CN117542528B - Radiomics-based risk labeling system for hip joint involvement in ankylosing spondylitis - Google Patents

Abstract

The application discloses a ankylosing spondylitis hip joint involvement risk labeling system based on image histology, and relates to the technical field of image histology. The system comprises a clinical information acquisition module, a nuclear magnetic resonance scanning module, an image histology feature extraction module, a hierarchical clustering module, an affected risk giving module and a risk labeling output module, wherein the magnetic resonance image histology features of all AS patients can be extracted firstly, hierarchical clustering analysis is carried out on all patients according to the extraction results to obtain a plurality of patient classes, AS hip joint affected risk labels are given to all the classes according to clinical feature data, then the distance to the center of each class is calculated according to the magnetic resonance image histology features of target patients, and finally the risk label of a certain class corresponding to the minimum distance is used AS a risk labeling result of the target patients and is output, so that the accuracy of the AS hip joint affected risk labeling result can be improved, and effective reference information is provided for doctor decision.

Description

Ankylosing spondylitis hip joint affected risk marking system based on image histology

Technical Field

The invention belongs to the technical field of image histology, and particularly relates to an ankylosing spondylitis hip joint involvement risk labeling system based on image histology.

Background

Ankylosing spondylitis (Ankylosing Spondylitis, AS) is a chronic inflammatory rheumatic disease that involves mainly the spine and peripheral joints, and structural damage and restricted movement of the affected parts may occur with progression of the disease. Hip joints are peripheral joints often involved in AS, and are also a major cause of disability in AS patients. In the last decade, although there has been a deeper understanding of AS and a great progress in therapeutic methods, there has been insufficient understanding of AS hip joint involvement, and there has not been even a uniform definition of AS hip joint involvement.

The currently most widely used scale for evaluating radiological changes in the AS hip joint involvement is the back ankylosing spondylitis radiological index (BASRI-hip), which classifies the AS hip joint involvement into a scale of 0-4, based mainly on the hip joint gap, with above 2 being considered AS hip joint involvement. Although this index is very reliable, it may not be satisfactory in describing the sensitivity of the radiological progression of the hip joint. MRI (Magnetic Resonance Imaging ) of AS hip involvement appears to be of two classes: acute phase manifestations and chronic phase manifestations, wherein MRI acute phase manifestations reflect the early stages of hip inflammation, may help to discover early hip involvement in AS patients. Although MRI has a high sensitivity for finding the involvement of the hip, the high sensitivity of MRI is prone to false positives due to many causes, such AS hip effusion and bone marrow edema, and thus overestimates the proportion of patients with AS that have the involvement of the hip. Image group science (Radiomics) first proposes the concept of image group science in 2012 by Lambin and the like of the netherlands scholars, namely, high-flux characteristics (namely, millions of image characteristic data are extracted at one time) are extracted from medical images, and key information which really plays a role is extracted and stripped from massive information by adopting diversified statistical analysis and data mining methods, so that the method is finally used for auxiliary diagnosis, classification or prediction of diseases.

In view of the fact that there is no widely accepted MRI definition of AS hip involvement so far, the application of image histology technology in AS hip involvement has not been fully studied, so how to provide a technical solution for implementing automatic labeling of AS hip involvement risk of a patient based on image histology technology is a subject of urgent need for research by those skilled in the art.

Disclosure of Invention

The invention aims to provide an image-based ankylosing spondylitis hip joint involvement risk labeling system, which is used for solving the problem that an existing evaluation and labeling method for the hip joint involvement risk of an AS patient does not have a reliable scheme.

