Screening and intervention system for outpatient colonoscopy pre-bowel preparation and method thereof

Abstract

This application relates to the field of medical data processing technology, and discloses a screening and intervention system and method for bowel preparation before outpatient colonoscopy, including: providing patients with standardized bowel preparation guidance information; receiving image data of excrement uploaded by patients after bowel preparation; inputting the image data into a pre-trained intelligent judgment model for bowel preparation quality to generate quantitative scores and pass / fail judgment results; and executing an intervention decision process based on the judgment results, including confirming the examination or triggering a rescheduling and supplemental bowel preparation medication instructions. Through the above technical solution, this application achieves objective and quantitative assessment of bowel preparation quality and closed-loop dynamic intervention before examination, effectively improving assessment accuracy, avoiding waste of medical resources, and systematically improving the overall efficiency and safety of outpatient colonoscopy.

Description

Technical Field

[0001] This invention belongs to the field of medical data processing technology, specifically relating to a screening and intervention system and method for bowel preparation before outpatient colonoscopy. Background Technology

[0002] Colonoscopy, as a core tool for screening and diagnosing colorectal diseases, relies heavily on the quality of bowel preparation for its clinical value. High-quality bowel preparation not only significantly improves endoscopic clarity and shortens procedure time, but also effectively increases the detection rate of adenomas and early lesions, while reducing procedure-related risks such as perforation and bleeding. Currently, bowel preparation before outpatient colonoscopy mainly relies on patients self-administering bowel-cleansing medications and adjusting their diet as prescribed by their doctor. The entire process lacks a real-time, objective assessment mechanism and dynamic intervention measures.

[0003] Current technologies still face multiple challenges in practical applications: First, patient compliance varies greatly, with some individuals failing to adequately prepare their bowels due to misunderstandings, improper medication, or inadequate dietary control. Second, there is a lack of objective and quantifiable assessment standards; doctors typically rely solely on patients' subjective descriptions to judge cleanliness, resulting in low accuracy and susceptibility to interference. Third, the existing process lacks a closed-loop feedback mechanism; if bowel preparation is insufficient, the procedure often has to be temporarily suspended on the day of the examination, leading to a waste of medical resources and repeated appointments for patients. Finally, an intelligent discrimination system based on real excrement images has not yet been established, making it impossible to achieve efficient and automated quality screening and intervention decisions before the examination.

[0004] These problems are particularly pronounced in high-load outpatient environments, directly leading to decreased colonoscopy efficiency, increased adenoma misdiagnosis rates, and a deterioration in patient experience. Therefore, there is an urgent need for an integrated approach that combines patient information management, medication guidance, intelligent analysis of excrement images, and dynamic intervention decisions to systematically improve the pass rate of bowel preparation before outpatient colonoscopy and the safety of the procedure. Summary of the Invention

[0005] The purpose of this invention is to provide a screening and intervention system and method for bowel preparation before outpatient colonoscopy, which can effectively solve the problems in the background art.

[0006] To achieve the above objectives, the present invention provides the following technical solution: A screening and intervention system for bowel preparation before outpatient colonoscopy, comprising the following components: Standardized guidance information push module: After a patient schedules a colonoscopy, standardized bowel preparation guidance information is provided to the patient through the patient terminal. The guidance information includes a pre-examination notice document, medication guidance videos and illustrated instructions, and a dietary control list. The medication guidance videos and illustrated instructions demonstrate in detail the preparation method, dosage, time interval, and water requirements of the bowel cleansing drugs. The dietary control list clearly lists the types of foods that should not be consumed and the recommended list of liquid foods to be consumed during a specific time period before the examination. The module records the patient's viewing status and duration of the guidance information and marks patients who do not view it in a timely manner as subjects of compliance risk. Excrement Image Acquisition and Upload Module: Within a preset time window before the examination, a reminder to upload excrement images is sent to the patient via the patient terminal, and the image acquisition module integrating an image quality pre-detection algorithm is activated. The algorithm analyzes the preview image captured by the camera in real time and automatically detects whether the image meets the preset clarity threshold, illumination uniformity threshold, and proportion threshold of the fecal sample in the image. Only when all detection items meet the threshold requirements can the image acquisition module allow the user to perform the shooting operation, and encrypt and upload the captured image data, shooting timestamp, and associated patient identifier to the backend server. The intelligent judgment module for bowel preparation quality receives the image data and inputs it into a pre-trained intelligent judgment model for bowel preparation quality. The model adopts a deep convolutional neural network architecture, and its training dataset consists of images of excrement labeled by endoscopists. During annotation, physicians consider the quality of the subsequent colonoscopy performed on the patient corresponding to the image (such as whether the operation was successful and whether the mucosal visibility was good) to comprehensively determine the cleanliness level of the image, thereby ensuring that the annotation results have clear clinical significance. The front-end convolutional layer of the model is used to extract multi-scale visual features from the input image. These features include color distribution features, texture complexity features, solid residue morphology features, and liquid transparency features. The back-end fully connected layer of the model fuses the extracted visual features and maps them to a continuous cleanliness score. The system automatically executes the judgment process based on the score: if the score is in the range of 1-2 points, the system automatically judges the bowel preparation quality as "excellent"; if the score is in the range of 8-10 points, the system automatically judges the bowel preparation quality as "failed"; if the score is in the range of 3-7 points, the system automatically triggers the manual judgment process, pushes the image data and score value to the medical staff terminal for review, and completes the judgment based on the final rating of the manual review. Intervention Decision and Execution Module: Based on the judgment result, executes a predefined intervention decision process. The process includes automatically sending an examination confirmation notification to the patient's terminal and synchronizing the patient's preparation status with the endoscopy center information system when the judgment is "excellent"; automatically triggering a rescheduling process when the judgment is "failed", releasing the original appointment time slot and notifying the patient to prepare again when the judgment result is "good" or "unqualified" after manual review; automatically triggering the intervention decision engine. The engine selects the optimal strategy from a pre-set intervention strategy library based on the final rating, the patient's historical compliance record, and the available appointment time slots for the day. The intervention strategy includes sending instructions to the patient's terminal to strengthen medication guidance and re-upload images, automatically generating rescheduling suggestions and synchronizing notifications to the endoscopy center appointment system, or generating supplementary bowel cleansing drug prescription suggestions and pushing them to the outpatient doctor's workstation for review. The full-process digital archive management module integrates the guidance information viewing records from the standardized guidance information push module, the image data and its metadata uploaded from the excrement image acquisition and upload module, the cleanliness score and judgment results generated by the intelligent bowel preparation quality judgment module and the manual review records, and the intervention decision records executed from the intervention decision and execution module. It constructs a digital archive of the patient's entire bowel preparation process organized in a timeline format, supporting multi-dimensional retrieval and statistical analysis by patient, by time period, and by preparation quality results.

