AI tumor immunotherapy response early warning device based on CTC dynamic monitoring

Abstract

The invention relates to the technical field of tumor immunotherapy monitoring, and discloses an AI tumor immunotherapy response early warning device based on CTC dynamic monitoring, and the device comprises a micro-fluidic chip, a graph acquisition module, a data acquisition module, a graph and data verification module and an early warning module. According to the method, the CTC samples are continuously collected, the Raman spectrum and multi-parameter detection are combined to achieve multi-dimensional data acquisition and quantified data consistency, the early warning probability is generated in a graded mode according to the comparison result of the graph number verification value and the dynamic threshold value, and dynamic monitoring, multi-source data fusion verification and graded early warning of tumor immunotherapy response are achieved; the limitation of traditional single-index monitoring is broken through, the comprehensiveness of treatment response evaluation and the reliability of risk prediction are improved, and technical support is provided for precise medical treatment.

Description

Technical Field

[0001] The present application relates to the technical field of tumor immunotherapy monitoring, and specifically to an AI tumor immunotherapy response early warning device based on CTC dynamic monitoring. Background Art

[0002] In the field of tumor immunotherapy, accurate monitoring and early warning are crucial to enhancing treatment efficacy and improving patient prognosis. Currently, existing technologies have many deficiencies in monitoring tumor immunotherapy responses, specifically:

[0003] On the one hand, traditional monitoring methods often rely on a single indicator, which makes it difficult to fully reflect the complex biological characteristics of tumors and changes in the immune microenvironment, resulting in inaccurate assessment of treatment effects and inability to timely detect potential treatment risks.

[0004] On the other hand, existing data processing systems are unable to achieve dynamic monitoring of circulating tumor cells (CTCs) and cannot fully utilize the tumor information carried by CTCs, limiting the in-depth understanding of tumor development and treatment response.

[0005] While the Chinese patent application CN110706216A, "A Rapid Processing System for Tumor Immune Microenvironment Data Based on AI Intelligent Identification," addresses the application of AI in the field of tumor immunity, it still suffers from significant flaws. The system lacks targeted monitoring of CTCs and fails to fully explore the value of CTC-related data in immunotherapy monitoring. Its data processing primarily relies on static analysis, making it difficult to adapt to the dynamic changes in the tumor immune microenvironment and unable to adjust monitoring and early warning strategies in real time. Furthermore, the system lacks an effective early warning mechanism and cannot provide clinicians with timely and accurate treatment risk alerts, leaving it deficient in guiding clinical decision-making.

[0006] In summary, there is an urgent need for a new technical solution for AI tumor immunotherapy response based on CTC dynamic monitoring, which can improve the accuracy and effectiveness of tumor immunotherapy through multi-dimensional data collection, analysis and intelligent early warning. Summary of the Invention

[0007] The purpose of this application is to provide an AI tumor immunotherapy response early warning device based on CTC dynamic monitoring to solve the technical problems raised in the above background technology.

[0008] To achieve the above objectives, the present application discloses the following technical solutions: an AI tumor immunotherapy response early warning device based on dynamic CTC monitoring, comprising: a microfluidic chip for continuously collecting CTC samples from the patient's blood before and after treatment, an image acquisition module for non-invasively acquiring the cell metabolic fingerprint of the CTC sample as image data through Raman spectroscopy, and a data acquisition module for acquiring time-series CTC counts, CTC typing data, PD-L1 expression levels, and T cell receptor diversity data as numerical data;

[0009] Processing the image data and the numerical data using an image-number verification module, and calculating an image-number verification value corresponding to a treatment cycle, the image-number verification value being used to characterize the degree of consistency between the image data and the numerical data;

[0010] The warning module of the built-in gradient boosting decision tree model is used to compare the image number verification value with the dynamic verification threshold. When the image number verification value is greater than or equal to the dynamic verification threshold of the corresponding treatment cycle, the first warning probability is calculated based on the image data and numerical data and the first warning information is output. Otherwise, the second warning probability is calculated based on the image number verification value and the first warning probability, and the second warning information is generated.

