Image-assisted analysis and diagnosis system for hospital department based on big data learning

Abstract

The invention discloses an image-assisted analysis and diagnosis system for a hospital department based on big data learning, and belongs to the technical field of medical image diagnosis, and the system comprises a big data storage and management module which is used for storing image data and clinical information of a patient in a preset region, and the clinical information comprises the age, gender and medical history of the patient; constructing a database according to the image data and the clinical information; the data pre-processing module responds to the database and is used for pre-processing the image data in the database, including de-noising, normalizing and segmenting the image; through big data learning and deep learning algorithms, automatic analysis and diagnosis suggestion generation of the medical image data are realized, the diagnosis efficiency and accuracy are improved, data management and utilization are optimized, clinical decision support is enhanced, and the medical resource utilization efficiency is improved.

Description

Technical Field

[0001] The present invention relates to the technical field of medical imaging diagnosis, and particularly to an image-assisted analysis and diagnosis system for hospital departments based on big data learning. Background Art

[0002] In recent years, medical imaging diagnosis and analysis based on big data have received increasing widespread attention in the medical field, which provides medical decision-making and personalized treatment plans for doctors and patients. In modern medicine, medical images (such as X-rays, CTs, MRIs, ultrasounds, etc.) are important bases for disease diagnosis.

[0003] Regarding the research on medical image analysis and diagnosis, the application document with the application number CN202410486788.5 provides an adjustable medical assistance diagnosis system and method based on big data. The technical solution includes a database for expanding and supporting the diagnostic reasoning module, a knowledge base management module, and a cycle timing module. The cycle timing module is used to obtain the physical condition information of the patient within a limited time period and use this time period as the cycle for judging the effectiveness of the medical plan; a diagnosis adjustment module. Adopting the technical solution of the present invention, after the doctor confirms the diagnosis and treatment plan, the diagnosis and treatment process is cyclically tracked and verified, so that the assisted diagnosis system is improved to the whole process of treating the patient.

[0004] Another technical application document with the application number CN202310023862.5 provides a diagnostic report auxiliary generation system based on medical image big data. The technical solution classifies the existing case data and related medical images to obtain different symptom sets, and creates index keywords to improve the retrieval efficiency; obtains the patient's medical examination images through an ultrasound device, and performs a complementary drawing operation on the target area according to the prior knowledge of the consistency between the density of the target area and the density of the lesions in the surrounding adjacent areas. The technical solution inputs the keyword of the confirmed symptom prompt information, matches the obtained medical examination images with the existing case medical images for the same points, and thus matches the existing case medical image with the most same points, and infers the diagnosis result of the obtained medical examination images based on the diagnosis results of the existing cases.

[0005] However, the existing diagnosis systems based on big data learning are difficult to identify the subtle abnormal features of the lesion sites in the images and are difficult to label the features of the lesion sites, resulting in deficiencies in the segmentation and comparison of the lesions in the images. Furthermore, the ability to identify unobvious abnormal conditions is insufficient, leading to misdiagnosis and missed diagnosis, and it is difficult to improve the accuracy of diagnosis. Summary of the Invention

[0006] In view of the above problems existing in the current technical field of medical imaging diagnosis, the present invention is proposed.

[0007] Therefore, one of the objectives of the present invention is to provide an image-assisted analysis and diagnosis system for hospital departments based on big data learning. Through big data learning and deep learning algorithms, it realizes the automated analysis of medical image data and the generation of diagnostic suggestions, improves the efficiency and accuracy of diagnosis, optimizes data management and utilization, enhances clinical decision-making support, and improves the utilization efficiency of medical resources.

[0008] To solve the above technical problems, the present invention provides the following technical solutions:

[0009] The present invention provides an image-assisted analysis and diagnosis system for hospital departments based on big data learning, including:

[0010] A big data storage and management module for storing the image data and clinical information of patients in a preset area. The clinical information includes the age, gender, and medical history of the patients; and constructing a database based on the image data and clinical information;

[0011] A data preprocessing module. The data preprocessing module responds to the database and is used for preprocessing the image data in the database, including denoising, normalizing, and segmenting the images;

[0012] An image data acquisition module for acquiring the image data of a preset patient and uploading the image data and the clinical information of the preset patient to the database for association with the image data;

[0013] A big data fusion learning module for using deep learning algorithms to perform feature recognition and / or learning on the image data in the database; the big data fusion learning module includes a feature extraction unit, an acquisition unit, a comparison unit, and an update unit;

