1 week, 4 weeks, 12 weeks) to perform temporal modeling of features at each stage, uncovering temporal patterns such as short-term efficacy fluctuations, mid-term efficacy stabilization trends, and long-term recurrence risk trends. The pre-training phase of the self-supervised model used unlabeled historical case data (data from 500 GBM patients) for unsupervised learning. A feature reconstruction task (predicting features for the next stage based on features from the previous stage) and a trend classification task (determining whether a feature sequence is effective, stable, or progressive) were designed to enable the model to automatically learn the spatiotemporal characteristics of brain tumor treatment. After pre-training, 60 cases from the 86 patients in this study (training set) were used for fine-tuning, and the remaining 26 cases were used as a validation set to optimize model parameters. The efficacy index was calculated using a weighted scoring method, summing the key indicators in the feature maps of each stage (e.g., tumor volume reduction rate ≥30% = 10 points, ADC value increase ≥20% = 8 points, Ki-67 index decrease ≥15% = 12 points, etc.) according to their weight coefficients, ultimately outputting an efficacy index of 0-100 points, where ≥80 points indicate excellent efficacy, 60-79 points indicate effective efficacy, 40-59 points indicate stable efficacy, and <40 points indicate ineffective efficacy. The recurrence risk value is represented by a probability value (0-1.0) output by the model. Combined with the weights of recurrence warning features in the feature map (such as increased ctDNA mutation frequency, abnormally decreased ADC value, and abnormal tumor margin enhancement), a logistic regression model is used to calculate the risk value, where ≥0.7 is high risk, 0.3-0.6 is medium risk, and <0.3 is low risk. Model validation results show that the consistency between the efficacy index assessment and the comprehensive assessment by clinicians reaches 0.85 (Kappa value), the AUC value for predicting 1-year recurrence by the recurrence risk value is 0.89, the sensitivity is 0.87, and the specificity is 0.83, meeting the needs of clinical application.

[0096] Personalized treatment parameters are adjusted by combining dynamic changes in the blood-brain barrier (BBB) ​​with imaging examinations, and a targeted treatment plan is formulated through multidisciplinary collaborative consultation. Dynamic monitoring of the BBB is quantitatively assessed using dynamic enhancement curves and permeability parameters (such as surface permeability PS values) from DCE-MRI. Measurements are performed before treatment, every 3 weeks during treatment, and monthly after treatment. By comparing changes in PS values ​​at different stages, the dynamic changes in BBB integrity are determined: a PS value increase of ≥30% indicates increased BBB permeability and improved drug delivery efficiency; a PS value decrease of ≥20% indicates BBB repair and decreased drug delivery efficiency. The generation of personalized treatment parameter adjustment recommendations adopts a multi-factor decision model. The input parameters include the rate of change of blood-brain barrier PS value, efficacy index, relapse risk value, and patient multi-omics characteristics (such as MGMT methylation status and IDH mutation type): For patients with MGMT methylation positive, efficacy index ≥80, low relapse risk, and elevated PS value, it is recommended to maintain the original temozolomide dose and extend the consolidation treatment period to 30 weeks; for patients with MGMT methylation negative, efficacy index 60-79, medium relapse risk, and decreased PS value, it is recommended to increase the temozolomide dose by 10% and combine it with bevacizumab targeted therapy; for patients with efficacy index <40, high relapse risk, and significantly elevated PS value, it is recommended to terminate the current chemotherapy regimen and revise the radiotherapy combined with immunotherapy regimen. The similar case matching system is based on cosine similarity calculation of case feature vectors. These feature vectors include 28 dimensions such as tumor pathological type, molecular markers, treatment plan, efficacy index, and recurrence risk value. The matching database contains complete case data of 1200 GBM patients from our hospital over the past 5 years, along with corresponding expert opinions (3 chief physicians each from neurosurgery, oncology, radiotherapy, radiology, and pathology). The matching threshold is set to 0.85, and each time 3-5 cases with the highest similarity are output along with expert comments. The 3D visualization platform is built using the VTK engine, supporting the reconstruction of multimodal image data into 3D tumor models, overlaying and displaying changes in tumor volume, blood-brain barrier permeability distribution, and dynamic curves of the efficacy index. It also features multi-user online collaborative annotation capabilities. The multidisciplinary collaborative consultation process is as follows: The consultation is initiated by the attending physician of the neuro-oncology center. The system automatically pushes the patient's characteristic atlas, efficacy index, recurrence risk value, blood-brain barrier monitoring results and similar case data to the terminals of experts in each department. Experts mark key areas of the tumor and put forward treatment adjustment opinions through the 3D visualization platform. The platform records all opinions and forms a summary report. Finally, the consultation chair (chief physician of neurosurgery) leads the discussion, and combined with personalized treatment parameter adjustment suggestions, comprehensively determines the target treatment plan. The entire consultation process does not exceed 48 hours.

