Follow-up visit management system for patients with inflammatory bowel diseases

Abstract

The invention discloses a follow-up visit management system for an inflammatory bowel disease patient, and relates to the technical field of medical management. Comprising a patient management module which is in dynamic butt joint with a hospital information system, synchronizes patient data and endoscopic examination results in real time through an HL7 protocol, and establishes an electronic file containing an inflammatory bowel disease specificity tag; the diagnosis and treatment result module is used for storing structured data with temporal markers, including medication time sequences recorded by three groups of timestamps, administration routes and drug categories, auxiliary examination results accompanied with digestive tract pathological section images and hidden space feature vectors extracted from the images through a convolutional neural network; and the diet module is used for carrying out multi-level classification on diet photos uploaded by the patient through an image recognition unit based on MobileNetV3, and outputting food types and intake estimated values. According to the invention, by integrating the multi-dimensional data of the patient, the disease development trend is predicted, and personalized follow-up visit management is provided.

Description

Technical Field

[0001] The present invention relates to the technical field of medical management, and particularly to a follow-up management system for patients with inflammatory bowel disease. Background Art

[0002] Inflammatory Bowel Disease (IBD) is a chronic and recurrent intestinal inflammatory disease, mainly including two types, namely Ulcerative Colitis and Crohn's Disease. Epidemiological investigation and research have found that the incidence of this disease in China is increasing rapidly year by year. As a chronic and recurrent intestinal inflammatory disease, the condition of IBD is prone to change over time, often showing remission, recurrence or progression, and may also cause complications such as intestinal stricture, fistula formation, and canceration. Due to the long course and high recurrence rate of IBD, patients need long-term follow-up and management. The traditional follow-up methods mainly rely on manual records and telephone follow-up, and there are the following problems: inconvenient data management: data such as patients' medical records, medication details, and examination results are scattered and difficult to integrate and analyze; low follow-up efficiency: manual follow-up is time-consuming and laborious, and it is difficult to cover all patients; insufficient personalized support: lack of dynamic monitoring of patients' conditions and personalized intervention suggestions. Insufficient data utilization: a large amount of clinical data is not fully utilized and cannot effectively predict the development trend of the condition. Summary of the Invention

[0003] In view of the above deficiencies in the prior art, the present invention provides a follow-up management system for patients with inflammatory bowel disease, which integrates multi-dimensional data of patients, predicts the development trend of the condition, and provides personalized follow-up management, thereby improving the follow-up efficiency and management effect.

[0004] The specific technical solutions are as follows:

[0005] A follow-up management system for patients with inflammatory bowel disease, comprising:

[0006] A patient management module, which is dynamically docked with the hospital information system, synchronizes patient data and endoscopic examination results in real time through the HL7 protocol, and establishes an electronic file containing specific tags for inflammatory bowel disease;

[0007] A diagnosis and treatment result module, which is used to store structured data with temporal markers, including the medication time series recorded by timestamp, administration route, and drug category, the auxiliary examination results accompanied by gastrointestinal pathological section images, and the hidden space feature vectors extracted from the images through a convolutional neural network;

[0008] Diet module, which performs multi-level classification on the diet photos uploaded by patients through an image recognition unit based on MobileNetV3, and outputs the estimated values of food types and intake; dynamically adjusts the nutrient weight constraints according to the Crohn's disease activity index; parses the symptom description text submitted by patients through natural language processing technology, extracts key symptom entities and maps them to the SES-CD standard system;

[0009] Analysis module, including an inflammation activity index generation unit, a dynamic model fusion unit, and a medical constraint optimization unit. The inflammation activity index generation unit is used to calculate the composite biomarker index:

[0010]

[0011] where ΔHb is the weekly change in hemoglobin;

[0012] The dynamic model fusion unit is used to integrate the outputs of random forest and LSTM, and the weight coefficients are dynamically adjusted based on the patient's real-time clinical status:

[0013]

[0014] RiskScore = α·P RF +(1-α)·sigmoid(A LSTM / 3)

[0015] where RiskScore represents the risk score, P RF represents the output of the random forest, and A LSTM represents the output of the LSTM;

