A clinical informed consent embedded insurance interaction method and system

Abstract

The application relates to the field of insurance software, and discloses a clinical informed consent embedded insurance interaction method and system, which comprises the following steps: acquiring patient multi-modal clinical data in real time and mapping the data into a standardized feature tensor; using a gradient back propagation technology to perform attribution calculation on a specific clinical risk, generating a risk heat distribution, and rendering a risk explanation anchor point in situ on the side of a corresponding risk clause of a clinical informed consent; in response to user triggering of the anchor point, through semantic alignment and compliance verification, a popularized guarantee description consistent with a legal clause is rendered in the form of a linear extension module on the side of the anchor point, realizing non-invasive fusion of insurance recommendation and diagnosis and treatment notification processes; and the whole-process interaction node, semantic similarity and user stress state are encrypted and stored as evidence. The application eliminates the trust barrier between an actuarial black box and a medical lesion, improves the substantial compliance and evidence storage and proof ability of insurance application while guaranteeing medical ethics.

Description

Technical Field

[0001] This invention relates to the field of insurance software, and more particularly to a clinical informed consent form embedded insurance interaction method and system. Background Technology

[0002] With the deep integration of internet insurance and healthcare scenarios, health insurance applications are extending to the clinical setting. However, existing technologies for generating insurance application interfaces still face deep-seated problems. Traditional solutions often employ pop-up ads that operate independently of the clinical system, forcibly interrupting the patient's reading flow when they are reading critical information such as surgical instructions and examination notices. This intrusive interference not only exacerbates patient anxiety but also easily leads to accidental closure, resulting in missed risk coverage.

[0003] In terms of risk assessment, existing AI models are mostly based on black-box calculations of actuarial big data. Premium pricing and claims conclusions lack intuitive medical basis. Patients cannot understand why their current imaging or test indicators lead to increased premiums or exclusions. There is a serious gap in interpretability between actuarial logic and clinical medical manifestations, which weakens the right to know and the willingness to purchase insurance.

[0004] Furthermore, the existing interactive interface lacks real-time monitoring of users' cognitive load and psychological stress. When patients are in a state of cognitive overload due to sudden changes in their condition or complex surgical procedures, the application portal remains active, which can easily induce irrational decisions, violate medical ethics, and create potential for claims disputes.

[0005] A deeper problem is that insurance interaction and clinical pathways are physically separated. Insurance modules often exist as independent H5 pages or pop-ups, failing to achieve native integration with core clinical processes such as informed consent forms. As a result, insurance coverage cannot serve as a logical extension of medical risk disclosure and lacks persuasiveness in a contextualized manner.

[0006] Existing technologies include studies that apply interpretability technologies such as Grad-CAM to lesion annotation in medical images (e.g., Selvaraju RR, et al. Grad-CAM: Visual Explanations from Deep Networks via Gradent-based Localization. CVPR 2017), as well as user emotion recognition solutions based on touch behavior analysis. However, none of these technologies address insurance interaction scenarios or solve the problem of non-invasive integration of clinical informed consent forms and insurance recommendations.

[0007] Therefore, there is an urgent need for an interactive interface generation method that can non-intrusively embed safeguard recommendations into the informed consent process and achieve a compliance closed loop through risk explanation and pressure blocking mechanisms. Summary of the Invention

[0008] To address the aforementioned technical problems, this invention provides a clinical informed consent form embedded insurance interaction method, comprising: Acquire the patient's multimodal clinical data and convert the multimodal clinical data into a standardized feature tensor; Clinical risk attribution calculations are performed based on the standardized feature tensor to generate a risk heat map distribution, and risk interpretation anchors are rendered in the area next to specific risk clauses in the clinical informed consent form. In response to the triggering operation of the risk interpretation anchor point, the semantic consistency of the preset protection description and the original text of the insurance terms is checked. After the check passes, the protection description is rendered in the form of a linear extension module in the side area of ​​the risk interpretation anchor point. Collect user interaction sequences on the front-end interface, calculate a real-time stress index based on the interaction sequences, generate a circuit breaker command to lock the insurance interaction entry on the front-end interface when the real-time stress index is greater than or equal to a preset trigger threshold, and clear the interaction feature buffer when a clinical pathway switching signal is received. Obtain user dwell time data for the risk interpretation anchor point, and encrypt and store the dwell time data, the interaction sequence, the semantic consistency verification result, and the real-time stress index.

