Endoscopic hemorrhagic spot detection device and system based on infrared thermal imaging

Abstract

The invention relates to the technical field of endoscope detection, in particular to an endoscopic hemorrhagic spot detection device and system based on infrared thermal imaging. The device is applied to an endoscope probe integrated with a microscopic image capturing module, and comprises an infrared thermal imaging positioning module used for determining at least one abnormal bleeding tissue part by using an infrared thermal imaging technology, obtaining position information of the abnormal bleeding tissue part and sending the position information; the endoscopic image acquisition control module is used for receiving the position information of the abnormal hemorrhage tissue part, controlling an endoscope module in the endoscope probe to move to a part corresponding to the position information provided by the infrared thermal imaging positioning module, performing image acquisition according to a predetermined endoscopic image acquisition strategy, and transmitting the acquired image to the endoscope probe; obtaining an endoscopic panoramic image of the abnormal bleeding tissue part, and compressing and sending the endoscopic panoramic image; the efficiency and precision of hemorrhagic spot detection are improved, and the reliability and operability of diagnosis are enhanced.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of endoscopic detection, and in particular to an endoscopic bleeding point detection device and system based on infrared thermal imaging. BACKGROUND

[0002] In recent years, with the continuous development of endoscopic technology, endoscopic bleeding point detection has become an important diagnostic method. Endoscopy relies on visible light imaging technology to capture internal tissue images through high-resolution cameras, helping doctors accurately determine the location and cause of bleeding.

[0003] Existing bleeding point detection devices based on endoscopic image processing rely on the naked eye observation of doctors, are limited by the personal experience and skill level of doctors, and when multiple bleeding points are present, the positioning and capturing of bleeding points are more time-consuming and laborious, increasing the complexity and uncertainty of the operation. SUMMARY

[0004] To solve the above technical problems, the present application provides an endoscopic bleeding point detection device and system based on infrared thermal imaging, which not only improves the efficiency and accuracy of bleeding point detection, but also enhances the reliability and operability of diagnosis.

[0005] In a first aspect, the present application provides an endoscopic bleeding point detection device based on infrared thermal imaging, which is applied to an endoscope probe integrated with a microscopic image taking module, and the device comprises:

[0006] An infrared thermal imaging positioning module for determining at least one abnormal bleeding tissue site using infrared thermal imaging technology, obtaining the position information of the abnormal bleeding tissue site, and sending it;

[0007] An endoscopic image acquisition control module for receiving the position information, controlling the endoscopic module in the endoscope probe to move to the site corresponding to the position information, performing image acquisition according to a predetermined endoscopic image acquisition strategy, obtaining an endoscopic panoramic image of the abnormal bleeding tissue site, and compressing and sending it;

[0008] A bleeding point recognition module for receiving and decompressing the endoscopic panoramic image, using a pre-stored bleeding point recognition model to recognize the endoscopic panoramic image for bleeding characteristics, identifying a suspected bleeding point set, and sending it; the suspected bleeding point set contains the orientation information of at least one suspected bleeding point;

[0009] A microscopic image acquisition control module for receiving the suspected bleeding point set, controlling the endoscope probe to switch to the microscopic image taking module, performing image acquisition on each suspected bleeding point according to the orientation information of each suspected bleeding point in the suspected bleeding point set, obtaining the corresponding microscopic image, and packaging and sending it;

[0010] a bleeding point analysis module, configured to receive the microscopic images, and perform feature analysis on each microscopic image of the suspected bleeding point by using a pre-stored bleeding point analysis model, calculate a bleeding probability value of each suspected bleeding point, and send the bleeding probability value;

[0011] a result output module, configured to receive the microscopic images and the corresponding bleeding probability values, and output a detection result after comprehensive analysis.

[0012] Further, the comprehensive analysis of the result output module comprises:

[0013] adjusting the multiple microscopic images in descending order according to the bleeding probability values of each suspected bleeding point, generating a bleeding probability point sequence, and taking the bleeding probability point sequence and the corresponding multiple microscopic images as the detection result.

[0014] Further, the infrared thermal imaging positioning module comprises:

[0015] an infrared data acquisition submodule, configured to acquire infrared thermal imaging data of a region to be detected of a patient, and send the data;

[0016] an abnormal region identification and segmentation submodule, configured to receive the thermal imaging data sent by the infrared data acquisition submodule, perform abnormal region identification and segmentation, and obtain an abnormal temperature region image; and send the image;

[0017] a temperature gradient division submodule, configured to receive the abnormal temperature region image sent by the abnormal region identification and segmentation submodule, and further divide the abnormal temperature region image into multiple temperature abnormal unit blocks based on a pre-set temperature gradient standard; and send the blocks after packaging; the pixel temperature values in each temperature abnormal unit block are within the same temperature gradient;

[0018] a temperature feature extraction submodule, configured to receive the temperature abnormal unit blocks sent by the temperature gradient division submodule, and perform temperature feature extraction on each temperature abnormal unit block to obtain a corresponding temperature feature vector, wherein the temperature feature vector comprises a temperature value feature and a temperature value distribution feature; and send the vector;

[0019] a temperature feature comprehensive analysis submodule, configured to receive the temperature feature vector sent by the temperature feature extraction submodule, input the temperature feature vector into a pre-stored temperature feature comprehensive analysis model, perform comprehensive analysis on the temperature feature vector, and output a temperature feature value capable of representing the temperature state of the corresponding temperature abnormal unit block; and send the value;

[0020] The abnormal block filtering and marking submodule is used to receive temperature feature values ​​sent by the temperature feature comprehensive analysis submodule. Among all temperature abnormal unit blocks, it filters out temperature abnormal unit blocks that exceed the preset temperature feature threshold based on the temperature feature values, and marks the corresponding tissue parts as abnormal bleeding tissue parts.

[0021] Furthermore, in the endoscopic image acquisition control module, the method for determining the endoscopic image acquisition strategy includes:

[0022] Using the location information and the corresponding temperature feature value as retrieval coordinates, the corresponding endoscopic image acquisition strategy is extracted from the endoscopic image acquisition strategy library;

[0023] The endoscopic image acquisition strategy includes image acquisition focal length, image acquisition step frequency, image acquisition angle, light source brightness, and color temperature.