In order to achieve the above purpose, the present invention adopts the following technical scheme:

in a first aspect, an image-histology-based ankylosing spondylitis hip joint involvement risk labeling system is provided, which comprises a clinical information acquisition module, a nuclear magnetic resonance scanning module, an image histology feature extraction module, a hierarchical clustering module, an involvement risk giving module and a risk labeling output module;

the clinical information acquisition module is used for acquiring clinical characteristic data of a plurality of AS patients;

the nuclear magnetic resonance scanning module is used for scanning to obtain hip joint magnetic resonance images of the AS patients;

the image histology feature extraction module is in communication connection with the nuclear magnetic resonance scanning module and is used for extracting corresponding magnetic resonance image histology features from corresponding hip joint magnetic resonance images of each AS patient in the plurality of AS patients after the hip joint magnetic resonance images of the plurality of AS patients are read;

the hierarchical clustering module is in communication connection with the image histology feature extraction module and is used for carrying out clustering analysis on the AS patients by adopting an unsupervised hierarchical clustering algorithm according to the magnetic resonance image histology features of the AS patients to obtain K patient classes, wherein K represents a positive integer greater than or equal to 2;

the affected risk giving module is respectively in communication connection with the hierarchical clustering module and the clinical information acquisition module, and is used for accessing the clinical information acquisition module for each patient class in the K patient classes, reading clinical characteristic data of corresponding AS patients, and giving AS hip joint affected risk labels of the corresponding classes according to corresponding data reading results;

the nuclear magnetic resonance scanning module is also used for scanning to obtain a hip joint magnetic resonance image of the target patient;

the image histology feature extraction module is further used for extracting magnetic resonance image histology features of the target patient from the hip magnetic resonance images of the target patient;

the risk labeling output module is respectively in communication connection with the affected risk giving module and the image histology feature extraction module, and is used for calculating the distance from the target patient to the class center of each patient class according to the magnetic resonance image histology feature of the target patient, and then taking the AS hip joint affected risk label of a certain patient class which is in the K patient classes and corresponds to the minimum distance AS an AS hip joint affected risk labeling result of the target patient and outputting the AS hip joint affected risk labeling result.

Based on the above-mentioned summary of the invention, a new scheme for realizing automatic labeling of AS hip joint affected risk of a patient based on an image histology technology and a hierarchical clustering algorithm is provided, namely, the scheme comprises a clinical information acquisition module, a nuclear magnetic resonance scanning module, an image histology feature extraction module, a hierarchical clustering module, an affected risk giving module and a risk labeling output module, and through the cooperation of the modules, the magnetic resonance image histology feature of each AS patient can be firstly extracted, then hierarchical clustering analysis is carried out on all patients according to the extraction result, a plurality of patient classes are obtained, AS hip joint affected risk labels are given to each class according to clinical feature data, then the distance to the center of each class is calculated according to the magnetic resonance image histology feature of a target patient, and finally, the risk label of a class corresponding to the minimum distance is used AS the AS hip joint affected risk labeling result of the target patient and is output, so that the AS hip joint affected risk labeling result is poor in accuracy due to the low specificity performance based on MRI, and the purpose of improving the AS hip joint affected risk labeling result is realized, so that effective reference information is provided for decision making and practical application is facilitated.

In one possible design, the clinical profile data comprises demographic information, AS-related clinical profile information and hip radiological damage information, wherein the demographic information comprises age, sex, body mass index, complications, smoking status, drinking status and/or vaccination status, the AS-related clinical profile comprises onset age, course of illness, medication regimen comprising non-steroidal anti-inflammatory drug therapy, DMARDs medication regimen, disease activity, body function status and/or physical movement status, and the hip radiological damage information comprises joint effusion inspection, bone marrow edema inspection, adnexitis inspection, joint sclerosis inspection, joint erosion inspection, fat lesion inspection and/or joint interstitial stenosis inspection, and the hip radiological damage information comprises hip radiological damage degree score based on the Bath ankylosing spondylitis radiological index.

In one possible design, for each AS patient in the plurality of AS patients, extracting the corresponding magnetic resonance imaging histology features from the corresponding hip magnetic resonance images includes:

for one AS patient in the plurality of AS patients, a corresponding region of interest is intercepted from a corresponding hip joint magnetic resonance image, wherein the region of interest is an image region containing acetabulum, femoral head and constituted hip joint capsule content;

extracting a plurality of T1WI/T2WI sequence features from a region of interest of the certain AS patient;

preprocessing the plurality of T1WI/T2WI sequence features to obtain a plurality of preprocessed features, wherein the preprocessing comprises removing features with zero inter-patient variance, removing features with intra-observer variability and inter-observer variability smaller than a preset threshold and/or eliminating batch effects;

and carrying out data standardization processing on the preprocessed features to obtain a plurality of standardized features which are suitable for distance calculation in a clustering algorithm and correspond to the preprocessed features one by one, and taking the standardized features AS the magnetic resonance image histology features of the AS patient.