[0009] This invention also relates to an embedded insurance interaction system for clinical informed consent forms, comprising: The medical data processing module is used to acquire patients' multimodal clinical data and convert the multimodal clinical data into standardized feature tensors; The risk attribution calculation module is used to perform clinical risk attribution calculation based on the standardized feature tensor, generate a risk heat map distribution, and render risk interpretation anchor points in the area next to specific risk clauses in the clinical informed consent form. The semantic anchoring mapping module is used to respond to the triggering operation of the risk interpretation anchor point, and to perform semantic consistency verification on the preset protection description and the original text of the insurance terms. After the verification is passed, the protection description is rendered in the form of a linear extension module in the side area of ​​the risk interpretation anchor point. The emotion monitoring and blocking module is used to collect the user's interaction sequence on the front-end interface, calculate the real-time stress index based on the interaction sequence, generate a circuit breaker command to lock the insurance interaction entry on the front-end interface when the real-time stress index is greater than or equal to a preset trigger threshold, and clear the interaction feature buffer when a clinical pathway switching signal is received. The compliance record storage module is used to acquire user dwell time data for the risk interpretation anchor point, and encrypt and store the dwell time data, the interaction sequence, the semantic consistency verification result, and the real-time stress index.

[0010] Compared with the prior art, the beneficial effects of the present invention are as follows: Compared to traditional pop-up insurance interaction solutions, this invention has significant advantages in reducing patient reading interruption rates, minimizing accidental insurance purchases, improving understanding of risk clauses, and curbing irrational insurance purchase behavior. The following specific embodiments further illustrate this.

[0011] This invention abandons the traditional pop-up ad interference model, designing the insurance interaction as a linear logical extension of the clinical informed consent form content. It renders protection suggestions in situ alongside the risk clauses, achieving simultaneous on-screen notification and recommendation, and a continuous cognitive interaction. Coupled with an emotion-blocking mechanism based on interaction characteristics, the system can automatically shut down the insurance application process when the patient experiences excessive psychological stress or cognitive overload, effectively preventing excessive sales pitches and ensuring that the interaction strictly complies with medical ethics and industry regulatory requirements.

[0012] This invention utilizes deep learning interpretability technologies such as Grad-CAM to directly link actuarial risk assessment with medically visible lesions or clinical indicators in anatomical space, generating risk topography maps and interactive anchor points. This gives the black-box risk model an intuitive medical interpretation. Patients can clearly identify the specific connection between the coverage and their own condition, significantly reducing the trust barrier caused by information asymmetry and improving the quality of informed decision-making.

[0013] This invention employs a semantic alignment mechanism to ensure a high degree of similarity between user-friendly language and the original insurance legal terms, preventing descriptive discrepancies. Simultaneously, the system fully records micro-behaviors during the interaction process, such as the duration of dwell time at risk anchor points, interaction depth, trajectory entropy, and real-time stress index, generating an immutable encrypted evidence package. Compared to traditional checkbox confirmations, this provides more compelling behavioral trace data, significantly enhancing the substantive compliance of the application process and the ability to trace and prove the source afterward.

[0014] This invention introduces a parameter reset unit that automatically clears the feature buffer when switching between different diagnostic and treatment tasks, avoiding feature contamination between heterogeneous data. This self-closing design enables the method to maintain stable operation in various differentiated clinical scenarios, from routine physical examinations to complex surgeries, demonstrating good versatility and robustness. Attached Figure Description

[0015] Figure 1 This is a system structure diagram provided for an embodiment of the present invention. Detailed Implementation

[0016] The terms first, second, third, fourth, etc. (if present) in the specification, claims, and accompanying drawings of this invention are used to distinguish similar objects and are not necessarily used to describe a particular order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments described herein can be implemented in orders other than those illustrated or described herein. Furthermore, the terms include or have, and any variations thereof, are intended to cover a non-exclusive inclusion; for example, a process, method, system, product, or apparatus that includes a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or apparatus.

[0017] A clinical informed consent form embedded insurance interaction method includes: Acquire the patient's multimodal clinical data and convert the multimodal clinical data into a standardized feature tensor; Clinical risk attribution calculations are performed based on the standardized feature tensor to generate a risk heat map, and risk interpretation anchors are rendered in the area next to specific risk clauses of the clinical informed consent form displayed on the front-end interface. In response to the triggering operation of the risk interpretation anchor point, the semantic consistency of the preset protection description and the original text of the insurance terms is checked. After the check passes, the protection description is rendered in the form of a linear extension module in the side area of ​​the risk interpretation anchor point. Collect user interaction sequences on the front-end interface, calculate a real-time stress index based on the interaction sequences, generate a circuit breaker command to lock the insurance interaction entry on the front-end interface when the real-time stress index is greater than or equal to a preset trigger threshold, and clear the interaction feature buffer when a clinical pathway switching signal is received. Obtain user dwell time data for the risk interpretation anchor point, and encrypt and store the dwell time data, the interaction sequence, the semantic consistency verification result, and the real-time stress index.

