Method and system for detecting OCT (optical coherence tomography) data in multi-mode special disease database

Abstract

The embodiment of the invention relates to the technical field of data processing, and discloses a method and system for detecting OCT data in a multi-modal special disease database, and the method comprises the steps: matching multi-modal feature data related to target OCT image data from the multi-modal special disease database based on the to-be-detected target OCT image data, the multi-modal feature data at least comprises text data, image data and gene data; performing key feature extraction on the target OCT image data to obtain feature data of at least one key feature; the key features are related to pathological features of a to-be-identified target disease; based on the feature data of the key features, performing quantitative analysis on the key features to obtain quantitative data of the key features; and performing type detection on the target OCT image data based on the multi-modal feature data and the quantized data. And the multi-modal feature data and the quantized data are utilized to perform type detection on the target OCT image data, so that the accuracy of the detection result is improved.

Description

Technical Field

[0001] This invention relates to the field of data processing technology, and in particular to a method and system for detecting OCT data in a multimodal disease database. Background Art

[0002] Optical coherence tomography (OCT) was first applied in ophthalmology. In recent years, with the maturity of the technology, it has been gradually applied to the coronary artery and intracranial fields. It has unique advantages in assessing the morphology of plaques in the vascular wall, evaluating the pre- and post-operative effects of stent implantation, and monitoring the follow-up of patients with intracranial arterial stenosis. The principle of OCT is to insert an imaging catheter into the blood vessel and analyze the time delay of the built-in light source reflecting to the vessel wall tissue to convert the internal structural information into a high-resolution image.

[0003] However, diseases often present with multiple pathological features, and different diseases may share the same pathological features. Therefore, relying solely on OCT images to detect disease types often results in the detection of multiple diseases, leading to inaccurate disease detection results and affecting the accuracy of the final diagnosis and treatment. Summary of the Invention

[0004] The purpose of this invention is to provide at least one OCT data detection method and system in a multimodal disease database, which can at least solve the problems of inaccurate disease detection results and poor user experience, and can at least improve the accuracy of disease detection results and improve the user experience.

[0005] To address the aforementioned technical problems, at least one embodiment of this application provides a method for OCT data detection in a multimodal disease database, comprising: matching multimodal feature data related to the target OCT image data from a multimodal disease database based on the target OCT image data to be detected, wherein the multimodal feature data includes at least text data, image data, and gene data; extracting key features from the target OCT image data to obtain feature data of at least one key feature; wherein the key feature is related to the pathological features of the target disease to be identified; performing quantitative analysis on the key feature based on the feature data of the key feature to obtain quantitative data of the key feature; and performing type detection on the target OCT image data based on the multimodal feature data and the quantitative data.

[0006] At least one embodiment of this application also provides an OCT data detection system in a multimodal disease database. The system is used to: match multimodal feature data related to the target OCT image data from a multimodal disease database based on the target OCT image data to be detected, wherein the multimodal feature data includes at least text data, image data, and gene data; extract key features from the target OCT image data to obtain feature data of at least one key feature; the key feature is related to the pathological features of the target disease to be identified; perform quantitative analysis on the key feature based on the feature data of the key feature to obtain quantitative data of the key feature; and perform type detection on the target OCT image data based on the multimodal feature data and the quantitative data.

[0007] The OCT data detection method provided in the embodiments of this application for a multimodal disease database utilizes multimodal feature data and quantified data to perform type detection on target OCT image data, thereby improving the accuracy of the detection results. Specifically, it leverages the accuracy and scientific rigor of genetic data, the multi-faceted nature of various image data, and the accuracy of text data, combined with quantified data, to enhance the accuracy of the detection results. Accurate detection results improve the user experience.

[0008] In some optional embodiments, matching multimodal feature data related to the target OCT from a multimodal disease database includes: obtaining an encrypted function image of the target OCT image data when it is stored in the multimodal disease database; and obtaining multimodal feature data associated with the target OCT image data based on the encrypted function image. Obtaining multimodal feature data through an encrypted function image improves data security and supports data privacy protection.

[0009] In some optional embodiments, key features are extracted from the target OCT image data to obtain feature data of at least one key feature, including: acquiring the pathological features of the target disease to be identified and its corresponding OCT image feature type; and extracting features from the target OCT image data based on each OCT image feature type to obtain feature data corresponding to each OCT image feature type. Utilizing OCT image feature types allows for precise extraction of the vital signs from the feature data, improving the accuracy of feature data acquisition.