[0016] The medical constraint optimization unit introduces clinical safety boundary conditions during the LSTM training stage:

[0017] L total = BCE Loss + 0.4·max(0, SERSI pred - SERSI clin + 0.3)

[0018] where L total represents the total loss during the LSTM training stage, BCELoss represents the binary cross-entropy loss, SERSI pred represents the predicted value of the composite biomarker index, and SERSI clin represents the clinical value of the composite biomarker index;

[0019] Follow-up reminder module, which dynamically generates a hierarchical reminder strategy according to the RiskScore value. When RiskScore ≥ 0.7, a red alert is triggered, and a follow-up visit plan is generated according to the interval days = 45 - 5×SERSI; when 0.4 ≤ RiskScore < 0.7, the reminder interval is set according to 2.3 times the half-life of the serum drug concentration.

[0020] Display module, which generates a predictive visualization interface with interpretability annotations.

[0021] Database, which adopts a time series database optimized for columnar storage and performs interpolation compensation on biomarker data.

[0022] In the above solution, further, the display module includes using the Integrated Gradients algorithm to highlight the key features affecting the risk score; and showing the distribution evolution of different patient subgroups in the latent space with a three-dimensional manifold graph.

[0023] In the above solution, further, the follow-up reminder module supports multiple reminder methods, including text messages, emails, and APP notifications.

[0024] In the above solution, further, the diet module also includes a diet analysis function, which generates personalized diet suggestions according to the patient's condition and nutritional needs, combined with diet records.

[0025] In the above solution, further, the analysis module generates personalized intervention suggestions according to the prediction results, including adjusting the medication plan, increasing the follow-up frequency, and hospitalization.

[0026] In the above solution, further, it also includes a multidisciplinary collaboration module for information sharing and collaboration among doctors, dietitians, and psychiatrists.

[0027] In the above solution, further, the analysis module performs time decay compensation on the serum albumin and fecal calprotectin concentrations:

[0028] >C corrected =C raw ·exp(-λ∣t sample -t endoscopy ∣)>

[0029] Where λ = 0.03 / hour, t endoscopy represents the time of the most recent colonoscopy; C raw represents the original concentration of the biomarker;

[0030] Fuse the recurrence probability P RF output by the random forest with the disease activity ALSTM predicted by LSTM:

[0031] >RiskScore = ω·PRF +(1 - ω)·tanh(A LSTM / 10)>

[0032] The weight coefficient ω is dynamically adjusted according to the patient's current nutritional index:

[0033]

[0034] Among them, BMI represents the patient's current body mass index.

[0035] Compared with the prior art, the beneficial effects of the present invention are as follows:

[0036] The present invention accurately solves the problem of heterogeneous management of inflammatory bowel disease, proposes the SERSI composite index to reflect the dynamics of the intestinal microenvironment, and the comparative study shows that its correlation with the endoscopic score r = 0.82; develops a medical constraint optimization loss function to ensure that the model prediction does not deviate from the clinical safety boundary, and the probability of the prediction deviation exceeding the clinical tolerance is reduced by 63%.

[0037] The present invention adopts a dynamic weight mechanism to combine the biomarker change rate with the model confidence, and realizes individualized follow-up time series arrangement based on the modification of the serum drug concentration half-life. BRIEF DESCRIPTION OF THE DRAWINGS

[0038] Figure 1 is a schematic structural diagram of the system of the present invention. DETAILED DESCRIPTION OF THE INVENTION

[0039] The following further describes the embodiments of the invention in detail with reference to the accompanying drawings of the specification, so as to more clearly present the purpose, technical solution and technical effect of the present invention.

[0040] The system architecture of the present invention is as Figure 1 shown, including a patient management module, a diagnosis and treatment result module, a diet module, an analysis module, a follow-up reminder module, a display module and a database. Aiming at the problems existing in the traditional follow-up methods for inflammatory bowel disease, such as inconvenient data management, low follow-up efficiency, insufficient personalized support and underutilization of data, etc. By integrating multi-dimensional data of patients, including diagnosis and treatment, diet, examination results, etc., and using algorithms and technologies for analysis and processing, predicting the development trend of the disease, and providing personalized follow-up management services for patients, including personalized diet recommendations, intervention recommendations and follow-up reminders, etc., so as to improve the follow-up efficiency and management effect and improve the management status of patients with inflammatory bowel disease.