[0024] Furthermore, the temperature feature vector is:

[0025] X = [T] avg ,T diff [,D,S];

[0026] Where X represents the temperature feature vector, T avg T represents the average temperature in the numerical temperature characteristics. diff denoted by , D represents the maximum and minimum temperature difference in the temperature numerical characteristics, D represents the dispersion of temperature values ​​at each point in the temperature numerical distribution characteristics, and S represents the distribution pattern of temperature values ​​in the temperature numerical distribution characteristics.

[0027] Furthermore, the calculation formula for the comprehensive temperature characteristic analysis model is as follows:

[0028]

[0029] Where K is the temperature feature value, which is mapped to the range [0,1] and represents the probability that the corresponding temperature abnormality unit block has a bleeding risk; X′ represents the standardized feature vector; W1 and W2 represent the weight matrix; (X′⊙X′) represents the element-wise square of the feature vector elements, used to introduce a quadratic term; b represents the bias term; and e is a constant.

[0030] Furthermore, the bleeding point recognition model extracts local and global features from the endoscopic panoramic image through multiple convolutional and pooling layers; uses a fully connected layer to determine whether each pixel belongs to a bleeding point based on the extracted features; and generates a probability map of the same size as the endoscopic panoramic image, where each pixel value represents the probability that the location is a bleeding point.

[0031] Furthermore, the operation of the bleeding point recognition module includes:

[0032] receiving and decompressing the endoscopic panoramic image sent by the endoscopic image acquisition control module;

[0033] preprocessing the endoscopic panoramic image;

[0034] inputting the preprocessed endoscopic panoramic image into the pre-trained bleeding point recognition model to obtain a probability map with the same size as the input image, wherein each pixel value represents the probability that the position is a bleeding point;

[0035] The probability threshold is pre-set in the bleeding point recognition module, and the probability threshold is used to traverse the probability map, and the pixels exceeding the probability threshold are marked as potential bleeding points;

[0036] All potential bleeding points are summarized to obtain a suspected bleeding point set.

[0037] Further, the preprocessing of the endoscopic panoramic image includes:

[0038] removing random noise in the image;

[0039] adjusting the contrast of the image to increase the difference between different tissue types;

[0040] adjusting all images to a uniform resolution and size.

[0041] In another aspect, the present application also provides an endoscopic bleeding point detection system based on infrared thermal imaging, which adopts the endoscopic bleeding point detection device based on infrared thermal imaging as described above, and further comprises:

[0042] an endoscope probe, which further integrates a microscopic image capturing module and an infrared thermal imaging sensor, for collecting images and capturing the temperature distribution in the patient's body;

[0043] a display for displaying the images collected by the endoscope probe and the microscopic image capturing module, the infrared thermal imaging map, and the detection results to the operating physician;

[0044] a user interface as an interactive platform for the operating physician, allowing the operating physician to browse images, adjust parameters, and confirm bleeding points;

[0045] a power management unit for providing power supply for all system components;

[0046] a communication interface for supporting data transmission between modules of the endoscopic bleeding point detection device based on infrared thermal imaging, and information exchange with the hospital information system.

[0047] Compared with existing technologies, the beneficial effects of this invention are as follows: First, this invention uses infrared thermal imaging technology to quickly locate abnormal bleeding tissue, shortening the detection time and reducing the reliance on doctors' visual observation, thus reducing human error; second, the device automatically acquires endoscopic panoramic and microscopic images, and combines them with a pre-trained model for bleeding feature recognition and analysis, realizing intelligent screening and probability assessment of bleeding points, improving the objectivity and accuracy of diagnosis; in summary, this device not only improves the efficiency and accuracy of bleeding point detection, but also enhances the reliability and operability of diagnosis. Attached Figure Description

[0048] Figure 1 This is a structural diagram of the endoscopic bleeding point detection device based on infrared thermal imaging in Embodiment 1;

[0049] Figure 2 This is a structural diagram of the endoscopic bleeding point detection system based on infrared thermal imaging in Example 2. Detailed Implementation

[0050] This application will now be described with reference to the accompanying drawings.

[0051] Example 1: As Figure 1 As shown, the endoscopic bleeding point detection device based on infrared thermal imaging of the present invention is applied to an endoscope probe integrated with a microscopic imaging module, and specifically includes the following modules:

[0052] The infrared thermal imaging positioning module is used to determine at least one site of abnormal bleeding tissue using infrared thermal imaging technology, acquire the location information of the abnormal bleeding tissue site, and send it.

[0053] The endoscopic image acquisition and control module is used to receive the location information of the abnormal bleeding tissue site, and control the endoscope module in the endoscope probe to move to the site corresponding to the location information provided by the infrared thermal imaging positioning module. According to the predetermined endoscopic image acquisition strategy, it performs image acquisition, obtains an endoscopic panoramic image of the abnormal bleeding tissue site, and compresses and sends it.

[0054] The bleeding point identification module is used to receive the decompressed endoscopic panoramic image and use a pre-stored bleeding point identification model to identify bleeding features in the endoscopic panoramic image, identify a set of suspected bleeding points, and send them out; the set of suspected bleeding points contains the location information of at least one suspected bleeding point.

[0055] The microscopic image acquisition and control module is used to receive the set of suspected bleeding points and control the endoscope probe to switch to the microscopic imaging module. Based on the orientation information of each suspected bleeding point in the set of suspected bleeding points, it acquires images of each suspected bleeding point, obtains the corresponding microscopic images, and packages and sends them.

[0056] The bleeding point analysis module is configured to receive the microscopic images of each suspected bleeding point, and perform feature analysis on each microscopic image of the suspected bleeding point by using a pre-stored bleeding point analysis model, calculate a bleeding probability value of each suspected bleeding point, and send the bleeding probability value.

[0057] The result output module is configured to receive the microscopic images and the corresponding bleeding probability values, and output a detection result after comprehensive analysis.