[0010] In some optional embodiments, feature extraction is performed on the target OCT image data to obtain feature data corresponding to each OCT image feature type. This includes: inputting the target OCT image data into an OCT feature extraction model to extract features from the target OCT image data using the OCT feature extraction model; wherein, the OCT feature extraction model is trained using labeled data from a multimodal disease database. The labeled data is obtained by doctors annotating the pathological features of the target disease when storing multimodal data in the multimodal disease database. Using labeled data from the multimodal disease database to train the OCT feature extraction model improves the accuracy of the OCT feature extraction model, thereby improving the accuracy of the OCT feature extraction model in extracting features from the target OCT image data.

[0011] In some optional embodiments, the annotation methods provided in the multimodal disease database correspond to the way pathological features are displayed in OCT image data. Annotated pathological features using these annotation methods reduce misidentification of pathological features and improve the readability of OCT image data.

[0012] In some optional embodiments, based on the feature data of key features, quantitative analysis is performed on the key features to obtain quantitative data of the key features, including: acquiring the quantitative index corresponding to the key feature; determining reference data corresponding to the feature data of the key feature based on the quantitative index corresponding to the key feature; and determining the quantitative data of the key feature based on the key feature and the reference data. By determining the quantitative data, the key features become calculable, measurable, and comparable, thereby improving the efficiency and accuracy of detection when performing detection based on the quantitative data.

[0013] In some optional embodiments, if the target disease to be identified is intracranial atherosclerotic stenosis, key features include at least one of a lipid core, calcified plaques, fibrous cap, macrophage infiltration, microchannels, and cholesterol crystals. Using multiple key features improves the accuracy of detection for the target disease and enhances the accuracy of the detection results.

[0014] In some optional embodiments, if the key feature is a lipid core, the quantification indicators are the number of lipid pixels and the total number of plaque pixels. Based on the quantification indicators corresponding to the key feature, reference data corresponding to the feature data of the key feature is determined, including: obtaining the lipid pixel value corresponding to the number of lipid pixels and the total number of plaque pixels corresponding to the total number of plaque pixels from the feature data of the lipid core; based on the key feature and the reference data, the quantification data of the key feature is determined, including: obtaining the quantification formula corresponding to the lipid core; and calculating the quantification data of the lipid core based on the quantification formula, the lipid pixel value, and the total number of plaque pixels. Quantifying the feature data corresponding to the key feature based on the quantification indicators and quantification formulas improves the accuracy and scientific nature of obtaining the quantification data.

[0015] In some optional embodiments, type detection is performed on the target OCT image data based on multimodal feature data and quantized data. This includes: fusing multimodal features of the multimodal feature data and quantized data to obtain multi-channel type data; and inputting the multi-channel type data into a trained OCT type detection model to obtain the type detection result corresponding to the target OCT image data. By fusing multimodal features of the multimodal feature data and quantized data, multi-channel type data containing a large amount of information and in a standard format is obtained, improving the readability, persuasiveness, and accuracy of the multi-channel type data. Inputting the multi-channel type data into the OCT type detection model improves the model's processing efficiency for multi-channel type data. Attached Figure Description

[0016] One or more embodiments are illustrated by way of example with reference to the accompanying drawings, and these illustrative descriptions do not constitute a limitation on the embodiments.

[0017] Figure 1 This is a flowchart illustrating an OCT data detection method in a multimodal disease database according to an embodiment of this application;

[0018] Figure 2 This is a schematic diagram illustrating the various data contained in multimodal feature data according to an embodiment of this application;

[0019] Figure 3 This is a schematic diagram of a process for obtaining feature data according to an embodiment of this application;

[0020] Figure 4 This is a schematic diagram of a process for determining quantitative data according to an embodiment of this application;

[0021] Figure 5 This is a block diagram of an OCT data detection system in a multimodal disease database provided in an embodiment of this application. Detailed Implementation

[0022] To make the objectives, technical solutions, and advantages of the embodiments of this application clearer, the various embodiments of this application will be described in detail below with reference to the accompanying drawings. However, those skilled in the art will understand that many technical details have been provided in the various embodiments of this application to help readers better understand this application. However, the technical solutions claimed in this application can be implemented even without these technical details and various changes and modifications based on the following embodiments. The division of the various embodiments below is for the convenience of description and should not constitute any limitation on the specific implementation of this application. The various embodiments can be combined with and referenced by each other without contradiction.