[0014] The feature extraction unit is used for extracting the features of the image data in the database, including the organizational structure, the shape, size, and texture of the lesion area;

[0015] The acquisition unit is used for obtaining the diagnostic results corresponding to the image data in the database;

[0016] The comparison unit is used for analyzing and comparing the features of the image data obtained in the future period with the features of the image data in the database, and generating a diagnostic suggestion for the patient matching the obtained image data according to the result of the analysis and comparison; the diagnostic suggestion includes the disease type and the degree of lesion of the patient;

[0017] The update unit is used for periodically updating the database, including introducing an online learning mechanism, and updating when the database receives new image data and diagnostic results.

[0018] As a preferred embodiment of the present invention, the following is provided: According to the diagnosis result, the image data in the database is distinguished. The distinguishing methods include distinguishing by diagnosis category, by case, and / or by chronological order; among them, for the diagnosis category distinction, according to the diagnosis result, the image data is divided into different categories; including dividing the image data into a normal category, a disease A category, and / or a disease B category;

[0019] For the case distinction, according to the diagnosis result, it is distinguished in units of patient cases, and a data set regarding the corresponding patient is constructed; for the chronological order distinction, it is distinguished according to the chronological order of image data acquisition, including using the image data collected early for a patient as the training set, the image data collected in the middle as the validation set, and the image data collected late as the test set.

[0020] As a preferred embodiment of the present invention, the following is provided: According to the case distinction, the number of different cases is counted in the database, and the characteristics of the image data of each case are obtained from the counted different cases, including extracting the shape characteristics of the lesion or organ in one and / or more cases in the database by means of contour detection and / or geometric analysis, and marking the extracted cases as reference cases; the shape characteristics include area, perimeter, and circularity; among different cases, the change rules of the shape characteristics are analyzed according to the image data obtained at different times.

[0021] As a preferred embodiment of the present invention, the following is provided: The image data obtained at different times is divided into θ1, θ2,..., θ n , where θ n represents the image data obtained at the nth time. At least 3 to 5 feature points regarding the lesion are selected from the image pictures corresponding to the divided image data, and the contour characteristics of each feature point are obtained. If the contour of the feature point becomes clear, the system determines that the corresponding feature point is developing in a stable and / or improving trend; otherwise, it is not determined.

[0022] As a preferred embodiment of the present invention, the following is provided: The selected feature points are divided into initial feature points, middle feature points, and late feature points. When new patient image data is collected in the future period, the lesion of the image corresponding to the image data is analyzed and compared with the initial feature points. If the contour characteristics of the lesion are the same as those of the initial feature points, the system determines that the patient and the corresponding case in the reference case are the same case; otherwise, it is not determined.

[0023] As a preferred embodiment of the present invention, it is as follows: Based on the initial feature points, mid-term feature points, and late feature points, the corresponding feature points are labeled with the lesion type and lesion characteristics; wherein, according to the contour characteristics, the lesion characteristics are labeled as blurred contour, irregular contour, absent or rough contour, dilated or displaced contour, lobulated or umbilicated contour, ring enhancement or halo sign of the contour, and serrated or ground-glass change of the contour, and according to the labeling of the lesion characteristics, the corresponding labeling of the lesion type is carried out; when the image data of a new patient is collected in the future period, the contour characteristics of the lesion in the corresponding image of the image data are analyzed, and the contour characteristics are matched with the labeled lesion characteristics. If the match is successful, the system determines that the patient has the lesion type corresponding to the labeled lesion characteristics; if the match fails, no determination is made.

[0024] As a preferred embodiment of the present invention, it is as follows: The labeling of the corresponding feature points with the lesion type and lesion characteristics further includes the labeling based on functional and physiological characteristics. Among them, the labeling of the functional characteristics includes labeling the features related to the organ function for the image data.

[0025] The labeling of the physiological characteristics includes labeling the features related to the physiological state.

[0026] It further includes the labeling based on the time series. The labeling based on the time series includes the dynamic change labeling and the time point labeling. Among them, the dynamic change labeling includes labeling the change of the lesion over time for the image data.

[0027] The time point labeling includes labeling the time point when the image data is collected.

[0028] It further includes the labeling based on multi-modal images. The labeling based on multi-modal images includes the multi-modal fusion labeling. In the multi-modal fusion labeling, for the image data containing multiple modalities, the lesion characteristics in different modalities are labeled, and a fusion analysis is carried out; and according to each labeling, the corresponding labeling of the lesion type is carried out.