[0018] As one implementation, the feature map of the calculated feature tensor is weighted according to the gradient contribution weights of the preset risk category, the feature map is weighted and summed according to the gradient contribution weights, and the weighted sum is input into an activation function for processing to output a heat distribution matrix, including: Extract multiple feature maps from the last convolutional layer of the feature tensor; Calculate the partial derivative of the predicted score of the preset risk category with respect to the activation value at each spatial location on each feature map; The partial derivatives of the activation values ​​at each spatial location on each feature map are accumulated along the spatial dimension and divided by the total number of pixels in the feature map to obtain the gradient contribution weight of each feature map. Each feature map is multiplied by its corresponding gradient contribution weight and then summed. The summation result is then input into the modified linear unit activation function for processing, and the thermal distribution matrix is ​​output.

[0019] As one implementation method, the semantic consistency check between the preset coverage description and the original insurance terms includes: The description of the coverage and the original text of the insurance terms are respectively input into the pre-trained language model; Obtain the first feature vector of the guarantee description output by the pre-trained language model, and the second feature vector of the original text of the insurance terms; Calculate the dot product of the first eigenvector and the second eigenvector; Calculate the product of the magnitude of the first eigenvector and the magnitude of the second eigenvector; Divide the dot product by the product to obtain the cosine similarity, and determine whether the semantic consistency check passes.

[0020] As one implementation method, the step of calculating the interaction trajectory entropy and touch jitter frequency based on the touch point position and timestamp in the interaction sequence, and calculating the real-time pressure index based on the interaction trajectory entropy and the touch jitter frequency, includes: Extract the positions of multiple touch points and their corresponding timestamps from the interaction sequence within the time sliding window; The interaction trajectory entropy is calculated based on the distribution dispersion of the touch point positions within the time sliding window. The touch jitter frequency is calculated based on the coordinate displacement change of the touch point position between adjacent timestamps; Obtain the first preset weighting coefficient and the second preset weighting coefficient; The real-time pressure index is obtained by adding the product of the interaction trajectory entropy and the first preset weighting coefficient to the product of the touch jitter frequency and the second preset weighting coefficient.

[0021] As one implementation, after the verification passes, following the rendering of the protection description in the form of a linear extension module in the area adjacent to the risk explanation anchor point, the method further includes: Construct a mapping linked list containing risk area parameters, the description of the coverage, and the index identifiers of the original text of the insurance terms; The process of concatenating the dwell time data, the interaction sequence, the semantic consistency verification result, and the real-time stress index, generating a digital digest using a hash algorithm, attaching a system timestamp, encapsulating it into a storage package, and storing it includes: The dwell time data, the interaction sequence, the semantic consistency verification result, the real-time pressure index, and the data of the mapping list are concatenated. The hash algorithm is called on the concatenated string to generate the digital digest; An asymmetric encryption algorithm is invoked to encrypt and encapsulate the digital digest and the system timestamp to generate the evidence storage package, which is then sent to a distributed storage node for persistent storage.

[0022] like Figure 1 The present invention also relates to an embedded insurance interaction system for clinical informed consent forms, characterized in that the system comprises: The medical data processing module is used to acquire patients' multimodal clinical data and convert the multimodal clinical data into standardized feature tensors; The risk attribution calculation module is used to perform clinical risk attribution calculation based on the standardized feature tensor, generate a risk heat map distribution, and render risk interpretation anchor points in the area next to specific risk clauses in the clinical informed consent form. The semantic anchoring mapping module is used to respond to the triggering operation of the risk interpretation anchor point, and to perform semantic consistency verification on the preset protection description and the original text of the insurance terms. After the verification is passed, the protection description is rendered in the form of a linear extension module in the side area of ​​the risk interpretation anchor point. The emotion monitoring and blocking module is used to collect the user's interaction sequence on the front-end interface, calculate the real-time stress index based on the interaction sequence, generate a circuit breaker command to lock the insurance interaction entry on the front-end interface when the real-time stress index is greater than or equal to a preset trigger threshold, and clear the interaction feature buffer when a clinical pathway switching signal is received. The compliance record storage module is used to acquire user dwell time data for the risk interpretation anchor point, and encrypt and store the dwell time data, the interaction sequence, the semantic consistency verification result, and the real-time stress index.

[0023] The present invention will be further described below with reference to the accompanying drawings and embodiments.