[0097] In summary, this application's embodiments, through a multi-system integrated data acquisition system, achieve comprehensive and standardized coverage of multi-dimensional data in diagnosis, imaging, and pathology. Strict data anonymization and quality control mechanisms ensure data compliance and accuracy, laying a solid foundation for subsequent analysis. Based on a unified feature map generated using multimodal feature fusion and attention mechanisms, the dynamic changes of tumors at different treatment stages are clearly presented. Combined with the spatiotemporal pyramid Transformer self-supervised model, accurate assessment of efficacy index and recurrence risk is achieved. The consistency of assessment and recurrence prediction efficacy both meet clinical application standards, effectively solving the problems of strong subjectivity and insufficient prediction accuracy in traditional assessment methods. Personalized treatment parameter adjustment suggestions based on dynamic blood-brain barrier monitoring results and a multi-factor decision-making model can flexibly optimize treatment plans according to individual patient characteristics, significantly improving the individualized fit of treatment. Simultaneously, the combination of a similar case matching system, a 3D visualization platform, and a multi-departmental collaborative consultation process not only shortens the diagnosis and treatment decision-making cycle but also integrates the experience of experts from multiple fields, effectively reducing the high risk of recurrence and improving the quality of patient prognosis management.

[0098] Figure 10 A schematic diagram of the structure of an electronic device provided in an embodiment of this application. The electronic device may include: The memory 1001, the processor 1002, and the computer program stored on the memory 1001 and capable of running on the processor 1002.

[0099] When the processor 1002 executes the program, it implements a brain tumor efficacy analysis method provided in the above embodiments.

[0100] Furthermore, electronic devices also include: Communication interface 1003 is used for communication between memory 1001 and processor 1002.

[0101] The memory 1001 is used to store computer programs that can run on the processor 1002.

[0102] The memory 1001 may include high-speed RAM (Random Access Memory) memory, and may also include non-volatile memory, such as at least one disk storage.

[0103] If the memory 1001, processor 1002, and communication interface 1003 are implemented independently, then the communication interface 1003, memory 1001, and processor 1002 can be interconnected via a bus to complete communication between them. The bus can be an ISA (Industry Standard Architecture) bus, a PCI (Peripheral Component Interconnect) bus, or an EISA (Extended Industry Standard Architecture) bus, etc. The bus can be divided into address bus, data bus, control bus, etc. For ease of representation, Figure 10 The bus is represented by a single thick line, but this does not mean that there is only one bus or one type of bus.

[0104] Optionally, in a specific implementation, if the memory 1001, processor 1002, and communication interface 1003 are integrated on a single chip, then the memory 1001, processor 1002, and communication interface 1003 can communicate with each other through an internal interface.

[0105] The processor 1002 may be a CPU (Central Processing Unit), an ASIC (Application Specific Integrated Circuit), or one or more integrated circuits configured to implement the embodiments of this application.

[0106] This application also provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the above-described method for analyzing the efficacy of brain tumor treatment.

[0107] Furthermore, this application also provides a computer program product, including a computer program or instructions, which, when executed, implement the aforementioned method for analyzing the efficacy of brain tumor treatment.

[0108] In the description of this specification, the references to "one embodiment," "some embodiments," "example," "specific example," or "some examples," etc., indicate that a specific feature, structure, material, or characteristic described in connection with that embodiment or example is included in at least one embodiment or example of this application. In this specification, the illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples. Moreover, without contradiction, those skilled in the art can combine and integrate the different embodiments or examples described in this specification, as well as the features of different embodiments or examples.