[0029] As a preferred embodiment of the present invention, it is as follows: According to the labeling of the lesion characteristics, the proportion of different labelings is calculated among all the labelings. According to the calculated result, the proportions of different labelings are divided step by step into high proportion, medium proportion, and low proportion. When the image data of a new patient is collected in the future period, the lesion characteristics of the image data are matched according to the corresponding division.

[0030] As a preferred embodiment of the present invention, the following is provided: If the system determines that the patient's case is the same as the corresponding case in the reference case, then for the inspection cycle of the patient's given image data, the determination of the inspection cycle includes being given based on the acquisition time of the image data corresponding to the mid-term feature points, calculating the time difference between the acquisition time of the image data corresponding to the mid-term feature points and the acquisition time of the image data corresponding to the initial feature points. During the time difference, the patient is acquired with at least 1 to 2 times of image data. Among the acquired image data, if the lesion of the image corresponding to the image data changes towards the contour feature corresponding to the mid-term feature points, the system maintains the original determination; otherwise, it does not maintain.

[0031] Beneficial effects:

[0032] 1. Through big data learning, the system of the present invention can extract the features of image data and generate diagnostic suggestions, improving the diagnostic efficiency.

[0033] 2. The present invention not only analyzes common features such as the contour, size, and shape of images, but also combines multi-dimensional information such as functional and physiological features, time series changes, and multi-modal images, thereby more comprehensively evaluating lesions. This comprehensive analysis method helps to improve the accuracy of diagnosis and reduce the possibility of misdiagnosis and missed diagnosis.

[0034] 3. The present invention differentiates image data according to the patient's case and constructs a data set for the corresponding patient. This patient-centered data management method enables the system to perform personalized analysis for each patient's specific situation and provides support for personalized medicine.

[0035] 4. The present invention labels the lesion features, including contour features, functional features, physiological features, etc. When new patient image data is input, the system will match the lesion features with the labeled features, thereby quickly giving diagnostic suggestions. This big data-based learning and matching mechanism provides strong support for clinical decision-making.

[0036] 5. The present invention can perform annotation and fusion analysis of multi-modal images, enabling doctors to simultaneously refer to the information of multiple image modalities, thereby more comprehensively evaluating the condition. This multi-modal analysis method not only improves the accuracy of diagnosis, but also reduces the dependence on a single-modal image and improves the utilization efficiency of medical resources. Description of the Drawings

[0037] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the following will briefly introduce the drawings required for the description of the embodiments. Obviously, the following drawings are only some embodiments of the present invention. For those of ordinary skill in the art, other drawings can be obtained based on these drawings without creative efforts.

[0038] Wherein:

[0039] Figure 1 is a schematic modular structure diagram of an image-assisted analysis and diagnosis system for hospital departments based on big data learning according to an embodiment of the present invention;

[0040] Figure 2 is a schematic flow structure diagram according to an embodiment of the present invention;

[0041] Reference numerals in the figure: 110 - big data storage and management module; 120 - data preprocessing module; 130 - image data acquisition module; 140 - big data fusion learning module; 1401 - feature extraction unit; 1402 - acquisition unit; 1403 - comparison unit; 1404 - update unit. Detailed implementation manners

[0042] In order to make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are some, but not all, of the embodiments of the present invention. All other embodiments obtained by those of ordinary skill in the art based on the described embodiments of the present invention fall within the scope of protection of the present invention.

[0043] Since existing diagnostic systems based on big data learning are difficult to identify subtle abnormal features in the lesion sites in images and are difficult to label the features of the lesion sites, there are still deficiencies in the segmentation and comparison of lesions in images. As a result, the ability to identify unobvious abnormal conditions is insufficient, leading to misdiagnosis and missed diagnosis, and it is difficult to improve the accuracy of diagnosis. Based on this, the present invention proposes an image-assisted analysis and diagnosis system for hospital departments based on big data learning, which realizes the automated analysis of medical image data and the generation of diagnostic suggestions through big data learning and deep learning algorithms, improves the efficiency and accuracy of diagnosis, optimizes data management and utilization, enhances clinical decision support, and improves the utilization efficiency of medical resources.

[0044] The following further specifically describes this solution through embodiments and with reference to the accompanying drawings.