[0024] This invention provides an embedded insurance interaction system for clinical informed consent forms, including: a medical data processing module, a risk attribution calculation module, a semantic anchoring mapping module, an emotion monitoring and blocking module, and a compliance record storage module.

[0025] 1. Medical Data Processing Module The medical data processing module is used to collect or receive patients' multimodal clinical data in real time. This data includes: (1) Structured data: such as laboratory test index sequences and time-series sampling of vital signs (heart rate, blood pressure).

[0026] (2) Unstructured data: such as medical images (CT, MRI, ultrasound reports) and electronic medical record texts.

[0027] The terminal is equipped with a preprocessing unit that performs data anonymization and normalization, mapping medical data of different dimensions into high-dimensional feature tensors. This provides a normalized input benchmark for subsequent algorithm analysis.

[0028] 2. Risk Attribution Calculation Module The risk attribution calculation module is the core of achieving risk interpretability, and it uses gradient backpropagation technology to identify high-risk areas.

[0029] The detailed technical logic of this unit is as follows: (1) Category activation mapping calculation: The system uses a convolutional neural network to extract image features, targeting specific actuarial risk categories. (e.g., postoperative bleeding risk), calculate the last convolutional layer's... Feature map In spatial location gradient contribution weights : in, To predict the score, This represents the total number of pixels in the feature map.

[0030] (2) Spatial attribution superposition: By weighted summation of feature maps and ReLU activation, a thermal distribution matrix representing the sources of risk is generated. : (3) Interactive anchor point generation: The system generates interactive anchor points. Perform binarization segmentation and identify weight thresholds exceeding [a certain threshold]. The connected components are used to generate risk interpretation anchor points at the corresponding coordinates of the front-end UI, directly linking the actuarial logic with medically visible lesions. As one implementation method, the threshold of the binarized segmentation... The pixel values ​​of the risk thermal distribution are adaptively determined based on their statistical characteristics, for example, by taking the top 20% quantiles of the pixel values ​​in the thermal distribution matrix.

[0031] 3. Semantic Anchor Mapping Module The semantic anchoring mapping module is used to translate actuarial terms into a more accessible, ethically sound explanation. Its core function is to configure insurance options as a linear extension of the clinical informed consent form, achieving in-situ rendering. The specific execution logic is as follows: (1) Fragmented semantic retrieval: Based on the risk type output by the risk attribution calculation module, retrieve matching colloquial descriptive phrases from the preset semantic database. .

[0032] (2) Compliance and consistency verification: Calculate the descriptive statements using pre-trained models such as BERT. According to the original text of the corresponding legal clauses The cosine similarity. Let... For its embedding vector, it must satisfy: in The preset compliance threshold ranges from 0.90 to 0.98, with 0.95 being preferred.

[0033] (3) In-situ embedded rendering: The system constructs a linked list containing [risk area - explanatory language - legal index]. Unlike traditional pop-ups, the output unit renders insurance suggestions in parallel next to the relevant risk clauses of the informed consent form, ensuring the linear continuity between insurance interaction and clinical treatment pathways.

[0034] 4. Emotional monitoring and blocking module The emotion monitoring and blocking module is used to sense the psychological stress of patients when making decisions, and the core of it is to implement the emotion blocking mechanism.

[0035] This unit is detailed as follows: (1) Behavior vector acquisition: Real-time acquisition of user interaction sequences ,in For the touch point location, For timestamps.

[0036] (2) Stress Index Inference: The stress index inference unit performs real-time inference based on the collected micro-behavioral feature vectors. To quantify the impact of anxiety state on interaction stability, this embodiment adopts the following stress assessment model: in, This is a real-time stress index; Entropy of user interaction trajectory is used to characterize the degree of disorder in click or swipe paths; The jitter frequency of the touch point is used to characterize the interaction instability caused by muscle tremors; and These are the preset weighting coefficients.

[0037] (3) Circuit breaker triggering logic: The preset triggering threshold Dynamically configured according to clinical scenarios, in informed consent settings during surgery. Take 0.75, in a routine physical examination scenario Take 0.85. If determined... (Trigger threshold) When the system determines that the user is in a state of anxiety or cognitive overload, the parameter reset unit immediately generates and issues a circuit breaker command, physically blocking the insurance entry and redirecting to the guidance page to achieve mandatory protection against irrational decisions.

[0038] 5. Compliance-compliant record-keeping storage module The compliance traceability storage module is used to persistently store the evidence records of the entire process. The specific logic is as follows: (1) Interactive evidence encapsulation: The system collects the user's dwell time at the risk anchor point. And the depth of interaction. A unique digital digest of the informative-understanding behavior sequence is generated using a hash algorithm and then timestamped. It forms an immutable evidence package, providing stronger evidentiary capabilities than traditional checkboxes.

