Inflammatory disease detection method for monitoring hypoxia characteristics of organ-like chip disease model based on endoscope

Abstract

The invention discloses an inflammatory disease detection method for monitoring hypoxia characteristics of an organ-like chip disease model based on an endoscope, which is applied to screening and prediction of inflammatory diseases and aims to solve the problem that related disease diagnosis indexes for reflecting biochemical parameters under the endoscope and used for screening and prediction of early inflammatory diseases are lacked in the prior art. The method comprises the following steps: firstly, constructing an in-vitro organ-like chip model of inflammatory diseases based on an organ-like culture technology and a microfluidic regulation system, and simulating an in-vivo tissue microenvironment and hypoxia heterogeneity; secondly, on the basis of an endoscope imaging system, full-life-cycle, real-time, dynamic, label-free and low-damage hypoxia characteristic monitoring is carried out on the organ-like chip disease model; then, fusing cross-scale multi-source information such as biological image morphological characteristics, tissue optical hypoxia characteristics and cell morphology and pathological characteristics of tissue pathological detection to form a comprehensive data set, and constructing a correlation model between the hypoxia characteristics and disease occurrence, development and outcome; and finally, based on the established correlation model, training a comprehensive data set, and developing an intelligent endoscopic detection algorithm for inflammatory diseases based on hypoxia feature monitoring of an organ-like chip disease model.

Description

TECHNICAL FIELD

[0001] The present application belongs to the field of inflammatory disease prediction and organ chip, and particularly relates to an endoscopic prediction technology for inflammatory diseases. BACKGROUND

[0002] Inflammation is a defense mechanism of the body, which can find and remove harmful exogenous substances, and initiate the self-healing process of the body. However, chronic inflammation is a long-term and slow progress of inflammation, which can last for months to years and cause various diseases. Chronic inflammation and related diseases are important causes of human death, and are called a great threat to human health by the World Health Organization (WHO). Globally, 3 out of 5 people are troubled by chronic inflammation, and these chronic inflammation diseases include cancer, chronic respiratory disease, stroke, cardiovascular disease, obesity and diabetes, etc. Inflammatory diseases (such as inflammatory bowel disease, gastric ulcer) and chronic inflammation related cancers (such as colorectal cancer, gastric cancer) are also proven to have a close relationship.

[0003] Among many factors affecting diseases, hypoxia as a key pathological factor deeply affects the occurrence and development of chronic inflammation by regulating cell metabolism, inflammatory response and oxidative stress, etc. In the diagnosis of digestive tract inflammatory diseases, the traditional method needs to determine the degree of inflammation or the nature of the disease by comprehensively judging the lesion location and lesion area by endoscopy and judging acute / chronic and benign / malignant by pathological examination. The reason why the traditional endoscopy technology cannot directly diagnose is that it mostly acquires morphological structure information, and many diseases only have physiological function changes in the early stage, and cannot achieve in vivo pathological diagnosis. Ex vivo biopsy or postoperative pathological examination is the gold standard in clinical practice. Therefore, in the early diagnosis and efficacy evaluation of diseases, clinical practice needs to develop new endoscopic diagnosis methods based on functional information and pathological characteristics.

[0004] Although, for early tumor functional diagnosis before structural changes occur, various functional imaging technologies have been developed, imaging machines have been developed, and in vitro test experiments have been carried out. However, new imaging technologies such as photoacoustic imaging are limited by hardware performance, which hinders the promotion of clinical application; PET imaging needs contrast agents, and the influence of reagents on the human body restricts its use scenarios; MRI equipment costs are high, and it is not easy to popularize to daily screening and diagnosis, and the pain point problem that the traditional in vitro model cannot accurately reflect the physiological / pathological phenomena of tissues / organs in the body has not been effectively solved. Therefore, to solve a more accurate and simulated in vitro evaluation system and an ethically friendly experimental platform is a necessary condition for exploring and verifying new endoscopic diagnosis methods.

[0005] In summary, the prior art has the following problems: (1) The research on the mechanism of hypoxia in chronic inflammatory diseases is mainly focused on the microscopic level of the influence of hypoxia on a single cell type or a few factors, and lacks macroscopic monitoring of the dynamic spatiotemporal variation of hypoxia in the entire disease development process. (2) In terms of monitoring hypoxic state, although various imaging techniques are available, low-cost, real-time, dynamic, and label-free monitoring methods for hypoxia in digestive tract inflammatory diseases still need to be developed and improved. (3) The traditional research model has limitations in simulating the real microenvironment in vivo, and it is difficult to accurately reflect the complex influence of hypoxia in diseases. (4) In terms of disease diagnosis, there is a need for the development of new endoscopic diagnostic techniques based on functional information and pathological characteristics. SUMMARY

[0006] To solve the above technical problems, the present application provides an inflammatory disease detection method based on endoscopic monitoring of hypoxic characteristics of organ chip disease models.