In one possible design, for a particular AS patient in the plurality of AS patients, intercepting a corresponding region of interest from a corresponding hip magnetic resonance image includes:

importing the hip joint magnetic resonance image of the certain AS patient into a pre-trained region-of-interest recognition model based on a YOLOv4 target detection algorithm, and outputting to obtain a region-of-interest recognition result;

and according to the identification result of the region of interest, cutting out the region of interest of the AS patient from the hip joint magnetic resonance image of the AS patient, wherein the region of interest refers to an image region containing acetabulum, femoral head and constituted hip joint capsule content.

In one possible design, extracting a plurality of T1WI/T2WI sequence features from the region of interest of the certain AS patient includes:

according to the radiation characteristics defined by the image biomarker standardization initiative, 18 first-order characteristics, 8 shape characteristics and 75 texture characteristics of the T1WI sequence and the T2WI sequence are respectively extracted from the region of interest of the certain AS patient, and different various image filtering algorithms are applied to decompose the original image of the region of interest, so that 1210 high-order characteristics are extracted, and 1412T 1WI/T2WI sequence characteristics are obtained in total.

In one possible design, the batch effect is eliminated during the pretreatment by means of comBat compensation.

In one possible design, the performing data normalization processing on the preprocessed features to obtain normalized features that are suitable for performing distance calculation in a clustering algorithm and are in one-to-one correspondence with the preprocessed features, where the normalizing includes:

and carrying out normalization processing on the preprocessed features by adopting a Z score normalization mode to obtain a plurality of normalization features which are suitable for distance calculation in a clustering algorithm and correspond to the preprocessed features one by one.

In one possible design, the hierarchical clustering algorithm employs an unsupervised clustering algorithm that performs distance computation based on the euclidean distance algorithm and hierarchical clustering based on Ward connection criteria.

In one possible design, the system further comprises a clinical statistics analysis module communicatively connected to the risk of involvement assignment module;

the clinical statistics analysis module is used for carrying out descriptive statistics analysis on each patient according to clinical characteristic data of corresponding AS patients by using an SPSS analysis tool to obtain corresponding clinical explanation.

In one possible design, the system further comprises a data display module in communication with the clinical statistics analysis module;

the data display module is used for displaying the occurrence probability of each clinical feature in the corresponding clinical explanation through a chord chart for each patient class.

The beneficial effect of above-mentioned scheme:

(1) The invention creatively provides a new scheme for realizing automatic labeling of AS hip joint affected risk of a patient based on an image histology technology and a hierarchical clustering algorithm, which comprises a clinical information acquisition module, a nuclear magnetic resonance scanning module, an image histology feature extraction module, a hierarchical clustering module, an affected risk giving module and a risk labeling output module, wherein the magnetic resonance image histology feature of each AS patient can be firstly extracted through the cooperation of the modules, then hierarchical clustering analysis is carried out on all patients according to the extraction results to obtain a plurality of patient classes, AS hip joint affected risk labels are given to each class according to clinical feature data, then distances to the centers of each class are calculated according to the magnetic resonance image histology feature of a target patient, and finally the risk label of a certain class corresponding to the minimum distance is taken AS the AS joint affected risk labeling result of the target patient and is output, so that the problem that the accuracy of the AS joint affected identification result is poor due to low specificity expression of the prior MRI can be avoided, the aim of improving the accuracy of the AS hip joint affected risk labeling result is fulfilled, so that effective reference information is provided for decision making, and popularization and application are convenient;

(2) Clinical interpretation of each patient class can also be obtained through statistical analysis and visualized, and effective reference information is further provided for doctor decision.

Drawings

In order to more clearly illustrate the embodiments of the invention or the technical solutions in the prior art, the drawings that are required in the embodiments or the description of the prior art will be briefly described, it being obvious that the drawings in the following description are only some embodiments of the invention, and that other drawings may be obtained according to these drawings without inventive effort for a person skilled in the art.