[0011] Preferably, the image number verification module includes:

[0012] a spatiotemporal alignment unit for matching metabolic characteristic peaks of image data with detection timestamps of numerical data based on treatment cycles;

[0013] A feature fusion unit is used to extract image features of image data and numerical features of numerical data to construct a joint feature matrix;

[0014] The verification calculation unit calculates the graph verification value by counting the similarity of the joint feature matrix of each treatment cycle.

[0015] Preferably, the calculation formula of the image number verification value is:

[0016]

[0017] in:

[0018] is the image feature of the i-th period;

[0019] The numerical features of the i-th cycle include CTC typing ratio, CTC count, PD-L1 expression and T cell receptor diversity data;

[0020] is the joint characteristic matrix of the i-th period;

[0021] is the similarity of the joint feature matrix of the i-th cycle (calculated based on the numerical features and image features containing the typing data);

[0022] n is the number of treatment cycles;

[0023] V i is the calculated verification value of the graph number of the i-th period.

[0024] Preferably, the similarity of the joint feature matrix is calculated as:

[0025] Calculate the similarity of the original joint feature matrix based on the preset similarity algorithm;

[0026] The fluctuation of the graph verification value is calculated, and the similarity of the original joint feature matrix is corrected based on the fluctuation to calculate the similarity of the joint feature matrix.

[0027] Preferably, the first warning probability is calculated as follows:

[0028] Calculate the sum of the weighted predictions of all decision trees in the gradient boosted decision tree model;

[0029] Determine the minimum and maximum sums of weighted prediction values (based on statistics from a historical training set containing CTC typing data);

[0030] A normalized calculation is performed based on the sum of the weighted prediction values and their corresponding minimum and maximum values to obtain the first warning probability.

[0031] Preferably, the sum of the weighted prediction values is calculated as:

[0032] Determine the total number of decision trees;

[0033] Determine the preset weight of each decision tree, which is used to adjust the contribution of the corresponding decision tree prediction function;

[0034] Input image features and numerical features (including CTC typing ratio, CTC count, etc.) into the prediction function preset by each decision tree to obtain a scalar prediction value;

[0035] All scalar prediction values are summed according to the preset weights to obtain the sum of the weighted prediction values.

[0036] Preferably, the calculation formula for the second warning probability is:

[0037] P2=P1*[1+λ*(V th_i -V i )]

[0038] in:

[0039] P2 is the calculated second warning probability;

[0040] P1 is the first warning probability;

[0041] λ is the preset correction coefficient;

[0042] V th_i is the dynamic verification threshold of the i-th cycle;

[0043] V i is the calculated verification value of the graph number of the i-th period.

[0044] Preferably, the generation rule of the second warning information is:

[0045] When P2≥P1+φ, the preset warning level is increased by one level; when P2≤P1-φ, the warning level is decreased by one level; otherwise, the original warning level is maintained; wherein φ is the preset second warning information output adjustment parameter.

[0046] Preferably, the calculation formula of the dynamic verification threshold is:

[0047]

[0048] in:

[0049] V th_i is the calculated dynamic verification threshold of the i-th cycle;

[0050] V th_i-1 is the dynamic verification threshold of the i-1th cycle;

[0051] n is the number of treatment cycles;

[0052] is the average value of the dynamic verification threshold over n cycles;

[0053] V th _0 is the initial value of the preset dynamic verification threshold;

[0054] V i is the graph number verification value of the i-th cycle;

[0055] is the average value of the graph verification value of n cycles;

[0056] Preferably, the first warning information and the second warning information are both displayed in real time on a preset visual interface, and the content of the real-time display at least includes:

[0057] Metabolic fingerprint heat map;

[0058] CTC count and typing distribution diagram;

[0059] PD-L1 expression trend curve;

[0060] Revised risk probability radar chart and decision tree path.

