Cancer patient prognosis prediction system based on multi-modal model

Abstract

According to the cancer patient prognosis prediction system based on the multi-modal model, multi-modal data of blood vessels, senescence and genomes are synchronously collected, a mathematical model is constructed to quantify time sequence characteristics of a blood vessel normalization index VNI and a senescence burden score SBS, and a cross-channel attention mechanism is combined to align a blood vessel function peak value and a senescence inflection point, so that the prognosis of a cancer patient is predicted. And the limitation of traditional single data prediction is solved. The dynamic monitoring characteristic of the system can capture the physiological index change of the whole treatment cycle, the optimal treatment window period is accurately identified through the TOW (t) state value, the senescence threshold value is individually adjusted based on the p53 gene mutation state, the prediction fits the individual difference of patients, and the accuracy and timeliness of prognosis prediction are improved through the system.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of cancer survival analysis and prognosis prediction, and in particular to a cancer patient prognosis prediction system based on a multi-modal model. BACKGROUND

[0002] In the field of cancer clinical treatment, patient prognosis prediction is a core link for guiding personalized treatment decisions and improving treatment effectiveness. The progress of deep learning in cancer pathology enables digital pathology images to be applied to the diagnosis and prognosis prediction of various cancer types.

[0003] Traditional cancer patient prognosis prediction methods generally rely on a single type of medical data, such as only based on genomic sequencing data or isolated imaging indicators, which has significant technical limitations: on the one hand, a single data dimension cannot comprehensively reflect the dynamic correlation of tumor microenvironment-patient physiological state-genetic background during cancer treatment. For example, only through genomic data can tumor mutation targets be identified, but real-time capture of treatment-induced changes in tumor vascular function, such as vascular permeability and blood supply capacity, is not possible, as vascular function directly determines whether chemotherapy / targeted drugs can effectively reach the lesion. Relying solely on imaging data (such as tumor size) makes it difficult to quantify the patient's overall burden of aging (such as accumulation of senescent cells in the tumor microenvironment and inflammation levels), resulting in an inability to early warn of the risk of adverse reactions caused by decreased treatment tolerance. On the other hand, traditional methods mostly use static analysis mode, and only based on data before treatment or at a fixed time point for prediction, which cannot adapt to the dynamics of cancer treatment.

[0004] Based on this, the present application provides a cancer patient prognosis prediction system based on a multi-modal model. SUMMARY

[0005] The present application provides a cancer patient prognosis prediction system based on a multi-modal model to solve the above technical problems.

[0006] The technical solutions adopted by the present application are as follows:

[0007] A cancer patient prognosis prediction system based on a multi-modal model, comprising:

[0008] An acquisition module for synchronously acquiring multi-modal data at preset time points during the treatment of a cancer patient, the multi-modal data including vascular-related data, senescence-related data, and genomic data;

[0009] A generation module for generating a vascular normalization index VNI time series based on the vascular-related data; generating a senescence burden score SBS time series based on the senescence-related data;

[0010] a coupling module, configured to input the vascular normalization index VNI time sequence and the senescence burden score SBS time sequence into a dual-channel time sequence coupling model, align a vascular function peak value and a senescence inflection point through a cross-channel attention mechanism, and output a cross-modality coupling feature;

[0011] a calculation module, configured to calculate a treatment opportunity window state value TOW(t) according to the cross-modality coupling feature: when the vascular normalization index VNI(t) is greater than a vascular threshold and the senescence burden score SBS(t) is less than a senescence threshold, TOW(t) = 1, otherwise TOW(t) = 0;

[0012] a determination output module, configured to trigger a permanent drug resistance state determination when a number of consecutive TOW(t) = 0 reaches a drug resistance early warning threshold, generate a TOW time sequence decision tree based on a TOW(t) state sequence, mark a window period of TOW(t) = 1 and a corresponding recommended treatment operation on a time axis, and output a permanent drug resistance risk probability and a treatment strategy adjustment suggestion.

[0013] Further, the blood vessel related data includes a Ktrans value of a tumor region acquired by dynamic contrast-enhanced MRI and a vascular endothelial growth factor VEGF concentration acquired by serum detection;

[0014] The senescence related data includes a senescent cell density acquired by SA-β-Gal staining of tumor biopsy tissue and an interleukin-6 concentration acquired by plasma sample detection.

[0015] The genomic data is a p53 gene mutation state of tumor tissue.