Fig. 1 is a schematic structural diagram of an imaging-based ankylosing spondylitis hip joint involvement risk labeling system according to an embodiment of the present application.

Fig. 2 is an exemplary diagram of a tree structure obtained after hierarchical clustering of all patients according to an embodiment of the present application.

Fig. 3 is a chord example graph presenting the probability of occurrence of individual MRI performances for different patient classes, provided in an embodiment of the present application.

Detailed Description

In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the present invention will be briefly described below with reference to the accompanying drawings and the description of the embodiments or the prior art, and it is obvious that the following description of the structure of the drawings is only some embodiments of the present invention, and other drawings can be obtained according to these drawings without inventive effort to a person skilled in the art. It should be noted that the description of these examples is for aiding in understanding the present invention, but is not intended to limit the present invention.

It should be understood that although the terms first and second, etc. may be used herein to describe various objects, these objects should not be limited by these terms. These terms are only used to distinguish one object from another. For example, a first object may be referred to as a second object, and similarly a second object may be referred to as a first object, without departing from the scope of example embodiments of the invention.

It should be understood that for the term "and/or" that may appear herein, it is merely one association relationship that describes an associated object, meaning that there may be three relationships, e.g., a and/or B, may represent: three cases of A alone, B alone or both A and B exist; as another example, A, B and/or C, can represent the presence of any one of A, B and C or any combination thereof; for the term "/and" that may appear herein, which is descriptive of another associative object relationship, it means that there may be two relationships, e.g., a/and B, it may be expressed that: the two cases of A and B exist independently or simultaneously; in addition, for the character "/" that may appear herein, it is generally indicated that the context associated object is an "or" relationship.

Examples

As shown in fig. 1, the imaging-histology-based ankylosing spondylitis hip joint involvement risk labeling system provided in the first aspect of the present embodiment includes, but is not limited to, a clinical information acquisition module, a nuclear magnetic resonance scanning module, an imaging-histology feature extraction module, a hierarchical clustering module, an involvement risk giving module, a risk labeling output module, and the like.

The clinical information acquisition module is used for acquiring clinical characteristic data of a plurality of AS patients. The clinical characteristic data is used for reflecting clinical characteristics of corresponding patients, and particularly comprises, but is not limited to, demographic information, AS related clinical characteristic information, hip joint radiological injury information and the like, wherein the demographic information comprises, but is not limited to, age, sex, body mass index, complications, smoking, drinking and/or vaccination conditions and the like; the AS-related clinical features include, but are not limited to, having a Disease age, a course of Disease, a medication regimen including, but not limited to, having a non-steroidal Anti-inflammatory drug treatment and/or a DMARDs (Disease-Modifying Anti-Rheumatic Drugs) medication treatment, etc., and a hip normal magnetic resonance examination result including, but not limited to, having an effusion examination result (which reflects acute inflammation), a bone marrow edema examination result (which reflects acute inflammation), an attachment point inflammation examination result (which reflects acute inflammation), an arthritic sclerosis examination result (which reflects chronic inflammation), a joint erosion examination result, a lipopathy examination result (which reflects chronic inflammation), and/or a joint interstitial stenosis examination result (which reflects chronic inflammation), etc.; the hip joint radiological injury information includes, but is not limited to, a hip joint structural injury degree score obtained based on the back ankylosing spondylitis radiological index, and the hip joint injury is considered to exist for more than 2 minutes. In addition, the clinical profile data may be routinely read from a database.

The nuclear magnetic resonance scanning module is used for scanning to obtain hip joint magnetic resonance images of the AS patients. The aforementioned hip magnetic resonance image can be obtained by scanning at the time of the corresponding patient's visit, but is not limited to, and is transmitted and stored by DICOM (Digital Imaging and Communications in Medicine, "digital imaging and communication in medicine", which is a standard for digital medical image transmission, display and storage). In addition, the nuclear magnetic resonance scanning module can be realized by adopting an existing nuclear magnetic resonance scanner.