[0058] In the embodiment, first, the infrared thermal imaging positioning module identifies abnormal bleeding tissue parts by using temperature differences and obtains position information of the abnormal bleeding tissue parts, and can discover potential bleeding sources earlier, and is more sensitive and reliable than traditional visible light imaging.

[0059] Then, the endoscopic image acquisition control module accurately moves the endoscope module to a target area for panoramic image acquisition according to the received position information, and compresses and sends the panoramic image to reduce data transmission burden. The bleeding point recognition module receives the decompressed image, automatically recognizes bleeding features by using a pre-trained model, forms a suspected bleeding point set, and reduces subjective errors caused by manual observation by doctors. The microscopic image acquisition control module switches to a high-resolution microscopic image acquisition module according to the position information of the suspected bleeding points, and collects detailed images of the suspected bleeding points, to ensure that key areas are paid attention to. The bleeding point analysis module further analyzes the microscopic images, calculates bleeding probability values, and provides a quantitative basis, which is helpful to improve the diagnosis accuracy.

[0060] More specifically, the comprehensive analysis of the result output module includes:

[0061] According to the bleeding probability value of each suspected bleeding point, the microscopic images are adjusted in descending order, a bleeding probability point sequence is generated, and the bleeding probability point sequence and the corresponding microscopic images are taken as the detection result. Then, an operating doctor manually determines a final bleeding point position according to information provided by the sorting display module.

[0062] The image sequence is adjusted in descending order of the bleeding probability value and displayed to the doctor, and the doctor finally determines the bleeding point position by using the manual confirmation module. This process not only improves the operation efficiency, but also reduces the complexity and uncertainty. The entire device integrates the advantages of infrared thermal imaging, machine learning, and high-resolution microscopic imaging, realizes the whole-process optimization from preliminary screening to accurate diagnosis, greatly improves the effect of endoscopic bleeding point detection, meets the demand for accurate positioning of bleeding points in clinical practice, reduces the work burden of doctors, and improves the consistency and reliability of diagnosis.

[0063] Further, the infrared thermal imaging positioning module includes:

[0064] An infrared data acquisition submodule is configured to acquire infrared thermal imaging data of a region to be detected of a patient. This submodule is responsible for receiving or collecting raw data from an infrared thermal imaging device, ensuring the accuracy and integrity of the data.

[0065] An abnormal region identification and segmentation submodule is configured to process the thermal imaging data provided by the infrared data acquisition submodule, identify and segment an abnormal temperature region image through a specific image recognition algorithm. This submodule can distinguish between normal temperature regions and abnormal high or low temperature regions.

[0066] A temperature gradient division submodule is configured to further divide the abnormal temperature region image output by the abnormal region identification and segmentation submodule into a plurality of temperature abnormal unit tiles based on a preset temperature gradient standard. The pixel temperature values in each temperature abnormal unit tile are within the same temperature gradient, which helps to more carefully analyze the characteristics of the abnormal temperature region.

[0067] A temperature feature extraction submodule is configured to extract temperature features for each temperature abnormal unit tile generated by the temperature gradient division submodule. This submodule generates corresponding temperature feature vectors, which include temperature value features and temperature value distribution features, for comprehensively describing the temperature state of the temperature abnormal unit tile.

[0068] A temperature feature comprehensive analysis submodule is configured to input the temperature feature vectors output by the temperature feature extraction submodule into a preset temperature feature comprehensive analysis model. This model is based on machine learning or deep learning algorithms to comprehensively analyze the temperature feature vectors and output temperature feature values that can represent the temperature state of the corresponding temperature abnormal unit tile.

[0069] An abnormal tile screening and marking submodule is configured to screen out temperature abnormal unit tiles that exceed a preset temperature feature threshold from all temperature abnormal unit tiles based on the temperature feature values output by the temperature feature comprehensive analysis submodule. The tissue sites corresponding to these tiles are marked as abnormal bleeding tissue sites, serving as a basis for subsequent endoscopic image acquisition and bleeding point identification.

[0070] Each module in the endoscopic bleeding point detection device based on infrared thermal imaging will be explained in detail below.

[0071] For the infrared thermal imaging positioning module, at least one abnormal bleeding tissue site is determined using infrared thermal imaging, including:

[0072] Step S11, obtain the infrared thermal imaging data of the patient's detection area; In this step, the endoscope probe integrated with the microscopic image capturing module is equipped with a high-performance infrared thermal imaging sensor; When the endoscope probe enters the patient's body, the infrared thermal imaging sensor starts working, capturing the temperature distribution map inside the digestive tract or other body cavities in real time; The temperature distribution map is a two-dimensional temperature field data set containing the temperature information of each point in the entire examination area; In order to ensure the integrity and accuracy of the data, multiple scans are needed, and the obtained data is averaged to reduce noise interference;

[0073] Step S12, identify and segment the abnormal area of the thermal imaging data to obtain an abnormal temperature area image; The thermal imaging data obtained in step S11 is preprocessed, including removing noise, correcting artifacts, etc.; Then, using image processing algorithms such as threshold segmentation, edge detection, etc., to identify the area that is significantly different from the surrounding tissue temperature, i.e. the abnormal temperature area; The abnormal temperature area is caused by temperature rise due to bleeding, inflammation or other pathological changes; Through this step, the area where the problem may exist can be preliminarily delineated, forming an abnormal temperature area image;

[0074] Step S13, according to the preset temperature gradient, for example, every 0.5℃ as a gradient, the abnormal temperature area image is subdivided into multiple small temperature anomaly unit blocks; The pixel temperature values in each block are within the same temperature gradient range, which helps subsequent feature extraction and analysis for different temperature levels and improves the understanding of complex temperature patterns;

[0075] Step S14, for each temperature anomaly unit block, the system will calculate its temperature value features, including average temperature and maximum and minimum temperature difference, etc.; And temperature value distribution features, including temperature value dispersion and distribution form, i.e. skewness or kurtosis, etc.; Temperature value features and temperature value distribution features are combined into temperature feature vectors to describe the temperature state in the block; The temperature feature vector not only reflects the absolute temperature level, but also contains the temperature distribution information, which helps to distinguish the real bleeding point from other factors that may cause temperature rise;

