A method for improving the accuracy of AI computing Boston score by assisting the withdrawal speed of enteroscope

Abstract

The application discloses a method for improving the precision of AI calculated Boston score by auxiliary colonoscope withdrawal speed, and belongs to the cross field of colonoscopy and artificial intelligence technology. The main steps include: in the colonoscope withdrawal stage, starting the AI calculated Boston score module and the withdrawal speed calculation module based on the preset conditions, synchronously obtaining the score and the withdrawal speed and storing them in order; accumulating the withdrawal speed to generate a coordinate sequence, determining the length of the intestine after sorting; dividing the intestine into multiple sections and assigning the corresponding coordinate sequence; using the integral algorithm to combine the coordinate difference in the paragraph and the corresponding score, calculating the average cleanliness of each section and finally obtaining the final score. The method solves the problem of score weight imbalance caused by the difference in regional residence time in the traditional technology, improves the precision and reliability of the Boston score, adapts to the individual differences of the patient's intestine, and provides reliable support for the examination and clinical diagnosis and treatment of intestinal lesions.

Description

Technical Field

[0001] This invention relates to the intersection of colonoscopy and artificial intelligence technology, and in particular to a method for improving the accuracy of AI-calculated Boston score by assisting in the speed of colonoscopy withdrawal. Background Technology

[0002] In the field of colonoscopy, intestinal cleanliness assessment is a crucial step in assisting subsequent lesion screening. Currently, the Boston score is widely used in the industry as the core assessment standard for intestinal cleanliness. With the development of artificial intelligence technology, AI technology can now automatically calculate the Boston score based on intestinal images captured by colonoscopy. This type of technology has been gradually integrated into the colonoscopy procedure, providing automated cleanliness assessment references for clinical practice and reducing the subjective differences in manual scoring.

[0003] Existing AI technologies for calculating the Boston Score do not incorporate location-related information during colonoscopy withdrawal during data acquisition and score calculation. They simply acquire scores periodically according to fixed rules and take the average as the final result. This approach cannot avoid score weighting imbalances caused by differences in residence time in different areas of the intestine, and it is difficult to ensure that the score covers the entire intestinal region through effective coordinate correlation. Ultimately, this results in insufficient accuracy of the Boston Score, which cannot truly and uniformly reflect the overall cleanliness of the intestine, affecting the reliability of subsequent clinical judgments. Summary of the Invention

[0004] The purpose of this invention is to overcome one or more shortcomings of the prior art and provide a method to improve the accuracy of AI-calculated Boston score by assisting in the withdrawal speed of colonoscopes.

[0005] The objective of this invention is achieved through the following technical solution:

[0006] A method is provided to improve the accuracy of AI-calculated Boston score by accelerating colonoscopy withdrawal speed. This method includes the following steps:

[0007] S1: During the colonoscopy withdrawal stage, the AI ​​Boston score calculation module and withdrawal speed calculation module are activated based on preset trigger conditions; the Boston score calculated by the AI ​​based on the current intestinal image and the withdrawal speed calculated based on the current intestinal image and the historical intestinal image are acquired synchronously according to preset rules, and the Boston score and the corresponding withdrawal speed are stored in sequence; when a preset stop condition is detected, the above calculation and storage operations are stopped.

[0008] S2: The withdrawal speeds stored in S1 are accumulated in the order of storage to obtain a sequence of intestinal coordinates with coordinates as indices and Boston scores as index values ​​at the corresponding coordinates; the coordinate values ​​of the intestinal coordinate sequence are sorted, and the length of the patient's intestine is determined by traversing the sorted coordinate values;

[0009] S3: Divide the intestine length determined in S2 into multiple segments, and assign the sorted intestine coordinate sequence to each segment according to the division;

[0010] S4: Using an integral algorithm, the average cleanliness of each segment is calculated by combining the difference between adjacent coordinates within each segment with the Boston score at that coordinate, where the difference is the local length of the corresponding segment of the intestine; the average cleanliness of all segments is summed to obtain the final colonoscopy Boston score.

[0011] Furthermore, the preset trigger condition for starting the AI ​​calculation Boston score module and the withdrawal speed calculation module in S1 is that the colonoscope lens reaches the end of the ileocecal junction.

[0012] Furthermore, in S1, the Boston Scores and exit speed calculated by AI are synchronously acquired according to preset rules by cyclically performing the acquisition operation at preset time intervals.

[0013] Furthermore, the preset time interval can be dynamically adjusted according to the peristalsis frequency and degree of intestinal stenosis during the current colonoscopy.

[0014] Furthermore, in S2, the coordinate values ​​of the intestinal coordinate sequence are sorted in ascending order to eliminate the problem of repeated calculation of the intestinal region caused by the insertion operation during the withdrawal of the endoscope.

[0015] Furthermore, in S2, determining the length of the patient's intestine by traversing and sorting the coordinate values ​​specifically involves taking the maximum coordinate value in the sorted intestine coordinate sequence as the length of the patient's intestine.