[0023] It should be noted that the acquisition or use of data in the embodiments of this application requires the user's consent. The relevant data can only be obtained after the user's authorization, and the acquisition or use of the data complies with the provisions of relevant laws and regulations.

[0024] To facilitate understanding of the embodiments of this application, the relevant content regarding the OCT data detection method in the multimodal disease database will be introduced first.

[0025] OCT imaging boasts advantages such as high speed and high resolution. Specifically, OCT image capture takes only a few seconds, while traditional CT imaging and MRI angiography require significantly longer times. Furthermore, OCT achieves a resolution of 10 μm by scattering near-infrared light, which is 10 times higher than intravascular ultrasound that reflects sound waves. Additionally, because OCT uses non-ionizing radiation, it is not affected by interference from metallic substances.

[0026] Intracranial atherosclerotic stenosis (ICAS) is a disease caused by abnormal lipid accumulation, leading to thickening and hardening of the arteries and narrowing of the lumen. Because the lipids accumulated in the arterial intima appear as a yellowish, porridge-like substance, it is called atherosclerosis. When it is combined with luminal narrowing, it becomes ICAS. For patients with ICAS, the structure and morphology of the plaque are important factors influencing stroke risk. Therefore, accurately assessing the pathological structure of the plaque can help doctors effectively predict the risk of stroke and take preventative measures in advance. While high-resolution MRI and intravascular ultrasound can detect plaques, their resolution is insufficient for accurate assessment. OCT, with its high resolution and ability to detect plaque structure, has stood out among various angiography techniques and become a powerful tool for accurate plaque assessment.

[0027] Specifically, the pathological structures of plaques characterizing intracranial atherosclerotic stenosis include various types, such as lipid cores (lipid plaques), calcified plaques, fibrous caps (fibrous plaques), macrophage infiltration, microchannels, and cholesterol crystals. Lipid plaques are characterized by blurred edges, high back reflection, and strong attenuation. Calcified plaques are characterized by well-defined, regular low signal or heterogeneous edges. Fiber plaques are characterized by homogeneity, high signal, and weak attenuation. Macrophage infiltration is characterized by highly reflective, strongly attenuated punctate or strip-like structures, often forming radial shadows behind high-signal punctate areas. Microchannels on OCT images are characterized by 50-300 μm in diameter, low signal, and well-defined black cavity-like structures, usually observable in multiple (three or more) continuous sections. Cholesterol crystals on OCT often appear as linear high-signal structures. The above-mentioned plaque pathological structures associated with intracranial atherosclerotic stenosis can all be obtained well using OCT technology. Therefore, based on the target OCT image data, it is possible to reflect whether there is intracranial atherosclerotic stenosis.

[0028] However, OCT is not accurate in determining the composition of new plaques, and the formation of plaques is related to intracranial atherosclerotic stenosis. Therefore, relying solely on OCT images to detect intracranial atherosclerotic stenosis will not yield accurate results.

[0029] To address the aforementioned technical problem of inaccurate detection results, this invention proposes an OCT data detection method in a multimodal disease database. The implementation details of the OCT data detection method in the multimodal disease database of this embodiment are described below. The following implementation details are provided for ease of understanding and are not essential for implementing this solution.

[0030] Example 1:

[0031] The OCT data detection method in the multimodal disease database of this embodiment can be applied to electronic devices with communication, computing, and data storage capabilities. Its specific process can be as follows: Figure 1 As shown, it includes:

[0032] Step 110: Based on the target OCT image data to be detected, match multimodal feature data related to the target OCT image data from the multimodal disease database. The multimodal feature data includes at least text data, image data, and genetic data.

[0033] Specifically, the principle of OCT optical coherence tomography is to insert an imaging catheter into the blood vessel and analyze the time delay of the built-in light source reflecting to the vessel wall tissue to convert the internal structural information into a high-resolution image.