[0076] Step S15, input the temperature feature vector into the preset temperature feature comprehensive analysis model to obtain a temperature feature value that can represent the temperature state of the corresponding temperature abnormal unit tile; the pre-trained temperature feature comprehensive analysis model receives the temperature feature vector from step S14 as input; the temperature feature comprehensive analysis model is constructed based on machine learning or deep learning, which can learn the differences between different types of temperature abnormalities and output a temperature feature value representing the temperature state; the temperature feature value comprehensively considers the temperature value feature and the temperature value distribution feature, and is a quantitative evaluation of the tile temperature characteristics, helping the system to determine whether there is a potential bleeding risk;

[0077] Step S16, among all the temperature abnormal unit tiles, filter out the temperature abnormal unit tiles whose temperature feature values exceed the preset temperature feature threshold, and mark the corresponding tissue sites as abnormal bleeding tissue sites; the system compares the temperature feature value of each temperature abnormal unit tile with the preset temperature feature threshold; if the temperature feature value of a certain tile exceeds the set threshold, it is considered that the tissue site represented by the tile has a higher bleeding possibility, and it is marked; the marked area is the abnormal bleeding tissue site, and its position information is recorded for further analysis and confirmation; in this way, the system can accurately locate the potential bleeding point at an early stage, providing reliable diagnostic basis for doctors.

[0078] More specifically, in step S13, the temperature gradient is set based on in-depth understanding and analysis of infrared thermal imaging data, and grasp of temperature change characteristics caused by bleeding, inflammation and other pathological processes, and is set as follows:

[0079] Determine the basic temperature range, which covers the temperature fluctuation of normal tissue in physiological state; the basic temperature range can be obtained by statistical analysis of a large amount of infrared thermal imaging data of normal tissue;

[0080] After determining the basic temperature range, it can be divided into multiple temperature gradients as needed; the division of temperature gradient should be detailed enough to capture subtle temperature changes, while avoiding too much complexity that makes the analysis process cumbersome; the temperature gradient division is divided by fixed temperature interval, for example, every 0.2℃ or 0.5℃ as a gradient; on the other hand, the division of temperature gradient can also be dynamically adjusted according to the temperature change characteristics caused by pathological processes; for example, if it is known that a certain pathological process usually causes temperature change within a certain range, then this range can be divided into more detailed gradients to better capture such changes;

[0081] The setting of the temperature gradient is not fixed, but needs to be optimized and adjusted according to the actual application; for example, the optimal temperature gradient division method can be selected by comparing the recognition effect of abnormal bleeding tissue parts under different temperature gradients; when setting the temperature gradient, the influence of individual differences on temperature distribution also needs to be considered; the normal tissue temperature of different patients may be different, so it needs to be adjusted appropriately according to the specific situation of the patient in actual application.

[0082] In step S15, the temperature feature comprehensive analysis model extracts the temperature feature value representing the temperature state of the temperature abnormal unit block from the temperature feature vector through a comprehensive calculation formula; the specific calculation process is as follows:

[0083] X = [T avg ,T diff ,D,S];

[0084] Wherein, X represents the temperature feature vector, T avg represents the average temperature in the temperature value feature, T diff represents the maximum and minimum temperature difference in the temperature value feature, D represents the dispersion degree of the temperature value of each point in the temperature value distribution feature, and S represents the distribution form of the temperature value in the temperature value distribution feature;

[0085] Standardize each feature to ensure that different features are compared on the same scale; the standardized temperature feature vector is denoted as:

[0086]

[0087] Wherein, μ and σ are the mean and standard deviation of the corresponding feature in the training set, respectively;

[0088] In order to better capture the relationship between the temperature value feature and the temperature value distribution feature, we introduce an interaction term in the weighted summation process, which is realized by cross multiplication or nonlinear combination; the specific calculation formula of the temperature feature comprehensive analysis model is:

[0089]

[0090] Wherein, K is the temperature feature value, which is mapped to the range [0, 1], representing the probability of bleeding risk of the corresponding temperature abnormal unit block; W1 and W2 represent the weight matrix; (X'⊙X') represents the element-wise square of the feature vector element, which is used to introduce the quadratic term; b represents the bias term; e is a constant.

[0091] Through the comprehensive calculation formula of the temperature feature comprehensive analysis model, not only the temperature value features and temperature value distribution features are considered, but also the interaction terms between them are introduced, so that the complex patterns of the temperature abnormal unit block can be more accurately captured; the temperature feature comprehensive analysis model is more comprehensive and flexible, which helps to improve the precision and reliability of the bleeding point detection.

[0092] For the endoscopic image acquisition control module, the endoscopic image acquisition control module extracts the appropriate endoscopic image acquisition strategy from the endoscopic imaging strategy library according to the position information and temperature feature value, so as to ensure that the endoscopic panoramic image of the abnormal bleeding tissue site is obtained with high quality.

[0093] In the endoscopic imaging strategy library, the most suitable endoscopic image acquisition strategy is extracted based on the search coordinates; the endoscopic imaging strategy library is a pre-constructed database, which stores a variety of different image acquisition parameter combinations, each of which is optimized for specific anatomical structures, temperature features and clinical needs; the endoscopic imaging strategy library is derived from a large amount of clinical experience and experimental data, aiming to ensure that high-quality endoscopic panoramic images can be obtained under different conditions; specifically, the endoscopic image acquisition strategy includes the following aspects:

[0094] Image acquisition focal length: according to the depth and size of the abnormal bleeding tissue site, select the appropriate focal length to ensure image clarity and detail visibility;

[0095] Image acquisition step frequency: set the speed and frequency of the endoscope probe movement to ensure coverage of the entire target area while avoiding missing important details;

[0096] Image acquisition angle: adjust the angle of the endoscope probe to capture the diseased tissue from different perspectives and improve diagnostic accuracy;

[0097] Light source brightness: adjust the light source brightness according to the reflection characteristics of the surrounding tissues to ensure sufficient illumination intensity without excessive exposure or shadows;

[0098] Light source color temperature: select the appropriate light source color temperature to enhance image contrast and make the differences between different tissue types more obvious.