[0016] Furthermore, in S3, the intestinal length determined in S2 is divided into multiple segments, specifically by dividing the intestinal length into three segments, corresponding to the left, middle, and right segments of the intestine, respectively.

[0017] Furthermore, the integral algorithm used in S4 to calculate the average cleanliness of each paragraph specifically includes: reading the coordinate values ​​of the corresponding paragraphs in sequence, calculating the difference between adjacent coordinate values ​​to obtain the local length of the intestine; multiplying the local length by the Boston score at the corresponding coordinate to obtain the local intestinal cleanliness integral value; accumulating all local intestinal cleanliness integral values ​​within the paragraph, and then dividing by the total length of the paragraph to obtain the average cleanliness of the paragraph.

[0018] Furthermore, the preset stop condition in S1 is the detection that the colonoscope lens has reached outside the body.

[0019] Furthermore, in S1, storing the Boston score and the corresponding withdrawal speed in sequence is specifically done by storing them in a preset data storage medium, and the data storage medium is associated with the timestamp information of each colonoscopy examination when the data is acquired.

[0020] The beneficial effects of this invention are:

[0021] (1) By simultaneously acquiring the score and exit speed during the exit phase, generating a coordinate sequence and performing piecewise integral calculation, the problem of uneven weighting in traditional scoring is solved, effectively improving the accuracy of the Boston Scoring System.

[0022] (2) By verifying data association, sorting coordinates to remove duplicates and filtering anomalies, the quality of data used for calculation is ensured, and the reliability of intestinal length determination and segment calculation is improved;

[0023] (3) By dynamically acquiring data, adapting to individual differences in the gut, and encrypting and storing data, the robustness of the solution is improved, and the security of medical data and its clinical reference value are guaranteed. Attached Figure Description

[0024] Figure 1 A flowchart outlining the specific steps of a method to improve the accuracy of AI-calculated Boston score by accelerating colonoscopy withdrawal.

[0025] Figure 2 The flowchart illustrates a method for improving the accuracy of AI-calculated Boston Score by accelerating colonoscopy withdrawal speed, as provided in this embodiment. Detailed Implementation

[0026] The technical solution of the present invention will be clearly and completely described below with reference to the embodiments. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0027] Example 1

[0028] See Figure 1 This paper provides a method to improve the accuracy of AI-calculated Boston Score by accelerating colonoscopy withdrawal, including the following steps:

[0029] S1: During the colonoscopy withdrawal stage, the AI ​​Boston score calculation module and withdrawal speed calculation module are activated based on preset trigger conditions; the Boston score calculated by the AI ​​based on the current intestinal image and the withdrawal speed calculated based on the current intestinal image and the historical intestinal image are acquired synchronously according to preset rules, and the Boston score and the corresponding withdrawal speed are stored in sequence; when a preset stop condition is detected, the above calculation and storage operations are stopped.

[0030] S2: The withdrawal speeds stored in S1 are accumulated in the order of storage to obtain a sequence of intestinal coordinates with coordinates as indices and Boston scores as index values ​​at the corresponding coordinates; the coordinate values ​​of the intestinal coordinate sequence are sorted, and the length of the patient's intestine is determined by traversing the sorted coordinate values;

[0031] S3: Divide the intestine length determined in S2 into multiple segments, and assign the sorted intestine coordinate sequence to each segment according to the division;

[0032] S4: Using an integral algorithm, the average cleanliness of each segment is calculated by combining the difference between adjacent coordinates within each segment with the Boston score at that coordinate, where the difference is the local length of the corresponding segment of the intestine; the average cleanliness of all segments is summed to obtain the final colonoscopy Boston score.

[0033] In some embodiments, before activating the AI-based Boston Performance Index (Boston) score calculation module and the retraction speed calculation module, a multi-dimensional quality check can be performed on the current intestinal image. This includes clarity check (analyzing the discernibility of intestinal mucosal texture in the image to remove images with blurred textures due to lens smudges or insufficient lighting) and stability check (determining whether there is inter-frame shift due to operation jitter; if the shift exceeds a preset range, image stabilization processing is triggered). When simultaneously acquiring the score and retraction speed, a data correlation check can be added to ensure that each acquired Boston Performance Index and corresponding retraction speed comes from the intestinal image at the same time point, avoiding data corruption. In the data storage phase, in addition to storing scores and speeds sequentially, real-time consistency checks can be performed on multiple consecutively stored data sets. If the fluctuation range of the withdrawal speed or Boston score in two adjacent data sets exceeds a preset threshold, abnormal data is automatically marked and the intestinal image characteristics at the time of the abnormality are recorded for subsequent manual review. Before stopping calculation and storage operations, a data integrity check can also be triggered to confirm that the stored data covers the complete withdrawal process from the meeting of the triggering condition to the triggering of the stopping condition. If there are missing data segments, it can prompt the re-collection of withdrawal data corresponding to the missing segments, further ensuring the integrity and reliability of the data used for subsequent calculations.