[0045] Refer to Figures 1 to 2 , which is an embodiment of the present invention. This embodiment provides an image-assisted analysis and diagnosis system for hospital departments based on big data learning, including:

[0046] A big data storage and management module 110, configured to store image data and clinical information of patients in a preset area, where the clinical information includes the age, gender, and medical history of the patients; and construct a database based on the image data and clinical information;

[0047] In this embodiment, a complete database is constructed, providing a rich data basis for subsequent analysis and diagnosis;

[0048] A data preprocessing module 120, which responds to the database and is used to preprocess the image data in the database, including denoising, normalizing, and segmenting the images;

[0049] In this embodiment, by preprocessing the image data in the database, the quality and analyzability of the images can be improved; it helps to enhance the accuracy and reliability of subsequent analysis;

[0050] An image data acquisition module 130, which is used to acquire the image data of a preset patient and upload the image data and the clinical information of the preset patient to the database for association with the image data;

[0051] In this embodiment, the image data of a preset patient is acquired, and the acquisition method includes connecting to the imaging devices in the hospital (such as X-ray machines, CT scanners, MRI devices, etc.) to acquire the patient's image data;

[0052] A big data fusion learning module 140, which is used to perform feature recognition and / or learning on the image data in the database using deep learning algorithms; the big data fusion learning module 140 includes a feature extraction unit 1401, an acquisition unit 1402, a comparison unit 1403, and an update unit 1404;

[0053] In this embodiment, the deep learning algorithm includes a deep learning algorithm based on a convolutional neural network;

[0054] The feature extraction unit 1401 is used to extract the features of the image data in the database, including the organizational structure, the shape, size, and texture of the lesion area;

[0055] The acquisition unit 1402 is used to obtain the diagnostic results corresponding to the image data in the database;

[0056] The comparison unit 1403 is used to analyze and compare the features of the image data obtained in the future with the features of the image data in the database, and generate a diagnostic suggestion for the patient matching the obtained image data according to the result of the analysis and comparison; the diagnostic suggestion includes the disease type and the degree of lesion of the patient;

[0057] The update unit 1404 is used to update the database regularly, including introducing an online learning mechanism, and when the database receives new image data and diagnostic results, it will be updated;

[0058] In this embodiment, the online learning mechanism means that every time new data is received, its own parameters or strategies are immediately updated without the need to regenerate the database to adapt to the continuous generation of data;

[0059] The big data fusion learning module adopts deep learning algorithms, which can automatically extract the features of image data, achieving automated feature recognition and learning, and generating diagnostic suggestions, thereby improving the efficiency and accuracy of diagnosis;

[0060] The update unit introduces an online learning mechanism, which can update the information in the database in real time, enabling the system to continuously learn and optimize, adapt to new image data and diagnostic results, and maintain the advancement of diagnosis;

[0061] Specifically in this embodiment, according to the diagnostic results, the image data in the database is differentiated, and the differentiation methods include differentiation by diagnostic category, by case, and / or by chronological order; among them,

[0062] Differentiation by diagnostic category: According to the diagnostic results, the image data is differentiated into different categories; including differentiating the image data into a normal category, a disease A category, and / or a disease B category;

[0063] In this embodiment, it is convenient to separately analyze and study the image data of different disease types;

[0064] Differentiation by case: According to the diagnostic results, differentiation is carried out in units of patient cases, and a data set corresponding to the patient is constructed; in this embodiment, this can ensure that all the image data of the same patient is assigned to the same data set, preventing the image data of the same patient from appearing in other data sets at the same time, keeping all the image data of the same case together, and facilitating longitudinal analysis;

[0065] Differentiation by chronological order: According to the chronological order of image data acquisition, it includes using the image data collected early for a patient as the training set, the image data collected in the middle as the validation set, and the image data collected late as the test set;

[0066] In this embodiment, this can be closer to the chronological order in actual clinical applications, can better evaluate the patient based on new image data, and is convenient for analyzing the change of the patient's disease over time;

[0067] The system can differentiate the image data in multiple ways such as by diagnostic category, case, or chronological order according to the diagnostic results, facilitating the analysis and research of different types of data, and meeting different application scenarios and requirements;

[0068] Moreover, through differentiation by diagnostic category, the image data of different diseases can be classified and analyzed, which helps to deeply understand the characteristics and laws of each disease; differentiation by case can analyze with the patient as the center and provide support for personalized medicine; differentiation by chronological order is convenient for studying the dynamic changes of lesions and evaluating the treatment effect;