[0099] The position information and temperature feature value obtained by the infrared thermal imaging positioning module are used as search coordinates for accurate search in the endoscopic imaging strategy library; the most suitable endoscopic image acquisition strategy is matched through an algorithm; the matching process involves multi-dimensional weight calculation and optimization algorithms to ensure that the selected strategy can best adapt to the current detection environment and requirements; once the optimal endoscopic image acquisition strategy is determined, the system will control the endoscope probe to perform image acquisition tasks according to these parameters, the specific operation is as follows:

[0100] Focus adjustment: Adjust the lens focus by electric or manual means to keep the target area in optimal focus;

[0101] Step frequency control: Start the automatic navigation system of the endoscope probe to smoothly scan the entire abnormal area at the set step frequency;

[0102] Angle adjustment: Use the fine adjustment mechanism on the endoscope probe to flexibly change the shooting angle to ensure full coverage;

[0103] Light source adjustment: Dynamically adjust the light source output according to the selected brightness and color temperature parameters to optimize image quality.

[0104] After completing the above steps, the endoscope probe will capture a series of continuous high-quality images and integrate them into a complete endoscopic panoramic image through stitching or other image processing techniques; the endoscopic panoramic image is a comprehensive display of the abnormal bleeding tissue site and its surrounding environment, providing rich visual information for subsequent bleeding point identification.

[0105] The endoscopic image acquisition control module extracts the most suitable endoscopic image acquisition strategy from the endoscope image acquisition strategy library based on the position information and temperature characteristic values determined by infrared thermal imaging, ensuring the relevance and effectiveness of image acquisition; at the same time, it also considers various image acquisition parameters such as focal length, step frequency, angle, light source brightness and color temperature to optimize image quality in all directions; this comprehensive image acquisition method not only improves the accuracy of diagnosis, but also provides high-quality visual information for subsequent bleeding point identification, which helps doctors make more accurate judgments and treatment plans.

[0106] For the bleeding point identification module, the bleeding point identification module is mainly used for bleeding feature identification of the endoscopic panoramic image, which implements the following:

[0107] Step S31, the endoscopic panoramic image is a comprehensive display of the abnormal bleeding tissue site and its surrounding environment; in order to ensure the accuracy of subsequent bleeding feature identification, the images first need to be preprocessed; the preprocessing steps include removing random noise in the image to improve image quality; adjusting the image contrast to make the differences between different tissue types more obvious; adjusting all images to a uniform resolution and size for input into the bleeding point identification model;

[0108] Step S32, input the pre-processed endoscopic panoramic image into the pre-trained bleeding point recognition model; the bleeding point recognition model is constructed based on deep learning (such as convolutional neural network CNN), and is trained by a large amount of labeled data, and can automatically recognize the characteristics of bleeding points; the specific operation is: through multiple convolutional layers and pooling layers, rich local and global features are extracted from the image, including shape, texture, color, etc.; using a fully connected layer or other classifier, it is judged whether each pixel or region belongs to a bleeding point according to the extracted features; a probability map with the same size as the input image is generated, where each pixel value represents the probability of the position being a bleeding point;

[0109] Step S33, according to the probability map, a threshold (for example, 0.5) is set, and the pixels exceeding the threshold are marked as potential bleeding points; then, through morphological operations (such as dilation, corrosion), small isolated regions are removed, and larger and connected suspected bleeding regions are retained; finally, a set containing multiple suspected bleeding points is formed, each suspected bleeding point is accompanied by its orientation information (such as coordinates, angles, etc.), so as to facilitate subsequent accurate navigation and microscopic imaging;

[0110] Step S34, in order to further improve the detection accuracy, the preliminary generated suspected bleeding point set is subjected to secondary screening and optimization, which specifically includes: combining the information of the surrounding tissue structure, excluding those suspected bleeding points that do not conform to the anatomical logic; applying different scale feature detection in the same region to ensure that small but important bleeding points are not missed; for dynamically acquired image sequences, consider the continuity and change trend in the time dimension to improve the reliability of identification.

[0111] Through the above process, the bleeding point recognition module realizes the automatic bleeding feature recognition of the endoscopic panoramic image, and generates a reliable suspected bleeding point set, which not only reduces the workload of doctors, but also improves the detection speed and accuracy; Especially with the help of the powerful representation ability of the deep learning model, even in a complex background, it can effectively distinguish bleeding points from other normal tissues, significantly improving the consistency and reliability of diagnosis; in addition, through the optimization processing of the suspected bleeding point set, the false positive rate is further reduced.

[0112] For the microscopic image acquisition control module, the suspicious bleeding point set has been generated by the bleeding point identification module, and detailed positional information (such as coordinates, angles, etc.) is provided for each suspicious bleeding point. These information will serve as the basis for navigation in subsequent microscopic imaging. According to the positional information in the suspicious bleeding point set, the device will automatically control the endoscope probe to switch from the conventional endoscope module to the integrated microscopic image acquisition module. Specifically, through electric or manual means, the optical components at the front end of the endoscope probe are switched from wide-angle lenses to high-resolution microscopes. According to the requirements of microscopic imaging, the brightness and color temperature of the light source are adjusted to ensure sufficient illumination intensity and appropriate contrast, meeting the requirements of high-resolution imaging.

[0113] To ensure that the microscopic image acquisition module can accurately focus on each suspicious bleeding point, the device needs to perform precise positioning and navigation operations, such as using the fine adjustment mechanism on the endoscope probe to make precise adjustments based on the specific positional information of the suspicious bleeding points, ensuring that the probe accurately aligns with the target area. Combining real-time image feedback and pre-set position information, the position of the probe is dynamically adjusted until the best imaging effect is achieved.

[0114] Once the endoscope probe is successfully positioned to the suspicious bleeding point, the microscopic image acquisition module will start high-resolution image acquisition. In order to fully capture the details of the bleeding point and its surrounding tissue, the device will take pictures from multiple angles to form a set of multi-view images. For larger bleeding areas, the device can automatically perform image stitching to generate a complete microscopic panoramic image. At the same time, image fusion technology is used to improve image quality and detail visibility.

