Medical instrument use supervision system based on image recognition

Abstract

The invention discloses a medical instrument use supervision system based on image recognition, particularly relates to the field of medical instrument supervision, is used for solving the problems of endoscope pollutant monitoring and use state early warning, and is used for automatically adjusting a time unit and eliminating interference in a non-operation stage through a dynamic time division technology and a system. The method comprises the following steps: calculating repeatability indexes and diffusion regularity indexes of adjacent time unit pollutant areas, and synthesizing the repeatability indexes and the diffusion regularity indexes to generate decision indexes, so as to effectively evaluate the early warning level of pollutant influence; meanwhile, the system can accurately judge whether the use state of the endoscope is in a stable decay or abrupt deterioration stage or not by combining the change trends of the short-term and long-term decision indexes; real-time and fine monitoring of pollutants on the surface of the medical instrument is achieved, sensitivity and accuracy of pollution detection are remarkably improved, and equipment performance reduction and infection risks can be pre-warned in advance, so that cleaning and maintenance strategies are optimized, and safety and accuracy of medical operation are ensured.

Description

Technical Field

[0001] The present invention relates to the field of medical device supervision, and more specifically, to a medical device use supervision system based on image recognition. Background Art

[0002] In modern medical environments, medical devices that require cameras, such as endoscopes, are widely used in various diagnostic and surgical operations. These devices rely on high-quality image transmission to ensure that medical staff can clearly observe the patient's internal organs and tissues, so as to perform accurate diagnosis and surgery. However, during frequent use of endoscopes, contaminants such as tissue fluid, blood particles and other biological residues easily accumulate on the lens and working end surface. These contaminants will not only gradually reduce the clarity and color reproduction of the image, affecting the accuracy of medical operations, but may also become a potential source of infection, threatening patient safety.

[0003] Existing endoscope cleaning and maintenance methods mainly rely on manual visual inspection and regular cleaning, which have many shortcomings such as inaccurate evaluation, insufficient real-time monitoring capabilities, and the inability to dynamically adjust the cleaning cycle. The specific technical problem is that with the increase in the number of uses, the accumulation of pollutants presents a complex repeatability and expansibility pattern. Traditional detection methods are difficult to effectively identify the slight changes and dynamic trends of pollutants, resulting in pollution problems that are often not discovered and handled in time before they seriously affect the performance of the equipment. In addition, the existing system lacks intelligent analysis capabilities based on image recognition, and cannot dynamically divide the use time units to accurately monitor the distribution and evolution of pollutants, which in turn affects subsequent cleaning and maintenance decisions. These technical challenges limit the comprehensiveness and reliability of the medical device use supervision system. There is an urgent need for an innovative solution based on image recognition to achieve efficient and accurate real-time monitoring and early warning of pollutants during the use of medical devices.

[0004] In order to solve the above problems, a technical solution is now provided. Summary of the invention

[0005] In order to overcome the above-mentioned defects of the prior art, the embodiments of the present invention provide a medical device use supervision system based on image recognition, which automatically adjusts the time unit through dynamic time division technology, ensures data stability, and eliminates interference in the non-operation stage; adopts multi-dimensional pollutant feature extraction, and calculates the repeatability index and diffusion regularity index of the pollutant area of ​​adjacent time units, and comprehensively generates a decision index to effectively evaluate the warning level of pollutant impact; combines the changing trends of short-term and long-term decision indexes to accurately judge whether the use status of the endoscope is in a stable decay or sudden deterioration stage. It realizes real-time and precise monitoring of pollutants on the surface of medical devices, significantly improves the sensitivity and accuracy of pollution detection, can warn of equipment performance degradation and infection risks in advance, optimize cleaning and maintenance strategies, extend the service life of equipment, and ensure the safety and accuracy of medical operations to solve the problems raised in the above background technology.

[0006] To achieve the above object, the present invention provides the following technical solutions: A medical device use supervision system based on image recognition includes: an image preprocessing module, a feature extraction module, a rule recognition module and a trend inference module.

[0007] Image preprocessing module: obtains imaging data during the use of the endoscope, dynamically divides the time unit to eliminate interfering images in the non-operation stage, and outputs the standardized imaging data and time unit division results to the feature extraction module.

[0008] Feature extraction module: extracts key features of pollutants based on image data within each time unit, outputs pollutant features of each time unit and passes them to the pattern recognition module.

[0009] Pattern recognition module: According to the overlapping area and diffusion characteristics of pollutants in adjacent time units, the repeatability index and diffusion regularity index of the pollutant area are calculated, and the decision index is generated by combining these two indexes. The output of the calculated decision index is passed to the trend inference module.

[0010] Trend inference module: Based on the short-term and long-term comprehensive decision index change trends, the usage status of the endoscope is analyzed as stable decay or sudden deterioration.

[0011] In a preferred embodiment, the image preprocessing module includes the following contents: The process of dynamically dividing time is divided into four steps, namely preliminary segmentation, data feature analysis, unit time adjustment and segmentation optimization.