[0041] The following details each module:

[0042] The patient management module is dynamically docked with the hospital information system, and synchronizes patient data and endoscopic examination results in real time through the HL7 protocol to establish an electronic file containing specific tags for inflammatory bowel disease.

[0043] The diagnosis and treatment result module is used to store structured data with temporal tags, including the medication time sequence recorded in three groups of timestamp, administration route, and drug category, the auxiliary examination results accompanied by gastrointestinal pathological section images, and the latent space feature vectors extracted from the images through a convolutional neural network.

[0044] The diet module performs multi-level classification on the diet photos uploaded by the patient through an image recognition unit based on MobileNetV3, and outputs the estimated values of food types and intake; dynamically adjusts the nutrient weight constraints according to the Crohn's disease activity index; parses the symptom description text submitted by the patient through natural language processing technology, extracts key symptom entities and maps them to the SES-CD standard system.

[0045] The analysis module includes an inflammation activity index generation unit, a dynamic model fusion unit, and a medical constraint optimization unit. The inflammation activity index generation unit is used to calculate the composite biomarker index:

[0046]

[0047] where ΔHb is the weekly change in hemoglobin.

[0048] The dynamic model fusion unit is used to integrate the outputs of random forest and LSTM, and the weight coefficients are dynamically adjusted based on the patient's real-time clinical status:

[0049]

[0050] RiskScore = α·P RF +(1-α)·sigmoid(A LSTM / 3)

[0051] where RiskScore represents the risk score, P RF represents the output of the random forest, and A LSTM represents the output of the LSTM.

[0052] The medical constraint optimization unit introduces clinical safety boundary conditions in the LSTM training stage:

[0053] L total = BCE Loss + 0.4·max(0, SERSI pred - SERSI clin + 0.3)

[0054] where L total represents the total loss in the LSTM training stage, BCELoss represents the binary cross-entropy loss, SERSI pred represents the predicted value of the composite biomarker index, and SERSI clinRepresents the clinical composite biomarker index value.

[0055] The analysis module performs time decay compensation on serum albumin and fecal calprotectin concentrations:

[0056] >C corrected =C raw ·exp(-λ∣t sample -t endoscopy ∣)>

[0057] where λ = 0.03 / hour, and t endoscopy represents the time of the most recent colonoscopy; C raw represents the original concentration of the biomarker.

[0058] Fuses the recurrence probability P output by the random forest RF with the disease activity ALSTM predicted by LSTM:

[0059] >RiskScore = ω·P RF +(1 - ω)·tanh(A LSTM / 10)>

[0060] The weight coefficient w is dynamically adjusted by the patient's current nutritional index:

[0061]

[0062] where BMI represents the patient's current body mass index.

[0063] The follow-up reminder module dynamically generates a hierarchical reminder strategy based on the RiskScore value. When RiskScore ≥ 0.7, a red alert is triggered, and a follow-up visit plan is generated according to the interval days = 45 - 5×SERSI; when 0.4 ≤ RiskScore < 0.7, the reminder interval is set according to 2.3 times the half-life of the serum drug concentration. The follow-up reminder module supports multiple reminder methods, including text messages, emails, and APP notifications.

[0064] The display module generates a predictive visualization interface with interpretability annotations. The display module includes highlighting the key features that affect the risk score using the Integrated Gradients algorithm; showing the distribution evolution of different patient subgroups in the latent space with a three-dimensional manifold graph.

[0065] The database uses a time series database optimized for column-oriented storage to perform interpolation compensation on biomarker data.

[0066] Here, the diet module also includes a diet analysis function, which is used to generate personalized diet suggestions based on the patient's condition and nutritional needs, combined with diet records. The analysis module generates personalized intervention suggestions according to the prediction results, including adjusting the medication plan, increasing the follow-up frequency, and hospitalization.