[0115] After preliminary processing, the collected microscopic images will be automatically stored by the device for subsequent analysis and manual confirmation. Preliminary processing includes removing image noise and enhancing image contrast to make bleeding features more obvious. Labeling information (such as bleeding point number, position, etc.) is added to each microscopic image, and classified management is performed according to the suspicious bleeding point set.

[0116] Through the above process, the microscopic image acquisition control module realizes high-resolution microscopic imaging of suspicious bleeding points, significantly improving detection accuracy and reliability. The entire switching and positioning process is highly automated, reducing the doctor's operation burden and human error. Combined with the real-time feedback mechanism, the endoscope probe can accurately align with each suspicious bleeding point, improving imaging quality. Through multi-angle shooting, image stitching and fusion, etc., high-quality microscopic images are obtained, which helps to more accurately identify bleeding features. Labeling and classification management are added during image processing and storage, which facilitates subsequent analysis and doctor's manual confirmation, improving work efficiency.

[0117] For the bleeding point analysis module, before this, the microscopic image acquisition control module has obtained the high-resolution microscopic image of each suspected bleeding point through the microscopic image acquisition module; In order to ensure the accuracy of subsequent analysis, the microscopic image needs to be preprocessed first; The preprocessing steps include but are not limited to denoising, contrast enhancement, size standardization;

[0118] Next, the preprocessed microscopic image is input into the pre-trained bleeding point analysis model; The bleeding point analysis model and the bleeding point recognition model are constructed in a similar way, both of which are based on deep learning (such as convolutional neural network CNN) and can automatically identify the characteristics of bleeding points after a large amount of training data; The bleeding point analysis model will extract and compare the features of the input microscopic image, and calculate the bleeding probability value of each suspected bleeding point through deep learning algorithm; The bleeding probability value reflects the possibility of the suspected bleeding point being a real bleeding point; Store the bleeding probability value of each suspected bleeding point and its corresponding microscopic image for subsequent sorting and manual confirmation.

[0119] In the implementation process of the bleeding point analysis module, preprocessing the microscopic image can significantly improve the accuracy of subsequent analysis; At the same time, using the pre-trained deep learning model for feature analysis and bleeding probability calculation can automatically identify the subtle features of bleeding points, overcoming the limitations of individual experience and skill level; In addition, the output of bleeding probability value provides more objective and reliable diagnostic basis for doctors, which helps doctors quickly and accurately determine the bleeding point, and improves the efficiency and accuracy of diagnosis.

[0120] For the sorting display module, before this, the bleeding point analysis module has generated a bleeding probability value for each suspected bleeding point; In the running process of the sorting display module, all suspected bleeding points and their corresponding bleeding probability values and microscopic images are organized into a data set; Each data item contains:

[0121] Bleeding probability value: indicates the possibility of the suspected bleeding point being a real bleeding point;

[0122] Microscopic image: the high-resolution microscopic image corresponding to the suspected bleeding point;

[0123] Orientation information: the specific location information (such as coordinates, angles, etc.) of the suspected bleeding point.

[0124] Next, the above data set is sorted in descending order according to the bleeding probability value to form an ordered bleeding probability point sequence; Select an efficient sorting algorithm (such as quicksort, mergesort, etc.) to ensure that the sorting process is efficient and accurate; For the case of having the same bleeding probability value, the sorting rule can be further refined according to other factors (such as image quality, orientation information, etc.);

[0125] After sorting, a bleeding probability point sequence is generated, where each element contains the bleeding probability value, microscopic image, and location information of the suspected bleeding point. The bleeding probability point sequence is arranged from high to low bleeding probability to ensure that doctors can prioritize viewing the most likely bleeding points.

[0126] To facilitate doctors' viewing and analysis, the sorted sequence of bleeding probability points is optimized for display. Specifically, an intuitive user interface is designed to allow doctors to easily browse microscopic images and related information of different bleeding points. Clear annotations and necessary notes are added to each microscopic image to help doctors quickly understand the image content. Interactive functions, such as zooming, rotation, and contrast adjustment, are provided to enhance doctors' ability to observe image details.

[0127] During the sorting and display module's operation, the descending order of bleeding probability values ​​ensures that doctors can prioritize the most likely bleeding points, reducing unnecessary examination time. The entire sorting process is highly automated, reducing the time and error associated with manual sorting and evaluation by doctors. Optimized display formats and interactive functions enable doctors to browse and analyze images more efficiently, improving work efficiency. The intuitive user interface and detailed annotation information provide doctors with scientific evidence, helping them make faster and more accurate clinical decisions. Based on quantitatively assessed bleeding probability values, the influence of subjective judgment is reduced, improving the consistency and reliability of diagnosis.

[0128] For the manual confirmation module, during the operation of the manual confirmation module, the previously obtained bleeding probability point sequence and the corresponding multiple microscopic images are presented to the operating doctor; the bleeding probability point sequence has been sorted according to the bleeding probability value, and the suspected bleeding point at the top is most likely to be the real bleeding point.

[0129] When doctors review these images and sequences, they can visually see the microscopic images of each suspected bleeding point and their corresponding bleeding probability values. Doctors can make a comprehensive judgment based on their professional knowledge and experience, combined with this information. If doctors have questions about a suspected bleeding point or need more detailed information, they can further review more images of that point or conduct other examinations.

[0130] After confirming the location of the bleeding point, the doctor can formulate an appropriate treatment plan based on the actual situation. If the bleeding point is clear and the situation is urgent, the doctor may immediately perform hemostasis. If the bleeding point is not obvious or requires further confirmation, the doctor may choose to conduct more examinations or observations.