[0039] (2) Semantic Consistency Index Storage: Stores the mapping linked list generated by the semantic anchoring mapping module and records the front-end display scripts. With corresponding actuarial terms The semantic consistency verification results (cosine similarity score) serve as the audit basis for fully informing and eliminating trust barriers.

[0040] (3) Distributed storage and encryption: Asymmetric encryption technology is used to process the data. In the event of a subscription dispute or regulatory audit, the interaction scenario at the time is restored through a specific authorized private key, ensuring that the entire process complies with regulatory requirements and medical ethics.

[0041] Specific workflow (such as) Figure 1 (As shown) The workflow provided by the embodiments of the present invention is as follows (e.g.) Figure 1 As shown), the aim is to achieve risk visualization, semantic alignment, and sentiment closure through multi-unit collaboration: Step S1: Clinical Data Acquisition and Preliminary Packaging The medical data processing module retrieves the patient's multimodal clinical data (such as images and laboratory indicators), performs data anonymization and normalization preprocessing to create standardized feature tensors, and sends them to the risk attribution calculation unit.

[0042] Step S2: Risk Attribution and Visual Anchor Generation The risk attribution calculation module performs Grad-CAM feature extraction and thermal mapping to calculate the gradient contribution weight of clinical features to a specific risk category.

[0043] Thermal mapping unit: The system converts the calculated risk thermal distribution into anatomical spatial weights, and the thermal mapping unit performs in-situ rendering next to the risk clauses of the informed consent form in the front-end UI to generate risk interpretation anchor points based on the spatial weights. This achieves initial alignment between the protection recommendations and the medical disclosure content.

[0044] Step S3: Interaction Triggering and Semantic Consistency Verification When a user focuses on or clicks on a rendered risk point while reading the informed consent form, a specific anchor point display request is triggered.

[0045] The semantic anchoring mapping module performs a statement consistency check by calculating the vector similarity between the colloquial description and the original legal clauses using the BERT model.

[0046] If the compliance threshold is met The informed consent form interface will then output a compliant fragmented guarantee description in the form of a linear extension module to ensure that the interaction process does not deviate from the treatment notification path.

[0047] Step S4: Real-time monitoring of behavioral characteristics and stress simulation During the user's browsing protection description, the emotion monitoring and blocking module uploads micro-behavioral feature vectors in real time, such as user interaction trajectory entropy, touch jitter frequency, and page dwell time.

[0048] The stress index extrapolation unit dynamically extrapolates the real-time stress index based on the collected compensatory characteristics. .

[0049] Step S5: Real-time decision-making and branching The system determines in real time whether the pressure index is below the preset safety threshold. .

[0050] Branch A (Activation Path): If satisfied The interface remains active and subscriptions are allowed. The compliance tracking and storage module simultaneously captures in-depth user behavior data at risk anchor points, then encapsulates it into an encrypted evidence package (containing interaction data, term IDs, and stress indices, etc.), and generates a successful evidence storage confirmation.

[0051] Branch B (Circuit Breaker Path): If satisfied Upon receiving the notification, the system sends a circuit breaker command, immediately locking the insurance application portal and forcibly redirecting the user to a doctor's consultation guidance page or a psychological recovery page. Simultaneously, the system records the cause of the blocking event and metadata related to the abnormal behavior, thus implementing medical ethics protection from a technical perspective.

[0052] Step S6: Complete feedback and system reset The system provides feedback to users and terminals on the final processing results.

[0053] After the process is completed, the parameter reset unit resets the sliding window parameters, clears the feature buffer of the current interaction cycle, ensures that the system state rolls back to the initial level, and prepares to respond to the interaction request of the next clinical pathway.

[0054] Example 1: Scenario of Informed Consent Form for Percutaneous Coronary Intervention (PCI) Stent Surgery Patient Zhang, male, 58 years old, was admitted to the hospital due to "recurrent chest pain for 3 months, worsening for 2 days". He was diagnosed with coronary atherosclerotic heart disease and unstable angina. Coronary angiography showed 85% stenosis in the proximal segment of the left anterior descending artery, and PCI stent implantation was planned.

[0055] Step S1: Clinical Data Acquisition and Standardization The medical data processing module retrieves the patient's multimodal clinical data, including: (1) Structured data: myocardial injury markers (troponin I 0.15 ng / mL, slightly elevated), blood lipid profile (LDL-C 3.2 mmol / L), vital signs (blood pressure 142 / 88 mmHg, heart rate 76 beats / min); (2) Unstructured data: coronary angiography DICOM images (showing localized stenosis in the proximal segment of the left anterior descending artery), electronic medical record text (including diagnosis of "unstable angina" and history of "grade 3 hypertension").