[0016] Further, a calculation formula of the vascular normalization index VNI time sequence is:

[0017]

[0018] In the formula, VNI(t) is the vascular normalization index at time t; t is a time point after treatment starts; Ktrans(t) is a Ktrans peak value of the tumor region at time t; is a vascular decay constant; is a time decay term;

[0019] A calculation formula of the senescence burden score SBS time sequence is:

[0020]

[0021] In the formula, SBS(t) is the senescence burden score at time t; is a weight coefficient of the normalized senescent cell density; is a weight coefficient of the plasma IL-6 concentration; is the normalized tumor tissue senescent cell density at time t; is the plasma interleukin-6 concentration at time t.

[0022] Further, the normalized senescent cell density SAβGal(t) is obtained by the following steps:

[0023] Counting the positive cells per unit area under high-power microscope on the SA-β-Gal stained section to obtain the original senescent cell density;

[0024] The original senescent cell density is subjected to Z-score standardization processing.

[0025] Further, the cross-channel attention mechanism is realized by the following steps:

[0026] Calculate the cosine similarity between the feature vector of the VNI time sequence at time point t and the feature vector of the SBS time sequence at time point t:

[0027]

[0028] In the formula, is the feature vector of the VNI time sequence at time t; is the feature vector of the SBS time sequence at time t;

[0029] Generate the attention weight at time point t:

[0030]

[0031] In the formula, is the attention weight at time t; is the natural exponential function;

[0032] Output the cross-modal coupling feature:

[0033]

[0034] In the formula, denotes the vector splicing operation, is the total number of time points.

[0035] Further, the calculation formula of the permanent drug resistance risk probability is:

[0036]

[0037] In the formula, is a sigmoid function, ; ​​​is the cumulative slope of the SBS time series; is the number of consecutive TOWs (t) = 0; is the weight coefficient of the cumulative slope of the SBS; is the weight coefficient of the number of consecutive TOWs = 0.

[0038] Further, the drug resistance early warning threshold is 3 consecutive ;

[0039] When the cumulative slope of the SBS time series > 0.05 / day, a high drug resistance risk early warning is triggered.

[0040] Advantages of the present application:

[0041] The present application synchronously collects vascular, aging, and genomic multi-modal data, constructs a mathematical model to quantify the time series characteristics of the vascular normalization index VNI and the aging burden score SBS, aligns the vascular function peak and the aging inflection point through a cross-channel attention mechanism, solves the limitations of traditional single data prediction, dynamically monitors physiological indicators throughout the treatment cycle, accurately identifies the optimal treatment window period through the TOW(t) state value, and individualizes the aging threshold based on the p53 gene mutation state, making the prediction fit individual differences, and improving the accuracy and timeliness of prognosis prediction through the system. BRIEF DESCRIPTION OF DRAWINGS

[0042] Figure 1 is a connection block diagram of a cancer patient prognosis prediction system based on a multi-modal model of an embodiment of the present application;

[0043] Figure 2 is a data acquisition flowchart of an embodiment of the present application. DETAILED DESCRIPTION

[0044] The technical solutions in the embodiments of the present application will be described clearly and completely below with reference to the accompanying drawings in the embodiments of the present application. Obviously, the described embodiments are only a part of the embodiments of the present application, rather than all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those of ordinary skill in the art without creative labor fall within the scope of protection of the present application.

[0045] As Figures 1-2 shown, a cancer patient prognosis prediction system based on a multi-modal model of an embodiment of the present application includes an acquisition module, a generation module, a coupling module, a calculation module, and a determination and output module.

[0046] The acquisition module is used to synchronously acquire multi-modal data at a preset time point during the treatment of a cancer patient, and the multi-modal data includes vascular related data, aging related data, and genomic data.

[0047] In an embodiment of the present application, the blood vessel related data includes Ktrans value of tumor region acquired by dynamic contrast-enhanced MRI and vascular endothelial growth factor (VEGF) concentration acquired by serum detection; specifically, the Ktrans value of tumor region is the blood vessel permeability transfer constant, reflecting the rate of contrast agent permeating from blood vessels to interstitial space, with the unit of min-1, which is acquired by collecting the tumor lesion and the surrounding normal tissue of the patient, using a 3.0T or above MRI device such as Siemens Prisma 3.0T, and continuously scanning for 5 minutes after injecting contrast agent, with 1 set of images acquired every 10 seconds; the tumor region ROI (region of interest) needs to be independently outlined by 2 imaging department attending physicians, and the intersection region is calculated for Ktrans value, and if the deviation exceeds 10%, it needs to be outlined again. In addition, the Ktrans value of the normal tissue around the tumor, such as normal lung tissue / liver tissue 2 cm away from the tumor edge, needs to be synchronously collected for subsequent background correction. The VEGF concentration reflects the cytokine of angiogenesis activity, and the higher the concentration, the more active the tumor angiogenesis, with the unit of pg / mL, which is acquired by collecting 5 mL of peripheral venous blood of the patient.