The image histology feature extraction module is in communication connection with the nuclear magnetic resonance scanning module and is used for extracting corresponding magnetic resonance image histology features from corresponding hip joint magnetic resonance images of each AS patient in the AS patients after the hip joint magnetic resonance images of the AS patients are read. The access reading mode can be realized by a DICOM mode. Specifically, for each AS patient in the plurality of AS patients, the corresponding magnetic resonance image histology features are extracted from the corresponding hip magnetic resonance images, including but not limited to the following steps S21 to S24.

S21, aiming at a certain AS patient in the AS patients, a corresponding region of interest is cut out from the corresponding hip joint magnetic resonance image, wherein the region of interest refers to an image region containing acetabulum, femoral head and constituted hip joint capsule content.

In the step S21, the region of interest may be manually drawn by a manual film reading manner, and in order to improve the efficiency and reduce the workload of personnel, the region of interest may also be automatically obtained by using an existing target detection algorithm, that is, preferably, for a certain AS patient in the plurality of AS patients, the region of interest is obtained by cutting out the corresponding region of interest from the corresponding hip magnetic resonance image, including but not limited to the following steps S211 to S212.

S211, importing the hip joint magnetic resonance image of the certain AS patient into a pre-trained region-of-interest recognition model based on a YOLOv4 target detection algorithm, and outputting to obtain a region-of-interest recognition result.

In the step S211, the specific model structure of the YOLOv4 object detection algorithm is composed of three parts, namely a backbone network back, a neck network back and a head network head. The Backbone network Backbone may employ a CSPDarknet53 (CSP representation Cross Stage Partial) network for extracting features. The neck network neg consists of SPP (Spatial Pyramid Pooling block) blocks for adding receptive fields and separating out the most important features and PANet (Path Aggregation Network) networks for ensuring that semantic features are accepted from the higher level layers and fine-grained features are accepted from the lower level layers of the transverse backbone network at the same time. The header network head is detected based on anchor boxes and detects three different sized 13×13, 26×26 and 52×52 feature maps for detecting large to small objects, respectively (here, the large sized feature map contains more information, and thus the 52×52 feature map is used for detecting small objects, and vice versa). The above-mentioned region-of-interest recognition model can be trained by a conventional sample training method, so that after the test image is input, the information such as whether the recognition result of the region-of-interest (if any, a region marking frame will be drawn) and the confidence prediction value of the region-of-interest can be output.

S212, according to the identification result of the region of interest, the region of interest of the AS patient is cut out from the hip joint magnetic resonance image of the AS patient, wherein the region of interest refers to an image region containing acetabulum, femoral head and content of the constituted hip joint capsule.

In the step S212, since the region of interest identification result is a region marker frame drawn when the region of interest is identified, the region marker frame may be used to extract an in-frame image from the hip magnetic resonance image of the certain AS patient AS the region of interest of the certain AS patient.

S22, extracting a plurality of T1WI/T2WI sequence features from the region of interest of the AS patient.

In said step S22, T1WI and T2WI are different sequences of magnetic resonance imaging, because the protons move in the magnetic field, the magnetic signals in different lateral and longitudinal directions are different, forming different weighted images. Specifically, a plurality of T1WI/T2WI sequence features are extracted from the region of interest of the certain AS patient, including but not limited to: according to the radiation characteristics defined by the image biomarker standardization initiative, 18 first-order characteristics, 8 shape characteristics and 75 texture characteristics of the T1WI sequence and the T2WI sequence are respectively extracted from the region of interest of the certain AS patient, and different various image filtering algorithms are applied to decompose the original image of the region of interest, so that 1210 high-order characteristics are extracted, and 1412T 1WI/T2WI sequence characteristics are obtained in total. The foregoing feature extraction process may be accomplished conventionally, particularly by the Pyradiomics package (version 3.0.1) of the Python platform (version 3.7).

S23, preprocessing the T1WI/T2WI sequence features to obtain preprocessed features, wherein the preprocessing includes, but is not limited to, removing the feature that the inter-patient variance is zero, removing the intra-observer variability, removing the feature that the inter-observer variability is smaller than a preset threshold, and/or eliminating batch effects.