[0069] In this embodiment, further, according to the case classification, the number of different cases is counted in the database, and the characteristics of the image data of each case are obtained from the counted different cases, including extracting the shape characteristics of lesions or organs in one and / or more cases in the database by means of contour detection and / or geometric analysis, and marking the extracted cases as reference cases; the shape characteristics include area, perimeter, and circularity; in different cases, the variation law of the shape characteristics is analyzed according to the image data obtained at different times;

[0070] In this embodiment, the shape characteristics are of great significance for the judgment of the benign and malignant nature of tumors;

[0071] It also includes using methods such as gray-level co-occurrence matrix (GLCM) and local binary pattern (LBP) to extract the texture characteristics of the image data; these characteristics can reflect the subtle structural changes of tissues and are applicable to lesion detection and tissue classification;

[0072] It also includes calculating statistical characteristics such as the average intensity, standard deviation, and skewness of the image data to reflect the density and uniformity of the tissue. It also includes using pre-trained CNN models (such as ResNet, VGG, etc.) to extract high-level semantic features of the images. These features can capture complex image patterns and are applicable to disease classification and diagnosis;

[0073] It also includes extracting features through self-supervised learning methods (such as contrast learning). This method does not require a large amount of labeled data and can better adapt to the feature changes of different cases;

[0074] It also includes obtaining features by means of principal component analysis (PCA). Through PCA, the high-dimensional feature space is reduced to a low-dimensional space while retaining the main variation information. PCA can help remove feature redundancy and improve the analysis efficiency;

[0075] Further, the system can count the number of different cases and extract the shape characteristics of the image data of each case, such as area, perimeter, circularity, etc., providing detailed feature information for subsequent analysis;

[0076] By analyzing the variation law of the shape characteristics of the image data obtained at different times, the development trend of the lesion can be better understood, providing an important reference for clinical diagnosis and treatment;

[0077] On the above basis, the image data obtained at different times are divided into θ1, θ2,..., θ n , where θ nDenote the imaging data obtained at the nth time. Select at least 3 to 5 feature points regarding the lesion from the imaging pictures corresponding to the divided imaging data, and obtain the contour features of each feature point. If the contour of the feature point becomes clear, the system determines that the corresponding feature point shows a stable and / or improving trend; otherwise, it does not determine.

[0078] In this embodiment, a clear contour usually indicates the stability or improvement of the lesion. For example, after the treatment of an inflammatory lesion, the boundary between the lesion and the surrounding normal tissue will become clearer.

[0079] On the contrary, if the contour shows missing or becomes rough, it usually indicates the invasive destruction of the lesion to the surrounding tissues. For example, when a malignant tumor infiltrates into the surrounding tissues, the interface between the normal tissue and the lesion will disappear.

[0080] The system selects the feature points in the imaging data at different times, analyzes the changes in their contour features, can judge whether the lesion is stable or improving, provides intuitive information on the development trend of the lesion for doctors, and helps to adjust the treatment plan in a timely manner.

[0081] At the same time, through the quantitative analysis of the contour features of the feature points, the evaluation of the lesion is made more objective and accurate, reducing the subjectivity of human judgment.

[0082] Furthermore, the selected feature points are divided into initial feature points, middle-stage feature points, and late-stage feature points. When the imaging data of a new patient is collected in the future time period, the lesion in the imaging picture corresponding to the imaging data is analyzed and compared with the initial feature points. If the contour features of the lesion and the initial feature points are the same, the system determines that the patient and the corresponding case in the reference case are the same case; otherwise, it does not determine.

[0083] In this embodiment, it provides a diagnostic reference based on historical cases for doctors, improving the accuracy and reliability of the diagnosis. At the same time, by comparing and matching the existing case data, it gives full play to the advantages of big data, enabling the system to learn and draw on from a large number of historical cases and providing strong support for the diagnosis of new cases.

[0084] It should be emphasized in this embodiment that according to the initial feature points, middle - stage feature points, and late - stage feature points, the corresponding feature points are labeled with the lesion type and lesion characteristics; among them, according to the contour characteristics, including labeling the lesion characteristics as blurred contour, irregular contour, absent or rough contour, dilated or displaced contour, lobulated or umbilicated contour, ring - enhanced or halo sign contour, and serrated or ground - glass - like change of the contour, and according to the labeling of the lesion characteristics, the corresponding labeling of the lesion type is carried out; when the image data of a new patient is collected in the future, the contour characteristics of the lesion in the corresponding image are analyzed, and the contour characteristics are matched with the labeled lesion characteristics. If the match is successful, the system determines that the patient has the lesion type corresponding to the labeled lesion characteristics; if the match fails, no determination is made.