[0048] The senescence related data includes senescent cell density acquired by SA-β-Gal staining of tumor biopsy tissue and interleukin-6 concentration acquired by detection of plasma samples; specifically, the senescent cell density refers to the number of stained senescent cells per unit area, with the original unit of mm², which is acquired by collecting the tumor biopsy tissue of the patient. The interleukin-6 concentration is a cytokine reflecting the systemic inflammatory response, and the higher the concentration, the more severe the inflammation level, with the unit of pg / mL, which is acquired by collecting 5 mL of peripheral venous blood of the patient.

[0049] The genomic data is the p53 gene mutation status of tumor tissue. The p53 gene mutation status of tumor tissue is a key genetic marker for judging the sensitivity of drug resistance risk of the patient, which is acquired by collecting tumor biopsy tissue / surgically resected tissue (DNA content ≥ 50 ng) of the patient, and the tissue sample needs to contain ≥ 30% of tumor cells.

[0050] The generating module is used to generate a blood vessel normalization index (VNI) time sequence based on the blood vessel related data, and generate a senescence burden score (SBS) time sequence based on the senescence related data. In this way, discrete original detection values are converted into quantitative indicators with time sequence continuity and clear clinical significance.

[0051] It should be noted that VNI focuses on tumor vascular normalization; the better the vascular function, the easier it is for drugs to reach the lesion. SBS focuses on the patient's aging burden; the lighter the aging burden, the stronger the treatment tolerance. The VNI and SBS time series of the same patient must strictly correspond to the preset time points in the acquisition module, such as t=0, 3, 7, and 14 days. One VNI value and one SBS value are generated at each time point, forming two time series curves aligned with the time axis, avoiding subsequent cross-modal coupling deviations due to time misalignment.

[0052] In one embodiment of the present invention, the VNI time series is a set of indicators quantifying the degree of tumor vascular normalization at time t. The larger the VNI(t) value at each time point, the closer the tumor vascular permeability and blood supply capacity are to normal tissue at that time, which is the core basis for judging whether the treatment is effective. The formula for calculating the vascular normalization index VNI time series is as follows:

[0053]

[0054] In the formula, The vascular normalization index at time t; This is the time point after the start of treatment, completely consistent with the preset data collection time point in the generation module, in days; The peak value of Ktrans in the tumor region at time t is obtained by taking the maximum value of Ktrans from 5 minutes of continuous MRI scans at time t. The unit is min⁻¹, which reflects the peak level of tumor vascular permeability at time t. The larger the blood vessel, the greater its ability to allow drugs to penetrate into the interstitial space; the unit is usually 0.1-2.0 min⁻¹. VEGF(t) represents the concentration of vascular endothelial growth factor in serum at time t, which reflects the activity of tumor angiogenesis at time t. The lower the VEGF(t), the weaker the angiogenic activity and the higher the potential for vascular normalization. The unit is usually 50-1500 pg / mL. This is the vascular attenuation constant; The term represents the time decay, which decreases monotonically with increasing treatment time t. This reflects the physiological law that vascular function naturally decays with the treatment cycle, ensuring that VN(t) truly reflects the treatment intervention effect rather than the natural influence of time.

[0055] For example, a lung cancer patient ( Taking the preset acquisition time points t=0, 3, 7, and 14 days as an example, the steps for generating the VNI time series are as follows:

[0056] Retrieve data from the database in the data acquisition module for each time point. and As shown in the table below:

[0057] T (days) (min"1) (pg / mL) 0 0.65 120 3 0.92 95 7 1.15 70 14 0.88 85

[0058] Calculate VNI(t) point by point:

[0059]

[0060]

[0061]

[0062]

[0063] Arrange the VNI(t) at each time point in chronological order to obtain the VNI time series sequence: .