[0007] Preferably, in the intelligent judgment module for intestinal preparation quality, the model incorporates an attention mechanism, and its core computation process is represented by the following formula: Let the feature map extracted from the input image after the convolutional layer be... , in , , These represent the height, width, and number of channels of the feature map, and the attention weight map, respectively. The following was calculated using a lightweight subnetwork: , in This represents a 1×1 convolution operation. This represents the Sigmoid activation function. Weighted feature map The calculation is as follows: ,in This indicates element-wise multiplication.

[0008] Preferably, in the intervention decision and execution module, the intervention strategy library built into the intervention decision engine contains at least three strategies: The first strategy applies to patients whose manual review rating is "good" and who have no record of compliance risk, and involves sending an examination confirmation notification to the patient's terminal. The second strategy applies to patients whose manual review rating is "unqualified" and who have a history of compliance risk. It automatically generates rescheduling suggestions, releases the original appointment time slot, and recommends a new examination time for the patient. The third strategy applies to situations where the manual review rating is "unacceptable," but the system detects that there are still available examination slots after the original appointment time on the same day. If the doctor selects this option, the system will perform intelligent calculations: based on the time of the new slot, it will work backwards to clearly indicate the latest deadline for taking the bowel cleansing medication within the prescribed time window; this time reminder and rescheduling suggestion will be provided to both the doctor and the patient to assist in decision-making; if the patient accepts and meets this time requirement, the system can assist them in rescheduling to that later slot; otherwise, it will proceed to the cross-day standard rescheduling process of the second strategy.

[0009] Preferably, in the excrement image acquisition and uploading module, the image quality pre-detection algorithm performs the following: the sharpness detection is evaluated by calculating the gradient magnitude of the local area of ​​the image; the illumination uniformity detection is determined by analyzing the image histogram distribution; and the fecal sample proportion detection is estimated by preliminary color and texture segmentation to estimate the coverage ratio of excrement in the central area of ​​the image. All three detections have quantified thresholds.

[0010] Preferably, in the full-process digital record management module, the digital record supports the use of desensitized aggregated data for multi-dimensional statistical analysis at the group level. The analysis includes calculating the average preparation score of patients in different age groups, analyzing the impact of different versions of guidance materials on compliance, and tracking the effectiveness of intervention strategies.

[0011] Preferably, the system further includes a data security and privacy protection module, which ensures that all data communication between the patient terminal and the backend server adopts an end-to-end encrypted transmission protocol, anonymizes uploaded image data on the server side to remove direct personal identifiers, and implements role-based access control policies to restrict data access permissions for medical staff in different roles.

[0012] This invention also discloses a screening and intervention method for bowel preparation before outpatient colonoscopy, the method comprising the following steps: S110, Before the examination, standardized bowel preparation guidance information is provided to the patient through the patient terminal. The guidance information includes a pre-examination notice document, medication guidance video and graphic instructions, and a diet control list. The patient's viewing status and duration of the guidance information are recorded, and patients who do not view the information in a timely manner are marked as subjects of compliance risk. S120, within a preset time window before the examination, receive image data of the last excrement after bowel preparation uploaded by the patient through the patient terminal. The uploading process is executed by an image acquisition module that integrates an image quality pre-detection algorithm. The algorithm ensures that the uploaded image meets preset clarity thresholds, illumination uniformity thresholds, and fecal sample proportion thresholds. S130, the image data is input into a pre-trained intelligent judgment model for intestinal preparation quality. The model generates a quantitative score for intestinal cleanliness based on big data learning, and the system automatically executes the judgment process according to the score value: If the score is in the range of 1-2 points, the system will automatically judge it as "excellent"; If the score is between 8 and 10, the system will automatically determine it as "failure"; If the score is in the range of 3-7, the system will automatically trigger the manual judgment process, push the image data and score value to the medical staff's terminal for review, and complete the judgment based on the final rating of the manual review. S140, based on the judgment result, execute a predefined intervention decision process, the process including confirming the examination appointment when the judgment is "excellent"; automatically triggering the rescheduling process when the judgment is "failed"; when the judgment result is "good" or "unsatisfactory" after manual review, triggering corresponding enhanced guidance, rescheduling or supplemental bowel cleansing medication intervention instructions based on the final rating, the patient's historical compliance record and the available appointment time slots on the same day. S150, integrate the guidance information review record in step S110, the image data and its metadata uploaded in step S120, the cleanliness score and judgment result and manual review record generated in step S130, and the intervention decision record executed in step S140 to establish a digital archive of the patient's entire bowel preparation process.

[0013] Preferably, in step S130, the intelligent judgment model for intestinal preparation quality introduces an attention mechanism, and its core operation process is represented by the following formula: Let the feature map extracted from the input image after the convolutional layer be... , in , These represent the height, width, and number of channels of the feature map, and the attention weight map, respectively. The following was calculated using a lightweight subnetwork: , in This represents a 1×1 convolution operation. This represents the Sigmoid activation function. Weighted feature map The calculation is as follows: This indicates element-wise multiplication.