[0034] The default trigger condition for starting the AI ​​calculation of the Boston Score module and the exit speed calculation module in S1 is that the colonoscope lens reaches the end of the ileocecal junction.

[0035] In some embodiments, the preset triggering condition can be set to a dual verification mechanism of "hardware sensor signal + AI image recognition".

[0036] The hardware sensors can be distance sensors or position encoders built into the colonoscope to collect real-time displacement data of the lens relative to the beginning of the intestine. When the displacement data reaches the typical distance range of the ileocecal junction, an initial trigger signal is generated. At the same time, the AI ​​image recognition module is activated to perform feature analysis on the intestinal image captured by the lens, focusing on identifying the unique physiological structures of the ileocecal junction (such as the leaflet morphology of the ileocecal valve, the difference in mucosal color at the junction of the ileum and colon, and the vascular distribution pattern). When the AI ​​image recognition module determines that the current image meets the feature matching criteria of the ileocecal junction (the matching degree exceeds the preset threshold), a secondary trigger signal is generated. Only when both the initial trigger signal and the secondary trigger signal are detected and the time difference is within the preset range, are the AI ​​Boston score calculation module and the withdrawal speed calculation module officially activated.

[0037] This dual verification method can effectively avoid displacement measurement deviations caused by intestinal folds when relying solely on hardware sensors, or feature misjudgments caused by intestinal inflammation when relying solely on image recognition. It is especially suitable for patients with intestinal structural abnormalities (such as intestinal malformations and postoperative adhesions), and significantly improves the accuracy and robustness of triggering conditions.

[0038] In S1, the Boston Scores and exit speed calculated by AI are synchronously acquired according to preset rules. Specifically, the acquisition operation is performed cyclically at preset time intervals.

[0039] In some embodiments, the synchronous acquisition of data according to preset rules can be replaced by an "adaptive triggering mechanism based on dynamic changes in intestinal images".

[0040] First, AI algorithms are used to analyze the feature changes in consecutive frames of intestinal images in real time. The calculation dimensions of feature changes include pixel differences in the mucosal region (such as changes in pixel grayscale values ​​caused by the movement of intestinal contents), morphological changes in the intestinal contour (such as changes in the cross-sectional shape of the intestine caused by camera movement), and positional offsets of key structures (such as folds and blood vessels). Multiple feature change thresholds are set. When the feature changes in consecutive frames exceed the high threshold (indicating that the camera movement speed is fast or the intestinal contents flow frequently and the image information is updated quickly), the data acquisition interval is shortened to collect Boston performance scores and exit speed at a higher frequency to ensure that information in key areas is not missed. When the feature changes are below the low threshold (indicating that the camera is stationary or moving slowly and the image information is stable), the data acquisition interval is extended to reduce the collection of repetitive data and reduce storage resource consumption and computational load.

[0041] Simultaneously, the initial setting of the characteristic change threshold can be preset according to different stages of colonoscopy (such as the rapid movement of the lens in the early stage of withdrawal and the slow movement of the lens when approaching the outside of the body), and then dynamically adjusted in combination with real-time image changes. Compared with fixed time intervals, this adaptive triggering mechanism can ensure the temporal resolution of data when the lens moves rapidly, and avoid invalid data redundancy when the lens is stationary, realizing "on-demand acquisition" and greatly improving the efficiency and effectiveness of data acquisition.

[0042] The preset time interval can be dynamically adjusted according to the peristalsis frequency and degree of intestinal stenosis during the current colonoscopy.

[0043] In some embodiments, the adjustment of the preset time interval can be further combined with "patient preoperative bowel assessment data + real-time examination feedback data" to construct a multi-dimensional adjustment model.

[0044] Preoperative bowel assessment data can be extracted from the patient's preoperative examination report, including the overall length of the intestine obtained by abdominal CT or intestinal angiography, the diameter of each segment of the intestine (used to predict the distribution of narrow areas), and the results of bowel motility function assessment (such as peristaltic frequency and amplitude obtained by gastrointestinal motility monitoring). Based on these preoperative data, initial time intervals are preset for different intestinal segments (such as shorter initial intervals for narrow segments and shorter initial intervals for frequently peristaltic segments).

[0045] During real-time examination, in addition to monitoring intestinal peristalsis frequency (through AI analysis of the contraction cycle of the intestinal wall in the image) and stenosis degree (through AI identification of the deviation of the intestinal cross-sectional diameter from the normal range), a "score fluctuation feedback" dimension is added. When the fluctuation amplitude of the Boston score acquired multiple times exceeds the preset value (indicating unstable cleanliness in that area, possibly with local contents residue or mucosal lesions), the data acquisition interval for that area is automatically shortened, and data is collected more intensively to more accurately reflect the true changes in cleanliness. When the score fluctuation amplitude remains within a stable range, the time interval can be appropriately widened. Simultaneously, the upper and lower limits of the interval adjustment can be set in the adjustment model to avoid excessively short or long intervals due to extreme conditions (such as vigorous peristalsis), ensuring the stability of data acquisition. This multi-dimensional adjustment model can fully adapt to the individual intestinal characteristics of different patients, making the adjustment of the time interval more targeted and further optimizing the quality of data acquisition.