In the step S23, the purpose of the preprocessing is to perform data cleansing on the extracted feature data so as to ensure data validity. Specifically, the time for removing the features with the variance of zero between patients is required to be performed after the plurality of T1WI/T2WI sequence features of each AS patient are extracted; the Intra-observer variability and inter-observer variability may be represented by Intra-group correlation numbers (Intra-class Correlation Coefficient, abbreviated ICC, used to evaluate the consistency of discrete ordered, continuous data for multiple measurement methods or multiple raters), with the preset threshold being exemplified by 0.75 to ensure good reproducibility of the remaining features. And to correct for differences in radiological characteristics caused by different scanners or imaging protocols, so-called batch effects, during which the batch effects are preferably eliminated using a combatt compensation approach, which is a practical rearrangement that rearranges image-derived data in a single space discarding batch effects without changing biological information (which has been successfully validated and implemented in previous studies involving MRI radiological characteristics); in detail, the comBat compensation mode can be implemented by using a "comBat" package (version 0.3.2) on the Python platform. In addition, the radiology feature may also be calculated in advance using a fixed bin size, for image discretization and at the same voxel size (1 x 1mm ³ ) All images are resampled to reduce the heterogeneity of different acquisition protocols or different scanners.

S24, carrying out data standardization processing on the preprocessed features to obtain a plurality of standardized features which are suitable for distance calculation in a clustering algorithm and correspond to the preprocessed features one by one, and taking the standardized features AS magnetic resonance image histology features of the AS patient.

In the step S24, specifically, the data normalization processing is performed on the plurality of preprocessed features to obtain a plurality of normalized features that are suitable for performing distance calculation in a clustering algorithm and are in one-to-one correspondence with the plurality of preprocessed features, including but not limited to: and carrying out normalization processing on the preprocessed features by adopting a Z score normalization mode to obtain a plurality of normalization features which are suitable for distance calculation in a clustering algorithm and correspond to the preprocessed features one by one. The aforementioned Z-Score, also called Standard Score, is a number-to-average difference divided by Standard deviation.

The hierarchical clustering module is in communication connection with the image histology feature extraction module and is used for carrying out clustering analysis on the AS patients by adopting an unsupervised hierarchical clustering algorithm according to the magnetic resonance image histology features of the AS patients to obtain K patient classes, wherein K represents a positive integer greater than or equal to 2. The hierarchical clustering algorithm is one of clustering algorithms, and creates a hierarchical nested cluster tree by calculating the similarity or distance between data points of different categories; the order of hierarchical decomposition can be divided into a Bottom-up and a Top-down, i.e., a condensed hierarchical clustering algorithm and a split hierarchical clustering algorithm (Agglerating and division), and can be also understood as a Bottom-up method (Bottom-up) and a Top-down method (Top-down); specifically, the hierarchical clustering algorithm may, but is not limited to, an unsupervised clustering algorithm that performs distance calculation based on the euclidean distance algorithm and hierarchical clustering based on Ward connection criteria (which is one of three existing connection criteria).

The affected risk giving module is respectively in communication connection with the hierarchical clustering module and the clinical information acquisition module, and is used for accessing the clinical information acquisition module for each patient class in the K patient classes, reading the clinical characteristic data of the corresponding AS patient, and giving an AS hip joint affected risk label of the corresponding class according to the corresponding data reading result. The specific details of the AS hip joint involvement risk label of the patient can be conventionally obtained based on the existing clinical experience, for example, if the occurrence rate of joint effusion of a certain patient is highest, the occurrence rate of osteoarthritis is higher, the occurrence rate of attachment inflammation is highest, the occurrence rate of erosive lesions is highest, and the occurrence rate of joint gap stenosis is highest, which reflects that they have severe acute and chronic inflammatory lesions at the same time, and can be given AS "high risk" AS hip joint involvement risk labels; and/or, if the radiation index of the back ankylosing spondylitis of a certain patient class is higher than grade 2, an AS hip joint involvement risk label of "high risk" is also assigned.

The nuclear magnetic resonance scanning module is also used for scanning to obtain a hip joint magnetic resonance image of the target patient. The target patient may or may not be an AS patient.

The image histology feature extraction module is further used for extracting magnetic resonance image histology features of the target patient from the hip magnetic resonance images of the target patient. The specific details of the magnetic resonance imaging histology features of the target patient extracted from the hip magnetic resonance image of the target patient can be found in the steps S21 to S24, and are not described herein.