[0064] In one embodiment of the present invention, the SBS time series is a set of indicators quantifying the overall aging burden of a patient at time t, integrating the density of senescent cells in the tumor and the level of systemic inflammation. The smaller the SBS(t) value at each time point, the lighter the aging burden of the patient and the stronger the tolerance to treatment. Specifically, the formula for calculating the aging burden score SBS time series is as follows:

[0065]

[0066] in, The aging burden score at time t; Weighting coefficients for standardizing the density of senescent cells; The weighting coefficient for plasma IL-6 concentration; Let be the senescent cell density of tumor tissue at time t. The standardized value range is usually -2 to 2, where a negative value indicates that the senescent cell density is lower than the historical average, and a positive value indicates that the senescent cell density is higher than the historical average. IL6(t) represents the concentration of interleukin-6 in plasma at time t, which reflects the systemic inflammatory state. The lower the IL6(t) level, the milder the systemic inflammatory response. The unit is usually 2-50 pg / mL, and about 2-10 pg / mL in healthy individuals.

[0067] Specifically, in this embodiment of the invention, the standardized senescent cell density SAβGal(t) is obtained through the following steps:

[0068] First, the SA-β-Gal stained sections were counted per unit area under a high-power microscope to obtain the original senescent cell density. Specifically, for the SA-β-Gal stained sections of tumor biopsy tissue at time t, three non-overlapping tumor parenchyma regions were selected under a 400x optical microscope, avoiding necrotic areas and marginal fibrous tissue. The number of blue positive cells (senescent cells) and the total number of cells were counted in each region, and the original senescent cell density of each region was calculated: Original density = (Number of positive cells / Total number of cells) × 100%. The average of the original densities of the three regions was recorded as . .

[0069] Then, the original senescent cell density is Z-score normalized. Specifically, this is done using the following formula: Convert to standardized value :

[0070]

[0071] In the formula, The percentage of primitive senescent cells at time t; This represents the average density of primitive senescent cells in patients with the same disease from a historical database. This represents the standard deviation of the original senescent cell density in patients with the same disease from the historical database.

[0072] For example, a lung cancer patient ( Taking the preset data collection time points t=0, 3, 7, and 14 days as an example, the steps for generating the SBS time series are as follows:

[0073] Extraction and standardization of raw data:

[0074] t (days) Density of original senescent cells (%) standardization Plasma IL6(t) (pg / mL) 0 30.5 (30.5-28.2) / 7.5≈0.31 18 3 26.8 (26.8-28.2) / 7.5≈-0.19 15 7 22.3 (22.3-28.2) / 7.5≈-0.79 12 14 29.1 (29.1-28.2) / 7.5≈0.12 16

[0075] Calculate SBS(t) point by point ( , ):

[0076]

[0077]

[0078]

[0079]

[0080] Arrange the SBS(t) at each time point in chronological order to obtain the SBS time series sequence: .

[0081] It should be further noted that, to ensure the reliability of the VNI and SBS time series, quality verification is required after generation to remove outlier data. This specifically includes the following steps:

[0082] First, perform a numerical validity check; if the calculation result... >2 or <0, needs to be checked. Does it exceed the normal range (e.g., greater than 2.0 min⁻¹)? If the result is 0 (detection error), recalculate after confirmation; if the calculation result is... >5 or <0, needs to be checked. Check if the sample exceeds the -2 to -2 range or if IL6(t) exceeds the detection linear range (e.g., >50 pg / mL). If necessary, recollect the sample.

[0083] Then compare the trends of the VNI time series and the SBS time series. When the treatment is effective, VNI(t) usually shows an initial increase followed by a stabilization trend, while SBS(t) shows an initial decrease followed by a stabilization trend. If abnormal trends occur, such as a continuous decrease in VNI(t) or a continuous increase in SBS(t), it is necessary to confirm whether there are interfering factors such as treatment interruption or complications in conjunction with clinical records, mark the abnormal time points and note the reasons.

[0084] Finally, data integrity is verified to ensure that VNI(t) and SBS(t) values ​​exist simultaneously at each preset time point without any missing values. If data is missing at a certain time point, such as when the patient did not collect data on time, it needs to be marked as missing in the time series and then filled in using linear interpolation when modeling the coupled module, such as estimating the missing value based on the numerical values ​​of adjacent time points.

[0085] The coupling module is used to input the vascular normalization index (VNI) time series sequence and the aging burden score (SBS) time series sequence into a dual-channel time-series coupling model, and align the vascular function peak and aging inflection point through a cross-channel attention mechanism to output cross-modal coupling features.

[0086] The dual-channel temporal coupling model employs a three-layer architecture of parallel input, attention interaction, and feature fusion to ensure accurate correlation and information complementarity between VNI and SBS time-series data in the time dimension.