[0046] In S2, the coordinate values ​​of the intestinal coordinate sequence are sorted in ascending order to eliminate the problem of repeated calculation of the intestinal region caused by the insertion operation during the withdrawal of the endoscope.

[0047] In some embodiments, sorting the coordinate values ​​of the intestinal coordinate sequence can be optimized into a three-step processing flow of "abnormal coordinate filtering → segmented sorting → secondary verification".

[0048] The first step is abnormal coordinate filtering: Based on the physiological length range of the human intestine (combined with the standard values ​​of adult or child intestinal length based on clinical statistics), a reasonable range of coordinate values ​​is set, and abnormal coordinates that exceed this range (such as excessively large or negative values ​​caused by sensor malfunction) are removed; at the same time, the difference between adjacent coordinate values ​​is calculated. If the absolute value of the difference exceeds the preset maximum reasonable step size (set according to the maximum single movement distance of the colonoscope lens), it is determined to be an abnormal jump coordinate. The abnormal coordinate is replaced by linear interpolation or the mean of adjacent valid coordinates to avoid abnormal data from interfering with the sorting results;

[0049] The second step is segmented sorting: Considering that there may be multiple local insertions during the withdrawal process (such as the lens repeatedly entering and exiting in a certain section of the intestine), the intestinal coordinate sequence can be divided into multiple continuous data segments according to the timestamp, with each data segment corresponding to one continuous withdrawal operation; the coordinate values ​​in each data segment are sorted separately from smallest to largest, and then the coordinate sequences after segmentation and sorting are spliced ​​together in chronological order to form a preliminary sorting result;

[0050] The third step is secondary verification: the continuity of the initially sorted coordinate sequence is checked, and the difference between adjacent coordinates is calculated to see if it is evenly distributed within a reasonable range. If there are multiple consecutive coordinate differences of zero (indicating that the same position is recorded repeatedly), the coordinates and corresponding scores of the first record are retained, and subsequent duplicates are deleted. If there are coordinate values ​​in reverse order (indicating that there are still unprocessed endoscopy operations), the local area is sorted and adjusted again to ensure that the final output coordinate sequence is strictly arranged in ascending order, without repetition or abnormal jumps, fundamentally eliminating the problem of repeated calculations caused by endoscopy operations, and providing a high-precision coordinate data foundation for subsequent determination of intestinal length and segmentation.

[0051] In S2, determining the length of the patient's intestines by traversing and sorting the coordinate values ​​specifically involves taking the maximum coordinate value in the sorted sequence of intestinal coordinates as the length of the patient's intestines.

[0052] In some embodiments, the length of a patient's intestine can be determined using a comprehensive method of "maximum coordinate value screening + multi-dimensional verification".

[0053] First, the sorted intestinal coordinate sequence is traversed, and the maximum value among all coordinates is extracted as the initial length value. Then, the first verification is initiated: combining the patient's preoperative baseline data (such as height, weight, and age), the expected intestinal length range for the patient is calculated using a clinically established regression model. If the initial length value is within the expected range, it is tentatively set as the intestinal length. If the initial length value exceeds the expected range, the second verification is initiated: the intestinal image corresponding to the initial length value is retrieved, and AI image recognition is used to determine whether the location is the intestinal exit point (e.g., identifying the air environment outside the lens, external tissue, or examination instruments). If it is determined to be the exit point, the initial length value is confirmed to be valid. If no exit point features are identified (possibly due to coordinate measurement errors), the last data segment of the coordinate sequence is further analyzed, and the mean of the last N consecutive coordinate values ​​is calculated (N can be dynamically set according to the total amount of data collected). If the deviation between the mean and the initial length value is within a preset range, the mean is taken as the intestinal length. If the deviation is large, manual review is prompted, and the final intestinal length is determined by combining the real-time operation record of the colonoscopy operator (e.g., the lens position feedback at the end of the endoscopy withdrawal). This comprehensive determination method can effectively avoid errors caused by relying solely on the maximum value of the coordinates (such as the coordinate values ​​being artificially high due to the lens repeatedly moving in and out of the end of the intestine, or the extreme value deviation caused by temporary sensor malfunctions). It is especially suitable for children or elderly patients with individual differences in intestinal length, and improves the accuracy and reliability of intestinal length determination.

[0054] In S3, the intestinal length determined in S2 is divided into multiple segments. Specifically, the intestinal length is divided into three segments on average, corresponding to the left, middle and right segments of the intestine.

[0055] In some embodiments, the intestinal segments may be divided using a "precise division method based on the physiological structure and functional partitioning of the intestine" rather than a simple average division.