[0012] Initial segmentation: The image data of the entire process used by the endoscope is collected in time series to form a continuous frame sequence. Each frame of the image contains the imaging timestamp and picture feature data. The acquisition process is divided into several initial segments using the timestamp, and each segment has an equal time span.

[0013] Data feature analysis: Extract features from the image data in each initial segment, including the frame-to-frame change rate, illumination change amplitude, and invalid frame ratio.

[0014] Inter-frame change rate: calculate the image change rate of adjacent frames within a clip to evaluate the dynamics of the image; Illumination variation: The illumination stability is assessed by the standard deviation of the illumination histogram mean.

[0015] Invalid Frame Ratio: The ratio of low-resolution frames or excessively blurred frames in the detected clip is defined as the invalid frame ratio.

[0016] Unit time adjustment: According to the data feature analysis results, the time span of the initial segment is dynamically adjusted; the segment division is adjusted to a unit time suitable for pollution feature extraction. The adjustment logic is as follows: When the inter-frame change rate and the illumination change amplitude are both higher than the corresponding thresholds, it means that the clip contains drastic dynamics or illumination fluctuations, and the clip time span needs to be shortened to filter out interference.

[0017] When the inter-frame change rate and illumination change amplitude are lower than the corresponding threshold and the invalid frame ratio is lower than the corresponding threshold, it means that the fragment data is stable and the time span needs to be extended to increase the unit data volume.

[0018] Segment optimization: Optimize the dynamically adjusted time units to ensure that the time units are representative and consistent. The optimization methods include feature weight smoothing, critical segment merging, and significant anomaly removal.

[0019] In a preferred embodiment, the feature extraction module includes the following contents: The key feature extraction process of pollutants is divided into the following steps, including pollutant region segmentation and pollutant region feature extraction.

[0020] Pollutant area segmentation: First, the image data of each time unit is processed by the image segmentation algorithm to identify and extract the pollutant area; a composite method based on thresholding and edge detection is used, combined with a local area adaptive threshold algorithm, to automatically identify the candidate areas of pollutants; this process calculates the grayscale mean and standard deviation of the local area and dynamically adjusts the threshold to identify the pollutant area under different lighting conditions; the specific steps are as follows: Local grayscale difference detection: calculate the grayscale value of each pixel The local area mean The difference is calculated by using the adaptive threshold formula: .

[0021] in, is the coefficient used to adjust the threshold sensitivity, is the standard deviation of the grayscale in the local area; if , the corresponding pixel is the polluted area.

[0022] Edge enhancement and regional connectivity analysis: The Canny edge detection algorithm is used to refine the segmentation results and enhance the clarity of the pollutant boundaries. Then, the connected domain analysis algorithm is used to detect the connectivity of the polluted area to ensure that the area is completely segmented.

[0023] Pollutant area feature extraction: After successfully segmenting the pollutant area, its geometric shape, area, boundary contour and center position are extracted.

[0024] In a preferred embodiment, the rule recognition module includes the following contents: The calculation process of repeatability index and diffusion regularity index includes pollutant area overlap detection, diffusion feature extraction, index calculation and standardization, and decision index definition.

[0025] Pollutant area overlap detection includes pollutant area matching and overlap rate calculation.

[0026] First, based on the pollutant area features within each time unit extracted by the feature extraction module, the matching relationship between pollutant areas in adjacent time units is determined; by calculating the geometric overlap degree of pollutant areas in adjacent time units, the same or similar pollution hotspots are identified.

[0027] Geometric overlap detection: for time units and , extract the boundary contour of its pollutant area and ; Use polygon intersection algorithm to calculate the overlapping area ,in: ;in Represents the intersection area.

[0028] Overlap rate calculation: Calculate the overlap rate of pollutant areas based on the area of ​​overlapping areas. , reflecting the repeatability of the pollutant: ;in, and The time units and The area of ​​polluted area.

[0029] Diffusion feature extraction includes calculation of geometric center displacement vector, area growth rate and diffusion regularity index.

[0030] Calculation of geometric center displacement vector: The geometric center position of the pollutant area provides information on the direction and speed of pollutant diffusion; calculation time unit and The geometric center position and , and obtain the displacement vector .

[0031] Area growth rate calculation: By comparing the pollutant area of ​​adjacent time units, the rate of pollutant diffusion is evaluated: ;in, Display from time unit and area growth rate.

[0032] Diffusion regularity index calculation: Combining the directional consistency of the displacement vector and the area growth rate, the diffusion regularity index is defined ;in, is the angle between two adjacent displacement vectors, reflecting the consistency of the diffusion direction.

[0033] Index calculation and standardization include repeatability index calculation and extension regularity index standardization.

[0034] Definition of decision index: The repeatability index and the standardized diffusion regularity index are calculated comprehensively to generate the decision index; according to the value of the decision index, the following warning levels are set: According to the value of the decision index, it is compared with the set first and second thresholds to obtain different risk levels; when the comprehensive index is lower than the first threshold, it means that the repeatability and diffusion regularity of the pollutants are low, which is a low-risk state and no immediate intervention is required; when the comprehensive index is between the first and second thresholds, it means that there is a certain degree of accumulation or diffusion of pollutants, and it is necessary to pay attention to its development trend and conduct appropriate inspections and evaluations; when the comprehensive index is higher than the second threshold, it indicates that the repeatability and diffusion regularity of the pollutants are significant, there is a high risk, and cleaning or maintenance measures must be taken immediately to avoid adverse effects on the use process.