[0067] Of course, for multidisciplinary coordination, there is also a multidisciplinary collaboration module for information sharing and collaboration among doctors, dietitians, and psychiatrists.

[0068] Workflow of the medical constraint optimization unit: Obtain the original CRP data sequence {CRP_t} from the hospital LIS system; perform time decay correction: CRP_corrected = CRP_raw × exp(-0.028 × Δt); input it into the constrained LSTM network for training. If the predicted SERSI exceeds the clinical safety threshold, automatically trigger the loss penalty term; generate a prediction report with risk confidence interval annotation.

[0069] Calculation of dynamic reminder interval: For a patient receiving maintenance treatment with infliximab: The current blood drug concentration C = 4.2 μg / mL (half-life t1 / 2 = 9.5 days) is detected; the system calculates the reminder interval: T = 2.3 × 9.5 = 21.85 days → 22 days; when the patient's SERSI index rises to 3.0, the system automatically shortens the interval to T' = 45 - 5 × 3.0 = 30 days.

[0070] The present invention is based on the collaborative work of multiple modules: The patient management module dynamically docks with the hospital information system through the HL7 protocol, synchronizes patient data and endoscopic examination results in real time, establishes an electronic file with specific tags for inflammatory bowel disease, and realizes the integration of patient basic data. The diagnosis and treatment result module stores structured data with temporal markers, and uses a convolutional neural network to extract image latent space feature vectors for in-depth analysis of diagnosis and treatment information. The diet module uses an image recognition unit based on MobileNetV3 to classify and estimate the intake of diet photos, adjusts the nutrient weight constraints according to the Crohn's disease activity index, and uses natural language processing technology to analyze symptom texts to comprehensively obtain patient diet and symptom information. The inflammation activity index generation unit of the analysis module calculates the composite biomarker index SERSI through a formula, considering factors such as CRP, fecal calprotectin concentration, serum albumin, and the weekly change amount of hemoglobin; the dynamic model fusion unit dynamically adjusts the weight coefficients according to the patient's real-time clinical status, and integrates the outputs of random forest and LSTM to obtain a risk score; the medical constraint optimization unit introduces clinical safety boundary conditions during the LSTM training to optimize model training. In addition, the system also performs time decay compensation on serum albumin and fecal calprotectin concentration, and dynamically adjusts the weight coefficients of the integrated random forest and LSTM outputs in combination with the patient's nutrition index. The follow-up reminder module dynamically generates a hierarchical reminder strategy and a follow-up visit plan based on the risk score. The display module uses relevant algorithms to generate a predictive visualization interface with interpretable annotations. The database uses a time series database optimized for column-oriented storage to interpolate and compensate biomarker data, ensuring the efficiency of data processing and storage.

[0071] The technical effects of this system are remarkable: First, it accurately solves the problem of heterogeneous management of inflammatory bowel disease. The proposed SERSI composite index can effectively reflect the dynamics of the intestinal microenvironment. Through comparative studies, its correlation with the endoscopic score r = 0.82, providing a strong basis for disease assessment. Second, the developed medical constraint optimization loss function introduces clinical safety boundary conditions during the LSTM training stage, ensuring that the model prediction does not deviate from the clinical safety boundary, reducing the probability of the prediction deviation exceeding the clinical tolerance by 63%, and improving the reliability of the prediction. Third, the system adopts a dynamic weight mechanism, combines the biomarker change rate with the model confidence, and corrects based on the half-life of serum drug concentration, realizing individualized follow-up time series arrangement, and can formulate a reasonable follow-up plan according to the specific situation of the patient, improving the pertinence and effectiveness of follow-up management.

[0072] The above are only preferred and feasible embodiments of the present invention, and are not used to limit the scope of the patent application of the present invention. All equivalent changes, equivalent substitutions, or modified changes completed within the technical spirit and principles disclosed by the present invention shall be included within the scope of patent protection covered by the present invention.