[0007] The technical scheme adopted by the present application is as follows: An inflammatory disease detection method based on endoscopic monitoring of hypoxic characteristics of organ chip disease models, comprising:

[0008] S1, constructing an in vitro organ chip model, the in vitro organ chip model comprising: a microfluidic chip, microchambers for simulating different tissues or organs arranged in the microfluidic chip, the microchambers being connected by microchannels; further comprising cells inoculated in the microchambers and corresponding culture substrates; the in vitro organ chip model further comprises a perfusion inflow channel and a perfusion outflow channel connected to a peristaltic pump;

[0009] S2, building an endoscopic detection system, the endoscopic detection system comprising: a laparoscope, a multi-spectral light source, a monochrome camera and a focusing lens, the multi-spectral light source being connected to the laparoscope through an optical fiber cable; the monochrome camera is used to take real-time images under the laparoscope; the monochrome camera is also connected to a computer;

[0010] S3, data acquisition and feature processing of the collected data;

[0011] The collected data specifically includes: first endoscopic imaging results collected by the endoscopic detection system built in step S2 from the in vitro organ chip model constructed in step S1, second endoscopic imaging results collected by a clinical endoscope from the in vitro organ chip model constructed in step S1, and results obtained by performing clinical ex vivo histopathological examination or histological examination on the in vitro organ chip model constructed in step S1;

[0012] The feature processing specifically includes:

[0013] Extracting tissue hypoxic feature information from the first endoscopic imaging results;

[0014] extracting tissue structure information from the second endoscopic imaging result;

[0015] extracting histopathological cell morphology and pathological feature information from the result obtained from clinical ex vivo histopathological examination or histological examination;

[0016] Based on the same in vitro organ chip model constructed in step S1, the tissue hypoxia feature information, the tissue structure information, the cell morphology and the pathological feature information are constructed into multi-modal data, and a training set is obtained based on the multi-modal data;

[0017] S4, constructing a correlation model, specifically: using a machine learning algorithm to learn the multi-modal data in the training set, constructing the mapping relationship between the tissue hypoxia feature information and the disease course classification result, thereby establishing a correlation model;

[0018] S5, for the organ chip model to be detected, input the corresponding multi-modal data into the correlation model constructed in step S4 to obtain the classification result.

[0019] The endoscopic imaging result specifically includes: obtaining the image gray value of the tissue under different wavebands according to the real-time multi-spectral image; then calculating the blood oxygen saturation according to the image gray value, forming an oxygen saturation map based on the blood oxygen saturation of each pixel point; finally, obtaining the oxygen saturation value, hypoxic position and hypoxic range according to the calculated oxygen saturation map, and recording the hypoxic duration at the same time; the oxygen saturation value, hypoxic position, hypoxic range and hypoxic duration are taken as the endoscopic imaging result.

[0020] The beneficial effects of the present application are: firstly, based on the organoid culture technology and the microfluidic regulation system, an in vitro organ chip model of inflammatory diseases is constructed to simulate the in vivo tissue microenvironment and hypoxic heterogeneity; secondly, based on the endoscopic imaging system, the organ chip disease model is monitored for hypoxia characteristics in the whole life cycle, real-time, dynamic, label-free and low-damage; then, the cross-scale multi-source information such as biological image morphological features, tissue optical hypoxia characteristics and cell morphology and pathological features of histopathological detection is fused to form a comprehensive data set, and a correlation model between hypoxia characteristics and disease occurrence, development and outcome is constructed; finally, based on the established correlation model, the comprehensive data set is trained, and an intelligent endoscopic detection algorithm for inflammatory diseases based on the hypoxia characteristic monitoring of the organ chip disease model is developed. The method of the present application has the following advantages:

[0021] (1) The organoid in the application can simulate the physiological structure and part of the function of the tissue, and the organoid chip technology combining the organoid with the microfluidic system can accurately control the oxygen concentration and other parameters in the microenvironment, simulate the more real microenvironment in the body. By setting different control groups (considering different disease sites, disease types, disease stages, etc.) and different experimental conditions (considering the introduction of hypoxic environment, reoxygenation environment, treatment operation, surgical resection operation, etc.), the dynamic process of disease occurrence and development and the hypoxic characteristics in vitro are simulated.

[0022] (2) The endoscopic imaging system in the application is calibrated for the monitoring object of the organoid chip disease model, and the hypoxic feature extraction algorithm parameters are optimized to ensure the imaging quality and measurement accuracy. The hypoxic characteristics in different disease models are monitored at multiple time points for a long time, and the key information of the hypoxic characteristics (such as hypoxic degree, distribution range, duration, etc.) is marked;

[0023] (3) The fused cross-scale multi-source information fusion in the application includes macro features such as morphological features of biological images and hypoxic features of tissue optics, and micro features such as cell morphology and pathological features of histopathological detection. The cross-scale multi-source feature information is extracted, aligned and fused to form a comprehensive data set, the mapping relationship between the hypoxic characteristics and the occurrence, development and prognosis of diseases is fitted by using an algorithm, and a correlation model is constructed.