[0056] The preprocessing unit performs anonymization on the above data (removing direct identifiers such as name and ID number, while retaining indirect identifiers such as age and gender for risk stratification) and normalization: laboratory indicators are mapped to the [0,1] interval according to clinical reference ranges, image data is uniformly scaled to 512×512 pixels, and text data is encoded after entity extraction using a medical NER model. Finally, the multimodal data is fused into a 768-dimensional normalized feature tensor.

[0057] Step S2: Risk Attribution and Anchor Rendering The risk attribution calculation module inputs standardized feature tensors into the pre-trained coronary intervention risk prediction model. The model outputs three preset risk categories and their predicted scores: intraoperative acute occlusion risk (0.23), postoperative in-stent restenosis risk (0.41), and contrast agent nephropathy risk (0.18).

[0058] Taking "postoperative in-stent restenosis risk" as the category of interest, Grad-CAM attribution calculation was performed: 512 feature maps (size 14×14×512) of the last convolutional layer of the model were extracted, the partial derivatives of the category prediction scores with respect to the activation values ​​of each feature map spatial location were calculated, the gradient contribution weights were obtained after spatial dimension accumulation and normalization, and the weighted superposition was then activated by ReLU to output the risk thermal distribution matrix (size 14×14).

[0059] The thermal distribution matrix was binarized and segmented (threshold τ was set to the top 20% of pixel values), and connected components were extracted. After screen coordinate mapping (14×14→1920×1200 resolution), risk interpretation anchors were rendered in the area next to the "Postoperative Complications" clause of the informed consent form (around coordinates x=680, y=920). The anchor icon was a red pulsed dot, indicating that there was an interpretable risk association for this clause.

[0060] Step S3: Semantic verification and guarantee of description rendering The patient clicks on the risk explanation anchor point, triggering the semantic anchoring mapping module. The system retrieves the matching coverage description S from the insurance semantic database: "If in-stent restenosis occurs within one year after the surgery and requires re-interventional treatment, this insurance covers the medical expenses for the second surgery and hospitalization allowance."

[0061] The script and the corresponding insurance clause text T (clause number CL-2025-PCI-03) were input into the BERT-base pre-trained model, and 768-dimensional feature vectors were extracted respectively. , Calculate the cosine similarity: because (Preset similarity threshold 0.95), semantic consistency verification passed. The system renders the guarantee description in the form of a linearly extended module to the right of the anchor point. The module width is 35% of the screen width, and it is distinguished by a light blue background and a white background of the informed consent form, but without obscuring the original text content, achieving non-intrusive integration.

[0062] Steps S4-S5: Pressure Monitoring and Branch Decisions During the patient's reading of the protection description, the emotion monitoring and blocking module collected the interaction sequence in real time. The time sliding window was set to 5 seconds, and the collected touch point sequence was: [(682,925,0ms), (685,928,120ms), (690,935,280ms), (688,930,450ms), (1200,800,1200ms)...].

[0063] Calculate the interaction trajectory entropy: Touch points are concentrated within the anchor point region (an 80-pixel diameter circle), with an entropy value of Entropy(U) = 0.32; Touch jitter frequency: The average displacement of adjacent point coordinates is 4.2 pixels / 100ms, and after normalization, f_jitter = 0.15. Using weighting coefficients w_1 = 0.6 and w_2 = 0.4, calculate the real-time pressure index: I_p = 0.6 × 0.32 + 0.4 × 0.15 = 0.252 Since I_p = 0.252 < Ω (the preset trigger threshold for PCI surgery scenarios is 0.75), the system determines that the patient is in a normal cognitive state and keeps the insurance interaction entry active. The compliance record storage module synchronously records: anchor point dwell time D = 8.7 seconds, interaction sequence, semantic verification result (passed, similarity 0.96), and real-time stress index (0.252).

[0064] Step S6: Evidence Preservation and Reset After the patient confirms the insurance application, the compliance record storage module concatenates the above data with the mapping list (risk area: proximal left anterior descending branch; insurance statement ID: S-PCI-02; clause index: CL-2025-PCI-03), calls the SHA-256 algorithm to generate a digital digest, attaches the system timestamp (2025-03-15T09:23:17.342Z), encapsulates it into a storage packet using RSA-2048 asymmetric encryption, and sends it to the distributed storage node (IPFS network) for persistent storage.