[0056] First, the AI ​​image recognition module analyzes the intestinal images corresponding to the sorted intestinal coordinate sequence frame by frame to identify the characteristic structural landmarks of each physiological region of the intestine. For example, the hepatic flexure of the colon (the junction of the ascending colon and transverse colon, with typical tortuous shape and mucosal color transition characteristics) is used as the dividing point between the "right segment (ascending colon)" and the "middle segment (transverse colon)". The splenic flexure of the colon (the junction of the transverse colon and descending colon, also with unique tortuous structure and vascular distribution) is used as the dividing point between the "middle segment (transverse colon)" and the "left segment (descending colon, sigmoid colon, rectum)". The process involves identifying the dividing points; simultaneously, combining the coordinate values ​​of the corresponding structural landmarks in the intestinal coordinate sequence, the specific coordinate positions of each dividing point are determined; if a structural landmark cannot be clearly identified due to intestinal inflammation, contents obscuring the view, or other reasons, the patient's preoperative intestinal imaging data (such as CT or MRI images) can be retrieved, and image registration technology can be used to map the physiological partition coordinates in the preoperative images to the current intestinal coordinate sequence to supplement the determination of the dividing point coordinates; finally, based on the coordinate values ​​of each dividing point, the intestinal coordinate sequence is divided into three segments corresponding to the right, middle, and left segments. This physiological structure-based division method perfectly matches the actual anatomical characteristics of the intestine, more accurately corresponds to the Boston Assessment's requirements for the cleanliness of different intestinal segments, and avoids the problem of "mismatch between physiological partitions and scoring segments" that may occur during average division (such as classifying a part of the transverse colon into the right or left segment), making the subsequent segmental scoring results more clinically valuable.

[0057] The S4 algorithm for calculating the average cleanliness of each segment specifically includes: reading the coordinate values ​​of the corresponding segments in sequence, calculating the difference between adjacent coordinate values ​​to obtain the local length of the intestine; multiplying the local length by the Boston score at the corresponding coordinate to obtain the local intestinal cleanliness integral value; accumulating all local intestinal cleanliness integral values ​​within a segment, and then dividing by the total length of the segment to obtain the average cleanliness of the segment.

[0058] In some embodiments, the average cleanliness of a paragraph can be calculated using a "weighted integral algorithm," which optimizes the integral calculation process by introducing regional weight coefficients, making the scoring results more in line with clinical diagnosis and treatment needs.

[0059] First, weighting coefficients are set according to the lesion risk levels of different intestinal regions: Clinical data shows that the rectum and sigmoid colon are high-incidence areas for colorectal cancer, and the ascending colon is a common area for adenomatous polyps. The weighting coefficients for these high-risk areas can be set higher than those for other ordinary areas (e.g., the weighting coefficient for ordinary areas is 1.0, and the weighting coefficient for high-risk areas is 1.2-1.5; the specific values ​​can be dynamically adjusted according to the latest clinical guidelines or hospital treatment data). Then, when calculating the local intestinal cleanliness score, in addition to multiplying the local length by the Boston score at the corresponding coordinate, it is also necessary to multiply by the weighting coefficient corresponding to the local area to obtain the weighted local score. All weighted local scores within a paragraph are summed, and then divided by the "weighted total length" of the paragraph (i.e., the sum of the product of each local length and its corresponding weighting coefficient) to finally obtain the average cleanliness of the paragraph. Furthermore, the weighting coefficients can be dynamically adjusted based on the patient's individual circumstances (such as family history of intestinal cancer or previous polyp removal): for patients with a family history of cancer, the weighting coefficients for the rectum and sigmoid colon can be further increased; for patients with a history of ascending colon polyp removal, the weighting coefficient for the ascending colon can be increased. This weighted integral algorithm makes the scoring results more prominent in terms of the cleanliness of high-risk areas, helping doctors to more accurately identify areas with potential lesions and providing more targeted reference for subsequent diagnosis and treatment decisions.

[0060] The preset stop condition in S1 is when the colonoscope lens is detected to have reached the outside of the body.

[0061] In some embodiments, the preset stop condition can be designed as a "multi-signal fusion judgment mechanism," where the fused signal types include environmental parameter signals from the colonoscope lens, hardware displacement signals, and feedback signals from the operating physician. First, the environmental sensor built into the colonoscope lens collects real-time parameters such as light intensity, media refractive index (the refractive index of air differs significantly from that of intestinal fluid), and temperature. When the detected values ​​of these parameters continuously match the characteristic range of the external environment for multiple frames (e.g., a sudden increase in light intensity, refractive index approaching that of air), an environmental judgment signal is generated. Second, the colonoscope's displacement sensor or encoder detects changes in lens displacement. When the displacement data shows that the lens has retracted from its maximum depth within the intestine to its initial position (close to the outside), and subsequent displacement changes approach zero, a displacement judgment signal is generated.