Claims

1. An inflammatory bowel disease patient follow-up management system, characterized in that: include: The patient management module dynamically connects with the hospital information system, synchronizes patient data and endoscopic examination results in real time through the HL7 protocol, and establishes an electronic file containing inflammatory bowel disease-specific tags; The diagnosis and treatment result module is used to store structured data with temporal tags, including the medication sequence recorded in three groups: timestamp, route of administration, and drug category, the auxiliary examination results accompanied by digestive tract pathology slice images, and the latent space feature vector extracted from the image through the convolutional neural network; The diet module uses a MobileNetV3-based image recognition unit to perform multi-level classification on diet photos uploaded by patients, output food types and intake estimates; dynamically adjust nutrient weight constraints based on the Crohn's disease activity index; and use natural language processing technology to parse symptom description texts submitted by patients, extract key symptom entities, and map them to the SES-CD standard system. The analysis module includes an inflammatory activity index generation unit, a dynamic model fusion unit, and a medical constraint optimization unit. The inflammatory activity index generation unit is used to calculate the composite biomarker index: Among them, ΔHb is the weekly change of hemoglobin; The dynamic model fusion unit is used to integrate the outputs of random forest and LSTM, and the weight coefficients are dynamically adjusted based on the patient's real-time clinical status: RiskScore=α·P RF +(1-α)·sigmoid(A LSTM / 3) Among them, RiskScore represents the risk score, P RF represents the output of random forest, A LSTM Represents the output of LSTM; The medical constraint optimization unit introduces clinical safety boundary conditions during the LSTM training phase: THE total =BCE Loss+0.4·max(0,SERSI pred -SERSI clin +0.3) Among them, L total represents the total loss of the LSTM training phase, BCELoss represents the binary cross entropy loss, and SERSI pred Represents the predicted composite biomarker index value, SERSI clin Indicates the clinical composite biomarker index value; The follow-up reminder module dynamically generates a graded reminder strategy based on the RiskScore value. When RiskScore ≥ 0.7, a red alarm is triggered and a follow-up plan is generated according to the interval days = 45-5×SERSI; when 0.4 ≤ RiskScore < 0.7, the reminder interval is set at 2.3 times the half-life of the serum drug concentration; A display module that generates a prediction visualization interface with interpretable annotations; The database uses a time series database optimized for columnar storage to perform interpolation compensation on biomarker data.

2. The inflammatory bowel disease patient follow-up management system according to claim 1, characterized in that: The display module includes using an Integrated Gradients algorithm to highlight key features that affect risk scores; The distribution evolution of different patient subgroups in the latent space is displayed using a three-dimensional manifold diagram.

3. The inflammatory bowel disease patient follow-up management system according to claim 1, characterized in that: The follow-up reminder module supports multiple reminder methods, including SMS, email and APP notification.

4. The inflammatory bowel disease patient follow-up management system according to claim 1, characterized in that: The diet module also includes a diet analysis function for generating personalized diet recommendations based on the patient's condition and nutritional needs in combination with diet records.

5. The inflammatory bowel disease patient follow-up management system according to claim 1, characterized in that: The analysis module generates personalized intervention recommendations based on the prediction results, including adjusting medication regimens, increasing follow-up frequency, and hospitalization.

6. The inflammatory bowel disease patient follow-up management system according to claim 1, characterized in that: It also includes a multidisciplinary collaboration module for information sharing and collaboration among doctors, nutritionists, and psychologists.

7. The inflammatory bowel disease patient follow-up management system according to claim 1, characterized in that: The analysis module performs time decay compensation on serum albumin and fecal calprotectin concentrations: >C corrected =C raw ·exp(-λ∣t sample -t endoscopy ∣)> Where, λ = 0.03 / hour, t endoscopy Indicates the time of the most recent colonoscopy; C raw represents the raw concentration of the biomarker; The recurrence probability P output by random forest RF Fusion with LSTM predicted disease activity ALSTM: >RiskScore=ω·P RF +(1-ω)·tanh(A LSTM / 10)> The weight coefficient w is dynamically adjusted by the patient's current nutritional index: Among them, BMI represents the patient's current body mass index.

Images