[0061] Beneficial effects: The AI tumor immunotherapy response early warning device based on CTC dynamic monitoring of the present application uses a microfluidic chip to continuously collect CTC samples, combines Raman spectroscopy with multi-parameter detection to achieve multi-dimensional data acquisition, matches the timestamps of images and numerical data through a spatiotemporal alignment unit, constructs a joint feature matrix and calculates similarity to quantify data consistency; relying on the gradient boosting decision tree model, the early warning probability is generated in a hierarchical manner according to the comparison results of the image verification value and the dynamic threshold; wherein, the first early warning probability realizes risk quantification through the normalized weighted prediction value, and the second early warning probability is dynamically corrected in combination with data consistency fluctuations to improve the accuracy of the early warning; the dynamic verification threshold is adaptively adjusted based on historical data to ensure the rationality of the threshold. The visual interface displays the metabolic map, indicator trends and decision paths in real time, enhancing clinical interpretability; thereby, dynamic monitoring of tumor immunotherapy response, multi-source data fusion verification and hierarchical early warning are achieved, breaking through the limitations of traditional single indicator monitoring, improving the comprehensiveness of treatment response evaluation and the reliability of risk prediction, and providing technical support for precision medicine. BRIEF DESCRIPTION OF THE DRAWINGS

[0062] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are only some embodiments of the present application. For those skilled in the art, other drawings can be obtained based on these drawings without paying any creative work.

[0063] Figure 1 This is a structural block diagram of the AI tumor immunotherapy response early warning device based on CTC dynamic monitoring provided in an embodiment of the present application. DETAILED DESCRIPTION

[0064] The following is a clear and complete description of the technical solutions in the embodiments of this application. Obviously, the embodiments described are only part of the embodiments of this application, not all of them. Based on the embodiments of this application, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of this application.

[0065] In this document, the term "comprising" is intended to encompass non-exclusive inclusion, such that a process, method, article, or apparatus that includes a list of elements includes not only those elements but also other elements not explicitly listed, or elements inherent to such process, method, article, or apparatus. In the absence of further limitations, an element defined by the phrase "comprising..." does not preclude the presence of additional identical elements in the process, method, article, or apparatus that includes the element.

[0066] The first aspect of this embodiment discloses Figure 1 The AI tumor immunotherapy response early warning device based on CTC dynamic monitoring shown in the figure includes: a microfluidic chip for continuously collecting CTC samples in the patient's blood before and after treatment, an image acquisition module for non-invasively obtaining the cell metabolic fingerprint of CTC samples through Raman spectroscopy as image data, and a data acquisition module for obtaining time-series CTC counts, CTC typing data, PD-L1 expression levels and T cell receptor diversity data as numerical data.

[0067] Specifically, the image data and the numerical data are processed using an image-number verification module, and an image-number verification value corresponding to the treatment cycle is calculated, where the image-number verification value is used to represent the degree of consistency between the image data and the numerical data;

[0068] The warning module of the built-in gradient boosting decision tree model is used to compare the image number verification value with the dynamic verification threshold. When the image number verification value is greater than or equal to the dynamic verification threshold of the corresponding treatment cycle, the first warning probability is calculated based on the image data and numerical data and the first warning information is output. Otherwise, the second warning probability is calculated based on the image number verification value and the first warning probability, and the second warning information is generated.

[0069] It should be noted that: the surface of the microfluidic chip of this embodiment is modified with nanomaterials to improve the capture efficiency of CTCs; the image acquisition module uses a Raman spectrometer to scan the CTC sample and non-invasively obtain the cell metabolic fingerprint map as image data; the number acquisition module uses immunofluorescence labeling technology to perform typing detection on CTCs (including epithelial type, mesenchymal type and mixed type), and records the number and proportion of each type, and at the same time obtains time-series CTC counts, PD-L1 expression levels and T cell receptor diversity data as numerical data; the acquisition rules can be but are not limited to acquisition frequency, typing labeling method, standardized processing flow, etc. The image acquisition module and the number acquisition module of this embodiment are data acquisitions in two relatively independent dimensions. Therefore, in the actual communication relationship, the microfluidic chip is connected to the image acquisition module and then to the image and number verification module, while the number acquisition module is directly connected to the image and number verification module, and the image and number verification module is connected to the early warning module.

[0070] Based on the above, this embodiment utilizes a microfluidic chip to achieve efficient and continuous collection of CTC samples. The image and data acquisition modules acquire multi-dimensional data, which is then quantified using the image and data verification module. The early warning module then relies on a gradient boosting decision tree model to compare the image and data verification values with dynamic verification thresholds, calculate early warning probabilities in a hierarchical manner, and output early warning information. This enables dynamic monitoring of tumor immunotherapy responses, data consistency verification, and precise, graded early warnings of hyperprogression risk, enhancing the comprehensiveness of clinical treatment monitoring and the reliability of risk prediction.