[0062] Meanwhile, a manual confirmation button is provided on the colonoscopy control console. The operator can trigger a manual judgment signal when the endoscope reaches the outside of the body, based on real-time observation of the lens and their tactile feedback. The preset stop condition trigger logic is as follows: when both the environmental judgment signal and the displacement judgment signal are met simultaneously for a preset duration (e.g., 1-2 seconds), or when either judgment signal is met and the manual judgment signal is triggered, the AI ​​calculation of the Boston score and the withdrawal speed is stopped, and the stored data is automatically saved. This multi-signal fusion mechanism effectively avoids the problem of misjudgment due to a single signal (such as false triggering of environmental signals due to brief contact with external light, or false judgment of displacement signals due to narrowing of the distal intestine), while retaining the doctor's manual intervention authority. It balances the efficiency of automated judgment with the reliability of manual confirmation, ensuring accurate timing of stopping the operation.

[0063] In S1, storing the Boston score and the corresponding withdrawal speed in sequence is specifically done by storing them in a preset data storage medium, and the data storage medium is associated with the timestamp information of each colonoscopy examination when the data is acquired.

[0064] In some embodiments, the data storage process can be expanded into a complete solution of "multi-dimensional data association storage + secure encryption + traceable management". First, in addition to the Boston score, endoscopy withdrawal speed, and timestamp, the stored data dimensions also need to be associated with and stored as follows: a complete intestinal image frame at each data acquisition (not just a thumbnail, for easy subsequent high-definition review), a unique identifier for the colonoscopy equipment (such as the equipment number, used to trace the equipment status), the operator's employee number (for easy accountability), and the patient's anonymized identifier (such as an encrypted medical record number, protecting privacy while supporting data association).

[0065] Secondly, a layered encryption mechanism is employed to protect the stored data: SSL / TLS encryption is used during transmission to the storage medium, and AES-256 encryption is used within the storage medium to encrypt sensitive information (such as patient identification and complete image frames). Only authorized personnel (such as attending physicians and quality control personnel) can decrypt and access the data after dual authentication (account password + dynamic verification code). Furthermore, a data traceability log is established to record every data access, modification, and deletion operation, including the operator, operation time, and operation content, ensuring traceability throughout the data's lifecycle. In addition, the storage medium can be designed with a dual storage architecture of "local cache + cloud backup": during the examination, data is first stored in real-time in the local cache (to avoid data loss due to network fluctuations), and automatically synchronized to the hospital's cloud server after the examination, with data integrity verification (such as using hash value verification). If inconsistencies are found between local and cloud data, automatic resynchronization is triggered to ensure data security and availability. This storage solution not only meets the privacy and compliance requirements of medical data but also provides complete and reliable data support for subsequent score result review and clinical research (such as AI model optimization and cleanliness-lesion correlation analysis).

[0066] Example 2

[0067] See Figure 2 This embodiment describes a specific implementation process of a method to improve the accuracy of AI-calculated Boston score by assisting in the withdrawal speed of a colonoscope, including the following steps:

[0068] Step 1: Startup of the calculation module and data acquisition and storage during the withdrawal phase:

[0069] When the colonoscopy enters the withdrawal stage and the colonoscope lens moves to the end of the ileocecal junction, the AI-based Boston Scores calculation module and the withdrawal speed calculation module are activated simultaneously.

[0070] The AI-powered Boston Scoring module captures the current intestinal image transmitted by the colonoscope in real time. It extracts features such as intestinal mucosal regions and residual contents through image recognition algorithms, and outputs the Boston Scoring Score corresponding to the current image based on the Boston Scoring Score (a grading system for intestinal cleanliness). Simultaneously, the endoscope withdrawal speed calculation module retrieves the current intestinal image and a historical intestinal image acquired 0.5 seconds earlier. By combining the pixel position offset of fixed intestinal feature points (such as mucosal fold apex and vascular branch points) in the two images with the imaging parameters of the colonoscope lens, it calculates the current endoscope withdrawal speed.

[0071] After each calculation of the Boston score and withdrawal speed, the two sets of data are stored sequentially into a data list. During storage, a data association index is established to ensure that each Boston score is accurately matched with its corresponding withdrawal speed, avoiding data misalignment. When the colonoscope lens is moved outside the body, the AI ​​Boston score calculation module and withdrawal speed calculation module immediately stop working after confirming this status via the lens position detection module, terminating the data acquisition and storage process.

[0072] Step 2: Generation of intestinal coordinate sequence and determination of intestinal length:

[0073] After data acquisition is terminated, the withdrawal speeds stored in the data list are accumulated in the order of storage: the initial position corresponding to the withdrawal speed acquired for the first time is taken as the origin of the coordinates, and the result of each subsequent accumulation is the real-time position coordinates of the colonoscope lens in the intestine, which finally forms an intestinal coordinate sequence containing position coordinates and corresponding Boston scores (in this sequence, the position coordinates are used as the index, and the Boston score at the corresponding coordinate is used as the index value).