[0065] When the clinical pathway switches to the postoperative follow-up module, the parameter reset unit clears the current interactive feature buffer and the system state rolls back to the initial level.

[0066] Example 2: Routine Annual Physical Examination Scenario Patient Li, female, 35 years old, underwent a routine annual health check-up and had no history of chronic diseases. The medical data processing module retrieved structured / unstructured data including complete blood count, liver and kidney function tests, and abdominal ultrasound, which were preprocessed to generate standardized feature tensors.

[0067] The risk attribution calculation module outputs preset risk category prediction scores all below 0.15 (e.g., "risk of malignancy in thyroid nodules" 0.08, "risk of progression of fatty liver" 0.12), which the system classifies as low-risk scenarios. The risk heatmap distribution is binarized and segmented (threshold). After taking the top 20% of pixel values, there are no connected regions with an area exceeding 100 pixels. The informed consent interface does not render the risk explanation anchor point, and only retains the basic insurance entry (a non-intrusive floating button with 30% transparency).

[0068] During the patient's browsing of the medical examination report, the emotion monitoring and blocking module detected the interaction trajectory entropy. Touch shake frequency Calculate the real-time pressure index =0.68. This is due to the preset trigger threshold in routine physical examination scenarios. The insurance interaction portal remains active.

[0069] This embodiment demonstrates the adaptive capability of the present invention in different clinical scenarios (surgery / physical examination): in high-risk surgical scenarios, risk interpretation anchors are actively pushed, while in low-risk physical examination scenarios, the entry point is silently retained to avoid excessive interference.

[0070] The two embodiments above demonstrate that the present invention achieves accurate integration of risk visualization and insurance recommendation in high-risk scenarios such as PCI surgery through scene-adaptive risk anchor rendering, semantic consistency verification, and pressure circuit breaker mechanism, while avoiding excessive interference in low-risk scenarios such as routine physical examinations, thus verifying the robustness and versatility of the method under different clinical pathways.

[0071] The above embodiments are only used to illustrate the technical solutions of the present invention, and are not intended to limit it. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A clinical informed consent form embedded insurance interaction method, characterized in that, The method includes: Acquire the patient's multimodal clinical data and convert the multimodal clinical data into a standardized feature tensor; Clinical risk attribution calculations are performed based on the standardized feature tensor to generate a risk heat map distribution, and risk interpretation anchors are rendered in the area next to specific risk clauses in the clinical informed consent form. In response to the triggering operation of the risk interpretation anchor point, the semantic consistency of the preset protection description and the original text of the insurance terms is checked. After the check passes, the protection description is rendered in the form of a linear extension module in the side area of ​​the risk interpretation anchor point. Collect user interaction sequences on the front-end interface, calculate a real-time stress index based on the interaction sequences, generate a circuit breaker command to lock the insurance interaction entry on the front-end interface when the real-time stress index is greater than or equal to a preset trigger threshold, and clear the interaction feature buffer when a clinical pathway switching signal is received. Obtain user dwell time data for the risk interpretation anchor point, and encrypt and store the dwell time data, the interaction sequence, the semantic consistency verification result, and the real-time stress index.

2. The method according to claim 1, characterized in that, The step of performing clinical risk attribution calculation based on the standardized feature tensor to generate a risk heatmap distribution includes: Extract multiple feature maps from the last convolutional layer of the standardized feature tensor; Calculate the partial derivative of the predicted score for each preset risk category with respect to the activation value at each spatial location on each feature map; The partial derivatives of the activation values ​​at each spatial location on each feature map are accumulated along the spatial dimension and divided by the total number of pixels in the feature map to obtain the gradient contribution weight of each feature map. Each feature map is multiplied by its corresponding gradient contribution weight and then summed. The summation result is then input into the modified linear unit activation function for processing, and the risk heat distribution is output.

3. The method according to claim 1, characterized in that, The semantic consistency check between the preset coverage description and the original insurance terms includes: The protection description and the original text of the insurance terms are respectively input into a pre-trained language model to obtain the first feature vector of the protection description and the second feature vector of the original text of the insurance terms output by the pre-trained language model. Calculate the cosine similarity between the first feature vector and the second feature vector; When the cosine similarity is greater than or equal to a preset similarity threshold, the semantic consistency check is deemed to have passed; wherein the preset similarity threshold ranges from 0.90 to 0.

98.

4. The method according to claim 1, characterized in that, The calculation of the real-time stress index based on the interaction sequence includes: Extract the positions of multiple touch points and their corresponding timestamps from the interaction sequence within the time sliding window; The interaction trajectory entropy is calculated based on the distribution dispersion of the touch point positions within the time sliding window. The touch jitter frequency is calculated based on the coordinate displacement change of the touch point position between adjacent timestamps; The real-time pressure index is obtained by adding the product of the interaction trajectory entropy and the first preset weighting coefficient to the product of the touch jitter frequency and the second preset weighting coefficient.