[0071] Specifically, the image number verification module includes:

[0072] a spatiotemporal alignment unit for matching metabolic characteristic peaks of image data with detection timestamps of numerical data based on treatment cycles;

[0073] A feature fusion unit is used to extract image features of image data and numerical features of numerical data (including CTC typing data) to construct a joint feature matrix;

[0074] The verification calculation unit calculates the graph verification value by counting the similarity of the joint feature matrix of each treatment cycle.

[0075] It should be noted that the spatiotemporal alignment unit of this embodiment sets the treatment cycle (e.g., each treatment cycle is 14 to 28 days), obtains the detection time of the metabolic characteristic peak in the image data, and extracts the detection timestamps of CTC count and PD-L1 expression in the numerical data. Through the existing time window matching method (the time window can be ±3 days), the metabolic characteristic peak of the image data within the same treatment cycle is matched with the detection timestamp of the numerical data. For example, in the third treatment cycle, the metabolic characteristic peak data on the 22nd day and the CTC count data on the 23rd day are matched to the same cycle.

[0076] It should be noted that the feature fusion unit of this embodiment uses an existing convolutional neural network to extract image features of the required dimensionality (such as 100 dimensions) from the cell metabolic fingerprint map, and performs existing standardization processing on the numerical data (including CTC typing ratio, CTC count, PD-L1 expression and T cell receptor diversity data) to obtain the numerical features of the required dimensionality (such as 60 dimensions). The standardization processing technology can be, but is not limited to, Z-score processing technology, and a 160-dimensional joint feature matrix is constructed by column splicing. For example, if the image feature [0.1, 0.3, ...], the normalized numerical feature vector is [0.2, -0.1, ...], based on this, it is fused into a joint feature matrix.

[0077] Through the above, this embodiment realizes the spatiotemporal alignment and fusion analysis of multimodal data, accurately quantifies the consistency of image data and numerical data, provides a reliable data basis for subsequent early warning, and improves the processing and verification accuracy of multi-source data in tumor immunotherapy response monitoring.

[0078] Specifically, the calculation formula of the image number verification value is:

[0079]

[0080] in:

[0081] is the image feature of the i-th period;

[0082] The numerical features of the i-th cycle include CTC typing ratio, CTC count, PD-L1 expression and T cell receptor diversity data;

[0083] is the joint characteristic matrix of the i-th period;

[0084] is the similarity of the joint feature matrix of the i-th cycle (calculated based on the numerical features and image features containing the typing data);

[0085] n is the number of treatment cycles;

[0086] V i is the calculated verification value of the graph number of the i-th period.

[0087] Based on the above, this embodiment achieves quantitative representation of the consistency between image data and numerical data by counting the similarities of the joint feature matrices of each treatment cycle and calculating the average, accurately calculating the image verification value, and providing data support for subsequent early warning modules to judge data credibility and adjust early warning strategies, thereby improving the accuracy and reliability of multi-source data fusion analysis in tumor immunotherapy response monitoring.

[0088] Specifically, the similarity of the joint feature matrix is calculated as:

[0089] The similarity of the original joint feature matrix is calculated based on a preset similarity algorithm (such as cosine similarity or Mahalanobis distance), where the numerical features include multivariate parameters such as CTC typing ratio and CTC count. The fluctuation of the image verification value is calculated, and the similarity of the original joint feature matrix is corrected based on this fluctuation to calculate the similarity of the joint feature matrix.

[0090] As a preferred implementation of this embodiment, the similarity of the joint feature matrix is calculated using the following formula:

[0091]

[0092] in:

[0093] is the similarity of the calculated joint feature matrix;

[0094] is the similarity of the original joint feature matrix; it should be noted that the similarity of the original joint feature matrix of this embodiment is based on any existing similarity calculation method, for example, a cosine similarity algorithm;

[0095] is the average value of the image number verification value of n cycles (the image number verification value is calculated using the joint feature matrix containing CTC typing data).