[0034] Specifically, target OCT image data refers to OCT image data of the user to be tested, acquired based on OCT; multimodal disease database refers to a database storing various types of modal data; and text data refers to data information that can be converted into text format, consisting of pathological information, identity information, etc., of the user to be tested.

[0035] Specifically, image data refers to image data obtained from other image detections performed on the user to be detected, including but not limited to X-ray imaging, magnetic resonance imaging (MRI), computed tomography (CT), ultrasound imaging, positron emission tomography (PET), nuclear medicine imaging, and pathological images.

[0036] Specifically, genetic data refers to the genetic data obtained from genetic testing performed on the user, including gene sequencing. The genetic data is integrated into the gene sequencing file. Genetic data can be obtained through the gene sequencing file.

[0037] For example, with Figure 2 For example, a multimodal disease database may include various modal feature data such as clinical text, imaging images, pathological images, and gene sequencing files.

[0038] In some examples, step 110 above, which involves matching multimodal feature data related to the target OCT image data from a multimodal disease database, includes: obtaining an encrypted function image of the target OCT image data when it is stored in the multimodal disease database; and obtaining multimodal feature data associated with the target OCT image data based on the encrypted function image.

[0039] Specifically, an encrypted function image is constructed using the target OCT image data as the x-axis and the ciphertext corresponding to the encrypted function image as the y-axis. Multimodal feature data with a similarity higher than a preset threshold to the encrypted function image is obtained from a multimodal disease database, thus obtaining multimodal feature data associated with the target OCT image data. Therefore, comparing the similarity between image encryption feature values ​​and text encryption feature values, and comparing the similarity of data while it is encrypted, ensures user privacy and improves data security.

[0040] In some examples, before matching multimodal feature data related to the target OCT image data from the multimodal disease database, the method also includes encrypting and storing the target OCT image data and the quantized data corresponding to each key feature in the multimodal disease database, and associating and integrating it with its related multimodal feature data.

[0041] In some examples, before matching multimodal feature data related to the target OCT image data from the multimodal disease database, the method also includes setting permission configurations for the multimodal disease database, assigning different permissions to different departments or doctors to restrict their access to the multimodal disease database. Therefore, based on permission settings, the data security of the multimodal disease database is ensured.

[0042] For example, permissions for multimodal disease databases can be set based on data needs within the healthcare system. For instance, neurological access and intensive care units may have higher access permissions, such as access to all data in the multimodal disease database, including but not limited to image data and genetic data.

[0043] In some examples, step 110 above, which involves matching multimodal feature data related to the target OCT image data from a multimodal disease database, includes: obtaining the unique identity information of the user to be tested; obtaining multimodal feature data associated with the unique identity information from the multimodal disease database based on the unique identity information; and using the multimodal feature data as the multimodal feature data associated with the target OCT image data. Therefore, obtaining multimodal feature data belonging to the user to be tested based on their unique identity information ensures that the obtained multimodal feature data uniquely belongs to the user to be tested, guaranteeing data uniqueness and preventing others from obtaining data that is not their own, thus improving data security.

[0044] Step 120: Extract key features from the target OCT image data to obtain feature data of at least one key feature; the key feature is related to the pathological features of the target disease to be identified.

[0045] Specifically, key features refer to features that can characterize the target disease to be identified.

[0046] In some cases, if the target disease to be identified is intracranial atherosclerotic stenosis, key features include at least one of the following: lipid core, calcified plaques, fibrous cap, macrophage infiltration, microchannels, and cholesterol crystals. Therefore, using multiple key features improves the accuracy of detecting the target disease and enhances the accuracy of the test results.

[0047] Specifically, feature data refers to image region data containing key features in the target OCT image data.

[0048] Specifically, pathological features refer to the characteristic manifestations of the target disease to be identified.

[0049] In some examples, step 120 above involves extracting key features from the target OCT image data to obtain feature data for at least one key feature, including steps 1201-1202, the specific process of which can be as follows: Figure 3 As shown:

[0050] Step 1201: Obtain the pathological features of the target disease to be identified and its corresponding OCT image feature types;

[0051] Specifically, OCT image feature types refer to feature types that correspond one-to-one with pathological features, which are characteristic manifestations related to the target disease.