[0109] Furthermore, the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of technical features indicated. Thus, a feature defined as "first" or "second" may explicitly or implicitly include at least one of that feature. In the description of this application, "multiple" means at least two, such as two, three, etc., unless otherwise explicitly specified.

[0110] Any process or method description in the flowchart or otherwise herein can be understood as representing a module, segment, or portion of code comprising one or more executable instructions for implementing custom logic functions or processes, and the scope of the preferred embodiments of this application includes additional implementations in which functions may be performed not in the order shown or discussed, including substantially simultaneously or in reverse order depending on the functions involved, as should be understood by those skilled in the art to which embodiments of this application pertain.

[0111] It should be understood that various parts of this application can be implemented using hardware, software, firmware, or a combination thereof. In the above embodiments, multiple steps or methods can be implemented using software or firmware stored in memory and executed by a suitable instruction execution system. For example, if implemented in hardware as in another embodiment, it can be implemented using any of the following techniques known in the art, or a combination thereof: discrete logic circuits having logic gates for implementing logical functions on data signals, application-specific integrated circuits (ASICs) having suitable combinational logic gates, programmable gate arrays (PGAs), field-programmable gate arrays (FPGAs), etc.

[0112] Those skilled in the art will understand that all or part of the steps of the methods described in the above embodiments can be implemented by a program instructing related hardware. The program can be stored in a computer-readable storage medium, and when executed, the program includes one or a combination of the steps of the method embodiments.

[0113] Although embodiments of this application have been shown and described above, it is understood that the above embodiments are exemplary and should not be construed as limiting this application. Those skilled in the art can make changes, modifications, substitutions and variations to the above embodiments within the scope of this application.

Claims

1. A brain tumor treatment efficacy analysis system, characterized in that, include: The system includes a tumor data perception layer, a tumor feature processing center, a dynamic efficacy analysis unit, a dynamic adaptation decision-making module, and an AI multidisciplinary consultation unit. The tumor data sensing layer is used to collect multidisciplinary diagnosis and treatment data, tumor multimodal imaging data, and patient pathological monitoring data. The tumor feature processing center is used to extract the core features of the multidisciplinary diagnosis and treatment data, tumor multimodal imaging data and patient pathological monitoring data, and to generate a unified feature map of the entire tumor cycle by associating the data before and after treatment with the core features through an attention mechanism. The efficacy dynamic analysis unit is used to evaluate the treatment effect in stages based on the unified feature map of the entire tumor cycle, and output the efficacy index and recurrence risk value by integrating the self-supervised model of the spatiotemporal pyramid Transformer architecture. The dynamic adaptation decision module generates personalized treatment parameter adjustment suggestions based on the dynamic changes of the blood-brain barrier observed in imaging examinations, combined with the efficacy index, recurrence risk value, and multi-omics characteristics of patients. The AI ​​multidisciplinary consultation unit is used to automatically match similar cases with opinions from experts in the field, conduct collaborative consultations among doctors from multiple departments through a 3D visualization platform, and formulate a target treatment plan based on the personalized treatment parameter adjustment suggestions.

2. The brain tumor treatment efficacy analysis system according to claim 1, characterized in that, The tumor data sensing layer includes a data acquisition module, a multimodal image integration unit, and a pathological data monitoring unit. The data acquisition module is used to collect diagnosis and treatment records, medication plans, and clinical examination data from multiple departments, including internal medicine, surgery, oncology, and radiology. The multimodal image integration unit is used to receive CT, MRI, and PET tumor multimodal image data and align and integrate them according to examination timestamps and scanning layers. The pathological data monitoring unit is used to continuously collect postoperative pathological slide analysis results, immunohistochemical indicators, and gene detection data from patients to establish a time-seriesd pathological monitoring archive.

3. The brain tumor treatment efficacy analysis system according to claim 1, characterized in that, The tumor feature processing center includes a core feature extraction unit, a cross-temporal data association module, and a feature map generation engine. The core feature extraction unit extracts treatment plan features from multidisciplinary diagnostic data, tumor size, enhancement level, and edema extent from multimodal images, and Ki-67 index and MGMT methylation status from pathological data. The cross-temporal data association module uses a multi-head attention mechanism to identify the temporal dependency between treatment intervention and changes in tumor features, focusing on key correlation points between pre-treatment baseline data and post-treatment follow-up data. The feature map generation engine binds multi-dimensional features to individual patients, constructing a unified tumor feature map encompassing time and indicator dimensions throughout the entire lifecycle.