[0096] Through the above, this embodiment realizes the dynamic adjustment of the similarity of the joint feature matrix, more accurately reflects the degree of correlation between image data and numerical data, improves the accuracy of the image verification value calculation, provides a more reliable data consistency judgment basis for subsequent tumor immunotherapy response warning, and enhances the reliability of the device for multi-source data fusion analysis.

[0097] In a simple example, the number of treatment cycles n=3, and the data and calculation process of each cycle are as follows:

[0098] Cycle 1: Original joint feature matrix similarity (This value is calculated based on the cosine similarity algorithm.) Average value of the number of images verified (Calculated based on historical data), the current periodogram verification value V1 = 0.72.

[0099] Similarly, process the second and third cycles respectively to obtain the corresponding and

[0100] Substituting the above similarity into Calculate the image verification value V3 = 0.773, and output this value as the image verification value for processing. When the image verification value is greater than or equal to the dynamic verification threshold, it indicates that the image data and the numerical data are highly consistent. The early warning module directly calculates the first warning probability based on the data and outputs reliable first warning information. When the image verification value is less than the dynamic verification threshold, it indicates that the data consistency is low. It is necessary to combine the image verification value to optimize the first warning probability and generate more accurate second warning information to assist doctors in judging the response status of tumor immunotherapy and adjusting treatment strategies.

[0101] Specifically, the first warning probability is calculated as follows:

[0102] Calculate the sum of the weighted predictions of all decision trees in the gradient boosted decision tree model;

[0103] Determine the minimum and maximum sums of weighted prediction values (the minimum and maximum values are statistically derived from a historical training set containing CTC typing data);

[0104] A normalized calculation is performed based on the sum of the weighted prediction values and their corresponding minimum and maximum values to obtain the first warning probability.

[0105] As a preferred implementation of this embodiment, the first warning probability is calculated using the following formula:

[0106]

[0107] in:

[0108] P1 is the calculated first warning probability, and its value range is [0,1];

[0109] S is the sum of the weighted prediction values of all decision trees;

[0110] min(S) is the minimum value of S;

[0111] max(S) is the maximum value of S.

[0112] It should be noted that the present embodiment utilizes the existing gradient boosting decision tree model to perform the calculations in the present embodiment.

[0113] Based on the above, this embodiment calculates the sum of the weighted prediction values of the gradient boosting decision tree model and normalizes it by combining the maximum value to generate a first warning probability. This achieves a quantitative assessment of the risk of superprogression in tumor immunotherapy, converts multi-source data features into standardized probability values, and provides a precise basis for the early warning module to output the first warning information, thereby improving the quantitative analysis capabilities and early warning reliability of tumor immunotherapy response monitoring.

[0114] Specifically, the sum of the weighted prediction values is calculated as follows:

[0115] Determine the total number of decision trees;

[0116] Determine the preset weight of each decision tree, which is used to adjust the contribution of the corresponding decision tree prediction function;

[0117] Input image features and numerical features (including CTC typing ratio, CTC count, etc.) into the prediction function preset by each decision tree to obtain a scalar prediction value;

[0118] All scalar prediction values are summed according to the preset weights to obtain the sum of the weighted prediction values.

[0119] As a preferred implementation of this embodiment, the sum of the weighted prediction values is calculated using the following formula:

[0120]

[0121] in:

[0122] S is the sum of the weighted prediction values of all decision trees calculated;

[0123] F g is the feature vector of image data;

[0124] F d is the eigenvector of numerical data;

[0125] h k (F g ,F d ) is the prediction function of the k-th decision tree, which inputs the feature vector and outputs a scalar prediction value;

[0126] α k is the preset weight of the kth decision tree, which is used to adjust h k (F g ,F d )’s contribution;

[0127] K is the total number of decision trees.

[0128] Through the above, this embodiment realizes the fusion of multiple decision tree prediction results, fully utilizes image and numerical feature information, provides comprehensive data support for the first warning probability calculation, improves the accuracy and reliability of tumor immunotherapy super-progression risk prediction, and enables the warning device to more accurately analyze the relationship between multi-source data features and treatment response.