[0024] (4) The inflammatory disease endoscopic intelligent detection algorithm in the application is different from the traditional clinical diagnosis which needs to comprehensively judge the lesion position and lesion area by the endoscope and judge the disease nature by the pathological examination. It is an inflammatory disease endoscopic intelligent detection algorithm based on hypoxic feature images, which can monitor the hypoxic characteristics of the lesion in situ and real time and judge the occurrence and development of different inflammatory diseases or chronic inflammation related cancers, and realize the endoscopic disease detection. BRIEF DESCRIPTION OF DRAWINGS

[0025] Figure 1 is a method flowchart of the application;

[0026] Figure 2 is an endoscopic hypoxic monitoring system and an organoid chip disease model schematic diagram of an embodiment of the application. DETAILED DESCRIPTION

[0027] In order to make the purpose, technical scheme and advantages of the application more clear and obvious, the application is further described in detail below through the drawings and embodiments. However, it should be understood that the specific embodiments described herein are only used to explain the application, and are not used to limit the scope of the application. In addition, in the following description, the description of known structures and technologies is omitted to avoid unnecessary confusion of the concept of the application.

[0028] AsFigure 1 The method of the present application comprises the following steps:

[0029] 1. Construction of organoid chip model

[0030] 11. Construction of organoid chip model

[0031] Based on the organoid chip technology, an in vitro model of inflammatory disease is constructed to simulate the in vivo hypoxic heterogeneity of the lesion. As shown in the perfusion type 3D co-culture medium in Figure 1 , the specific steps are as follows:

[0032] 111. Material preparation: select appropriate organoid culture matrix to support the three-dimensional growth of tissue epithelial cells or stem cells. Prepare culture solution to simulate the tissue microenvironment, which contains necessary growth factors, cytokines and nutritional ingredients.

[0033] Taking the digestive tract disease in this embodiment as an example, the culture matrix here is specifically: natural extracellular matrix (ECM) derived materials such as Matrigel, collagen or synthetic and semi-synthetic hydrogels such as sodium alginate / gelatin composite gel, polyethylene glycol (PEG) based hydrogel, etc.

[0034] 112. Chip design: design a microfluidic chip, which contains microchambers simulating different tissues or organs in the chip, and the microchambers are connected by microchannels for culturing organoids and controlling the microenvironment.

[0035] 113. Organoid culture: seed tissue epithelial cells or stem cells in the microchamber of the chip, add culture matrix, and place it in a suitable condition incubator for culture. Regularly check the growth status of the organoids, observe cell differentiation and tissue structure formation by cell labeling and imaging technology, and analyze the functional state of the cells by biochemical and molecular biology methods.

[0036] Taking the digestive tract organoid chip as an example, the tissue epithelial cells or stem cells, such as: small intestine organoid intestinal stem cells or intestinal epithelial cell lines, gastric organoid gastric stem cells, colorectal organoid colorectal stem cells, are mostly obtained from the crypts of the digestive tract, or are obtained by inducing differentiation of pluripotent stem cells (iPSC).

[0037] As shown in Figure 2 , this embodiment adopts perfusion type 3D co-culture: cells and culture matrix are added to the reactor, and part of the conditioned medium is continuously taken out during the culture process, while new culture medium is continuously perfused. The advantage of this method is that it can maintain a high cell density, the reaction rate is easy to control and the culture period is longer.

[0038] The method of moving cells into the microchamber includes the following steps:

[0039] 1) Injected by micropipette;

[0040] 2) Designing chemical concentration gradient (such as chemotactic factor) in microfluidic chip to induce cell autonomous migration to target chamber;

[0041] 3) Mixing cells with biocompatible hydrogel (such as gelatin, sodium alginate) and precisely depositing into microchamber by 3D bioprinter;

[0042] The implementation process of moving cells into microchamber is a known technology, which is not described in detail here.

[0043] The present application uses microfluidic technology to precisely control oxygen concentration, nutrient supply and auxiliary treatment intervention, etc. environmental conditions, to simulate more realistic microenvironment in vivo, and observe the changes of the model after the conditions change.

[0044] 12, Control group setting

[0045] According to the actual situation, each part has normal non-diseased experimental group, and different disease and disease staging control group. Taking digestive tract diseases as an example, different control groups are set during organoid culture, considering different disease sites, disease subtypes, disease stages and other characteristics:

[0046] Disease site control: Set up organoid culture groups simulating different parts of esophagus, stomach, small intestine, large intestine, etc.

[0047] Disease type control: Set up organoid culture groups simulating different diseases such as inflammatory bowel disease, gastric ulcer, colorectal cancer, etc.

[0048] Disease staging control: Set up organoid culture groups simulating different stages of TNM staging or clinical staging, etc.