[0014] Preferably, in step S140, when the judgment result is "good" or "unqualified" after manual review, the triggered intervention instruction is selected according to a preset strategy library, which includes: If the manual review rating is "good" and the patient has no risk of non-compliance, a confirmation notification will be sent to the patient's terminal. If the manual review rating is "unqualified" or the patient's compliance record is poor, a rescheduling suggestion will be automatically generated and the endoscopy center's appointment system will be notified simultaneously. For patients whose manual review rating is "unqualified" but the system determines that there is still a sufficient time window between the current time and the original appointment time, the system will automatically check if there are any available examination slots at the endoscopy center that are later than the current time and meet the bowel preparation time requirements. If so, a "same-day rescheduling" suggestion will be generated, including the new slot information and the latest medication deadline calculated based on that slot, for the doctor to review and then send to the patient.

[0015] Preferably, in step S120, the image quality pre-detection algorithm performs sharpness detection, illumination uniformity detection, and fecal sample proportion detection in parallel, and the shooting operation is only allowed when the real-time status of all detection items meets their corresponding quantization thresholds.

[0016] The technical effects and advantages of the present invention in the above technical solution are as follows: By combining standardized digital guidance with image-driven intelligent discrimination, an objective and quantitative assessment system for bowel preparation quality was constructed, overcoming the limitations of relying on patients' subjective descriptions and significantly improving the accuracy of the assessment.

[0017] 2. It has achieved closed-loop management throughout the entire process from guidance, implementation, evaluation to intervention, which can promptly identify cases of insufficient preparation before the examination and take dynamic intervention measures, effectively avoiding last-minute cancellations on the day of the examination, reducing the waste of medical resources and the cost of repeated appointments for patients.

[0018] 3. The intelligent discrimination model based on big data learning can be continuously optimized, and its discrimination ability is constantly enhanced with the accumulation of data, providing a scalable technical foundation for standardized screening of intestinal preparation quality.

[0019] 4. The integrated intervention decision engine can automatically execute differentiated processing strategies based on specific circumstances, improving the efficiency and accuracy of personalized interventions, thereby systematically improving the overall efficiency of outpatient colonoscopy, adenoma detection rate, and treatment safety.

[0020] 5. A comprehensive, multi-dimensional quantitative management system was established, realizing a paradigm shift from subjective experience to objective data-driven approaches. This system fully quantifies the assessment, decision-making, and management of gut health preparation. The visual characteristics of excrement are converted into a continuous cleanliness score of 1-10 using an intelligent model, replacing subjective descriptions with objective data. Based on quantitative scoring ranges and preset rules, the automatic and standardized triggering of judgment and intervention strategies is realized; Data recording and analysis of patient compliance, image quality, and final results form a measurable, optimizable, and traceable quality control closed loop. Attached Figure Description

[0021] Figure 1 This is a schematic diagram of the overall technical solution architecture of the screening and intervention method for bowel preparation before outpatient colonoscopy proposed in this invention; Figure 2 This is a schematic diagram of the core principle framework of the intelligent judgment model for intestinal preparation quality in this invention; Figure 3 This is a flowchart illustrating the main stages of the process from pushing guidance information to image data acquisition and uploading in this invention. Figure 4 This is a flowchart illustrating the logical process of executing predefined intervention decisions based on intelligent discrimination results in this invention. Detailed Implementation

[0022] To further illustrate the technical means and effects adopted by the present invention to achieve the intended purpose, the specific embodiments according to the present invention will be described in detail below with reference to the accompanying drawings and preferred embodiments.

[0023] The following examples illustrate the quantitative management system involved in this solution in detail. This system permeates the entire process of bowel preparation, and its core lies in quantification at three levels: quantifying cleanliness assessment into scores; quantifying intervention decisions into automated processes based on scores and rules; and quantifying patient behavior and quality control into analyzable data. Through this quantification, a shift from traditional experience-based models to objective data-driven models is achieved.

[0024] Example 1 Patients who schedule colonoscopies often face the common problem of inconsistent bowel preparation quality. Traditional methods relying on verbal or written instructions and patient self-reporting suffer from inaccurate information delivery, difficulty in monitoring compliance, and strong subjectivity in quality assessment. This directly leads to a high rate of cancellations or reschedulings on the day of the examination due to insufficient bowel preparation, resulting in a serious waste of medical resources and negatively impacting the patient experience.

[0025] This invention provides a screening and intervention method for bowel preparation before outpatient colonoscopy, aiming to achieve closed-loop management of the entire process from preparation guidance, execution monitoring, quality assessment to dynamic intervention through digital and intelligent means.

[0026] See Figure 1 The core technical architecture of this method consists of a patient terminal, a backend server, and an associated medical information system. The patient terminal is typically a dedicated application installed on the patient's personal smartphone. The backend server is deployed on the hospital's intranet or in a cloud environment that complies with medical data security standards, integrating a business logic processing engine, a database, and a core intelligent judgment model for bowel preparation quality. The medical information system includes a hospital appointment system, an endoscopy center information system, and outpatient doctor workstations. The entire process begins with appointment scheduling and ends with examination execution or rescheduling, ensuring that each patient receives examination in a qualified bowel preparation state through standardized information interaction and intelligent decision-making.

[0027] The specific implementation of this method involves a series of strictly defined steps. Step S110 involves providing standardized bowel preparation guidance information to the patient via their terminal before the examination. This step is fundamental to ensuring the quality of preparation. Specifically, after the patient successfully schedules a colonoscopy and completes payment, the hospital's appointment system synchronizes the patient's appointment record and linked mobile phone number information to the backend server of this invention. The server then triggers a guidance information push task. This task is not simply sending a text message, but rather pushing a structured "bow preparation guidance package" to the patient's terminal application. This guidance package is a packaged data set, the content of which has been reviewed by a clinical expert committee to ensure its authority and consistency. The guidance package contains at least three core parts: First, a pre-examination information document, which details the purpose, procedure, potential risks, precautions, and a checklist of pre-hospital preparations in a visually appealing format; second, a medication guidance video with accompanying text and images, the video being animated and the text linking to relevant articles on the endoscopy center's official WeChat account; and third, a dietary control checklist, presented in a timeline format, clearly listing prohibited foods at key time points such as 72 hours, 48 ​​hours, and 24 hours before the examination, including high-fiber vegetables, seeded fruits, and red liquids, while simultaneously providing a transparent list of permitted liquid foods, such as clear broth, sports drinks, and fruit juice without pulp. Upon receiving the guidance package, the patient's terminal application will remind the patient to review it through pop-ups, notification messages, and other means. The system backend continuously monitors the patient's interaction with each component of the guidance package, including clicks and browsing time, and records timestamps. If the patient has not opened any core content within 24 hours of the push notification, the system automatically marks the patient as a "compliance risk subject," which will affect subsequent decision-making logic.