[0052] For example, if the target disease is intracranial atherosclerotic stenosis, the pathological features include: Feature 1, characterized by blurred edges, high back reflection, and strong attenuation; Feature 2, characterized by clear boundaries and low signal; Feature 3, characterized by homogeneous, high signal, and weak attenuation; Feature 4, characterized by punctate high reflection; Feature 5, characterized by black cavities of 50-300 μm; and Feature 6, characterized by linear high signal. Among these, Feature 1 corresponds to a lipid core type in OCT images, Feature 2 corresponds to a calcified plaque type, Feature 3 corresponds to a fibrous cap type, Feature 4 corresponds to macrophage infiltration type, Feature 5 corresponds to a microchannel type, and Feature 6 corresponds to cholesterol crystallization type.

[0053] Step 1202: Based on each OCT image feature type, feature extraction is performed on the target OCT image data to obtain the feature data corresponding to each OCT image feature type.

[0054] Following the previous example, if the OCT image feature type is lipid core type, then its corresponding feature data is lipid core image region data; if the OCT image feature type is calcified plaque type, then its corresponding feature data is calcified plaque image region data; if the OCT image feature type is fibrous cap type, then its corresponding feature data is fibrous cap image region data; if the OCT image feature type is macrophage infiltration type, then its corresponding feature data is macrophage infiltration image region data; if the OCT image feature type is microchannel type, then its corresponding feature data is microchannel image region data; if the OCT image feature type is cholesterol crystal type, then its corresponding feature data is cholesterol crystal image region data.

[0055] In some examples, step 1202 above involves feature extraction from the target OCT image data to obtain feature data corresponding to each OCT image feature type. This includes: inputting the target OCT image data into an OCT feature extraction model to extract features from the target OCT image data using the OCT feature extraction model; wherein, the OCT feature extraction model is trained using labeled data from a multimodal disease database. The labeled data is obtained by doctors annotating the pathological features of the target disease when storing multimodal data in the multimodal disease database. Therefore, using labeled data from a multimodal disease database to train the OCT feature extraction model improves the accuracy of the OCT feature extraction model, thereby improving the accuracy of the OCT feature extraction model in extracting features from the target OCT image data.

[0056] Specifically, the OCT feature extraction model can extract feature data corresponding to various OCT image feature types from the target OCT image data.

[0057] In some cases, the annotation methods provided by the multimodal disease database correspond to the way pathological features are displayed in OCT image data. This indicates that annotating pathological features through these methods reduces misidentification and improves the readability of OCT image data.

[0058] Specifically, annotation methods include segmentation marking, detection marking, boundary tracking, contour drawing, region painting, single-layer expansion, multi-layer expansion, and intelligent segmentation. Among them, the full-scene intelligent annotation tool supports segmentation marking and detection marking, while the professional annotation tool supports boundary tracking, contour drawing, region painting, single-layer expansion, multi-layer expansion, and intelligent segmentation.

[0059] Specifically, when using professional annotation tools, one can use the tools to achieve automatic annotation, or combine them with manual annotation to achieve semi-automatic and manual annotation.

[0060] Step 130: Based on the feature data of the key features, perform quantitative analysis on the key features to obtain the quantitative data of the key features.

[0061] Specifically, quantifiable data refers to data that can be counted or measured.

[0062] In some examples, step 130 above involves quantitative analysis of the key features based on their feature data to obtain quantitative data for the key features, including steps 1301-1303. The specific process can be as follows: Figure 4 As shown:

[0063] Step 1301: Obtain the quantitative indicators corresponding to the key features;

[0064] Specifically, quantitative indicators refer to the indicators required when quantifying the feature data corresponding to key features.

[0065] Step 1302: Based on the quantitative indicators corresponding to the key features, determine the reference data corresponding to the feature data of the key features;

[0066] Specifically, reference data refers to the numerical values ​​of quantitative indicators contained in the feature data; that is, reference data refers to the specific numerical values ​​of quantitative indicators.

[0067] In some examples, the aforementioned step 130 is performed on the feature data of each key feature to obtain the quantified data of each key feature.

[0068] Step 1303: Based on the key features and reference data, determine the quantitative data of the key features.

[0069] Therefore, by defining quantitative data, key features can be calculated, measured, and compared, thus improving the efficiency and accuracy of detection when based on quantitative data.

[0070] Specifically, the image data of the regions corresponding to the key features are the feature data.