[0049] Those skilled in the art should know that one organoid chip model provided in the present application only corresponds to one of the multiple situations classified by site, disease type corresponding to each site, and disease staging corresponding to each type, that is, one organoid chip model is either a normal experimental group according to site, or a control group according to site, disease type corresponding to each site, and disease staging corresponding to each type.

[0050] 13, Experimental condition setting

[0051] Different experimental settings such as hypoxic environment, reoxygenation environment, treatment operation, surgical resection operation are introduced in the monitoring process. Different operations are performed according to the type of disease, and the decision of clinical treatment is referred to. If the corresponding operation is performed, the corresponding multi-modal data after the operation needs to be collected; in this embodiment, four kinds of operations of hypoxic environment, reoxygenation environment, treatment operation and surgical resection are mainly used, and each operation is set with a normal group and an operation control group, and the normal group does not perform specific operation.

[0052] Simulate different environmental changes during the development of the disease:

[0053] Hypoxic environment setting: Place the organoid chip in a hypoxic culture condition to simulate the in vivo hypoxic environment.

[0054] Reoxygenation environment setting: After a period of hypoxic culture, transfer the chip to a normoxic environment to simulate the reoxygenation process.

[0055] Therapeutic operation setting: Add different concentrations of drugs to the culture medium, or use photoelectric therapy, photodynamic therapy, etc. to simulate the treatment process.

[0056] Surgical resection operation setting: Remove part of the organoid to simulate surgical resection.

[0057] 2. Build an endoscopic detection system

[0058] As shown in Figure 2 , based on the principle of multispectral imaging, an endoscopic imaging system capable of real-time and dynamic monitoring of hypoxic characteristics is built. For the monitoring object of the organoid chip disease model, system calibration and algorithm optimization are performed to ensure imaging quality and measurement accuracy. Long-term, multi-time point hypoxic characteristic monitoring is performed on different disease models, and key information such as hypoxic degree, distribution range, and duration is marked. The specific steps are as follows:

[0059] Hardware setup: As shown in Figure 2 , the hardware specifically includes: laparoscope, optical fiber cable, focusing lens, multispectral light source, monochrome camera, and computer;

[0060] The working principle of the endoscope detection system of the present application is as follows: the laparoscope is the main mechanical structure of the system, including an optical channel (transmitting light and image), a working channel (inserting surgical instruments) and an illumination channel (introducing light source), the objective lens group (including focusing lens) at the front end of the laparoscope is responsible for capturing the image of the surface of the organ, and the rear end is connected with the optical fiber cable and the monochrome camera. The multi-spectral light source provides specific wavelength illumination for enhancing the identification of tissue characteristics (such as blood vessel distribution), and the multi-spectral light emitted by the light source is transmitted to the front end of the endoscope through the light guide fiber in the optical fiber cable (which is built in the illumination channel of the laparoscope), and the light is uniformly irradiated to the target tissue through the light dispersing element (such as lens group). The optical fiber cable is the transmission medium of image and light, and the illumination fiber transmits the light emitted by the multi-spectral light source to the surface of the tissue. The focusing lens (objective lens and eyepiece group) is the core of optical imaging, the objective lens group (front end lens) is located at the front end of the laparoscope, close to the target tissue, responsible for collecting reflected light and focusing imaging. Adjust the lens spacing by manual or automatic focusing to make the reflected light of the tissue surface clearly focused on the camera sensor. The eyepiece group (rear end lens) is located at the rear end of the laparoscope, further magnifies the image transmitted by the optical fiber, and is convenient for the monochrome camera to capture. The monochrome camera is used for photoelectric signal conversion, receives the optical image focused by the lens group, and converts the optical signal into digital electric signal through the CCD / CMOS photoelectric sensor. The monochrome camera has higher sensitivity to single spectrum, and can realize spectrum-specific imaging in cooperation with the multi-spectral light source. The working process of the monochrome camera is that the sensor collects light intensity signals pixel by pixel to generate a gray-scale image, and then the gray-scale image is processed by computer for pseudo-color processing or multi-spectral fusion. The computer performs subsequent steps such as image reconstruction, processing and analysis, and displays the processed image on the display in real time for the doctor to observe; at the same time, the data is stored for subsequent diagnosis or scientific research.

[0061] The laparoscope in the embodiment specifically adopts Qingdao Omer 680 series laparoscope, model: 680-331000H, mirror body outer diameter: 10mm, working length: 331mm, viewing angle: 0°, field of view angle: 75°, optical working distance: 50mm, effective depth of field range: 25-150mm, illumination variation rate: 20%, edge uniformity: 25%, the endoscope has good parfocal property in the range of 400nm-850nm, and there are three kinds of accessory light guide beam adapters: Wolf adapter, Storz adapter and ACMI.

[0062] In actual application, the endoscope monitoring system can be built by selecting a suitable multi-spectral light source. There is no special requirement for other hardware. Those skilled in the art should know that multi-spectral imaging is also called electronic staining technology, which uses blood vessels, blood, mucosal surface lipids and other specific markers, and uses the optical properties of biological tissues (such as the difference of absorption coefficient and scattering coefficient parameters at different wavelengths) to develop more specific imaging methods and imaging technology.