[0035] In a preferred embodiment, the trend inference module includes the following contents: Using short-term and long-term decision index change trends, the processing of the trend inference module includes trend data collection, trend model construction, trend classification and status evaluation.

[0036] Trend data collection: First, collect and organize decision index data within multiple time units to form short-term and long-term time series data sets.

[0037] Short-term time series: defined as the most recent Decision index data within a time unit is used to capture recent pollutant trends.

[0038] Long-term time series: defined as the past Decision index data within a time unit is used to analyze long-term accumulation and expansion trends.

[0039] Trend model construction: Use short-term and long-term time series data to construct corresponding trend models to quantify and predict the dynamic changes of pollutant accumulation; short-term trend model: Use Gaussian process regression model to fit short-term time series.

[0040] Long-term trend model: Long short-term memory network model is used to analyze long-term time series.

[0041] Trend classification and status assessment: Based on the output of short-term and long-term trend models, a comprehensive assessment of trend classification and endoscope usage status is performed.

[0042] Define trend classification rules, and classify the endoscope usage status as stable decay or sudden deterioration based on the matching degree and change rate of short-term and long-term trends.

[0043] Stable decay: When both the short-term trend and the long-term trend show that the slope of the decision index is greater than zero and the prediction error is within an acceptable range, it indicates that the accumulation of pollutants is in a stable decay state; prompting regular maintenance and cleaning, optimizing the cleaning cycle to slow down the accumulation of pollutants and ensure long-term stable operation of the equipment; Sudden deterioration: When the slope difference between the short-term trend and the long-term trend is greater than the corresponding threshold, and the prediction error is beyond the normal range, it indicates that there is a risk of sudden deterioration due to the accumulation of pollutants. A high-priority warning is issued, and it is recommended to immediately suspend the use of the equipment for comprehensive cleaning and maintenance to prevent the continued expansion of pollutants and affect medical operations.

[0044] The technical effects and advantages of the medical device use supervision system based on image recognition of the present invention are as follows: The present invention uses dynamic time division technology to automatically adjust the time unit, ensure data stability, and eliminate interference in the non-operation phase; it uses multi-dimensional pollutant feature extraction, and calculates the repeatability index and diffusion regularity index of the pollutant area of ​​adjacent time units, and generates a decision index by combining them to effectively evaluate the warning level of pollutant impact; combined with the change trend of short-term and long-term decision indexes, it accurately judges whether the use status of the endoscope is in a stable decay or mutation deterioration stage. It realizes real-time and precise monitoring of pollutants on the surface of medical devices, significantly improves the sensitivity and accuracy of pollution detection, can warn of equipment performance degradation and infection risks in advance, optimize cleaning and maintenance strategies, extend the service life of equipment, and ensure the safety and accuracy of medical operations. BRIEF DESCRIPTION OF THE DRAWINGS

[0045] Figure 1 This is a structural schematic diagram of a medical device use supervision system based on image recognition according to the present invention. DETAILED DESCRIPTION

[0046] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0047] Embodiment 1: Figure 1 The present invention provides a medical device use supervision system based on image recognition, which includes: an image preprocessing module, a feature extraction module, a rule recognition module and a trend inference module.

[0048] Image preprocessing module: obtains imaging data during the use of the endoscope, dynamically divides the time unit to eliminate interfering images in the non-operation stage, and outputs the standardized imaging data and time unit division results to the feature extraction module.

[0049] Feature extraction module: extracts key features of pollutants based on image data within each time unit, outputs pollutant features of each time unit and passes them to the pattern recognition module; Pattern recognition module: Calculate the repeatability index and diffusion regularity index of the pollutant area according to the overlapping area and diffusion characteristics of pollutants in adjacent time units, and combine these two indexes to generate a decision index. The output of the calculated decision index is passed to the trend inference module; Trend inference module: Based on the short-term and long-term comprehensive decision index change trends, the usage status of the endoscope is analyzed as stable decay or sudden deterioration.

[0050] During the use of endoscopes, due to differences in operation during the surgical phase and fluctuations in the state of equipment startup or stop, the collected image data often contains a lot of interference information. Such interference information, such as transitional blurred frames, areas of drastic light fluctuations, or invalid static background images, may lead to deviations in subsequent pollutant feature extraction and trend analysis. Therefore, it is necessary to dynamically group image data by reasonably dividing time units to eliminate interference during non-operation phases while ensuring the stability and comparability of image features per unit time.

[0051] The image preprocessing module includes the following: The process of dynamically dividing time is divided into four steps, namely preliminary segmentation, data feature analysis, unit time adjustment and segmentation optimization.

[0052] Preliminary segmentation: by collecting the image data of the entire process of endoscope use in time sequence to form a continuous frame sequence Each frame contains the imaging timestamp and screen feature data The acquisition process is divided into several initial segments using timestamps. Each segment has an equal time span (e.g., 1 minute). The set of initial segment data is defined as This process provides a preliminary segmentation framework for subsequent dynamic adjustments.