[0131] In this embodiment, firstly, the infrared thermal imaging positioning module uses infrared thermal imaging technology to determine the location of abnormal bleeding tissue and acquire location information, enabling earlier detection of potential bleeding sources. This is more sensitive and reliable than traditional visible light imaging. Next, the endoscopic image acquisition control module controls the endoscopic probe to move to the target area for panoramic image acquisition according to a pre-determined acquisition strategy, reducing the time and error associated with manual positioning by the doctor. The bleeding point identification module uses a pre-trained bleeding point identification model to automatically analyze the endoscopic panoramic image, forming a set of suspected bleeding points, reducing reliance on the doctor's experience and skill level and improving diagnostic consistency. Subsequently, the microscopic image acquisition control module switches to a high-resolution microscopic imaging module based on the location information of the suspected bleeding points, ensuring focused attention on key areas and enhancing the ability to observe details. The bleeding point analysis module applies the bleeding... The point analysis model performs feature analysis on microscopic images and calculates bleeding probability values, providing doctors with quantitative evidence to aid in accurate diagnosis. The sorting and display module adjusts the image sequence in descending order of bleeding probability values, allowing doctors to prioritize viewing the most likely bleeding points and optimizing the operation process. Finally, the manual confirmation module displays the sorted images and bleeding probability point sequence to the doctor, and combined with manual confirmation, the bleeding point location is finally determined, which not only improves operational efficiency but also reduces complexity and uncertainty. The entire method integrates the advantages of infrared thermal imaging, machine learning, and high-resolution microscopic imaging, realizing full automation and intelligence from initial screening to accurate diagnosis. It greatly improves the effect of endoscopic bleeding point detection, meets the need for accurate bleeding point localization in clinical practice, and at the same time reduces the workload of doctors and improves the consistency and reliability of diagnosis.

[0132] Example 2: Figure 2 As shown, the present invention provides an endoscopic bleeding point detection system based on infrared thermal imaging. This system combines traditional optical imaging technology with advanced infrared thermal imaging technology, and improves the accuracy of bleeding point diagnosis through intelligent image processing algorithms. The system specifically includes an endoscope probe, a display, a user interface, a power management unit, and a communication interface.

[0133] More specifically, the endoscope probe, integrated with a microscopic imaging module, is used to acquire images. Specifically, the endoscope probe's front end is equipped with a high-resolution CMOS camera and LED illumination source, enabling it to capture clear endoscopic images. The microscopic imaging module includes a miniaturized microscope lens with adjustable focus, which can switch to high-magnification mode when needed to obtain detailed images at the cellular level. An infrared thermal imaging sensor, using an uncooled microbolometer, is installed inside the endoscope probe, accurately capturing the temperature distribution within the body, especially areas of localized temperature elevation caused by bleeding.

[0134] The display uses a high-definition LCD or OLED screen with touch function, which can display the color images, infrared thermal images and processed bleeding probability distribution images collected by the endoscope in real time.

[0135] The user interface allows the doctor to directly browse different types of images through the touch screen, adjust the position and angle of the probe, set parameters such as exposure time, contrast, etc., and finally confirm the bleeding point position.

[0136] The power management unit is used to provide power supply for all system components; it contains a large-capacity lithium battery, which cooperates with the intelligent charging circuit to ensure long-time work without interruption; at the same time, it is provided with overcurrent protection and temperature monitoring mechanism to ensure the safe operation of the equipment.

[0137] The communication interface is equipped with multiple communication methods for supporting data transmission between modules of the infrared thermal imaging-based endoscopic bleeding point detection device, including but not limited to USB, HDMI, Wi-Fi, Bluetooth, etc., which not only supports high-speed data transmission between modules, but also can connect to the hospital information management system, facilitating medical record keeping and remote collaboration.

[0138] Specifically, the workflow of the system is as follows: before starting the system, the doctor first selects the corresponding examination mode through the user interface and makes necessary settings according to the specific situation of the patient; the endoscope probe is gently inserted into the patient's body and scanning begins; the infrared thermal imaging sensor continuously monitors the temperature changes in the body, and once an abnormally high temperature area is found, it is immediately marked and sent to the control console through wireless signal; according to the received position information, the control console guides the endoscope probe to move to the suspected bleeding point, and then uses the microscopic image acquisition module to collect detailed images; the AI algorithm built-in the system automatically analyzes the collected images, calculates the probability values of each suspected bleeding point, and sorts them according to the likelihood size; all analysis results are presented on the display together with the original images for the doctor to review; finally, the doctor makes a judgment according to the system's suggestions and personal experience to determine the real bleeding point position.

[0139] The above is only the preferred embodiment of the present application, it should be noted that for those skilled in the art, without departing from the technical principles of the present application, a number of improvements and modifications can be made, which should be considered as the protection scope of the present application.

Claims

1. An infrared thermography-based endoscopic bleeding spot detection device, characterized by, The device is applied to an endoscope probe integrated with a microscopic image capturing module, and the device comprises: an infrared thermal imaging positioning module, configured to determine at least one abnormal bleeding tissue site by using an infrared thermal imaging technology, acquire position information of the abnormal bleeding tissue site, and send the position information; an endoscopic image acquisition control module, configured to receive the position information, control an endoscope module in the endoscope probe to move to a site corresponding to the position information, perform image acquisition according to a predetermined endoscopic image acquisition strategy, acquire an endoscopic panoramic image of the abnormal bleeding tissue site, and send the endoscopic panoramic image after compression; a bleeding point identification module, configured to receive the endoscopic panoramic image after decompression, perform bleeding feature identification on the endoscopic panoramic image by using a pre-stored bleeding point identification model, identify a suspicious bleeding point set, and send the suspicious bleeding point set; the suspicious bleeding point set comprises azimuth information of at least one suspicious bleeding point; a microscopic image acquisition control module, configured to receive the suspicious bleeding point set, control the endoscope probe to switch to the microscopic image capturing module, perform image acquisition on each suspicious bleeding point according to azimuth information of each suspicious bleeding point in the suspicious bleeding point set, acquire a corresponding microscopic image, and send the microscopic image after packaging; a bleeding point analysis module, configured to receive the microscopic image, perform feature analysis on the microscopic image of each suspicious bleeding point by using a pre-stored bleeding point analysis model, calculate a bleeding probability value of each suspicious bleeding point, and send the bleeding probability value; a result output module, configured to receive the microscopic image and the corresponding bleeding probability value, and output a detection result after comprehensive analysis.