[0063] Software development: Develop image acquisition, preprocessing and statistical analysis to realize real-time processing and analysis of multispectral images.

[0064] System calibration: Calibrate the system to ensure imaging quality and measurement accuracy.

[0065] 3. Data collection and feature processing

[0066] Based on the built endoscope detection system and the clinical routine endoscope, data collection is performed on the constructed organoid model, and feature processing is performed on the collected data. The specific process is as follows:

[0067] 31. Data collection

[0068] Using the endoscope monitoring system to observe the oxygenation state of various organ chip disease models, realizing the monitoring of the hypoxic characteristics of the model in the whole life cycle. That is, long-term, multi-time point hypoxic characteristic monitoring of different organ disease models in the whole life cycle is realized, and traditional endoscopy, pathological examination or histological examination are simultaneously performed as detection basis, which specifically includes: the computer obtains the first endoscopic imaging result according to the real-time multispectral images captured by the monochrome camera; it also includes extracting clinical detection related information from the collected different modal results, which specifically includes: tissue structure information extracted from the second endoscopic imaging result obtained from the clinical routine endoscope and cell morphology and pathological characteristic information extracted from the histopathological examination or histological examination result. Combine the hypoxic characteristics with the clinical detection results to obtain multi-modal data.

[0069] The data collection steps are as follows:

[0070] Image acquisition: Collect multispectral images of organoid chips through the endoscope monitoring system, set appropriate exposure time and frame rate, and collect imaging results through clinical routine endoscope. In the same position, the information provided in two modes is different, and the specific information is as follows:

[0071] The information collected by the clinical routine endoscope includes:

[0072] (1) Tissue morphology and structure information: Using a wide-spectrum light source close to natural light for illumination, it can clearly show the details of tissue surface morphology, contour, color and texture, etc., helping doctors to observe the macrostructure and morphological changes of tissues, such as the smoothness of mucosa, the morphology of folds, the presence or absence of ulcers, tumors, etc.

[0073] (2) Vascular structure information: The vascular distribution and morphology on the mucosal surface can be observed, but the display effect of blood vessels may not be as prominent as the special blood vessel imaging mode.

[0074] (3) Preliminary screening of lesions: By observing the color changes of the tissue, such as redness, whitening, yellowing, blackening, etc., clues for the preliminary diagnosis of certain lesions are provided, for example, mucosal congestion and redness caused by inflammation, mucosal pallor caused by atrophic lesions, etc.

[0075] The information collected by the multispectral endoscope of the present application includes:

[0076] (1) Tissue oxygenation state information: Real-time tissue oxygenation state information can be provided, showing the oxygen concentration distribution of the tissue.

[0077] (2) Tumor microenvironment information: Hypoxic regions often exist in tumor tissues, and the hypoxia detection mode can help evaluate the microenvironment of the tumor. The degree of hypoxia is closely related to the invasiveness of the tumor, resistance to treatment, prognosis, etc., which helps doctors understand the biological behavior of the tumor and develop treatment plans.

[0078] (3) Tissue metabolic and functional state information: The hypoxia detection mode can reflect the metabolic and functional state of the tissue, providing important evidence for the study of the physiological and pathological processes of the tissue.

[0079] The frame rate is generally in the range of 25-30 frames per second (fps) to achieve real-time dynamic observation. This frame rate can ensure a smooth view for the doctor during operation. The exposure time is usually in the order of milliseconds (e.g., 1-10 milliseconds), depending on the intensity of the light source and the sensitivity of the photosensitive element. Short exposure time can reduce image blur caused by gastrointestinal peristalsis or hand tremor, but requires a high-sensitivity sensor to maintain picture brightness. The above data is ultimately determined according to the model and hardware parameters of the actual product, but the range can be referred to the above data.

[0080] Data processing: The multispectral images collected are preprocessed using image processing algorithms to remove noise and artifacts. The specific image processing algorithm is a known technology that can achieve the removal of image noise and artifacts. The present application does not elaborate on the image processing process in detail.

[0081] Time series data: The organoid chip disease model is dynamically, label-free, and low-damage monitored for the whole life cycle of hypoxia characteristics, and clinical endoscopy and pathological biopsy are performed at the hypoxia characteristic event. The time series data specifically includes time + parameters, forming a time series, or is recorded in table form for subsequent data processing.

[0082] 32、Feature processing

[0083] The tissue hypoxia feature information is extracted from the endoscopic imaging result of the real-time multispectral image taken by the monochrome camera in the endoscope detection system based on the application; the tissue structure information is extracted from the clinical routine endoscopic imaging result; and the cell morphology and pathological features of histopathology are extracted from the clinical ex vivo tissue pathological examination or histological examination result. Specifically,

[0084] 321. Hypoxia feature extraction: through spectral analysis and feature extraction algorithm, hypoxia feature parameters such as oxygen saturation, hypoxia position, hypoxia range, and hypoxia duration are calculated.