[0085] In this embodiment, according to the labeling of the lesion characteristics, the corresponding labeling of the lesion type is carried out, including labeling the lesion type as tumor, inflammation, fibrosis, etc. according to the labeling of the lesion characteristics. For example, a blurred contour usually indicates an inflammatory reaction or edema around the lesion. In pulmonary images, inflammatory exudation will cause the boundary between the lesion and the surrounding normal tissue to become blurred, which is common in infectious diseases (such as pneumonia), early neoplastic lesions or inflammatory lesions. This change may indicate that the lesion is in the active or progressive stage and requires further observation and treatment.

[0086] An irregular contour may indicate invasive growth or fibrosis of the lesion. For example, in chronic inflammatory lesions of the lung, fibrosis around the bronchi and blood vessels will cause the contour to be serrated; it is common in malignant tumors (such as lung cancer), chronic inflammatory diseases (such as pulmonary fibrosis) or certain infectious diseases (such as tuberculosis). This change may indicate a higher degree of malignancy or a more complex condition of the lesion, and further pathological examination and treatment are required.

[0087] A dilated or displaced contour usually indicates compressive growth of the lesion on the surrounding tissues. For example, a benign tumor or a tumor - like lesion will compress the surrounding tissues, resulting in a dilated or displaced contour; it is common in benign tumors (such as pulmonary hamartoma), certain inflammatory lesions or cystic lesions. This change may indicate a slower growth rate of the lesion, but it still requires regular follow - up to observe its changes.

[0088] At the same time, the labeled data is classified and stored according to the lesion type for subsequent query and analysis.

[0089] The system has detailedly labeled the lesion characteristics, including various situations such as blurred, irregular, absent or rough contours, and corresponding labeling of the lesion type according to the labeling, enriching the information of the image data and providing a more detailed basis for subsequent diagnosis and research.

[0090] When new patient imaging data is input, the system can match the lesion features with the labeled features, quickly give diagnostic suggestions, improve the efficiency and accuracy of diagnosis, and reduce the workload of doctors;

[0091] Furthermore, the corresponding feature points are labeled with lesion types and lesion features, and it also includes labeling based on functional and physiological features. Among them, the labeling of functional features includes labeling the features related to organ functions in the imaging data;

[0092] In this embodiment, for example, in cardiac imaging, the systolic function of the myocardium, the movement of heart valves, etc. can be labeled; the labeling of physiological features includes labeling the features related to physiological states;

[0093] In this embodiment, for example, in PET imaging, the metabolic activity of the lesion area can be labeled;

[0094] It also includes labeling based on time series. The labeling based on time series includes dynamic change labeling and time point labeling. Among them, dynamic change labeling includes labeling the change of the lesion over time in the imaging data;

[0095] In this embodiment, the change of the lesion over time can be labeled, such as the enlargement, shrinkage, morphological change of the lesion, etc. For example, in the follow-up of tumor patients, the volume change of the tumor is labeled;

[0096] Time point labeling includes labeling the time point when the imaging data is collected;

[0097] In this embodiment, it is convenient to analyze the progression speed of the lesion;

[0098] It also includes labeling based on multi-modal imaging. The labeling based on multi-modal imaging includes multi-modal fusion labeling. In multi-modal fusion labeling, for imaging data containing multiple modalities, the lesion features in different modalities are labeled and fused for analysis; and according to each label, the corresponding label of the lesion type is carried out;

[0099] In this embodiment, the imaging data of multiple modalities includes CT, MRI, PET, etc.; for example, in PET / CT imaging, the anatomical structure in the CT image and the metabolic situation in the PET image are labeled;

[0100] In addition to contour features, the system also adds labeling based on functional and physiological features, as well as labeling based on time series and multi-modal imaging, making the labeling information more comprehensive and rich, and being able to reflect the features and changes of the lesion from multiple angles;

[0101] And the multi-dimensional labeling information provides more-dimensional data for the comprehensive analysis of the lesion, which helps doctors understand the condition more comprehensively and make more accurate diagnostic and treatment decisions;