[0087] Specifically, the input layer is divided into two parallel channels, which respectively input the VNI timing sequence. With SBS time series Each channel extracts feature vectors from the raw time-series data, transforming the value at a single time point into a vector containing temporal context information. and .

[0088] The attention interaction layer calculates the attention weights at each time point t through a cross-channel attention mechanism. The weights reflect the correlation between the VNI and SBS features at that time point. The stronger the correlation, the higher the weight. The two types of feature vectors are weighted and focused based on the weights to highlight the key time points corresponding to the peak of vascular function and the inflection point of aging.

[0089] The feature fusion layer concatenates the weighted VNI feature vector and SBS feature vector at each time point, and then integrates the weighted concatenated features from all time points through a summation operation, ultimately outputting cross-modal coupled features. .

[0090] In one embodiment of the present invention, the cross-channel attention mechanism is implemented through the following steps:

[0091] To capture the contextual relevance of time-series data, such as the changing trends of the VNI value at time t compared to the VNI values ​​at times t-1 and t+1, the VNI(t) and SBS(t) of a single time point need to be transformed into feature vectors containing information from multiple time points. This invention employs a sliding window method with a window size of 3, meaning it includes the values ​​at the current time point t and the values ​​at the two adjacent time points t-1 and t+1. and All are 3-dimensional vectors. If the total number of time points T is small, such as T=4, zero-padding is applied to the first and last time points. For example, when t=0, the window contains the values ​​of t=0, t=0, and t=3.

[0092] The feature vector of the VNI time series at time point t The eigenvector of the SBS time series at time point t The formula for calculating cosine similarity is:

[0093]

[0094] Let be the feature vector of the VNI time series at time t; Let be the eigenvector of the SBS time series at time t. The closer the value is to 1, the better the trend of vascular function changes at that time point matches the trend of aging burden changes. For example, when vascular function increases, the aging burden decreases, which is in line with the physiological law of effective treatment.

[0095] For example, , ,but:

[0096]

[0097] The cosine similarity value is close to 1, indicating that the vascular function and the aging burden trend are highly matched at time t=3.

[0098] The cosine similarity at all time points is normalized using the softmax function to obtain the attention weight at each time point. Ensure that the sum of the weights at all time points is 1, that is... The calculation formula is as follows:

[0099]

[0100] In the formula, The attention weight at time t is the higher the weight, the greater the contribution of VNI and SBS features to prognosis prediction at that time point, and the more attention should be paid to them. It is a natural exponential function used to amplify the differences in cosine similarity at different time points, enhance the distinguishability of weights, and avoid weight averaging caused by similarity.

[0101] Output cross-modal coupling characteristics:

[0102]

[0103] In the formula, This represents a vector concatenation operation. This represents the total number of time points.

[0104] The calculation module is used to calculate the treatment opportunity window state value TOW(t) based on the cross-modal coupling characteristics: when the vascular normalization index VNI(t) is greater than the vascular threshold and the aging burden score SBS(t) is less than the aging threshold, TOW(t) = 1, otherwise TOW(t) = 0.

[0105] It should be noted that the logic behind TOW(t) is based on the core physiological mechanism of cancer treatment. Only when tumor vascular function is normalized (drugs can effectively reach the lesion) and the patient's aging burden is controllable (the body can tolerate the treatment) can treatment simultaneously achieve both efficacy and safety. Its binary criterion is as follows:

[0106]

[0107] VNI(t) quantifies the degree of vascular normalization, when At this time, the tumor's vascular permeability and blood supply capacity are close to those of normal tissue, allowing radiotherapy rays to precisely target tumor cells and chemotherapy / targeted drugs to effectively penetrate the lesion, avoiding treatment ineffectiveness due to drugs failing to reach the lesion; SBS(t) quantifies the overall aging burden, At that time, the patient's tumor microenvironment has fewer senescent cells and a lower level of systemic inflammation, resulting in a lower risk of adverse reactions such as bone marrow suppression and decreased immune function after treatment, thus avoiding treatment interruption due to poor patient tolerance.

[0108] In this invention, two conditions must be met simultaneously. If only VNI(t) is within the target range while SBS(t) is above the target range, the treatment may cause serious side effects due to poor patient tolerance. If only SBS(t) is within the target range while VNI(t) is below the target range, the therapeutic drug cannot effectively act on the tumor, resulting in ineffective treatment.