[0085] Hypoxia feature extraction is to establish a model for calculating blood oxygen saturation from pixel values based on the optical properties of biological tissues. The specific blood oxygen saturation calculation algorithm includes the following steps: (1) obtaining the image gray value of blood at different wavebands, which is the real-time image of different wavebands taken by the monochrome camera, (2) analyzing the relationship between image gray value and blood oxygen saturation, (3) calculating blood oxygen saturation through image gray value, and forming a blood oxygen saturation atlas based on the blood oxygen saturation value of each pixel point. According to the calculated blood oxygen saturation atlas, the oxygen saturation value, hypoxia position and hypoxia range are obtained, and the hypoxia duration is recorded.

[0086] The blood oxygen saturation atlas is an image, and the image gray value represents the degree of hypoxia (i.e. blood oxygen saturation value), and the pixel position represents the hypoxia position. The spatial distribution can naturally show the distribution of hypoxia position.

[0087] For example: the blood oxygen saturation is roughly divided into five degrees of hypoxia, 100% is completely non-hypoxic, 75% is mild hypoxia, 50% is moderate hypoxia, 25% is severe hypoxia, and 0% is completely anoxic. It can also be more refined into 10 degrees or 20 degrees, etc. In specific applications, the oxygen saturation corresponding to different disease classifications in the correlation model can be used to determine how to divide.

[0088] 322. Taking digestive tract diseases in this embodiment as an example, the tissue structure information comes from the digestive tract endoscopy, which is used for visual positioning and preliminary evaluation. The surface morphology of the digestive tract mucosa is directly observed by endoscopy, combined with staining, magnification, narrow-band imaging (NBI), etc. The lesion area is visually identified and positioned, and the lesion nature (such as inflammation, polyp, precancerous lesion or malignant tumor) is preliminarily judged, for example, morphological classification judgment based on the degree of lesion elevation / depression, or according to the color and blood vessel structure of the tissue: normal mucosa appears light pink, and the blood vessel texture is regular; the lesion area may show hyperemia, pallor, nodular elevation or disappearance of mucosal blood vessel network (such as atrophic gastritis). Visual identification and positioning of the lesion area by staining, magnification, and narrow-band imaging (NBI) are existing known technologies, and this application will not be described in detail.

[0089] Taking the digestive tract disease in the embodiment as an example, the histopathological cell morphology and pathological features extracted based on the histopathological examination or histological examination results of the organ chip disease model include the following two types:

[0090] Digestive tract disease features: endoscopic mucosal erythema, ulcer degree, pathological inflammation score (such as chronic inflammatory cell infiltration, crypt abscess, etc.), serum inflammation marker; the digestive tract disease features here are the results obtained according to the clinical diagnostic indicators, which belong to the existing known conventional clinical detection technology, and the present application does not make a detailed description here.

[0091] Microbiome data: Helicobacter pylori infection status, intestinal flora diversity (such as Shannon index), and specific pathogenic bacteria abundance (such as Klebsiella pneumoniae); the microbiome data here is the detection result obtained by detecting the omics analysis, and the skilled person in the art should know that the omics analysis here is the existing known technology, and the present application does not make a detailed description here.

[0092] In actual application, other organ chip disease model basic information and the like can be added as needed.

[0093] 33. Multimodal data fusion

[0094] The multiscale information of different modalities such as the tissue structure information, the tissue hypoxia feature information, and the histopathological cell morphology and pathological features is subjected to spatiotemporal alignment and feature matching and fusion, so that different modalities are time-aligned and space-aligned through spatiotemporal alignment. The hypoxia features corresponding to the key events of pathological diagnosis are marked to form a comprehensive data set. Image registration and feature matching include but are not limited to image morphological features extracted based on morphological theory, including color features, shape features, texture features, edge features, and deep morphological features. Image registration and feature matching are existing known technologies, and the present application does not make a detailed description here.

[0095] The specific process of spatiotemporal alignment and feature matching and fusion of the multiscale information of different modalities is as follows: first, the clinical endoscopy image is subjected to image registration and feature matching with the hypoxia feature information of the multispectral endoscope, and then the histopathological cell morphology and pathological features are marked in either of the two images after image registration and feature matching, to realize multimodal data fusion. The result of fusion is an image in which the hypoxia feature information, the tissue structure information, and the histopathological cell morphology and pathological features are marked.

[0096] Take the endoscopic examination results of digestive tract diseases (i.e. collected real-time images) as an example, the clinical endoscopic examination results are time and space registered with the endoscopic imaging results of the present application, and the oxygen saturation values, hypoxic positions, hypoxic ranges and hypoxic durations obtained by the present application according to the oxygen saturation map are marked on the clinical test results. The basis of this step is the time and space registration, so the same position at the same time can be found to correspond to different test results.