[0102] On the above basis, according to the annotation of lesion characteristics, calculate the proportion of different annotations among all the annotations. According to the calculation results, divide the proportions of different annotations step by step into high proportion, medium proportion and low proportion. When the image data of a new patient is collected in the future, match the lesion characteristics of the image data according to the corresponding division; in this embodiment, the system makes a step-by-step division according to the proportion of different annotations and matches the lesion characteristics according to the corresponding division during the matching process, which can more reasonably evaluate the importance of different characteristics and improve the reliability of the diagnostic suggestions; at the same time, by considering the proportion of different annotations, the system can more accurately match the lesion characteristics, avoid misjudgment caused by accidental similarity of some characteristics, and improve the accuracy of the diagnosis;

[0103] Further, if the system determines that the patient and the corresponding case in the reference case are the same case, then for the inspection cycle of the given image data of the patient, the given inspection cycle includes being given based on the acquisition time of the image data corresponding to the mid-term feature points, calculate the time difference between the acquisition time of the image data corresponding to the mid-term feature points and the acquisition time of the image data corresponding to the initial feature points. During the time difference, obtain the image data of the patient at least 1 to 2 times. Among the obtained image data, if the lesion of the image corresponding to the image data changes towards the contour feature corresponding to the mid-term feature points, the system maintains the original determination; otherwise, it does not maintain.

[0104] In this embodiment, after the system determines that the patient and the reference case are the same case, it will give the inspection cycle of the image data according to the acquisition time of the image data corresponding to the mid-term feature points and perform dynamic monitoring within the cycle. If the lesion changes in the expected direction, the original determination is maintained; otherwise, it is not maintained. This dynamic monitoring mechanism can timely detect the changes in the condition and ensure the accuracy and timeliness of the diagnosis;

[0105] Moreover, by obtaining image data and analyzing the lesion changes within the inspection cycle, the system can evaluate the treatment effect, provide a basis for doctors to adjust the treatment plan, and help improve the treatment effect and the prognosis of the patient.

[0106] In summary, through big data learning and deep learning algorithms, the present invention realizes the automated analysis of medical image data and the generation of diagnostic suggestions, improves the efficiency and accuracy of the diagnosis, optimizes data management and utilization, enhances clinical decision support, and improves the utilization efficiency of medical resources.

[0107] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention rather than to limit them. Although the present invention has been described in detail with reference to the preferred embodiments, those of ordinary skill in the art should understand that the technical solutions of the present invention can be modified or equivalently replaced without departing from the spirit and scope of the technical solutions of the present invention, and all of them should be covered by the scope of the claims of the present invention.

Claims

1. An image-assisted analysis and diagnosis system for hospital departments based on big data learning, characterized in that, Including: A big data storage and management module for storing the imaging data and clinical information of patients in a preset area, where the clinical information includes the age, gender, and medical history of the patients; And constructing a database based on the imaging data and clinical information; A data preprocessing module, which responds to the database and is used for preprocessing the imaging data in the database, including denoising, normalizing, and segmenting the images; An imaging data acquisition module for acquiring the imaging data of a preset patient and uploading the imaging data and the clinical information of the preset patient to the database for association with the imaging data; A big data fusion learning module for performing feature recognition and / or learning on the imaging data in the database using deep learning algorithms; the big data fusion learning module includes a feature extraction unit, an acquisition unit, a comparison unit, and an update unit; the feature extraction unit is used for extracting the features of the imaging data in the database, including the organizational structure, the shape, size, and texture of the lesion area; The acquisition unit is used for obtaining the diagnostic results corresponding to the imaging data in the database; The comparison unit is used for analyzing and comparing the features of the imaging data obtained in a future period with the features of the imaging data in the database, and generating a diagnostic suggestion for the patient matching the obtained imaging data according to the results of the analysis and comparison; the diagnostic suggestion includes the disease type and the degree of lesion of the patient; The update unit is used for periodically updating the database, including introducing an online learning mechanism, and when the database receives new imaging data and diagnostic results, it will be updated.

2. The image-assisted analysis and diagnosis system for hospital departments based on big data learning according to claim 1, characterized in that, According to the diagnostic results, the imaging data in the database is distinguished, and the distinguishing methods include distinguishing by diagnostic category, by case, and / or by chronological order; among them, For the diagnostic category distinction, according to the diagnostic results, the imaging data is divided into different categories; including dividing the imaging data into a normal category, a disease A category, and / or a disease B category; For the case distinction, according to the diagnostic results, it is distinguished by patient case, and a data set about the corresponding patient is constructed; For the chronological order distinction, it is distinguished according to the chronological order of the acquisition of the imaging data, including using the imaging data acquired early for a patient as the training set, the imaging data acquired in the middle as the validation set, and the imaging data acquired late as the test set.