[0028] Step S120: Receive image data of the patient's last bowel movement after bowel preparation, uploaded by the patient via the patient terminal. This step is a crucial data acquisition step for achieving objective assessment. See also... Figure 2The system sends an image upload reminder to the patient via a patient terminal application within a preset time window before the scheduled examination time, such as 6 to 2 hours before the examination. This reminder includes not only text prompts but also links to an integrated image acquisition module. This module does not use the phone's general camera but is a dedicated shooting interface with a built-in image quality pre-detection algorithm. When the patient activates this module, the pre-detection algorithm begins working as the camera preview is displayed in real time. The algorithm performs multiple tests in parallel: sharpness detection evaluates the image by calculating the gradient magnitude of local image regions to ensure image details are discernible; illumination uniformity detection analyzes the image histogram distribution to determine if there is overexposure, underexposure, or severe shadows; and fecal sample proportion detection estimates the coverage ratio of excrement in the central area of ​​the image through preliminary color and texture segmentation, requiring the sample to occupy the main area of ​​the image but not too far to the edge. All three tests have quantified thresholds, such as a sharpness value greater than 0.5, an illumination uniformity variance less than 0.1, and a sample proportion between 30% and 70%. A pass / fail icon is displayed next to the preview screen in real time, guiding the user to adjust the shooting distance, angle, or ambient light. The camera button is activated only when all detection items are in "pass" status in real time, allowing operation. This mandatory pre-screening ensures the data quality of uploaded images from the source, laying a reliable foundation for subsequent intelligent analysis. After the user takes a picture, the image data is immediately bound to the precise timestamp of the shot and the patient's unique anonymized identifier, and uploaded to the designated storage area of ​​the backend server through an end-to-end encrypted channel. All raw image data is encrypted during storage, ensuring security during transmission and static storage.

[0029] Step S130: The image data is input into a pre-trained intelligent judgment model for bowel preparation quality. This model generates a quantitative score (range 1-10) for bowel cleanliness based on large-scale data learning. The system automatically executes the judgment process based on the score value. If the score is in the range of 1-2 points, the system will automatically determine the quality of bowel preparation as "excellent"; If the score is in the range of 8-10, the system automatically determines that the quality of bowel preparation has "failed". If the score falls within the 3-7 range, the system automatically triggers a manual review process. The system encrypts and pushes the image data, score, and associated patient information to the endoscopist's or designated medical staff's terminal for manual review. The manual review outputs a score based on a clinical experience reference model and ultimately labels the score as "Good" (corresponding to a score of 3-5) or "Unsatisfactory" (corresponding to a score of 6-7). The review results are sent back to the system in real time.

[0030] The above scoring ranges are based on explicit visual criteria and model features, as detailed below: 1-2 points: Excellent (automatically judged as qualified) Judgment criteria: The excrement is colorless or pale yellow transparent watery stool; there are no visible solid fecal residues, flocculent matter or suspended matter; the liquid has extremely high transparency, and the bottom of the toilet can be clearly seen.

[0031] Model scoring characteristics: The color is a bright, low-saturation yellow or colorless; the texture is close to pure water and free of particles; the proportion of solid residue is about 0%; and the liquid transparency is ≥95%.

[0032] 3-7 points: Medium (all scores will be subject to manual review) Judgment criteria: Excrement is yellow, watery, or slightly cloudy and pasty; a small amount of fecal residue, fine flocculent matter, or slight suspended matter may be visible; the liquid is semi-transparent, without a large number of formed fecal masses. This range indicates that bowel preparation has not reached excellent levels, but has not completely failed.

[0033] Model scoring characteristics: color is medium-brightness yellow to light brown; texture shows a small to moderate amount of flocculent matter; solid residue accounts for <20%; liquid transparency is between 30% and 94%.

[0034] Manual review task: Within this score range, doctors need to make a comprehensive judgment based on the images and give a final rating of "good" or "unsatisfactory" to drive different subsequent intervention strategies.

[0035] 8-10 points: Failure (automatically judged as unqualified) Judgment criteria: The presence of formed soft stools or obvious fecal masses; stools that are thick and pasty containing a large amount of solid feces; and completely opaque liquids with dense residue. These indicate that bowel preparation was inadequate and an effective examination could not be performed.

[0036] Model scoring characteristics: solid residue ratio ≥20%; liquid transparency <30%; the model detected formed fecal morphology; color is dark yellow or brown.

[0037] The model analyzes the input image and extracts four core visual features: color distribution, texture complexity, solid residue ratio, and liquid transparency. It then uses an attention mechanism to focus on key areas of fecal residue and finally merges and outputs a continuous cleanliness score of 1-10. The model is then automatically judged or manually reviewed and sorted according to the above interval standards.

[0038] The intelligent model for judging bowel preparation quality is a deep convolutional neural network. Its training data comes from a massive amount of historically accumulated anonymized fecal images. Each image is initially screened by at least two senior endoscopists based on the characteristics of the last bowel movement. The specific screening criteria are as follows:

[0039] The labels are reviewed and annotated by pathology experts to ensure accuracy and reliability. The label output results are quantified into a score of 0 to 10, and simultaneously provide a four-level judgment of "excellent / good / unqualified / failure". It supports real-time embedding into the colonoscopy appointment system, automatically intercepting unqualified patients and triggering bowel preparation reminders.