[0071] In some examples, when the key feature is the lipid core, the quantification indicators are the number of lipid pixels and the total number of plaque pixels. In step 1302, based on the quantification indicators corresponding to the key feature, reference data corresponding to the feature data of the key feature is determined, including: obtaining the lipid pixel value corresponding to the number of lipid pixels and the total number of plaque pixels corresponding to the total number of plaque pixels from the feature data of the lipid core. In step 1303, based on the key feature and the reference data, the quantification data of the key feature is determined, including: obtaining the quantification formula corresponding to the lipid core; and calculating the quantification data of the lipid core based on the quantification formula, the lipid pixel value, and the total number of plaque pixels. Therefore, quantifying the feature data corresponding to the key feature based on the quantification indicators and quantification formula improves the accuracy and scientific rigor of the obtained quantification data.

[0072] Specifically, the quantification formula for the lipid core is that the ratio of the number of lipid pixels to the total number of pixels in the plaque is equal to the quantification data of the lipid core. The quantification data of the lipid core is calculated based on the quantification formula, the lipid pixel values, and the total number of pixels in the plaque, including using the ratio of the lipid pixel values ​​to the total number of pixels in the plaque as the quantification data of the lipid core.

[0073] For the methods of obtaining quantitative data for other key features, please refer to the methods of obtaining quantitative data for lipid core. We will not go into too much detail here. The difference between the quantitative data of different key features in the quantitative process is that the quantitative formulas and quantitative indicators corresponding to each key feature are different.

[0074] In one specific embodiment, taking the lipid core as an example of a key feature, the light attenuation coefficient corresponding to each pixel (x, y) in the target OCT image data can be used, and the pixel points with light attenuation coefficients greater than the lipid core attenuation threshold can be used as the pixel features corresponding to the lipid core:

[0075]

[0076] Ω lipid ={(x,y)|μ(x,y)>μ thresh}

[0077] Where μ(x,y) is the light attenuation coefficient at pixel position (x,y), and I(x,y) is the pixel intensity in the target OCT image data, such as the original OCT value. max Let d(y) be the maximum pixel intensity among the 8 adjacent pixels at pixel position (x, y), d(y) be the tissue depth, ∈ be the stability constant, and μ be the maximum pixel intensity. thresh Ω represents the decay threshold corresponding to the lipid core. lipid This is the feature data set corresponding to the lipid core.

[0078] Furthermore, the area ratio of the lipid core can be calculated using the following formula, which is to perform quantitative analysis on the key feature lipid core and obtain quantitative data of the lipid core.

[0079]

[0080] Among them, R lipid The percentage of lipid core area, Ω lipid Ω represents the feature data set corresponding to the lipid core. lumen This is the set of feature data corresponding to the lumen region.

[0081] It should be understood that the feature data set of the lumen region can be obtained by using image recognition models to perform image processing operations such as feature extraction and mask segmentation on the lumen region.

[0082] Step 140: Based on multimodal feature data and quantization data, perform type detection on the target OCT image data.

[0083] In some cases, the type detection is a dual-head type detection method, which performs type detection on target OCT image data based on multimodal feature data and quantized data. This includes: using a dual-head type detection method based on multimodal feature data and quantized data to perform type detection on target OCT image data to determine whether it is a target disease.

[0084] In some examples, step 140 above involves performing type detection on the target OCT image data based on multimodal feature data and quantized data. This includes: fusing multimodal features of the multimodal feature data and quantized data to obtain multi-channel type data; and inputting the multi-channel type data into a trained OCT type detection model to obtain the type detection result corresponding to the target OCT image data. Therefore, using multimodal feature data to perform feature correction on the quantized data to obtain accurate type detection results improves the accuracy of the detection.

[0085] Specifically, the type detection result is used to indicate whether it is the target disease.

[0086] For example, if quantitative data indicates that a certain location is a target disease, and multimodal feature data and quantitative data are fused to obtain multi-channel type data, and the multi-channel type data indicates that the location is not a target disease, then the type detection result is that the location is not a target disease.