[0109] In a specific embodiment of the present invention, the vascular threshold is set to 0.8; the aging threshold θ is determined by the 25th percentile of the SBS value of patients with the best treatment response in historical data; when there is a p53 gene mutation in tumor tissue, the aging threshold is adjusted. Adjusted to It should be noted that the vascular threshold and aging threshold θ are objectively determined based on historical clinical data using statistical methods, ensuring the thresholds have universality and clinical guidance. For example, patients with a tumor shrinkage rate ≥50% and no grade III or higher adverse reactions 3 months after treatment were selected from the historical database and defined as optimal response patients, totaling 189 cases. The SBS values ​​of these optimal response patients at various time points after treatment were extracted, resulting in 1246 valid data points. These 1246 SBS values ​​were sorted from smallest to largest, and the 25th percentile was used as the baseline θ, meaning that more than 80% of the optimal response patients had SBS values ​​below this threshold. For example, if the SBS value at the 25th percentile in the sorted data is 0.6, then the baseline θ = 0.6.

[0110] p53 is an important tumor suppressor gene. Mutations in p53 can lead to abnormal regulation of tumor cell senescence, accelerating the accumulation of senescent cells in the tumor microenvironment. This significantly increases the sensitivity of patients to treatment response due to senescence burden; that is, at the same SBS value, p53-mutant patients have poorer treatment tolerance. Therefore, setting an senescence threshold... Adjusted to .

[0111] The decision output module is used to trigger the determination of permanent drug resistance status when the number of consecutive TOW(t)=0 reaches the drug resistance warning threshold; it generates a TOW time series decision tree based on the TOW(t) state sequence, marks the window period of TOW(t)=1 and the corresponding recommended treatment operation on the time axis; and outputs the probability of permanent drug resistance risk and treatment strategy adjustment suggestions.

[0112] It should be noted that permanent drug resistance is a key early warning event in cancer treatment, indicating that the patient has completely lost response to the current treatment regimen, requiring an urgent adjustment of the treatment strategy. This invention achieves early and accurate determination of drug resistance status by linking the number of consecutive TOW(t) = 0 with the rate of change in aging burden as two indicators.

[0113] Specifically, the drug resistance warning threshold is 3 consecutive times. When the cumulative slope of the SBS time series When the concentration is greater than 0.05 / day, a high drug resistance risk warning is triggered.

[0114] The cumulative slope reflects the rate of change of the aging burden over time, and the formula is:

[0115]

[0116] In the formula, t represents the treatment start time (usually t=0 days). Baseline aging burden score before treatment; t represents the current time point. Score the burden of aging at the current point in time; For time intervals (unit: days).

[0117] TOW time-series decision trees are tools that transform TOW(t) state sequences into clinically interpretable time-to-treatment operation correspondences. The core functionality involves marking the window of TOW(t)=1 and its corresponding recommended treatment operation on the timeline, helping physicians intuitively grasp the treatment rhythm. The construction steps are as follows:

[0118] The horizontal axis is set to the time (in days) since the start of treatment, with the time scale matching the preset collection time points of the collection module, such as 0, 7, 14, and 21 days. The vertical axis is divided into two columns: TOW status (1 / 0) and recommended treatment operations.

[0119] On the horizontal axis, all consecutive time periods with TOW(t) = 1 can be marked with a solid green line segment. For example, if t = 7-14 days, TOW(t) = 1, it can be marked with a green line segment. Consecutive time periods with TOW(t) = 0 can be marked with a solid red line segment.

[0120] A red triangle can be used to mark the point when drug resistance warnings are triggered, along with the risk of permanent drug resistance, and an adjustment to the treatment plan is recommended.

[0121] In one embodiment of the present invention, to quantify the severity of drug resistance risk, the present invention uses a sigmoid function to convert the rate of change of aging burden and the number of consecutive inappropriate treatments into a probability value between 0 and 1, wherein the probability of permanent drug resistance risk is... The calculation formula is:

[0122]

[0123] In the formula, For the sigmoid function, To ensure that the value of P is in the range of [0,1], the closer P is to 1, the higher the risk of drug resistance. This represents the cumulative slope of the SBS time series. The number of consecutive TOW(t) = 0; The weighting coefficient for the cumulative slope of SBS; The weighting coefficient is the number of consecutive TOW=0 times.

[0124] In this invention, the treatment strategy adjustment recommendations are generated based on three dimensions: the TOW window period distribution, the probability of drug resistance risk, and the individual patient's condition (such as cancer type and gene targets), to ensure the personalization and operability of the recommendations.