The risk labeling output module is respectively in communication connection with the affected risk giving module and the image histology feature extraction module, and is used for calculating the distance from the target patient to the class center of each patient class according to the magnetic resonance image histology feature of the target patient, and then taking the AS hip joint affected risk label of a certain patient class which is in the K patient classes and corresponds to the minimum distance AS an AS hip joint affected risk labeling result of the target patient and outputting the AS hip joint affected risk labeling result. The foregoing distances may also be calculated based on the Euclidean distance algorithm.

The novel scheme for automatically labeling the AS hip joint affected risk of the patient based on the image histology technology and the hierarchical clustering algorithm is provided, namely the novel scheme comprises a clinical information acquisition module, a nuclear magnetic resonance scanning module, an image histology feature extraction module, a hierarchical clustering module, an affected risk giving module and a risk labeling output module, and through cooperation of the modules, the magnetic resonance image histology feature of each AS patient can be firstly extracted, hierarchical clustering analysis is carried out on all patients according to the extraction result to obtain a plurality of patient classes, AS hip joint affected risk labels are given to each class according to clinical feature data, then the distance to each class center is calculated according to the magnetic resonance image histology feature of a target patient, and finally the risk label of a certain class corresponding to the minimum distance is used AS the AS hip joint affected risk labeling result of the target patient and is output, so that the accuracy of the AS hip joint affected risk labeling result is poor due to low-specificity performance of the prior art, the accuracy of the AS joint affected risk labeling result is improved, and the AS hip joint affected risk labeling result is convenient for providing effective reference information and practical application.

Preferably, the system further comprises a clinical statistics analysis module which is in communication connection with the affected risk giving module; the clinical statistics analysis module is used for carrying out descriptive statistical analysis on each patient category by using SPSS (Statistical Product and Service Solutions, statistical product and service solution software, which is a total name of a series of software products and related services for statistical analysis operation, data mining, predictive analysis and decision support tasks, which are proposed by IBM company, according to clinical feature data of corresponding AS patients, and version of Windows and Mac OS X and the like) analysis tools to obtain corresponding clinical explanation. The clinical interpretation is used to present clinical differences of the corresponding class from other classes in order to also provide valid reference information for physician decisions. For example, based on hip magnetic resonance images and clinical characterization data of 167 hip pain AS patients who received pelvic MRI examination at a medical center from 1 st 2019 to 9 st 2022 and/or the path ankylosing spondylitis radiology index based on pelvic X-ray images, four patient classes (whose tree structures are shown in fig. 2) can be hierarchically clustered, and by descriptive statistical analysis, the following but not limited clinical explanation can be obtained: (1) a first patient class: the highest incidence of joint effusion (36 [94.7% ]), osteoarthritis at 22[57.9% ], attachment points at the highest incidence of attachment points (specifically, attachment points-t, attachment points-i and attachment points-p are 16[42.1%, 6[15.8% ] and 12[31.6% ], respectively), joint erosion at the highest incidence of 23[60.5% ], joint gap stenosis at the highest incidence of 17[44.7% ], reflecting that they have severe acute and chronic inflammatory lesions at the same time; (2) a second patient class: joint effusion (29 [85.3% ]), bone marrow edema (13 [38.2% ]) and attachment point inflammation (12 [35.3% ], 0 and 1[2.9% ], respectively, attachment point-t, attachment point-i and attachment point-p), indicating a lower incidence of active lesions, but a second structural failure rate, including joint face sclerosis (3 [8.8% ]), joint erosion (13 [34.2% ]) and joint gap stenosis (7 [20.6% ]; (3) third patient class: joint surface sclerosis (0), joint erosion (5 [20.8% ]) and joint stenosis (2 [8.3% ]), i.e. the prevalence in each structural injury is minimal, while joint effusion (20 [8.3% ]) and attachment points inflammation (6 [25.0% ], 0 and 4[16.7% ] at attachment point-t, attachment point-i and attachment point-p, respectively) also rarely occur, thus including patients with minimal MRI injury, with less medical burden of diseases associated with hip joint involvement; (4) fourth patient class: the level of structural damage was also lower, although it was slightly higher than the third patient class, showing a significantly increased prevalence of attachment site inflammation (attachment site-t, attachment site-i, and attachment site-p were 27[38.0% ], 4[5.6% ], and 17[23.9% ], respectively).