[0097] Before fusion, it also includes dividing the clinical endoscopic examination images and the hypoxic feature information of the multispectral endoscope into regions of interest to obtain the test results at each position on the image. The division of the region of interest belongs to the existing known technology, and the present application does not elaborate on it.

[0098] Before fusion, it also includes numerical processing of the extracted features, mainly involving numerical processing of cell morphology and pathological features of histopathology. The specific numerical processing process is the existing known technology, which can include the following numerical processing methods:

[0099] Feature standardization: Z-score standardization for continuous variables (such as hypoxic degree, inflammatory factor concentration); category variable encoding: such as Helicobacter pylori infection status (positive / negative) converted to one-hot encoding; oxygen saturation, i.e. hypoxic degree.

[0100] Time series feature extraction: calculate the sliding window statistics (mean, variance, maximum decline) of the long-term monitored hypoxic indicators (such as oxygen saturation value);

[0101] The present embodiment also includes feature screening: screening of hypoxic features strongly related to disease progression by LASSO regression or mutual information method (Mutual Information), such as the correlation of hypoxic time and hypoxic range with gastric mucosal injury in digestive tract diseases.

[0102] 4. Model construction and training

[0103] 41. Association model

[0104] Long-term, multi-time point synchronous traditional pathological examination or histological examination of different organ disease models throughout the life cycle is used as an evaluation standard, and an algorithm is used to fit the mapping relationship between hypoxic features and disease occurrence, development and outcome, to construct an association model. Specifically:

[0105] Machine learning, deep learning and other algorithms are used to train and learn the fused data to construct an association model between hypoxic features and the occurrence, development and outcome of chronic inflammatory diseases of the digestive tract.

[0106] The specific steps are as follows:

[0107] Model construction: Select appropriate machine learning, deep learning, etc. Algorithm, build correlation model.

[0108] Model training: Use comprehensive data set to train the model, get the classifier, and optimize the parameter activation of the classifier. Input the fusion information feature into the activated and optimized classifier to get the analysis result. The comprehensive data set is divided into training set, test set and validation set. First, use the training set to get the correlation model, then use the correlation model to test the test set. Here, the parameter activation optimization is to continue training and testing the training set and the test set to improve the performance of the model. Finally, the performance of the model is verified by the validation set.

[0109] Model evaluation: Use independent data set to evaluate the model, verify the accuracy and reliability of the model.

[0110] The above machine learning includes but is not limited to one or more traditional machine learning methods such as support vector machine, decision tree, random forest, logistic regression, principal component analysis, K nearest neighbor, Bayesian method; Deep learning methods include but are not limited to one or more deep learning methods such as convolutional neural network, recurrent neural network, natural language processing, long short term memory network, autoencoder, generative adversarial network, clustering, self attention mechanism, reinforcement learning, transfer learning.

[0111] 42, Parameter and threshold

[0112] Select the key parameters of the model, determine the corresponding relationship between the parameter threshold and the clinical pathological diagnosis. Organize the comprehensive data set formed by cross-scale and multi-modal information, and develop the corresponding intelligent diagnosis algorithm. The specific steps are as follows:

[0113] Parameter selection: according to the analysis result of step 41 model training, select the key parameters closely related to disease diagnosis.

[0114] Threshold determination: according to the analysis result of step 41 model training, determine the threshold of the parameter, and correspond with the clinical pathological diagnosis result.

[0115] The following takes random forest (Random Forest) algorithm as an example to explain how to build the correlation model between hypoxia features and the occurrence, development and outcome of chronic inflammatory diseases in digestive tract (such as Helicobacter pylori gastritis, inflammatory bowel disease, etc.):

[0116] Algorithm selection: Random forest, advantages: (1) processing high-dimensional, multi-source data; (2) capturing nonlinear relationship (such as threshold effect of hypoxia and inflammatory factors); (3) analyzing key driving factors through feature importance ranking.

[0117] Model design of this embodiment:

[0118] (1) Target variable:

[0119] Classification task: Predicting the risk of disease progression (e.g., low / medium / high risk);

[0120] Regression task: Predicting the inflammation score (e.g., pathological activity index) or the time to outcome (e.g., ulcer healing period).

[0121] (2) Hyperparameter optimization: Adjusting parameters using Grid Search, for example:

[0122] Number of decision trees (n_estimators): 100-500;

[0123] Maximum depth (max_depth): 5-20;

[0124] Minimum number of leaf samples (min_samples_leaf): 1-5.

[0125] (3) Training strategy

[0126] Stratified sampling: Stratified division of training set, test set, and validation set according to disease stage (early / advanced / late);

[0127] Handling imbalanced data: Using SMOTE oversampling or adjusting class weights for rare events (e.g., cancer);

[0128] Cross-validation: 5-fold cross-validation to evaluate model stability.

[0129] In summary, the correlation model constructed by random forest can integrate multi-dimensional data such as hypoxia features and biopsy pathology, analyze the dynamic mechanism of hypoxia in chronic inflammation of the digestive tract, and provide whole-chain decision support from risk warning to disease diagnosis and individualized treatment.