[0053] Data feature analysis: Perform feature analysis on the image data in each initial segment and extract stability parameters and validity parameters to evaluate the representativeness of the segment.

[0054] The feature extraction content includes the frame-to-frame change rate, illumination change amplitude and invalid frame ratio.

[0055] Inter-frame change rate: Calculate the image change rate of adjacent frames within a clip to evaluate the dynamics of the image. Use the average value of the pixel gray value difference Quantify the changes between frames: ;in. Indicates the frame-to-frame change rate. Reflects the picture changes between adjacent frames. The larger the value, the stronger the dynamics. Indicates the frame number.

[0056] Illumination variation: The illumination stability is assessed by the standard deviation of the illumination histogram mean.

[0057] Invalid Frame Ratio: The ratio of low-resolution frames or excessively blurred frames in the detected clip is defined as the invalid frame ratio.

[0058] Unit time adjustment: Dynamically adjust the time span of the initial segment based on the data feature analysis results. Adjust the segment division to a unit time that is more suitable for pollution feature extraction. , the adjustment logic is as follows: When the inter-frame change rate and the illumination change amplitude are both higher than the corresponding thresholds, it means that the clip contains drastic dynamics or illumination fluctuations, and the clip time span needs to be shortened to filter out interference.

[0059] When the inter-frame change rate and illumination change amplitude are lower than the corresponding threshold and the invalid frame ratio is lower than the corresponding threshold, it indicates that the fragment data is stable and it is suitable to extend the time span to increase the unit data volume.

[0060] Dynamically adjusted time unit set In , the duration of each unit changes adaptively according to its internal characteristics.

[0061] Segment optimization: Optimize the dynamically adjusted time units to ensure that the time units are representative and consistent. The optimization methods include: Feature weight smoothing: For different feature dimensions (such as the rate of change between frames, the amplitude of illumination change, and the proportion of invalid frames), a smoothing function is constructed to adjust the feature changes of the transition segment to adapt it to the adjacent time units.

[0062] Critical segment merging: For units with too short a time span or too discrete eigenvalues, they are merged with adjacent units to form more stable analysis segments.

[0063] Significant anomaly removal: Exclude extreme dynamic feature fragments (such as high-frequency invalid frames caused by equipment adjustment or failure) to avoid deviation in the extraction of contamination features.

[0064] The image preprocessing module collects continuous image data during the use of the endoscope, first divides it into several initial segments according to the timestamp, and then performs data feature analysis on each segment, extracts stability and effectiveness parameters including frame-to-frame change rate, illumination change amplitude, and invalid frame ratio, and evaluates the representativeness of the segment through these parameters. Based on the analysis results, the time span of each segment is dynamically adjusted, the time length of segments containing violent dynamics or illumination fluctuations is shortened, and the time span of stable data segments is extended to ensure that the image data within each time unit has high stability and comparability. On this basis, through optimization methods such as feature smoothing, critical segment merging, and anomaly removal, a dynamic time unit suitable for pollutant feature extraction is finally obtained, providing an accurate data basis for subsequent pollution monitoring and analysis.

[0065] The feature extraction module includes the following: In the image preprocessing module, the image data during the use of the endoscope has been divided into stable time segments by dynamically dividing the time units, and the interference of the non-operation stage has been removed. Next, for the image data of each time unit, it is necessary to identify and extract the key features of the pollutant area in order to further analyze the impact of pollutants on the observation performance of the endoscope. The accumulation and distribution of pollutants may show different geometric shapes, texture features and regional changes during use. Therefore, accurately extracting the various features of pollutants, especially the area, geometry, boundary contour, center position and texture information, is crucial for subsequent pollution monitoring and early warning.

[0066] The key feature extraction process of pollutants is divided into the following steps, including pollutant region segmentation and pollutant region feature extraction.

[0067] Pollutant area segmentation: First, the image data of each time unit is processed by the image segmentation algorithm to identify and extract the pollutant area. A composite method based on thresholding and edge detection is used, combined with a local area adaptive threshold algorithm, to automatically identify candidate pollutant areas. This process calculates the grayscale mean and standard deviation of the local area and dynamically adjusts the threshold to identify pollutant areas under different lighting conditions.

[0068] The specific steps are as follows: Local grayscale difference detection: calculate the grayscale value of each pixel The local area mean The difference is calculated by using the adaptive threshold formula: ;in, is the coefficient used to adjust the threshold sensitivity, is the standard deviation of the gray level in the local area. , the corresponding pixel is the polluted area.

[0069] Edge enhancement and regional connectivity analysis: Apply the Canny edge detection algorithm to refine the segmentation results to further enhance the clarity of the pollutant boundary. Then, use the connected domain analysis algorithm (such as FloodFill) to detect the connectivity of the polluted area to ensure that the area is completely segmented.