[0087] In summary, this solution, based on the target OCT image data to be detected, matches multimodal feature data related to the target OCT image data from a multimodal disease database. This multimodal feature data includes at least text data, image data, and genetic data. Key features are extracted from the target OCT image data to obtain feature data for at least one key feature. These key features are related to the pathological features of the target disease to be identified. Based on the feature data of the key features, quantitative analysis is performed on the key features to obtain quantitative data. Based on the multimodal feature data and the quantitative data, type detection is performed on the target OCT image data. Utilizing multimodal feature data and quantitative data for type detection of the target OCT image data improves the accuracy of the detection results. Specifically, by leveraging the accuracy and scientific validity of genetic data, the multi-faceted nature of various image data, and the accuracy of text data, combined with quantitative data, the accuracy of the detection results is improved. This avoids the significant psychological pressure and poor experience caused to the testee due to inaccurate disease detection results, thus improving the user experience.

[0088] Example 2:

[0089] Another embodiment of this application relates to an OCT data detection system in a multimodal disease database. The implementation details of the OCT data detection system in the multimodal disease database of this embodiment are described in detail below. The following implementation details are provided for ease of understanding and are not necessary for implementing this solution.

[0090] Among them, the OCT data detection system 10 in the multimodal disease database includes:

[0091] The matching module 11 is used to match multimodal feature data related to the target OCT image data from a multimodal disease database based on the target OCT image data to be detected. The multimodal feature data includes at least text data, image data, and gene data.

[0092] Extraction module 12 is used to extract key features from the target OCT image data to obtain feature data of at least one key feature; the key feature is related to the pathological features of the target disease to be identified;

[0093] The quantization module 13 is used to perform quantization analysis on the key features based on the feature data of the key features, and obtain the quantized data of the key features;

[0094] The detection module 14 is used to perform type detection on the target OCT image data based on the multimodal feature data and the quantization data.

[0095] In some embodiments, the matching module 11 is further configured to:

[0096] Obtain the encrypted function image of the target OCT image data when it is stored in the multimodal disease database;

[0097] Based on the encrypted function image, the multimodal feature data associated with the target OCT image data is obtained.

[0098] In some embodiments, the extraction module 12 is further configured to:

[0099] Obtain the pathological features of the target disease to be identified and its corresponding OCT image feature types;

[0100] Based on each of the OCT image feature types, feature extraction is performed on the target OCT image data to obtain the feature data corresponding to each of the OCT image feature types.

[0101] In some embodiments, the extraction module 12 is further configured to:

[0102] The target OCT image data is input into the OCT feature extraction model so that the target OCT image data can be feature extracted by the OCT feature extraction model;

[0103] The OCT feature extraction model is trained using labeled data from the multimodal disease database. The labeled data is obtained by doctors annotating the pathological features of the target disease when the multimodal data is stored in the multimodal disease database.

[0104] In some embodiments, the annotation method provided in the multimodal disease database corresponds to the way the pathological features are displayed in the OCT image data.

[0105] In some embodiments, the quantization module 13 is used for:

[0106] Obtain the quantitative indicators corresponding to the key features;

[0107] Based on the quantitative indicators corresponding to the key features, determine the reference data corresponding to the feature data of the key features;

[0108] Based on the key features and the reference data, the quantitative data of the key features are determined.

[0109] In some embodiments, if the target disease to be identified is intracranial atherosclerotic stenosis, the key features include at least one of a lipid core, calcified plaques, fibrous caps, macrophage infiltration, microchannels, and cholesterol crystals.

[0110] In some embodiments, the quantization module 13 is used for:

[0111] Obtain the lipid pixel value corresponding to the number of lipid pixels and the total pixel value of the plaque corresponding to the total number of plaque pixels from the feature data of the lipid core;

[0112] The process of determining the quantified data of the key features based on the key features and the reference data includes:

[0113] Obtain the quantitative formula corresponding to the lipid core;

[0114] The quantitative data of the lipid core are calculated based on the quantification formula, the lipid pixel value, and the total pixel value of the plaque.

[0115] In some embodiments, the detection module 14 is configured to:

[0116] The multimodal feature data and the quantized data are fused to obtain multi-channel type data;

[0117] The multi-channel type data is input into a trained OCT type detection model to obtain the type detection result corresponding to the target OCT image data.

[0118] It is worth mentioning that the system can implement all the steps proposed in the aforementioned method.

[0119] Those skilled in the art will understand that the above embodiments are specific embodiments for implementing this application, and in practical applications, various changes can be made to them in form and detail without departing from the spirit and scope of this application.