[0074] Because there may be partial retraction operations during the retraction process (such as retracting the lens to adjust the observation angle), the coordinates of the same intestinal region may be recorded repeatedly. Therefore, all position coordinates in the intestinal coordinate sequence are reordered in ascending order of value. By sorting, redundant data corresponding to duplicate coordinates are eliminated, ensuring that the coordinate sequence completely covers the entire intestinal region without duplicate records.

[0075] Traverse the sorted sequence of intestinal coordinates and extract the maximum coordinate value, which directly corresponds to the total length of the patient's intestine in this examination, thus completing the accurate determination of intestinal length.

[0076] Step 3: Intestinal segmentation and coordinate sequence allocation:

[0077] The total length of the intestine determined by the test is divided into three equal segments, corresponding to the left, middle and right segments of the intestine, respectively. The equal division of length ensures that the weight of the cleanliness score of each segment is consistent and avoids the scoring deviation caused by the difference in segment length.

[0078] Based on the lengths of the left, middle, and right segments of the intestine, the sorted intestinal coordinate sequence is proportionally divided so that each segment corresponds to all location coordinates of that region and the associated Boston score, forming the left segment intestinal coordinate sequence, the middle segment intestinal coordinate sequence, and the right segment intestinal coordinate sequence, providing a data basis for segmental cleanliness calculation.

[0079] Step 4: Calculation of average cleanliness across segments and summary of final scores:

[0080] For the coordinate sequence of the left intestinal segment, adjacent coordinate values ​​are read in ascending order. The difference between the next coordinate value and the previous coordinate value is calculated. This difference is the local intestinal length between the two coordinate segments. The local intestinal length is multiplied by the Boston score at the corresponding coordinate to obtain the cleanliness score of the local intestinal segment.

[0081] Calculate the local intestinal cleanliness integral value corresponding to all adjacent coordinates in the coordinate sequence of the left intestinal segment in turn. Summate all local integral values ​​to obtain the total cleanliness integral value of the left intestinal segment. Then divide the total integral value by the total length of the left intestinal segment to obtain the average cleanliness of the left intestinal segment.

[0082] Using the same calculation logic, the coordinate sequences of the middle segment of the intestine and the right segment of the intestine are processed respectively: the local length is calculated by the difference between adjacent coordinates, the local length is multiplied by the score to obtain the local integral, and the local integrals are accumulated and divided by the total length of the segment to obtain the average cleanliness of the middle segment of the intestine and the average cleanliness of the right segment of the intestine.

[0083] The average cleanliness of the left, middle, and right segments of the intestine is summed to obtain the final Boston score for this colonoscopy. This score can uniformly reflect the cleanliness of the entire intestine, effectively avoiding scoring deviations caused by differences in the time spent in different areas of the intestine, and improving scoring accuracy.

[0084] This embodiment effectively solves the problem of uneven weighting caused by differences in the dwell time of different intestinal regions in the traditional Boston score for colonoscopy by mapping the withdrawal speed with intestinal coordinates and combining a multi-dimensional data optimization processing mechanism, significantly improving the accuracy and reliability of the score. During the data acquisition phase, a dynamic triggering mechanism, dual verification, and multi-dimensional validation ensure the complete capture of key regional information while avoiding redundant invalid data. Abnormal and misaligned data are filtered out, laying a high-quality data foundation for subsequent calculations. In the coordinate processing and intestinal length determination stages, sorting for deduplication, anomaly filtering, and multi-dimensional validation eliminate the influence of insertion operation and measurement errors, achieving accurate determination of intestinal length. Segment division supports both uniform segmentation to ensure consistent weights and precise partitioning that conforms to the physiological structure of the intestine, adapting to different clinical assessment needs. Integral calculation (including weighted optimization) not only evenly covers the entire intestinal region but also highlights the cleanliness weight of high-risk lesion areas, enhancing the clinical reference value of the score. Data storage employs encrypted backup and traceability design to ensure the security and availability of medical data. The overall solution is adapted to individual differences in the intestines of different patients (such as structural abnormalities, age-related length differences, etc.), and is highly robust. The final Boston score can accurately reflect the overall cleanliness of the intestines, providing reliable technical support for the accurate screening, diagnosis and treatment decisions and clinical research of intestinal lesions.

[0085] The above description is merely a preferred embodiment of the present invention. It should be understood that the present invention is not limited to the forms disclosed herein and should not be construed as excluding other embodiments. It can be used in various other combinations, modifications, and environments, and can be altered within the scope of the concept described herein through the above teachings or related technologies or knowledge. Modifications and variations made by those skilled in the art that do not depart from the spirit and scope of the present invention should be within the protection scope of the appended claims.