5. The method according to claim 1, characterized in that, The encrypted evidence storage includes: Construct a mapping linked list containing risk area parameters, the description of the coverage, and the index identifiers of the original text of the insurance terms; The dwell time data, the interaction sequence, the semantic consistency verification result, the real-time pressure index, and the data of the mapping list are concatenated. A hash algorithm is called on the concatenated string to generate a digital digest. An asymmetric encryption algorithm is then called to encrypt and encapsulate the digital digest and the system timestamp to generate a certificate storage package. The certificate storage package is then sent to a distributed storage node for persistent storage.

6. A clinical informed consent form embedded insurance interaction system, characterized in that, The system includes: The medical data processing module is used to acquire patients' multimodal clinical data and convert the multimodal clinical data into standardized feature tensors; The risk attribution calculation module is used to perform clinical risk attribution calculation based on the standardized feature tensor, generate a risk heat map distribution, and render risk interpretation anchor points in the area next to specific risk clauses in the clinical informed consent form. The semantic anchoring mapping module is used to respond to the triggering operation of the risk interpretation anchor point, and to perform semantic consistency verification on the preset protection description and the original text of the insurance terms. After the verification is passed, the protection description is rendered in the form of a linear extension module in the side area of ​​the risk interpretation anchor point. The emotion monitoring and blocking module is used to collect the user's interaction sequence on the front-end interface, calculate the real-time stress index based on the interaction sequence, generate a circuit breaker command to lock the insurance interaction entry on the front-end interface when the real-time stress index is greater than or equal to a preset trigger threshold, and clear the interaction feature buffer when a clinical pathway switching signal is received. The compliance record storage module is used to acquire user dwell time data for the risk interpretation anchor point, and encrypt and store the dwell time data, the interaction sequence, the semantic consistency verification result, and the real-time stress index.

7. The clinical informed consent embedded insurance interaction system according to claim 6, characterized in that, The risk attribution calculation module is specifically used for: Extract multiple feature maps from the last convolutional layer of the standardized feature tensor; Calculate the partial derivative of the predicted score of the preset risk category with respect to the activation value at each spatial location on each feature map; The partial derivatives of the activation values ​​at each spatial location on each feature map are accumulated along the spatial dimension and divided by the total number of pixels in the feature map to obtain the gradient contribution weight of each feature map. Each feature map is multiplied by its corresponding gradient contribution weight and then summed. The summation result is then input into the modified linear unit activation function for processing, and the thermal distribution matrix is ​​output.

8. The embedded insurance interaction system for clinical informed consent according to claim 6, characterized in that, The semantic anchoring mapping module is specifically used for: The description of the coverage and the original text of the insurance terms are respectively input into the pre-trained language model; Obtain the first feature vector of the guarantee description output by the pre-trained language model, and the second feature vector of the original text of the insurance terms; Calculate the cosine similarity between the first feature vector and the second feature vector; When the cosine similarity is greater than or equal to a preset similarity threshold, the semantic consistency check is deemed to have passed; wherein the preset similarity threshold ranges from 0.90 to 0.

98.

9. The embedded insurance interaction system for clinical informed consent according to claim 6, characterized in that, The emotion monitoring and blocking module is specifically used for: Extract the positions of multiple touch points and their corresponding timestamps from the interaction sequence within the time sliding window; The interaction trajectory entropy is calculated based on the distribution dispersion of the touch point positions within the time sliding window. The touch jitter frequency is calculated based on the coordinate displacement change of the touch point position between adjacent timestamps; The real-time pressure index is obtained by adding the product of the interaction trajectory entropy and the first preset weighting coefficient to the product of the touch jitter frequency and the second preset weighting coefficient.

10. The embedded insurance interaction system for clinical informed consent according to claim 6, characterized in that: The semantic anchoring mapping module is also used to construct a mapping linked list containing risk area parameters, the protection description language, and the index identifier of the original text of the insurance terms; The compliance trace storage module is specifically used to concatenate the dwell time data, the interaction sequence, the semantic consistency verification result, the real-time stress index, and the data of the mapping list. The hash algorithm is called on the concatenated string to generate the digital digest; the asymmetric encryption algorithm is called to encrypt and encapsulate the digital digest and the system timestamp to generate the evidence storage package, and the evidence storage package is sent to the distributed storage node for persistent storage.

Images