[0040] The model's front end consists of multiple stacked convolutional and pooling layers, used to automatically extract multi-level, multi-scale visual features from the input RGB image. These features are not manually designed but are learned autonomously by the network. Their physical meaning can be interpreted as patterns highly correlated with cleanliness, such as: color distribution features to distinguish clear water, yellow turbid liquid, and dark liquid containing solid residue; texture complexity features to identify foam, flocculent matter, or undissolved drug particles in the liquid; solid residue morphology features to detect the size, shape, and quantity of fecal residue; and liquid transparency features to assess the clarity of the liquid. These abstract features are effectively combined and abstracted in the deeper layers of the network.

[0041] To further enhance the model's ability to discriminate key regions and suppress irrelevant background interference, an attention mechanism is introduced into the model. Its core computation process is represented by the following formula: Let the feature map extracted from the input image after passing through the front-end convolutional layer be: , in , , These represent the height, width, and number of channels of the feature map, respectively. Attention weight map. The calculation was performed using a lightweight subnetwork, which typically consists of 1×1 convolutional layers and a sigmoid activation function: , in This represents a 1×1 convolution operation. Denotes the Sigmoid activation function, such that The value for each spatial location in the image ranges from 0 to 1, representing the importance weight of that location. (Weighted feature map) The calculation is as follows: ,in Representation of feature map With attention weight map Element-wise multiplication in the spatial dimension. This mechanism enables the model to adaptively focus on key areas in the image related to fecal residue, such as suspended matter near the toilet waterline, while ignoring interfering information such as toilet walls and water reflections, thereby significantly improving the accuracy and robustness of the discrimination.

[0042] The weighted feature maps, after being flattened, are fed into the backend fully connected layer network. The role of the fully connected layer is to fuse high-dimensional visual features and ultimately map them to a scalar value, namely, the cleanliness score. The system presets a passing threshold, which is not fixed but dynamically optimized based on clinical guidelines and the correlation between cleanliness scores and the actual endoscopic adenoma detection rate in historical data from the hospital. When the cleanliness score output by the model is greater than or equal to the passing threshold, the system determines that the patient's bowel preparation quality is "qualified"; when the score is less than the passing threshold, it is determined to be "unqualified". The determination result, along with the specific score and determination time, is recorded.

[0043] Step S140: Execute the predefined intervention decision-making process based on the judgment result: If the system automatically determines that the result is "qualified" (score of 1-2 points), it will automatically send an examination confirmation notification to the patient's terminal and synchronize it to the endoscopy center's information system. If the system automatically determines it as "failure" (score of 8-10), it will automatically trigger the rescheduling process, release the original appointment time slot, and send a notification to the patient to prepare again. If the score is between 3 and 7, intervention will be implemented based on the review results after manual review is completed. When manually judged as "good" (score of 3-5 points), it generally does not affect the examination, and the doctor can irrigate during the procedure; If the result is manually determined to be "unqualified" (score of 6-7 points), it is recommended to postpone the examination and, if necessary, to strengthen bowel cleansing.

[0044] See Figure 4 This step establishes a closed loop from evaluation to action. The decision-making process is fully automated, executed by the intervention decision engine on the backend server. The engine has a built-in rule- and policy-based decision tree.

[0045] When the engine receives a "qualified" result, process branch one is triggered: the system automatically sends a confirmation notification to the patient's terminal application stating "Ready to proceed, please arrive at the hospital on time as scheduled." This notification may include supplementary information such as the route to the hospital and the reporting location. Simultaneously, the engine synchronizes the patient's readiness status to the endoscopy center's information system in real time via a standard data interface. On the nurse's workstation interface of the endoscopy center's information system, a green "Ready Confirmed" indicator will appear next to the patient's name, allowing them to enter the examination queue for the day.

[0046] When the engine receives a judgment result regarding the quality of bowel preparation, it will trigger the corresponding intervention process based on the result type and source. This is the core step in achieving precise intervention.

[0047] If the result is determined automatically by the system (i.e., the score falls within the "pass" range of 1-2 points or the "fail" range of 8-10 points), the engine will execute a preset standardized process: For the "qualified" determination: the system automatically sends an examination confirmation notification to the patient's terminal and synchronizes it to the endoscopy center's information system.

[0048] For a "failure" result: the system automatically triggers a rescheduling process, releases the original appointment time slot, and notifies the patient to prepare again.

[0049] If the result is derived from manual review (i.e., the initial system score falls within the 3-7 range and is ultimately labeled "Good" or "Unsatisfactory" by a human reviewer), the engine initiates a more complex decision-making process. The engine first analyzes three key input parameters: first, the final rating determined by the human reviewer ("Good" or "Unsatisfactory"); second, the patient's historical "compliance risk" record as marked in step S110; and third, the time interval between the current moment and the scheduled examination start time. Based on the combination of these parameters, the engine selects and executes the optimal intervention strategy from a pre-set strategy library.

[0050] Strategy 1: Applicable to patients whose manual review rating is "good" (corresponding to a score of 3-5) and who have no record of compliance risk. It basically does not affect the examination, and the doctor can irrigate during the procedure.

[0051] Strategy Two: Applicable to patients whose manual review rating is "Unqualified" (corresponding to a score of 6-7) and who have a history of compliance risk. Doctors can make a decision to "suggest delaying the examination." The engine will send a command to the endoscopy center's appointment system to release the patient's original appointment slot and mark it as "available for appointment." Simultaneously, based on the appointment system's available slots, new optional examination times will be intelligently recommended to the patient.

[0052] Strategy 3: Applicable when the manual review rating is "unacceptable," but the system detects available appointment slots after the original scheduled time. If the doctor selects this option, the system will perform intelligent calculations: based on the new time slot, it will work backwards to specify the latest deadline for taking bowel cleansing medication (e.g., at least 6 hours before the examination). This time reminder will be provided to the doctor and patient along with rescheduling suggestions as key information to aid decision-making. If the patient accepts and meets this time requirement, the system can assist in rescheduling to that later time slot; otherwise, it will proceed to the standard cross-day rescheduling process of Strategy 2.

[0053] All intervention actions, including the content of the notifications sent, the time of rescheduling, and the recommended prescriptions, will be recorded in detail.