2. The infrared thermography-based endoscopic bleeding spot detection device of claim 1, wherein, The comprehensive analysis of the result output module comprises: adjusting a plurality of microscopic images in a descending order according to the bleeding probability value of each suspicious bleeding point, generating a bleeding probability point sequence, and taking the bleeding probability point sequence and the corresponding plurality of microscopic images as the detection result.

3. The infrared thermography-based endoscopic bleeding spot detection device of claim 1, wherein, The infrared thermal imaging positioning module comprises: an infrared data acquisition submodule, configured to acquire infrared thermal imaging data of a region to be detected of a patient, and send the infrared thermal imaging data; an abnormal region identification and segmentation submodule, configured to receive the thermal imaging data sent by the infrared data acquisition submodule, perform abnormal region identification and segmentation, acquire an abnormal temperature region image, and send the abnormal temperature region image; a temperature gradient division submodule, configured to receive the abnormal temperature region image sent by the abnormal region identification and segmentation submodule, further divide the abnormal temperature region image into a plurality of temperature abnormal unit blocks based on a preset temperature gradient standard, and send the temperature abnormal unit blocks after packaging; pixel temperature values in each temperature abnormal unit block are in the same temperature gradient; a temperature feature extraction submodule, configured to receive the temperature abnormal unit blocks sent by the temperature gradient division submodule, perform temperature feature extraction on each temperature abnormal unit block, acquire a corresponding temperature feature vector, and send the temperature feature vector; the temperature feature vector comprises a temperature value feature and a temperature value distribution feature. The temperature feature comprehensive analysis submodule is configured to receive the temperature feature vector sent by the temperature feature extraction submodule, input the temperature feature vector into a pre-stored temperature feature comprehensive analysis model, comprehensively analyze the temperature feature vector, output a temperature feature value capable of representing a temperature state of a corresponding temperature abnormal unit tile, and send the temperature feature value. The abnormal tile screening and marking submodule is configured to receive the temperature feature value sent by the temperature feature comprehensive analysis submodule, screen out a temperature abnormal unit tile exceeding a preset temperature feature threshold from all temperature abnormal unit tiles according to the temperature feature value, and mark a corresponding tissue site of the temperature abnormal unit tile as an abnormal bleeding tissue site.

4. The infrared thermography-based endoscopic bleeding spot detection device of claim 3, wherein, In the endoscopic image acquisition control module, a determination method of the endoscopic image acquisition strategy includes: The position information and the corresponding temperature feature value are used as retrieval coordinates to extract a corresponding endoscopic image acquisition strategy from an endoscopic image acquisition strategy library. The endoscopic image acquisition strategy includes an image acquisition focal length, an image acquisition step frequency, an image acquisition angle, a light source brightness, and a color temperature.

5. The infrared thermography-based endoscopic bleeding spot detection device of claim 3, wherein, The temperature feature vector is: X = [T avg ,T diff ,D,S] ; where X represents the temperature feature vector, T avg represents the average temperature in the temperature value feature, T diff represents the maximum-minimum temperature difference in the temperature value feature, D represents the dispersion degree of the temperature values at each point in the temperature value distribution feature, and S represents the distribution form of the temperature values in the temperature value distribution feature.

6. The infrared thermography-based endoscopic bleeding spot detection device of claim 5, wherein, The calculation formula of the temperature feature comprehensive analysis model is: wherein K is the temperature feature value, the temperature feature value is mapped to a range of [0, 1], and represents a probability of a bleeding risk of the corresponding temperature abnormal unit tile; X' represents a standardized feature vector; W1 and W2 represent weight matrices; (X'⊙X') represents an element-wise square of a feature vector element, used to introduce a quadratic term; b represents a bias term; and e is a constant.

7. The infrared thermography-based endoscopic bleeding site detection device of claim 6, wherein, The bleeding point recognition model extracts local features and global features from the endoscopic panoramic image through multiple convolutional layers and pooling layers. A full connection layer is used to determine whether each pixel belongs to a bleeding point according to the extracted features, and a probability map with the same size as the endoscopic panoramic image is generated, wherein each pixel value represents a probability that the position is a bleeding point.

8. The infrared thermography-based endoscopic bleeding spot detection device of claim 7, wherein, The bleeding point recognition module running process includes: receiving and decompressing the endoscopic panoramic image sent by the endoscopic image acquisition control module; preprocessing the endoscopic panoramic image; inputting the preprocessed endoscopic panoramic image into a pre-trained bleeding point recognition model to obtain a probability map with the same size as the input image, wherein each pixel value represents a probability that the position is a bleeding point; a probability threshold is pre-set in the bleeding point recognition module, and the probability threshold is used to traverse the probability map, and pixels exceeding the probability threshold are marked as potential bleeding points; all potential bleeding points are summarized to obtain a suspicious bleeding point set.

9. The infrared thermography-based endoscopic bleeding spot detection device of claim 8, wherein, The preprocessing of the endoscopic panoramic image includes: removing random noise in the image; adjusting the contrast of the image to increase the difference between different tissue types; adjusting all images to a uniform resolution and size.

10. An infrared thermography-based endoscopic bleeding spot detection system, characterized by, The endoscopic bleeding point detection device based on infrared thermal imaging according to any one of claims 1-9 further includes: an endoscope probe, the endoscope probe further integrated with a microscopic image acquisition module and an infrared thermal imaging sensor, configured to acquire images and capture temperature distribution in a patient's body; and an endoscopic image acquisition control module, configured to receive the endoscopic panoramic image, determine an endoscopic image acquisition strategy according to the position information of the endoscopic panoramic image, and send the endoscopic panoramic image and the endoscopic image acquisition strategy to the endoscope probe. a display for showing the images captured by the endoscope probe and the microscopic image module, the infrared thermal image and the detection result to the operating physician; a user interface as an interactive platform for the operating physician to browse the images, adjust parameters and confirm the bleeding point; a power management unit for providing power supply to all system components; a communication interface for supporting data transmission among the modules of the infrared thermal imaging based endoscopic bleeding point detection device and information exchange with the hospital information system.

Images