[0070] Pollutant area feature extraction: After successfully segmenting the pollutant area, extract its geometric shape, area, boundary contour and center position. The specific steps are as follows: Area calculation: Polluted area It is obtained by the pixel counting method, that is, calculating the number of pixels belonging to the pollutant area in the segmented image: ;in, is the total number of pixels in the image, Indicates whether the pixel belongs to the polluted area, if so, it is 1, otherwise it is 0.

[0071] Geometric analysis: by calculating the circumscribed rectangle, minimum circumscribed circle and perimeter of the pollutant area , aspect ratio, and shape factor to evaluate the geometric shape of the pollutant. The specific calculation formula is as follows: ;in, Represents the pixel difference between contour points.

[0072] The shape factor is defined as: ; This factor reflects the regularity of the pollutant shape. The closer the value is to 1, the closer the shape is to a circle, and the smaller the value is, the more complex the shape is.

[0073] Center position calculation: The center position of the pollutant area , Calculated by weighted average method, the coordinates of each pixel are weighted according to its area contribution: .

[0074] In the feature extraction module, detailed feature vectors of the pollutant area in each dynamic time unit have been obtained through image segmentation and feature extraction, including area, geometry, boundary contour, center position and texture information. These feature vectors provide basic data for subsequent pollutant monitoring and analysis. In order to further understand the accumulation and expansion trends of pollutants, it is necessary to analyze the overlap of pollutant areas between adjacent time units and their diffusion characteristics. By comparing the pollutant characteristics of adjacent time units, the repeatability and diffusion regularity of pollutants can be quantified, and the repeatability index and diffusion regularity index can be calculated to provide a quantitative basis for comprehensive decision-making.

[0075] The pattern recognition module includes the following: The calculation process of repeatability index and diffusion regularity index includes pollutant area overlap detection, diffusion feature extraction, index calculation and standardization, and decision index definition. The specific processing process of each sub-step is described in detail below.

[0076] Pollutant area overlap detection includes pollutant area matching and overlap rate calculation.

[0077] First, based on the pollutant area features in each time unit extracted by the feature extraction module, the matching relationship between pollutant areas in adjacent time units is determined. By calculating the geometric overlap degree of pollutant areas in adjacent time units, the same or similar pollution hot spots are identified.

[0078] Geometric overlap detection: for time units and , extract the boundary contour of its pollutant area and . Use polygon intersection algorithm to calculate the overlapping area ,in: ;in Represents the intersection area. The overlapping area represents the overlapping part of the pollutant area in two adjacent time units.

[0079] Overlap rate calculation: Based on the area of ​​overlapping areas, calculate the overlap rate of pollutant areas. , reflecting the repeatability of the pollutants: ;in, and The time units and Overlap rate The value range is from 0 to 1, and higher values ​​indicate more repeatability of the pollutant areas.

[0080] Diffusion feature extraction includes calculation of geometric center displacement vector, area growth rate and diffusion regularity index.

[0081] Calculation of geometric center displacement vector: The geometric center position of the pollutant area provides information on the direction and speed of pollutant diffusion. Calculation time unit and The geometric center position and , and obtain the displacement vector ; This vector describes the moving direction and distance of the center of the pollutant area between two time units.

[0082] Area growth rate calculation: By comparing the pollutant area of ​​adjacent time units, the rate of pollutant diffusion is evaluated: ;in, Indicates the time unit and A positive value indicates that the polluted area is expanding, while a negative value indicates that it is shrinking.

[0083] Diffusion regularity index calculation: Combining the directional consistency of the displacement vector and the area growth rate, the diffusion regularity index is defined ;in, is the angle between two adjacent displacement vectors, reflecting the consistency of the diffusion direction. The smaller the angle, the more consistent the direction. is the logarithmic transformation of the area growth rate to smooth out extreme growth values. The value range is close to 0 to 1. The higher the value, the stronger the diffusion regularity and the more obvious the pollutant expansion trend.

[0084] Index calculation and standardization include repeatability index calculation and extension regularity index standardization.

[0085] Index calculation and standardization: Defining repeatability index based on overlap ratio ; This transformation weakens the impact of high overlap rate on the index through the square root function, making the index more sensitive and discriminative.

[0086] Standardization of diffusion regularity index: To ensure the comparability of diffusion regularity index between different time units, standardization is performed: ;in, and are the mean and standard deviation of the diffusion regularity index in all time units. The standardized diffusion regularity index makes its distribution have zero mean and unit variance, which is convenient for subsequent comprehensive analysis.

[0087] Definition of decision index: The repeatability index and the standardized diffusion regularity index are calculated comprehensively to generate a decision index, which can be calculated using the following formula: ; This formula combines the nonlinear interaction of repeatability and diffusion regularity, and introduces nonlinear regulation through an exponential function, so that the comprehensive index can respond more sensitively to changes in pollutant characteristics.

[0088] According to the value of the decision index, the following warning levels are set: According to the value of the decision index, it is compared with the set first and second thresholds to obtain different risk levels. When the comprehensive index is lower than the first threshold, it means that the repeatability and diffusion regularity of the pollutants are low, which is a low-risk state, and no immediate intervention is required; when the comprehensive index is between the first and second thresholds, it means that the pollutants have a certain degree of accumulation or diffusion, and it is necessary to pay attention to their development trends and conduct appropriate inspections and assessments; when the comprehensive index is higher than the second threshold, it indicates that the repeatability and diffusion regularity of the pollutants are significant, there is a high risk, and cleaning or maintenance measures must be taken immediately to avoid adverse effects on the use process.