[0054] Furthermore, the method includes step S150, establishing a digital archive of the entire patient bowel preparation process. This digital archive is a time-series data set constructed around a single patient's examination event. It integrates the time points and total duration of the patient's review of each component in the guidance package in step S110; metadata such as all image data uploaded in step S120, image capture time, and quality pre-screening pass indicators; the cleanliness score, pass / fail result, and model version number calculated by the model for each image in step S130; and the type, time, and execution result of each intervention decision triggered in step S140. All data is organized in a timeline format, forming a complete chain of evidence from "guidance push" -> "image upload" -> "intelligent scoring" -> "intervention execution". This digital archive supports rapid retrieval and export by patient ID, by examination date range, and by final preparation quality result. At the population level, these desensitized aggregated data can be used for multi-dimensional statistical analysis, such as calculating the average preparation score of patients in different age groups, analyzing the impact of different versions of guidance materials on compliance, and tracking the effectiveness of intervention strategies. This provides a solid data foundation for clinical research, incremental training and iterative optimization of intelligent discrimination models, and continuous improvement of the overall quality of medical care in the hospital.

[0055] Regarding data security and privacy protection, all communication between the patient terminal and the backend server, including instruction package downloads, image uploads, and notification distribution, adopts an end-to-end encrypted transmission protocol based on TLS to prevent data theft or tampering during transmission. Raw image data uploaded to the server undergoes an anonymization pipeline before intelligent discrimination analysis. This pipeline automatically removes potential device and geographic location information from image metadata and manages the association between image files and patient identities through an encrypted mapping table, achieving "usable but not visible" data. The dataset used for incremental model training uses thoroughly anonymized image features and corresponding scoring labels, containing no information that can directly or indirectly identify patients. The system backend implements role-based access control policies. For example, endoscopy center nurses can only view the preparation status results of patients awaiting examination that day; requesting doctors can view the preparation files of all patients under their name; system administrators can access aggregated statistical data but cannot view specific patient images; auditors can view all operation logs. Through both technical and managerial measures, the security of patient privacy data is ensured.

[0056] Example 2 In a medical consortium model, the method of this invention can be extended to central hospitals for unified remote preparation quality control of patients scheduled for colonoscopy examinations at their subordinate community health service centers. In this scenario, patients complete their initial consultation and appointment at the community center, and the information system of the community center connects with the backend server of this invention through a regional medical information platform.

[0057] In step S110, standardized guidance information is uniformly generated and managed by the central hospital's back-end server and distributed to patients' mobile applications in community centers through the regional platform. Community doctors can view the access status of their patients to the guidance information on their workstations and conduct telephone follow-ups with patients marked as "adherence risk subjects" to provide personalized explanations and encouragement, thus compensating for the shortcomings of purely digital push notifications.

[0058] In step S120, the image data acquisition and uploading process remains unchanged, but the backend server needs to process patient data from multiple community centers. Therefore, a microservice design is adopted in the architecture. The image upload interface has high concurrency processing capabilities, and the data from different community centers is logically partitioned and stored according to the data source.

[0059] In step S130, the intelligent judgment model for intestinal preparation quality adopts a centralized deployment mode, meaning that images uploaded by all community centers are sent to the central hospital's server for unified analysis and scoring. This ensures a high degree of consistency in evaluation standards across the entire medical consortium. The central hospital can utilize a larger volume of data to continuously incrementally train the model and periodically synchronize the optimized model, thereby improving the screening level of the entire medical consortium.

[0060] The intervention decision-making process in step S140 needs to be adapted to the tiered medical system. When a decision is deemed "unqualified," the system-generated intervention instructions (such as a reminder or rescheduling suggestion) will be simultaneously sent to the patient and the responsible physician in their community, who will then assist with subsequent communication and monitoring. For decisions triggering a "rescheduling suggestion," the engine needs to interact with the appointment systems of both the central hospital and the community center to coordinate the release of the original time slot and the booking of the new time slot. During this process, if there is a possibility of rescheduling on the same day, the system will calculate and clearly indicate the latest medication deadline corresponding to the new time slot as key decision information.

[0061] The complete digital record established in step S150 can be securely shared between the central hospital and community doctors with the patient's authorization, forming a continuous medical record that facilitates referrals and follow-ups.

[0062] Example 3 This embodiment, based on the method described in Embodiment 1, further enhances the monitoring of patient behavioral compliance after medication. By introducing exercise data to quantitatively assess the patient's activity, it provides an auxiliary reference for judging the quality of bowel preparation.

[0063] In step S110, when the system pushes standardized guidance information to the patient, it will specifically emphasize the importance of "appropriate exercise after medication" and add a "medication management" function entry to the patient's terminal application. After the patient finishes taking the bowel cleansing medication, they can actively click on this entry. The system will then prompt the patient to grant permission to access the patient's system-supported health data platform (such as a built-in health application on their phone) to read their step count. After the patient grants permission, the system can securely obtain the patient's step count data within a specific time window after medication in the background.

[0064] The "medication management" function works in conjunction with the image upload step S120. The system sets a motion monitoring period, for example, from the time the patient takes the first dose of the bowel cleansing medication until the last image of the excrement is uploaded. During this period, the system periodically (e.g., hourly) collects the patient's anonymized step count data and calculates the total number of steps or the average number of steps per hour. The system presets a reasonable reference threshold for exercise volume (e.g., a total step count of 2000-5000 after medication). The patient's actual motion data will be recorded along with image data, guidance information viewing records, etc.

[0065] In the intelligent discrimination step S130, the intelligent discrimination model for bowel preparation quality, in addition to analyzing image features, can also consider motion data as an auxiliary feature in its decision-making process. Although cleanliness scores are mainly based on visual features, for scores at critical values ​​(e.g., the boundary between "good" and "unsatisfactory" in manual review), the system can use the patient's exercise compliance as a reference factor. For example, if the image score is at the lower limit of the "good" (3-5 points) range, but the patient's motion data is significantly higher than the reference threshold, the system is more inclined to maintain the "good" judgment; conversely, if the image score is at the upper limit of the "unsatisfactory" (6-7 points) range and the patient's motion data is extremely low, the system may strengthen the confidence of the "unsatisfactory" judgment or prompt the system to add exercise guidance to the intervention recommendations. This process can be achieved by introducing motion data as an auxiliary input node in the fully connected layer at the back end of the model, or by setting rule-based post-processing logic in the decision engine.