Claims

1. A method for detecting OCT data in a multimodal disease database, characterized in that, include: Based on the target OCT image data to be detected, multimodal feature data related to the target OCT image data is matched from a multimodal disease database. The multimodal feature data includes at least text data, image data, and gene data. Key features are extracted from the target OCT image data to obtain feature data of at least one key feature; the key feature is related to the pathological features of the target disease to be identified. Based on the feature data of the key features, quantitative analysis is performed on the key features to obtain the quantitative data of the key features; Based on the multimodal feature data and the quantization data, type detection is performed on the target OCT image data.

2. The OCT data detection method in a multimodal disease database according to claim 1, characterized in that, The step of matching multimodal feature data related to the target OCT from a multimodal disease database includes: Obtain the encrypted function image of the target OCT image data when it is stored in the multimodal disease database; Based on the encrypted function image, the multimodal feature data associated with the target OCT image data is obtained.

3. The OCT data detection method in a multimodal disease database according to claim 1, characterized in that, The step of extracting key features from the target OCT image data to obtain feature data of at least one key feature includes: Obtain the pathological features of the target disease to be identified and its corresponding OCT image feature types; Based on each of the OCT image feature types, feature extraction is performed on the target OCT image data to obtain the feature data corresponding to each of the OCT image feature types.

4. The OCT data detection method in a multimodal disease database according to claim 3, characterized in that, The step of extracting features from the target OCT image data to obtain feature data corresponding to each type of OCT image feature includes: The target OCT image data is input into the OCT feature extraction model so that the target OCT image data can be feature extracted by the OCT feature extraction model; The OCT feature extraction model is trained using labeled data from the multimodal disease database. The labeled data is obtained by doctors annotating the pathological features of the target disease when the multimodal feature data is stored in the multimodal disease database.

5. The OCT data detection method in a multimodal disease database according to claim 4, characterized in that, The annotation method provided in the multimodal disease database corresponds to the way the pathological features are displayed in the OCT image data.

6. The OCT data detection method in a multimodal disease database according to claim 1, characterized in that, The quantitative analysis of the key features based on the feature data of the key features to obtain the quantitative data of the key features includes: Obtain the quantitative indicators corresponding to the key features; Based on the quantitative indicators corresponding to the key features, determine the reference data corresponding to the feature data of the key features; Based on the key features and the reference data, the quantitative data of the key features are determined.

7. The OCT data detection method in a multimodal disease database according to claim 6, characterized in that, If the target disease to be identified is intracranial atherosclerotic stenosis, then the key features include at least one of lipid core, calcified plaque, fibrous cap, macrophage infiltration, microchannels, and cholesterol crystals.

8. The OCT data detection method in a multimodal disease database according to claim 7, characterized in that, If the key feature is a lipid core, the quantification indicators are the number of lipid pixels and the total number of pixels in the plaque. The step of determining reference data corresponding to the feature data of the key feature based on the quantification indicators corresponding to the key feature includes: Obtain the lipid pixel value corresponding to the number of lipid pixels and the total pixel value of the plaque corresponding to the total number of plaque pixels from the feature data of the lipid core; The process of determining the quantified data of the key features based on the key features and the reference data includes: Obtain the quantitative formula corresponding to the lipid core; The quantitative data of the lipid core are calculated based on the quantification formula, the lipid pixel value, and the total pixel value of the plaque.

9. The OCT data detection method in a multimodal disease database according to claim 1, characterized in that, The step of performing type detection on the target OCT image data based on the multimodal feature data and the quantized data includes: The multimodal feature data and the quantized data are fused to obtain multi-channel type data; The multi-channel type data is input into a trained OCT type detection model to obtain the type detection result corresponding to the target OCT image data.

10. An OCT data detection system for a multimodal disease database, characterized in that, The system includes: The matching module is used to match multimodal feature data related to the target OCT image data from a multimodal disease database based on the target OCT image data to be detected. The multimodal feature data includes at least text data, image data, and gene data. The extraction module is used to extract key features from the target OCT image data to obtain feature data of at least one key feature; the key feature is related to the pathological features of the target disease to be identified. The quantization module is used to perform quantization analysis on the key features based on the feature data of the key features, and obtain the quantized data of the key features; The detection module is used to perform type detection on the target OCT image data based on the multimodal feature data and the quantization data.

Images