[0089] The regularity recognition module calculates the repeatability index and diffusion regularity index of pollutants by analyzing the overlapping area and diffusion characteristics of pollutant areas between adjacent time units. The process first evaluates the degree of overlap between pollutant areas in different time units through geometric overlap detection, and then obtains the repeatability of pollutants; then, the diffusion trend of pollutant areas is analyzed by geometric center displacement and area growth rate, so as to extract diffusion regularity. Through the standardization and nonlinear comprehensive calculation of these two indexes, a comprehensive decision index is generated, and the risk level of pollutants is judged according to the set threshold rules, providing a quantitative basis for subsequent cleaning or maintenance decisions. This step provides a scientific and accurate analysis framework for monitoring pollutant accumulation and diffusion trends.

[0090] In the regularity recognition module, the decision index is calculated and generated by analyzing the repeatability and diffusion regularity of the endoscope contaminant area. This index reflects the accumulation and expansion trend of contaminants on the surface of the endoscope, and serves as an important basis for judging the use status of the equipment. However, the decision index at a single time point is not enough to fully reflect the long-term use status changes of the equipment. Therefore, it is necessary to further analyze the changing trend of the comprehensive decision index on different time scales to identify the stable decay or sudden deterioration of the use status of the endoscope, and provide scientific guidance for maintenance and cleaning strategies.

[0091] The trend inference module includes the following: Using the short-term and long-term decision index change trends, the processing of the trend inference module includes three main stages: trend data collection, trend model construction, trend classification and status evaluation. Each stage is subdivided into specific calculation and analysis steps to ensure accurate judgment of the endoscope usage status.

[0092] Trend data collection: First, collect and organize decision index data within multiple time units to form short-term and long-term time series data sets.

[0093] Short-term time series: defined as the most recent Decision index data within a time unit is used to capture recent pollutant trends.

[0094] Long-term time series: defined as the past Decision index data within a time unit is used to analyze long-term accumulation and expansion trends.

[0095] These time series data ensure that trend analysis covers enough historical information to accurately identify trend patterns.

[0096] Trend model construction: Use short-term and long-term time series data to construct corresponding trend models to quantify and predict the dynamic changes of pollutant accumulation.

[0097] Short-term trend model: Gaussian process regression model is used to fit short-term time series. This model can flexibly capture local trend changes in time series and adapt to possible fluctuations in the short term.

[0098] Model building steps: 1. Local trend capture: Use the Gaussian process regression model to fit the short-term time series, extract the recent data change trend, and capture the short-term fluctuation characteristics.

[0099] 2. Dynamic adjustment: Combined with the sliding window technology, the kernel function parameters in the Gaussian process regression model are dynamically adjusted to adapt to the changing trends in different time periods.

[0100] 3. Forecast and error assessment: Generate short-term trend forecasts based on model fitting results and calculate forecast errors to provide data support for subsequent trend classification and assessment.

[0101] Long-term trend model: The long short-term memory network model is used to analyze long-term time series. Through its unique memory unit design, this model can effectively capture the long-term dependencies and trend patterns in time series.

[0102] Model building steps: 1. Sequence input processing: Standardize the long-term time series data to eliminate the difference in numerical magnitude, and input the processed data into the long short-term memory network model.

[0103] 2. Feature extraction and memory: The model’s forget gate and input gate mechanisms are used to extract long-term trends and nonlinear changes in sequence data, ensuring that the model can retain key time-dependent information.

[0104] 3. Trend prediction and evaluation: Generate long-term trend predictions based on the extracted features, evaluate the prediction results through the loss function, and optimize the model parameters to improve the prediction accuracy.

[0105] Trend classification and status assessment: Based on the output of short-term and long-term trend models, a comprehensive assessment of trend classification and endoscope usage status is performed.

[0106] Define trend classification rules, and classify the endoscope usage status as stable decay or sudden deterioration based on the matching degree and change rate of short-term and long-term trends.

[0107] Stable decay: When both the short-term trend and the long-term trend show that the decision index slope is greater than zero and the prediction error is within an acceptable range, it indicates that the pollutant accumulation is in a stable decay state. It is recommended to perform regular maintenance and cleaning, optimize the cleaning cycle to slow down the accumulation of pollutants and ensure long-term stable operation of the equipment.

[0108] Sudden deterioration: When the slope difference between the short-term trend and the long-term trend is greater than the corresponding threshold, and the prediction error exceeds the normal range, it indicates that there is a risk of sudden deterioration of pollutant accumulation. A high-priority warning is issued, and it is recommended to immediately suspend the use of the equipment for comprehensive cleaning and maintenance to prevent further spread of pollutants and affect medical operations.