[0066] In the intervention decision-making stage of step S140, exercise data is applied more directly. The policy library of the intervention decision engine is expanded, adding policy branches related to exercise compliance: For patients who are deemed "unqualified" but have sufficient remaining time, if their exercise data is far below the reference threshold, the "enhanced guidance instructions" generated by the system will particularly emphasize the importance of exercise and provide specific exercise suggestions (such as "it is recommended to walk back and forth or gently massage the abdomen for 30 minutes").

[0067] Patients' exercise data are incorporated into their comprehensive "adherence risk" assessment system. Patients with poor exercise adherence over a long period or with repeated examinations can be identified as having a specific risk type, thereby triggering earlier or more aggressive interventions in subsequent intervention decisions.

[0068] In the full-process digital archive of step S150, a new field, "Post-medication exercise data," is added to record the step count change curve, total steps, and comparison with reference thresholds during the monitoring period. This makes the archive data more comprehensive and provides a data foundation for subsequent research on the correlation between exercise and intestinal cleansing effects.

[0069] By introducing motion monitoring, this embodiment establishes a linkage mechanism of "medication-exercise-image assessment," which can detect and intervene earlier in potential risks of insufficient preparation due to lack of activity, further improving the accuracy of system early warning and personalized guidance.

[0070] The above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention in any way. Although the present invention has been disclosed above with reference to preferred embodiments, it is not intended to limit the present invention. Any person skilled in the art can make some modifications or alterations to the above-disclosed technical content to create equivalent embodiments without departing from the scope of the present invention. Any simple modifications, equivalent changes and alterations made to the above embodiments based on the technical essence of the present invention without departing from the scope of the present invention shall still fall within the scope of the present invention.

Claims

1. A screening and intervention system for bowel preparation before outpatient colonoscopy, characterized in that, The system includes the following components: Standardized guidance information push module: After a patient schedules a colonoscopy, standardized bowel preparation guidance information is provided to the patient through the patient terminal. The guidance information includes a pre-examination notice document, medication guidance videos and illustrated instructions, and a dietary control list. The medication guidance videos and illustrated instructions demonstrate in detail the preparation method, dosage, time interval, and water requirements of the bowel cleansing drugs. The dietary control list clearly lists the types of foods that should not be consumed and the recommended list of liquid foods to be consumed during a specific time period before the examination. The module records the patient's viewing status and duration of the guidance information and marks patients who do not view it in a timely manner as subjects of compliance risk. Excrement Image Acquisition and Upload Module: Within a preset time window before the examination, a reminder to upload excrement images is sent to the patient via the patient terminal, and the image acquisition module integrating an image quality pre-detection algorithm is activated. The algorithm analyzes the preview image captured by the camera in real time and automatically detects whether the image meets the preset clarity threshold, illumination uniformity threshold, and proportion threshold of the fecal sample in the image. Only when all detection items meet the threshold requirements can the image acquisition module allow the user to perform the shooting operation, and encrypt and upload the captured image data, shooting timestamp, and associated patient identifier to the backend server. The intelligent bowel preparation quality assessment module receives the image data and inputs it into a pre-trained intelligent bowel preparation quality assessment model. This model employs a deep convolutional neural network architecture, and its training dataset consists of images of excrement annotated by endoscopists. During annotation, the physician considers the quality of the subsequent colonoscopy performed on the corresponding patient to comprehensively determine the cleanliness level of the image, thus ensuring that the annotation results have clear clinical significance. The model's front-end convolutional layer extracts multi-scale visual features from the input image, including color distribution features, texture complexity features, solid residue morphology features, and liquid transparency features. The model's back-end fully connected layer fuses the extracted visual features and maps them to a continuous cleanliness score. The system automatically executes the judgment process based on the score: If the score is in the range of 1-2 points, the system automatically determines the quality of bowel preparation as "excellent"; if the score is in the range of 8-10 points, the system automatically determines the quality of bowel preparation as "failed"; if the score is in the range of 3-7 points, the system automatically triggers the manual judgment process, pushes the image data and score value to the medical staff's terminal for review, and completes the judgment based on the final rating of the manual review. Intervention Decision and Execution Module: Based on the judgment result, executes a predefined intervention decision process. The process includes automatically sending an examination confirmation notification to the patient's terminal and synchronizing the patient's preparation status with the endoscopy center information system when the judgment is "excellent"; automatically triggering a rescheduling process when the judgment is "failed", releasing the original appointment time slot and notifying the patient to prepare again; and automatically triggering the intervention decision engine when the judgment result is "good" or "unqualified" after manual review. The engine selects the optimal strategy from a pre-set intervention strategy library based on the final rating, the patient's historical compliance record, and available appointment time slots for the day. The intervention strategy includes sending instructions to the patient's terminal to strengthen medication instructions and re-upload images, automatically generating rescheduling suggestions and synchronizing notifications to the endoscopy center appointment system, or generating supplementary bowel cleansing drug prescription suggestions and pushing them to the outpatient doctor's workstation for review. The full-process digital archive management module integrates the guidance information viewing records from the standardized guidance information push module, the image data and its metadata uploaded from the excrement image acquisition and upload module, the cleanliness score and judgment results generated by the intelligent bowel preparation quality judgment module and the manual review records, and the intervention decision records executed from the intervention decision and execution module. It constructs a digital archive of the patient's entire bowel preparation process organized in a timeline format, supporting multi-dimensional retrieval and statistical analysis by patient, by time period, and by preparation quality results.

2. The screening and intervention system for bowel preparation before outpatient colonoscopy according to claim 1, characterized in that, In the intelligent judgment module for intestinal preparation quality, the model introduces an attention mechanism, and its core operation process is represented by the following formula: Let the feature map extracted from the input image after the convolutional layer be... ， in , , These represent the height, width, and number of channels of the feature map, and the attention weight map, respectively. The following was calculated using a lightweight subnetwork: ， in This represents a 1×1 convolution operation. This represents the Sigmoid activation function. Weighted feature map The calculation is as follows: ,in This indicates element-wise multiplication.