[0125] In a specific embodiment of the present invention, the TOW(t)=1 window period can be marked as a priority for radiotherapy / chemotherapy / targeted therapy. For example, for solid tumors, targeted therapy or chemotherapy is preferred. If the tumor volume is large, local radiotherapy can be arranged simultaneously within the window period. The specific treatment type needs to be combined with the patient's cancer type and gene target. For example, targeted therapy (trastuzumab) is marked for HER2-positive breast cancer. When SBS(t)>the aging threshold, senolytic drugs can be recommended next to the red line segment, such as the combination of dasatinib and quercetin.

[0126] The cancer patient prognosis prediction system based on a multimodal model according to embodiments of the present invention simultaneously collects vascular, aging, and genomic multimodal data, constructs a mathematical model to quantify the temporal characteristics of the Vascular Normalization Index (VNI) and the Aging Burden Score (SBS), and combines a cross-channel attention mechanism to align the peak of vascular function with the aging inflection point, overcoming the limitations of traditional single-data prediction. Its dynamic monitoring characteristics can capture changes in physiological indicators throughout the treatment cycle, accurately identify the optimal treatment window through TOW(t) state values, and personalize the aging threshold based on the p53 gene mutation status, making the prediction more aligned with individual patient differences. This system improves the accuracy and timeliness of prognosis prediction.

[0127] In the description of this invention, 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 indicated technical features. Thus, a feature defined as "first" or "second" may explicitly or implicitly include one or more of that feature. "A plurality of" means two or more, unless otherwise explicitly specified.

[0128] In this invention, unless otherwise explicitly specified and limited, the terms "installation," "connection," "linking," and "fixing," etc., should be interpreted broadly. For example, they can refer to a fixed connection, a detachable connection, or an integral part; they can refer to a mechanical connection or an electrical connection; they can refer to a direct connection or an indirect connection through an intermediate medium; they can refer to the internal communication of two components or the interaction between two components. Those skilled in the art can understand the specific meaning of the above terms in this invention according to the specific circumstances.

[0129] In this invention, unless otherwise explicitly specified and limited, "above" or "below" the second feature can mean that the first feature is in direct contact with the second feature, or that the first feature is in indirect contact with the second feature through an intermediate medium. Furthermore, "above," "over," and "on top" of the second feature can mean that the first feature is directly above or diagonally above the second feature, or simply that the first feature is at a higher horizontal level than the second feature. "Below," "below," and "under" the second feature can mean that the first feature is directly below or diagonally below the second feature, or simply that the first feature is at a lower horizontal level than the second feature.

[0130] In the description of this specification, the references to terms such as "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 the present invention. In this specification, the illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Moreover, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples. Furthermore, 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.

[0131] 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 a particular logical function or process, and the scope of the preferred embodiments of the invention 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 will be understood by those skilled in the art to which embodiments of the invention pertain.

[0132] The logic and / or steps represented in the flowchart or otherwise described herein, for example, can be considered as a ordered list of executable instructions for implementing logical functions, and can be embodied in any computer-readable medium for use by, or in conjunction with, an instruction execution system, apparatus, or device (such as a computer-based system, a processor-included system, or other system that can fetch and execute instructions from, an instruction execution system, apparatus, or device). For the purposes of this specification, "computer-readable medium" can be any means that can contain, store, communicate, propagate, or transmit programs for use by, or in conjunction with, an instruction execution system, apparatus, or device. More specific examples (a non-exhaustive list) of computer-readable media include: an electrical connection having one or more wires (electronic device), a portable computer disk drive (magnetic device), random access memory (RAM), read-only memory (ROM), erasable and editable read-only memory (EPROM or flash memory), fiber optic devices, and portable optical disc read-only memory (CDROM). Alternatively, the computer-readable medium may be paper or other suitable media on which the program can be printed, since the program can be obtained electronically, for example, by optically scanning the paper or other medium, followed by editing, interpreting, or otherwise processing as necessary, and then stored in a computer memory.

[0133] It should be understood that various parts of the present invention can be implemented in hardware, software, firmware, or a combination thereof. In the above embodiments, multiple steps or methods can be implemented in 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 one or a combination of the following techniques known in the art: 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.

[0134] Those skilled in the art will understand that all or part of the steps of the methods 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.

[0135] Furthermore, the functional units in the various embodiments of the present invention can be integrated into a processing module, or each unit can exist physically separately, or two or more units can be integrated into a module. The integrated module can be implemented in hardware or as a software functional module. If the integrated module is implemented as a software functional module and sold or used as an independent product, it can also be stored in a computer-readable storage medium.