[0129] In a simple example, there are three decision trees with corresponding weights of 0.4, 0.3, and 0.3, and the image data feature vector F g =[0.9,0.7], numerical data feature vector F d =[0.8,-0.2]. The first decision tree prediction function outputs h1(F g ,F d )=0.75, the second h2(F g ,F d )=0.6, the third h2(F g ,F d) = 0.8. The calculated S = 0.72. Obtaining min(S) = 0.5 and max(S) = 0.9 from the historical data, the calculated P1 = 0.55. The prediction results of multiple decision trees are integrated through weighted summation, balancing the contributions of different decision trees to the final prediction, thereby more comprehensively reflecting the combined information of image and numerical features. Normalization is then performed, mapping the weighted prediction value to a first warning probability within the range of [0, 1]. This quantifies the risk of overprogression in tumor immunotherapy, provides a standardized basis for warning information output, and helps doctors more intuitively judge the risk of treatment response.

[0130] Specifically, the calculation formula for the second warning probability is:

[0131] P2=P1*[1+λ*(V th_i -V i )]

[0132] in:

[0133] P2 is the calculated second warning probability;

[0134] P1 is the first warning probability;

[0135] λ is the preset correction coefficient;

[0136] V th_i is the dynamic verification threshold of the i-th cycle;

[0137] V i is the calculated verification value of the graph number of the i-th period.

[0138] In a simple example, in a corresponding treatment cycle, the first warning probability is 0.6; the preset correction coefficient is 0.2; the dynamic verification threshold for the corresponding treatment cycle is 0.7; and the image verification value for the corresponding treatment cycle is 0.65. The calculated second warning probability is 0.606, which is used to generate the second warning information.

[0139] Through the above, this embodiment realizes the dynamic optimization of the warning probability. When the image verification value is lower than the dynamic verification threshold, the warning probability is adjusted according to the data consistency difference, so that the warning result is more in line with the actual treatment response, thereby improving the accuracy of the tumor immunotherapy response warning and providing a more reliable risk assessment basis for clinical treatment decisions.

[0140] Specifically, the generation rule of the second warning information is:

[0141] When P2≥P1+φ, the preset warning level is increased by one level; when P2≤P1-φ, the warning level is decreased by one level; otherwise, the original warning level is maintained; wherein φ is the preset second warning information output adjustment parameter.

[0142] In a simple example, the adjustment parameter is 0.1, the warning levels are divided into low risk, medium risk and high risk, the first warning probability is 0.5, and the calculated second warning probability is 0.65. Based on this, it can be judged that the original warning level of medium risk will be increased by one level to generate a second warning information of high risk.

[0143] Through the above, this implementation achieves the refined output of early warning information, making the risk assessment of tumor immunotherapy response more in line with actual data changes, improving the accuracy of early warning and clinical guidance value, and providing doctors with more accurate reference basis for adjusting treatment strategies.

[0144] Specifically, the calculation formula of the dynamic verification threshold is:

[0145]

[0146] in:

[0147] V th_i is the calculated dynamic verification threshold of the i-th cycle;

[0148] V th_i-1 is the dynamic verification threshold of the i-1th cycle;

[0149] n is the number of treatment cycles;

[0150] is the average value of the dynamic verification threshold over n cycles;

[0151] V th_ 0 is the initial value of the preset dynamic verification threshold;

[0152] V i is the graph number verification value of the i-th cycle;

[0153] is the average value of the image number verification value of n cycles (the image number verification value is calculated using the joint feature matrix containing CTC typing data).

[0154] Through the above, this embodiment utilizes the adaptive adjustment of the verification threshold to make it fit the data change pattern during the treatment cycle, provide a more reasonable comparison benchmark for the graph verification value, improve the accuracy of data consistency judgment in tumor immunotherapy response warning, and enable the warning module to output warning information more accurately, providing a reliable basis for clinical treatment decision-making.

[0155] Specifically, the first warning information and the second warning information are both displayed in real time on a preset visual interface, and the content of the real-time display includes at least:

[0156] Metabolic fingerprint heat map;

[0157] CTC count and typing distribution diagram;

[0158] PD-L1 expression trend curve;

[0159] Revised risk probability radar chart and decision tree path.