3. The screening and intervention system for bowel preparation before outpatient colonoscopy according to claim 1, characterized in that, In the intervention decision and execution module, the intervention strategy library built into the intervention decision engine contains at least three strategies: The first strategy applies to patients whose manual review rating is "good" and who have no record of compliance risk. It involves sending an examination confirmation notification to the patient's terminal. The second strategy applies to patients whose manual review rating is "unqualified" and who have a history of compliance risk. It automatically generates rescheduling suggestions, releases the original appointment time slot, and recommends a new examination time for the patient. The third strategy applies to cases where the manual review rating is "unqualified," but the system detects that there are still available examination slots after the original appointment time on the same day. If the doctor selects this option, the system will perform intelligent calculations: based on the time of the new slot, it will work backwards to clearly indicate the latest deadline for taking the bowel cleansing medication within the time window; this time reminder and rescheduling suggestion will be provided to the doctor and patient together to assist in decision-making; if the patient accepts and meets the time requirement, the system can assist them in rescheduling to the later slot; otherwise, it will proceed to the cross-day standard rescheduling process of the second strategy.

4. The screening and intervention system for bowel preparation before outpatient colonoscopy according to claim 1, characterized in that, In the excrement image acquisition and uploading module, the image quality pre-detection algorithm performs the following: sharpness detection is evaluated by calculating the gradient magnitude of the local area of ​​the image; illumination uniformity detection is determined by analyzing the distribution of the image histogram; and fecal sample proportion detection is estimated by preliminary color and texture segmentation to estimate the coverage ratio of excrement in the central area of ​​the image. All three detections have quantified thresholds.

5. The screening and intervention system for bowel preparation before outpatient colonoscopy according to claim 1, characterized in that, In the full-process digital record management module, the digital record supports the use of desensitized aggregated data for multi-dimensional statistical analysis at the group level. The analysis includes calculating the average preparation score of patients in different age groups, analyzing the impact of different versions of guidance materials on compliance, and tracking the effectiveness of intervention strategies.

6. The screening and intervention system for bowel preparation before outpatient colonoscopy according to claim 1, characterized in that, The system also includes a data security and privacy protection module, which ensures that all data communication between the patient terminal and the backend server uses an end-to-end encrypted transmission protocol, anonymizes uploaded image data on the server side to remove direct personal identifiers, and implements role-based access control policies to restrict data access permissions for medical staff in different roles.

7. A screening and intervention method for bowel preparation before outpatient colonoscopy, characterized in that, The method includes the following steps: S110, Before the examination, standardized bowel preparation guidance information is provided to the patient through the patient terminal. The guidance information includes a pre-examination notice document, medication guidance video and graphic instructions, and a diet control list. The patient's viewing status and duration of the guidance information are recorded, and patients who do not view the information in a timely manner are marked as subjects of compliance risk. S120, within a preset time window before the examination, receive image data of the last excrement after bowel preparation uploaded by the patient through the patient terminal. The uploading process is executed by an image acquisition module that integrates an image quality pre-detection algorithm. The algorithm ensures that the uploaded image meets preset clarity thresholds, illumination uniformity thresholds, and fecal sample proportion thresholds. S130, the image data is input into a pre-trained intelligent judgment model for intestinal preparation quality. The model generates a quantitative score for intestinal cleanliness based on big data learning, and the system automatically executes the judgment process according to the score value: If the score is in the range of 1-2 points, the system will automatically judge it as "excellent"; If the score is between 8 and 10, the system will automatically determine it as "failure"; If the score is in the range of 3-7, the system will automatically trigger the manual judgment process, push the image data and score value to the medical staff's terminal for review, and complete the judgment based on the final rating of the manual review. S140, based on the judgment result, execute a predefined intervention decision process, the process including confirming the examination appointment when the judgment is "excellent"; automatically triggering the rescheduling process when the judgment is "failed"; and triggering corresponding enhanced guidance, rescheduling or supplemental bowel cleansing medication intervention instructions based on the final rating, the patient's historical compliance record and the available appointment time slots on the same day. S150, integrate the guidance information review record in step S110, the image data and its metadata uploaded in step S120, the cleanliness score and judgment result and manual review record generated in step S130, and the intervention decision record executed in step S140 to establish a digital archive of the patient's entire bowel preparation process.

8. The screening and intervention method for bowel preparation before outpatient colonoscopy according to claim 7, characterized in that, In step S130, the intelligent judgment model for intestinal preparation quality introduces an attention mechanism, and its core operation process is represented by the following formula: Let the feature map extracted from the input image after the convolutional layer be... ， in , , These represent the height, width, and number of channels of the feature map, and the attention weight map, respectively. The following was calculated using a lightweight subnetwork: ， in This represents a 1×1 convolution operation. This represents the Sigmoid activation function. Weighted feature map The calculation is as follows: ,in This indicates element-wise multiplication.

9. The screening and intervention method for bowel preparation before outpatient colonoscopy according to claim 7, characterized in that, In step S140, when the judgment result is "good" or "unqualified" after manual review, the triggered intervention instruction is selected according to a preset strategy library, which includes: If the manual review rating is "good" and the patient has no risk of non-compliance, a test confirmation notification will be sent to the patient's terminal. If the manual review rating is "unqualified" or the patient's compliance record is poor, a rescheduling suggestion will be automatically generated and the endoscopy center's appointment system will be notified simultaneously. For patients whose manual review rating is "unqualified" but who have enough time remaining, a supplemental bowel cleansing medication prescription will be generated and sent for review.

10. The screening and intervention method for bowel preparation before outpatient colonoscopy according to claim 7, characterized in that, In step S120, the image quality pre-detection algorithm performs sharpness detection, illumination uniformity detection, and fecal sample proportion detection in parallel. The shooting operation is only allowed when the real-time status of all detection items meets their corresponding quantization thresholds.

Images