[0130] Those skilled in the art will realize that the embodiments described herein are for the purpose of helping the reader understand the principles of the present application and should be understood as not limiting the scope of protection of the present application to such specific statements and embodiments. The present application can have various modifications and changes for those skilled in the art. Any modification, equivalent replacement, improvement, etc. within the spirit and principles of the present application should be included in the scope of protection of the claims of the present application.

Claims

1. A method for detecting inflammatory diseases based on endoscopic monitoring of hypoxia characteristics in organoid chip disease models, characterized in that: include: S1. Constructing an in vitro organ-on-a-chip model, the in vitro organ-on-a-chip model comprising: a microfluidic chip, microchambers disposed within the microfluidic chip for simulating different tissues or organs, the microchambers being connected by microchannels; cells seeded in the microchambers, and corresponding culture matrices; the in vitro organ-on-a-chip model further comprising a perfusion inflow channel and a perfusion outflow channel connected to a peristaltic pump; S2. Build an endoscopic detection system, comprising: a laparoscope, a multispectral light source, and a monochrome camera, wherein the multispectral light source is connected to the laparoscope via an optical fiber cable; the monochrome camera is used to capture real-time images under the laparoscope; and the monochrome camera is also connected to a computer; S3, data collection and feature processing of the collected data; The collected data specifically includes: a first endoscopic imaging result collected from the in vitro organ-on-a-chip model constructed in step S1 by the endoscopic detection system constructed in step S2, a second endoscopic imaging result collected from the in vitro organ-on-a-chip model constructed in step S1 by clinical endoscopy, and results obtained from clinical ex vivo tissue pathological examination or histological examination of the in vitro organ-on-a-chip model constructed in step S1; Feature processing specifically includes: extracting tissue hypoxia characteristic information from the first endoscopic imaging result; extracting tissue structure information from the second endoscopic imaging result; Extracting histopathological cell morphology and pathological characteristics information from the results of clinical in vitro tissue pathology or histological examination; Based on the same in vitro organoid chip model constructed in step S1, tissue hypoxia characteristic information, tissue structure information, cell morphology and pathological characteristic information are constructed into multimodal data, and a training set is obtained based on the plurality of multimodal data; S4. Constructing an association model. Specifically, a machine learning algorithm is used to learn the multimodal data in the training set, and a mapping relationship between tissue hypoxia feature information and disease course classification results is constructed to establish an association model. S5. For the organoid chip model to be tested, the corresponding multimodal data is input into the association model constructed in step S4 to obtain a classification result.

2. The method for detecting inflammatory diseases based on endoscopic monitoring of hypoxia characteristics of organ-on-a-chip disease models according to claim 1, characterized in that: For the in vitro organoid chip model in step S1, control groups are set according to the site, the disease type corresponding to each site, and the disease stage corresponding to each type.

3. The method for detecting inflammatory diseases based on endoscopic monitoring of hypoxia characteristics of organ-on-a-chip disease models according to claim 2, characterized in that: Step S3 also includes collecting data under different experimental conditions; the different experimental conditions specifically include: hypoxic environment, reoxygenation environment, treatment operation, and surgical resection operation.

4. The method for detecting inflammatory diseases based on endoscopic monitoring of hypoxia characteristics of organ-on-a-chip disease models according to claim 3, characterized in that: The first endoscopic imaging result in step S3 is a real-time image captured by a monochrome camera.

5. The method for detecting inflammatory diseases based on endoscopic monitoring of hypoxia characteristics of organ-on-a-chip disease models according to claim 4, characterized in that: Extracting tissue hypoxia characteristic information from the first endoscopic imaging result in step S3, specifically, the process is as follows: The grayscale values ​​of the blood images in different bands are obtained according to the real-time images of different bands taken by the monochrome camera; Calculate the blood oxygen saturation based on the grayscale value of the image, and form a blood oxygen saturation map based on the blood oxygen saturation value of each pixel; The oxygen saturation value, hypoxia location and hypoxia range are obtained according to the blood oxygen saturation spectrum, and the hypoxia duration is recorded through monitoring.

6. The method for detecting inflammatory diseases based on endoscopic monitoring of hypoxia characteristics of organ-on-a-chip disease models according to claim 5, characterized in that: In step S3, based on the same in vitro organoid chip model constructed in step S1, tissue hypoxia characteristic information, tissue structure information, cell morphology and pathological characteristic information are constructed into multimodal data. The specific implementation process is as follows: A1. Perform image registration and feature matching on the second endoscopic imaging result corresponding to the same in vitro organoid chip model and the tissue hypoxia feature information; A2. marking tissue hypoxia characteristic information, histopathological cell morphology, and pathological features onto the second endoscopic imaging result after image registration and feature matching; A3. After processing according to step A2, the obtained fusion result is: a second endoscopic imaging result marked with tissue hypoxia characteristic information, tissue pathology cell morphology and pathological characteristics.

Images