[0160] It should be noted that the visualization interface and the real-time display in this embodiment are implemented based on any existing visualization technology, such as a graphics rendering engine.

[0161] Through the above, this embodiment realizes the intuitive presentation of tumor immunotherapy data, helping doctors to quickly grasp the CTC metabolic characteristics, indicator change trends and risk probabilities. At the same time, it reveals the decision logic of the AI model through the decision tree path, improves the clinical interpretability of the warning information, provides a clear and visual basis for treatment decisions, and enhances doctors' understanding and confidence in the application of warning results.

[0162] In summary, the AI tumor immunotherapy response early warning device based on dynamic CTC monitoring in this embodiment uses a microfluidic chip to continuously collect CTC samples, combines Raman spectroscopy with multi-parameter detection to achieve multi-dimensional data acquisition, matches the timestamps of images and numerical data through a spatiotemporal alignment unit, constructs a joint feature matrix and calculates similarity to quantify data consistency; relying on a gradient boosting decision tree model, it generates early warning probabilities in a hierarchical manner based on the comparison results of the image verification value and the dynamic threshold; wherein, the first early warning probability realizes risk quantification through the normalized weighted prediction value, and the second early warning probability is dynamically corrected based on data consistency fluctuations to improve early warning accuracy; the dynamic verification threshold is adaptively adjusted based on historical data to ensure the rationality of the threshold. The visual interface displays metabolic maps, indicator trends, and decision paths in real time, enhancing clinical interpretability; thus, dynamic monitoring of tumor immunotherapy response, multi-source data fusion verification, and hierarchical early warning are achieved, breaking through the limitations of traditional single indicator monitoring, improving the comprehensiveness of treatment response assessment and the reliability of risk prediction, and providing technical support for precision medicine.

[0163] In the embodiments provided herein, it should be understood that the embodiments described herein can be implemented in hardware, software, firmware, middleware, code, or any appropriate combination thereof. For hardware implementation, the processor can be implemented in one or more of the following units: an application-specific integrated circuit (ASIC), a digital signal processor (DSP), a digital signal processing device (DSPD), a programmable logic device (PLD), a field programmable gate array (FPGA), a processor, a controller, a microcontroller, a microprocessor, other electronic units designed to implement the functions described herein, or a combination thereof. For software implementation, part or all of the processes of the embodiments can be completed by instructing the relevant hardware through a computer program. When implemented, the above program can be stored in a computer-readable storage medium or transmitted as one or more instructions or codes on a computer-readable storage medium. Computer-readable storage media include computer storage media and communication media, wherein the communication media include any medium that facilitates the transmission of a computer program from one place to another. The storage medium can be any available medium that a computer can access. The computer-readable storage medium can include, but is not limited to, RAM, ROM, EEPROM, CD-ROM or other optical disk storage, magnetic disk storage media or other magnetic storage devices, or any other medium that can be used to carry or store the desired program code in the form of instructions or data structures and can be accessed by a computer.

[0164] Finally, it should be noted that the above is only a preferred embodiment of the present application and is not intended to limit the present application. Although the present application has been described in detail with reference to the aforementioned embodiments, those skilled in the art can still modify the technical solutions described in the aforementioned embodiments or make equivalent replacements for some of the technical features therein. Any modifications, equivalent replacements, improvements, etc. made within the spirit and principles of the present application should be included in the scope of protection of the present application.

Claims

1. An AI-based tumor immunotherapy response early warning device based on CTC dynamic monitoring, comprising: A microfluidic chip for continuously collecting CTC samples from the patient's blood before and after treatment; an image acquisition module for non-invasively acquiring the cell metabolic fingerprint of CTC samples through Raman spectroscopy as image data; and a data acquisition module for acquiring time-series CTC counts, CTC typing data, PD-L1 expression levels, and T cell receptor diversity data as numerical data; characterized by: Processing the image data and the numerical data using an image-number verification module, and calculating an image-number verification value corresponding to a treatment cycle, the image-number verification value being used to characterize the degree of consistency between the image data and the numerical data; The warning module of the built-in gradient boosting decision tree model is used to compare the image number verification value with the dynamic verification threshold. When the image number verification value is greater than or equal to the dynamic verification threshold of the corresponding treatment cycle, the first warning probability is calculated based on the image data and numerical data and the first warning information is output. Otherwise, the second warning probability is calculated based on the image number verification value and the first warning probability, and the second warning information is generated.