[0136] Although embodiments of the present invention have been shown and described above, it is understood that the above embodiments are exemplary and should not be construed as limiting the present invention. Those skilled in the art can make changes, modifications, substitutions and variations to the above embodiments within the scope of the present invention.

Claims

1. A cancer patient prognosis prediction system based on a multimodal model, characterized in that, include: The acquisition module is used to synchronously acquire multimodal data at preset time points during the treatment of cancer patients. The multimodal data includes vascular-related data, aging-related data, and genomic data. The generation module is used to generate a time-series sequence of the Vascular Normalization Index (VNI) based on the vascular-related data; and to generate a time-series sequence of the Age Burden Score (SBS) based on the age-related data. The coupling module is used to input the time series of the Vascular Normalization Index (VNI) and the time series of the SBS (Severity of Aging) score into a dual-channel time-series coupling model, align the peak of vascular function with the inflection point of aging through a cross-channel attention mechanism, and output cross-modal coupling features. The calculation module is used to calculate the treatment opportunity window state value TOW(t) based on the cross-modal coupling characteristics: when the vascular normalization index VNI(t) is greater than the vascular threshold and the aging burden score SBS(t) is less than the aging threshold, TOW(t) = 1; otherwise, TOW(t) = 0. The determination output module is used to trigger the determination of permanent drug resistance status when the number of consecutive TOW(t)=0 reaches the drug resistance warning threshold; A TOW time-series decision tree is generated based on the TOW(t) state sequence. The window period when TOW(t)=1 is marked on the time axis and the corresponding recommended treatment operation is marked. The probability of permanent drug resistance risk and suggestions for adjusting treatment strategy are output.

2. The cancer patient prognosis prediction system based on a multimodal model according to claim 1, characterized in that, The vascular-related data include Ktrans values ​​of the tumor region obtained by dynamic contrast-enhanced MRI and vascular endothelial growth factor (VEGF) concentration obtained by serum detection. The aging-related data include the density of senescent cells obtained by SA-β-Gal staining of tumor biopsy tissue and the concentration of interleukin-6 obtained by plasma sample testing; The genomic data represents the p53 gene mutation status in tumor tissue.

3. The cancer patient prognosis prediction system based on a multimodal model according to claim 2, characterized in that, The formula for calculating the Vascular Normalization Index (VNI) time series is as follows: In the formula, The vascular normalization index at time t; This refers to the point in time after the start of treatment; The peak value of Ktrans in the tumor region at time t; This is the vascular attenuation constant; This is the time decay term; The formula for calculating the SBS (Spiritual Burden of Aging) time series is as follows: in, The aging burden score at time t; Weighting coefficients for standardizing the density of senescent cells; The weighting coefficient for plasma IL-6 concentration; The density of senescent cells in the tumor tissue after standardization at time t; The concentration of interleukin-6 in plasma at time t is given.

4. The cancer patient prognosis prediction system based on a multimodal model according to claim 3, characterized in that, The standardized senescent cell density SAβGal(t) was obtained through the following steps: The density of primitive senescent cells was obtained by counting positive cells per unit area under a high-power microscope on SA-β-Gal stained sections. The density of the original senescent cells was normalized using Z-score.

5. The cancer patient prognosis prediction system based on a multimodal model according to claim 4, characterized in that, The cross-channel attention mechanism is implemented through the following steps: Calculate the feature vector of the VNI time series at time point t. The eigenvector of the SBS time series at time point t Cosine similarity: In the formula, Let t be the feature vector of the VNI time series sequence; Let t be the feature vector of the SBS time series sequence; Attention weights at time point t: In the formula, Let be the attention weight at time t; It is a natural exponential function; Output cross-modal coupling characteristics: In the formula, This represents a vector concatenation operation. This represents the total number of time points.

6. The cancer patient prognosis prediction system based on a multimodal model according to claim 5, characterized in that, The formula for calculating the probability of permanent drug resistance is: In the formula, For the sigmoid function, ; This represents the cumulative slope of the SBS time series. The number of consecutive TOW(t) = 0; The weighting coefficient for the cumulative slope of SBS; The weighting coefficient is the number of consecutive TOW=0 values.

7. The cancer patient prognosis prediction system based on a multimodal model according to claim 6, characterized in that, The drug resistance warning threshold is 3 consecutive times. ; When the cumulative slope of the SBS time series When the concentration is greater than 0.05 / day, a high drug resistance risk warning is triggered.

Images