3. The image-assisted analysis and diagnosis system for hospital departments based on big data learning according to claim 2, wherein According to the case distinction, the number of different cases is counted in the database, and the features of the imaging data of each case are obtained from the counted different cases, including extracting the shape features of the lesion or organ in one and / or more cases in the database by means of contour detection and / or geometric analysis, and marking the extracted cases as reference cases; the shape features include area, perimeter, and circularity; in different cases, the change rules of the shape features are analyzed according to the imaging data obtained at different times.

4. The image-assisted analysis and diagnosis system for hospital departments based on big data learning according to claim 3, wherein Divide the image data obtained at different times into θ1, θ2, ..., θ n , where θ n represents the image data obtained at the nth time. Select at least 3 to 5 feature points regarding the lesion from the image pictures corresponding to the divided image data, and obtain the contour features of each of the feature points. If the contour of the feature point becomes clear, the system determines that the corresponding feature point is developing in a stable and / or improved trend; otherwise, it does not make a determination.

5. The image-assisted analysis and diagnosis system for hospital departments based on big data learning according to claim 4, wherein The selected feature points are divided into initial feature points, intermediate feature points, and late feature points. When the image data of a new patient is collected in the future, the lesions in the image corresponding to the image data are analyzed and compared with the initial feature points. If the contours of the lesions are the same as those of the initial feature points, the system determines that the patient and the corresponding case in the reference case are the same case; otherwise, it does not make a determination.

6. The image-assisted analysis and diagnosis system for hospital departments based on big data learning according to claim 5, characterized in that Based on the initial feature points, intermediate feature points, and late feature points, the corresponding feature points are labeled with lesion types and lesion characteristics. Among them, according to the contour characteristics, the lesion characteristics are labeled as blurred contour, irregular contour, missing or rough contour, dilated or displaced contour, lobulated or umbilicated contour, ring-enhanced or halo sign contour, and serrated or ground-glass-like contour change. And according to the labeling of the lesion characteristics, the corresponding lesion types are labeled. When the image data of a new patient is collected in the future, the contour characteristics of the lesions in the image corresponding to the image data are analyzed and matched with the labeled lesion characteristics. If the match is successful, the system determines that the patient has the lesion type corresponding to the labeled lesion characteristics; if the match fails, it does not make a determination.

7. The image-assisted analysis and diagnosis system for hospital departments based on big data learning according to claim 6, characterized in that, The labeling of the corresponding feature points with lesion types and lesion characteristics also includes labeling based on functional and physiological characteristics. Among them, the labeling of functional characteristics includes labeling the features related to organ function in the image data. The labeling of physiological characteristics includes labeling the features related to physiological states. It also includes labeling based on time series. The time series labeling includes dynamic change labeling and time point labeling. Among them, the dynamic change labeling includes labeling the change of lesions over time in the image data. The time point labeling includes labeling the time point when the image data is collected. It also includes labeling based on multi-modal images. The multi-modal image labeling includes multi-modal fusion labeling. In the multi-modal fusion labeling, for the image data containing multiple modalities, the lesion characteristics in different modalities are labeled and fused for analysis. And according to each labeling, the corresponding lesion types are labeled.

8. The image-assisted analysis and diagnosis system for hospital departments based on big data learning according to any one of claims 6 to 7, characterized in that According to the labeling of the lesion characteristics, the proportions of different labelings are calculated among all the labelings. According to the calculated results, the proportions of different labelings are divided step by step into high proportion, medium proportion, and low proportion. When the image data of a new patient is collected in the future, the lesion characteristics of the image data are matched according to the corresponding division.

9. The image-assisted analysis and diagnosis system for hospital departments based on big data learning according to claim 5, wherein If the system determines that the patient and the corresponding case in the reference cases are the same case, then for the inspection cycle of the given image data of the patient, the determination of the inspection cycle includes being determined based on the acquisition time of the image data corresponding to the mid-term feature points, calculating the time difference between the acquisition time of the image data corresponding to the mid-term feature points and the acquisition time of the image data corresponding to the initial feature points, within the time difference, at least 1 to 2 times of image data acquisition are performed on the patient. Among the acquired image data, if the lesion of the image corresponding to the image data changes towards the contour feature corresponding to the mid-term feature points, the system maintains the original determination; otherwise, it does not maintain.

Images