4. The brain tumor treatment efficacy analysis system according to claim 1, characterized in that, The efficacy dynamic analysis unit includes a phased efficacy assessment module, a self-supervised model training unit, and a risk and efficacy quantification engine. The phased efficacy assessment module divides the entire treatment cycle into initial treatment, intermediate treatment, late treatment, and follow-up periods, analyzing the adaptability of tumor characteristic changes and treatment response at each stage. The self-supervised model training unit performs self-supervised training based on unlabeled clinical data, capturing the spatiotemporal multi-scale changes in tumor characteristics and predicting tumor state evolution trends through a self-supervised model that integrates a spatiotemporal pyramid Transformer architecture. The risk and efficacy quantification engine generates an efficacy index based on the tumor state evolution trend, integrating tumor shrinkage rate and characteristic stability indicators, and calculates recurrence risk values ​​by combining historical recurrence data to form a quantitative result.

5. The brain tumor treatment efficacy analysis system according to claim 1, characterized in that, The dynamic adaptation decision module includes a blood-brain barrier dynamic monitoring unit and a treatment parameter optimization engine. The blood-brain barrier dynamic monitoring unit is used to monitor the dynamic changes of the blood-brain barrier in real time during treatment by dynamically comparing the Ktrans value and vascular permeability parameters of enhanced MRI. The multi-dimensional data fusion module is used to weightedly fuse blood-brain barrier change data, efficacy index, recurrence risk value, and patient's multi-omics genetic characteristics and basic clinical information. The treatment parameter optimization engine is used to generate personalized treatment parameter adjustment suggestions based on the fusion results, including drug dosage adjustment, radiotherapy target area correction, and treatment plan switching.

6. The brain tumor treatment efficacy analysis system according to claim 1, characterized in that, The AI-powered multidisciplinary consultation unit includes a similar case matching module, an expert opinion integration unit, and a 3D visualization collaboration platform. The similar case matching module automatically retrieves historical similar cases and prognostic results based on tumor type, stage, treatment plan, and core feature maps using a cosine similarity algorithm. The expert opinion integration unit captures and integrates the diagnostic and treatment consensus, relevant guidelines, and previous consultation opinions from authoritative experts in the field. The 3D visualization collaboration platform utilizes WebGL to construct a 3D reconstruction model of the tumor and brain tissue, enabling doctors from multiple departments to annotate key areas online, exchange diagnostic and treatment opinions in real time, and simultaneously display personalized treatment parameter adjustment suggestions.

7. A method for applying a brain tumor efficacy analysis system according to any one of claims 1-6, characterized in that, The method includes: Acquire multidisciplinary diagnostic and treatment data, tumor multimodal imaging data, and patient pathological monitoring data; The core features of the multidisciplinary diagnosis and treatment data, tumor multimodal imaging data and patient pathological monitoring data are extracted, and the data before and after treatment and the core features are associated through the attention mechanism to generate a unified feature map of the entire tumor cycle. Based on the unified feature map of the entire tumor cycle, the treatment effect is evaluated in stages. By integrating the self-supervised model of the spatiotemporal pyramid Transformer architecture, the efficacy index and recurrence risk value are output. Based on the dynamic changes of the blood-brain barrier observed in imaging examinations, combined with the efficacy index, recurrence risk value, and multi-omics characteristics of patients, personalized treatment parameter adjustment suggestions are generated. At the same time, similar cases and opinions from experts in the field are automatically matched, and multi-disciplinary physicians conduct collaborative consultations through a three-dimensional visualization platform. Based on the personalized treatment parameter adjustment suggestions, a target treatment plan is formulated.

8. An electronic device, characterized in that, It includes a memory, a processor, and a computer program stored in the memory and executable on the processor. The processor executes the program to implement the brain tumor efficacy analysis method as described in claim 7.

9. A computer-readable storage medium having a computer program or instructions stored thereon, characterized in that, When a computer program or instruction is executed, it implements the brain tumor efficacy analysis method as described in claim 7.

10. A computer program product, comprising a computer program or instructions, characterized in that, When a computer program or instruction is executed, it implements the brain tumor efficacy analysis method as described in claim 7.

Images