[0109] The trend inference module analyzes the stability, decay, or sudden deterioration of the endoscope's usage status based on the changing trends of the short-term and long-term decision-making indices. By performing trend analysis on the dynamic changes of the short-term and long-term decision-making indices, it is possible to identify potential abnormal states during the use of the endoscope and determine whether it is in a stage of decay, stability, or sudden deterioration. This analysis provides an accurate basis for the use decision of the equipment by comparing the trend changes on different time scales and combining the model prediction results, and guides whether further preventive or adjustment measures are needed to ensure the stable performance and safe use of the endoscope.

[0110] The above formulas are all dimensionless and numerical calculations. The formula is a formula for the most recent real situation obtained by collecting a large amount of data and performing software simulation. The preset parameters in the formula are set by technicians in this field according to actual conditions.

[0111] The above description is only by way of illustration of certain exemplary embodiments of the present invention. It is undoubted that those skilled in the art can modify the described embodiments in various ways without departing from the spirit and scope of the present invention. Therefore, the above drawings and descriptions are illustrative in nature and should not be construed as limiting the scope of protection of the claims of the present invention.

[0112] It should be noted that, in this article, if there are relational terms such as first and second, etc., they are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Moreover, the terms "include", "comprise" or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or device. In the absence of further restrictions, the elements defined by the sentence "including a..." do not exclude the existence of other identical elements in the process, method, article or device including the elements.

[0113] The above is only a specific implementation of the present application, but the protection scope of the present application is not limited thereto. Any person skilled in the art who is familiar with the present technical field can easily think of changes or substitutions within the technical scope disclosed in the present application, which should be included in the protection scope of the present application. Therefore, the protection scope of the present application should be based on the protection scope of the claims.

Claims

1. A medical device use supervision system based on image recognition, characterized in that: include: Image preprocessing module, feature extraction module, pattern recognition module and trend inference module; Image preprocessing module: obtains imaging data during the use of the endoscope, dynamically divides the time unit to eliminate interference images in the non-operation stage, outputs the standardized imaging data and the time unit division results to the feature extraction module; Feature extraction module: extracts key features of pollutants based on image data within each time unit, outputs pollutant features of each time unit and passes them to the pattern recognition module; Pattern recognition module: Calculate the repeatability index and diffusion regularity index of the pollutant area according to the overlapping area and diffusion characteristics of pollutants in adjacent time units, and combine these two indexes to generate a decision index. The output of the calculated decision index is passed to the trend inference module; Trend inference module: Based on the short-term and long-term comprehensive decision index change trends, the usage status of the endoscope is analyzed as stable decay or sudden deterioration.

2. According to claim 1, a medical device use supervision system based on image recognition is characterized in that: The image preprocessing module includes the following: The process of dynamically dividing time is divided into four steps, namely, preliminary segmentation, data feature analysis, unit time adjustment and segment optimization; Preliminary segmentation: by collecting the image data of the entire process of endoscope use in time sequence, a continuous frame sequence is formed, and each frame image contains the imaging timestamp and picture feature data; Use timestamps to divide the acquisition process into several initial segments, each with equal time span; Data feature analysis: Extract features from the image data in each initial segment, including the frame-to-frame change rate, illumination change amplitude, and invalid frame ratio; Inter-frame change rate: calculate the image change rate of adjacent frames within a clip to evaluate the dynamics of the image; Illumination variation: The illumination stability is evaluated by the standard deviation of the illumination histogram mean; Invalid frame ratio: The ratio of low-resolution frames or excessively blurred frames in the detection clip is defined as the invalid frame ratio; Unit time adjustment: According to the data feature analysis results, the time span of the initial segment is dynamically adjusted; the segment division is adjusted to a unit time suitable for pollution feature extraction. The adjustment logic is as follows: When the inter-frame change rate and illumination change amplitude are both higher than the corresponding thresholds, it means that the clip contains violent dynamics or illumination fluctuations, and the clip time span needs to be shortened to filter out interference; When the inter-frame change rate and illumination change amplitude are lower than the corresponding threshold and the invalid frame ratio is lower than the corresponding threshold, it means that the fragment data is stable and the time span needs to be extended to increase the unit data volume.

3. The medical device use supervision system based on image recognition according to claim 2, characterized in that: Segment optimization: Optimize the dynamically adjusted time units to ensure that the time units are representative and consistent. The optimization methods include feature weight smoothing, critical segment merging, and significant anomaly removal.

4. The medical device use supervision system based on image recognition according to claim 3 is characterized in that: The feature extraction module includes the following: The key feature extraction process of pollutants is divided into the following steps, including pollutant region segmentation and pollutant region feature extraction; Pollutant area segmentation: First, the image data of each time unit is processed by the image segmentation algorithm to identify and extract the pollutant area; a composite method based on thresholding and edge detection is used, combined with a local area adaptive threshold algorithm, to automatically identify the candidate areas of pollutants; this process calculates the grayscale mean and standard deviation of the local area and dynamically adjusts the threshold to identify the pollutant area under different lighting conditions; the specific steps are as follows: Local grayscale difference detection: calculate the grayscale value of each pixel The local area mean The difference is calculated by using the adaptive threshold formula: ; in, is the coefficient used to adjust the threshold sensitivity, is the standard deviation of the grayscale in the local area; if , then the corresponding pixel point is the pollutant area; Edge enhancement and regional connectivity analysis: The Canny edge detection algorithm is used to refine the segmentation results and enhance the clarity of the pollutant boundaries. Then, the connected domain analysis algorithm is used to detect the connectivity of the polluted area to ensure that the area is completely segmented.