2. The AI tumor immunotherapy response early warning device based on CTC dynamic monitoring according to claim 1, characterized in that: The image number verification module includes: a spatiotemporal alignment unit for matching metabolic characteristic peaks of image data with detection timestamps of numerical data based on treatment cycles; A feature fusion unit is used to extract image features of image data and numerical features of numerical data to construct a joint feature matrix; The verification calculation unit calculates the graph verification value by counting the similarity of the joint feature matrix of each treatment cycle.

3. The AI tumor immunotherapy response early warning device based on CTC dynamic monitoring according to claim 2, characterized in that: The calculation formula of the image number verification value is: in: is the image feature of the i-th period; is the numerical feature of the i-th cycle, including CTC typing ratio, CTC count, PD-L1 expression and T cell receptor diversity data; is the joint characteristic matrix of the i-th period; is the similarity of the joint feature matrix of the i-th period; n is the number of treatment cycles; V i is the calculated verification value of the graph number of the i-th period.

4. The AI tumor immunotherapy response early warning device based on CTC dynamic monitoring according to claim 2, characterized in that: The similarity of the joint feature matrix is calculated as: Calculate the similarity of the original joint feature matrix based on the preset similarity algorithm; The fluctuation of the graph verification value is calculated, and the similarity of the original joint feature matrix is corrected based on the fluctuation to calculate the similarity of the joint feature matrix.

5. The AI tumor immunotherapy response early warning device based on CTC dynamic monitoring according to claim 1, characterized in that: The calculation of the first warning probability is: Calculate the sum of the weighted predictions of all decision trees in the gradient boosted decision tree model; Determine the minimum and maximum values of the sum of weighted prediction values; A normalized calculation is performed based on the sum of the weighted prediction values and their corresponding minimum and maximum values to obtain the first warning probability.

6. The AI tumor immunotherapy response early warning device based on CTC dynamic monitoring according to claim 5, characterized in that: The sum of the weighted predicted values is calculated as: Determine the total number of decision trees; Determine the preset weight of each decision tree, which is used to adjust the contribution of the corresponding decision tree prediction function; Input the image features and numerical features into the prediction function preset by each decision tree to obtain a scalar prediction value; All scalar prediction values are summed according to the preset weights to obtain the sum of the weighted prediction values.

7. The AI tumor immunotherapy response early warning device based on CTC dynamic monitoring according to claim 3, characterized in that: The calculation formula of the second warning probability is: P2=P1*[1+λ*(V th_i -IN i )] in: P2 is the calculated second warning probability; P1 is the first warning probability; λ is the preset correction coefficient; V th_i is the dynamic verification threshold of the i-th cycle; V i is the calculated verification value of the graph number of the i-th period.

8. The AI tumor immunotherapy response early warning device based on CTC dynamic monitoring according to claim 7, characterized in that: The generation rule of the second warning information is: When P2≥P1+φ, the preset warning level is raised by one level; when P2≤P1-φ, the warning level is lowered by one level; Otherwise, the original warning level is maintained; wherein, φ is the preset second warning information output adjustment parameter.

9. The AI tumor immunotherapy response early warning device based on CTC dynamic monitoring according to claim 1, characterized in that: The calculation formula of the dynamic verification threshold is: in: V th_i is the calculated dynamic verification threshold of the i-th cycle; V th_i-1 is the dynamic verification threshold of the i-1th cycle; n is the number of treatment cycles; is the average value of the dynamic verification threshold over n cycles; V th _0 is the initial value of the preset dynamic verification threshold; V i is the graph number verification value of the i-th cycle; is the average value of the validation value of the graph number for n periods.

10. The AI tumor immunotherapy response early warning device based on CTC dynamic monitoring according to claim 1, characterized in that: The first warning information and the second warning information are both displayed in real time on a preset visual interface, and the content of the real-time display includes at least: Metabolic fingerprint heat map; CTC count and typing distribution diagram; PD-L1 expression trend curve; Revised risk probability radar chart and decision tree path.

Images