Claims

1. A method for improving the accuracy of AI-calculated Boston score by accelerating colonoscope withdrawal, characterized in that... Includes the following steps: S1: During the colonoscopy withdrawal stage, the AI ​​Boston score calculation module and withdrawal speed calculation module are activated based on preset trigger conditions; the Boston score calculated by the AI ​​based on the current intestinal image and the withdrawal speed calculated based on the current intestinal image and the historical intestinal image are acquired synchronously according to preset rules, and the Boston score and the corresponding withdrawal speed are stored in sequence; when a preset stop condition is detected, the above calculation and storage operations are stopped. S2: The withdrawal speeds stored in S1 are accumulated in the order of storage to obtain a sequence of intestinal coordinates with coordinates as indices and Boston scores as index values ​​at the corresponding coordinates; the coordinate values ​​of the intestinal coordinate sequence are sorted, and the length of the patient's intestine is determined by traversing the sorted coordinate values; S3: Divide the intestine length determined in S2 into multiple segments, and assign the sorted intestine coordinate sequence to each segment according to the division; S4: Using an integral algorithm, the average cleanliness of each segment is calculated by combining the difference between adjacent coordinates within each segment with the Boston score at that coordinate, where the difference is the local length of the corresponding segment of the intestine; The average cleanliness of all segments is summed to obtain the final Boston score for colonoscopy. Specifically, in S4, the integral algorithm is used to calculate the average cleanliness of each paragraph, which includes: reading the coordinate values ​​of the corresponding paragraphs in sequence, calculating the difference between adjacent coordinate values ​​to obtain the local length of the intestine; multiplying the local length by the Boston score at the corresponding coordinate to obtain the local intestinal cleanliness integral value; accumulating all local intestinal cleanliness integral values ​​within the paragraph, and then dividing by the total length of the paragraph to obtain the average cleanliness of the paragraph. Specifically, in S1, storing the Boston score and the corresponding withdrawal speed in sequence is done by storing them in a preset data storage carrier, and the data storage carrier is associated with the timestamp information of each colonoscopy examination when the data is acquired. The sorting of the coordinate values ​​of the intestinal coordinate sequence includes the following steps: The first step is abnormal coordinate filtering: a reasonable range of coordinate values ​​is set based on the physiological length range of the human intestine, and abnormal coordinates that exceed the range are removed; at the same time, the difference between adjacent coordinate values ​​is calculated. If the absolute value of the difference exceeds the preset maximum reasonable step size, it is determined to be an abnormal jump coordinate. The abnormal coordinate is replaced by linear interpolation or the mean of adjacent valid coordinates to avoid abnormal data from interfering with the sorting results. The second step is segmented sorting: Considering that there may be multiple partial re-entries during the retraction process, the intestinal coordinate sequence can be divided into multiple continuous data segments according to the timestamp, with each data segment corresponding to one continuous retraction operation; the coordinate values ​​in each data segment are sorted separately from smallest to largest, and then the coordinate sequences after segmentation and sorting are spliced ​​together in chronological order to form a preliminary sorting result; The third step is secondary verification: the continuity of the initially sorted coordinate sequence is checked, and the difference between adjacent coordinates is calculated to see if it is evenly distributed within a reasonable range. If there are multiple consecutive coordinate differences of zero, the coordinates and corresponding scores of the first record are retained, and subsequent duplicates are deleted. If there are coordinate values ​​in reverse order, the local area is sorted and adjusted again to ensure that the final output coordinate sequence is strictly arranged in ascending order, without repetition or abnormal jumps. This fundamentally eliminates the problem of repeated calculations caused by the endoscope insertion operation and provides a high-precision coordinate data foundation for subsequent determination of intestinal length and segmentation.

2. The method according to claim 1, characterized in that, The default trigger condition for starting the AI ​​calculation of the Boston Score module and the exit speed calculation module in S1 is that the colonoscope lens reaches the end of the ileocecal junction.

3. The method according to claim 1, characterized in that, In S1, the Boston Scores and exit speed calculated by AI are synchronously acquired according to preset rules. Specifically, the acquisition operation is performed cyclically at preset time intervals.

4. The method according to claim 3, characterized in that, The preset time interval can be dynamically adjusted according to the peristalsis frequency and degree of intestinal stenosis during the current colonoscopy.

5. The method according to claim 1, characterized in that, In S2, the coordinate values ​​of the intestinal coordinate sequence are sorted in ascending order to eliminate the problem of repeated calculation of the intestinal region caused by the insertion operation during the withdrawal of the endoscope.

6. The method according to claim 1, characterized in that, In S2, determining the length of the patient's intestines by traversing and sorting the coordinate values ​​specifically involves taking the maximum coordinate value in the sorted sequence of intestinal coordinates as the length of the patient's intestines.

7. The method according to claim 1, characterized in that, In S3, the intestinal length determined in S2 is divided into multiple segments. Specifically, the intestinal length is divided into three segments, corresponding to the left, middle and right segments of the intestine.

8. The method according to claim 1, characterized in that, The preset stop condition in S1 is when the colonoscope lens is detected to have reached the outside of the body.

Images