5. The medical device use supervision system based on image recognition according to claim 4 is characterized in that: Pollutant area feature extraction: After successfully segmenting the pollutant area, its geometric shape, area, boundary contour and center position are extracted.

6. The medical device use supervision system based on image recognition according to claim 5, characterized in that: The pattern recognition module includes the following: The calculation process of repeatability index and diffusion regularity index includes pollutant area overlap detection, diffusion feature extraction, index calculation and standardization, and decision index definition; Pollutant area overlap detection includes pollutant area matching and overlap rate calculation; First, based on the pollutant area features within each time unit extracted by the feature extraction module, the matching relationship between pollutant areas in adjacent time units is determined; by calculating the geometric overlap degree of pollutant areas in adjacent time units, the same or similar pollution hot spots are identified; Geometric overlap detection: for time units and , extract the boundary contour of its pollutant area and ; Calculate the overlapping area using polygon intersection algorithm ,in: ;in represents the intersection area; Overlap rate calculation: Based on the area of ​​overlapping areas, calculate the overlap rate of pollutant areas. , reflecting the repeatability of the pollutants: ;in, and The time units and The area of ​​pollutant area.

7. The medical device use supervision system based on image recognition according to claim 6, characterized in that: Diffusion feature extraction includes calculation of geometric center displacement vector, area growth rate and diffusion regularity index; Calculation of geometric center displacement vector: The geometric center position of the pollutant area provides information on the direction and speed of pollutant diffusion; calculation time unit and The geometric center position and , and obtain the displacement vector ; Area growth rate calculation: By comparing the pollutant area of ​​adjacent time units, the rate of pollutant diffusion is evaluated: ;in, Indicates the time unit and The area growth rate; Diffusion regularity index calculation: Combining the directional consistency of the displacement vector and the area growth rate, the diffusion regularity index is defined ;in, is the angle between two adjacent displacement vectors, reflecting the consistency of the diffusion direction.

8. The medical device use supervision system based on image recognition according to claim 7, characterized in that: Index calculation and standardization include repeatability index calculation and expansion regularity index standardization; Definition of decision index: The repeatability index and the standardized diffusion regularity index are calculated comprehensively to generate the decision index; according to the value of the decision index, the following warning levels are set: According to the value of the decision index, it is compared with the set first and second thresholds to obtain different risk levels; when the comprehensive index is lower than the first threshold, it means that the repeatability and diffusion regularity of the pollutants are low, which is a low-risk state and no immediate intervention is required; when the comprehensive index is between the first and second thresholds, it means that there is a certain degree of accumulation or diffusion of pollutants, and it is necessary to pay attention to its development trend and conduct appropriate inspections and evaluations; when the comprehensive index is higher than the second threshold, it indicates that the repeatability and diffusion regularity of the pollutants are significant, there is a high risk, and cleaning or maintenance measures must be taken immediately to avoid adverse effects on the use process.

9. The medical device use supervision system based on image recognition according to claim 8, characterized in that: The trend inference module includes the following: Using short-term and long-term decision index change trends, the processing of the trend inference module includes trend data collection, trend model construction, trend classification and status assessment; Trend data collection: First, collect and organize decision index data within multiple time units to form short-term and long-term time series data sets; Short-term time series: defined as the most recent Decision index data within time units is used to capture recent pollutant trends; Long-term time series: defined as the past Decision index data within a time unit is used to analyze long-term accumulation and expansion trends; Trend model construction: Use short-term and long-term time series data to build corresponding trend models to quantify and predict the dynamic changes of pollutant accumulation; Short-term trend model: Use Gaussian process regression model to fit short-term time series; Long-term trend model: Long short-term memory network model is used to analyze long-term time series.

10. A medical device use supervision system based on image recognition according to claim 9, characterized in that: Trend classification and status assessment: Based on the output of short-term and long-term trend models, comprehensive assessment of trend classification and endoscope usage status is performed; Define trend classification rules to classify the endoscope usage status as stable decay or sudden deterioration based on the matching degree and change rate of short-term and long-term trends; Stable decay: When both the short-term trend and the long-term trend show that the slope of the decision index is greater than zero and the prediction error is within an acceptable range, it indicates that the accumulation of pollutants is in a stable decay state; prompting regular maintenance and cleaning, optimizing the cleaning cycle to slow down the accumulation of pollutants and ensure long-term stable operation of the equipment; Sudden deterioration: When the slope difference between the short-term trend and the long-term trend is greater than the corresponding threshold, and the prediction error is beyond the normal range, it indicates that there is a risk of sudden deterioration due to the accumulation of pollutants; a high-priority warning is issued, and it is recommended to immediately suspend the use of the equipment for comprehensive cleaning and maintenance to prevent the continued expansion of pollutants and affect medical operations.

Images