The system further specifically comprises a data display module which is in communication connection with the clinical statistics analysis module; the data display module is used for displaying the occurrence probability of each clinical feature in the corresponding clinical explanation through a chord chart for each patient class. Also based on the above example, the probability of occurrence of each MRI performance (i.e., joint effusion, bone marrow edema, attachment inflammation, joint sclerosis, joint erosion, lipopathy, joint gap stenosis, etc.) in each of the four patient classes may be presented with a chord chart as shown in FIG. 3 to provide a physician with an intuitive clinical explanation.

In summary, the ankylosing spondylitis hip joint involvement risk labeling system provided by the embodiment has the following technical effects:

(1) The embodiment provides a new scheme for realizing automatic labeling of AS hip joint affected risk of a patient based on an image histology technology and a hierarchical clustering algorithm, which comprises a clinical information acquisition module, a nuclear magnetic resonance scanning module, an image histology feature extraction module, a hierarchical clustering module, an affected risk giving module and a risk labeling output module, wherein the magnetic resonance image histology feature of each AS patient can be firstly extracted through cooperation of the modules, hierarchical clustering analysis is carried out on all patients according to the extraction result, a plurality of patient classes are obtained, AS hip joint affected risk labels are given to each class according to clinical feature data, then distances to centers of each class are calculated according to the magnetic resonance image histology feature of a target patient, and finally the risk label of a certain class corresponding to the minimum distance is taken AS an AS hip joint affected risk labeling result of the target patient and is output, so that the AS hip joint affected risk labeling result is poor in accuracy due to low specificity expression of MRI, the aim of improving the accuracy of the AS hip joint affected risk labeling result is fulfilled, effective reference information is provided for decision making, and practical application and popularization are facilitated;

(2) Clinical interpretation of each patient class can also be obtained through statistical analysis and visualized, and effective reference information is further provided for doctor decision.

Finally, it should be noted that: the foregoing description is only of the preferred embodiments of the invention and is not intended to limit the scope of the invention. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention should be included in the protection scope of the present invention.

Claims (10)

1. The ankylosing spondylitis hip joint involvement risk labeling system based on image histology is characterized by comprising a clinical information acquisition module, a nuclear magnetic resonance scanning module, an image histology feature extraction module, a hierarchical clustering module, an involvement risk giving module and a risk labeling output module;

the clinical information acquisition module is used for acquiring clinical characteristic data of a plurality of AS patients;

the nuclear magnetic resonance scanning module is used for scanning to obtain hip joint magnetic resonance images of the AS patients;

the image histology feature extraction module is in communication connection with the nuclear magnetic resonance scanning module and is used for extracting corresponding magnetic resonance image histology features from corresponding hip joint magnetic resonance images of each AS patient in the plurality of AS patients after the hip joint magnetic resonance images of the plurality of AS patients are read;

the hierarchical clustering module is in communication connection with the image histology feature extraction module and is used for carrying out clustering analysis on the AS patients by adopting an unsupervised hierarchical clustering algorithm according to the magnetic resonance image histology features of the AS patients to obtain K patient classes, wherein K represents a positive integer greater than or equal to 2;

the affected risk giving module is respectively in communication connection with the hierarchical clustering module and the clinical information acquisition module, and is used for accessing the clinical information acquisition module for each patient class in the K patient classes, reading clinical characteristic data of corresponding AS patients, and giving AS hip joint affected risk labels of the corresponding classes according to corresponding data reading results;

the nuclear magnetic resonance scanning module is also used for scanning to obtain a hip joint magnetic resonance image of the target patient;

the image histology feature extraction module is further used for extracting magnetic resonance image histology features of the target patient from the hip magnetic resonance images of the target patient;

the risk labeling output module is respectively in communication connection with the affected risk giving module and the image histology feature extraction module, and is used for calculating the distance from the target patient to the class center of each patient class according to the magnetic resonance image histology feature of the target patient, and then taking the AS hip joint affected risk label of a certain patient class which is in the K patient classes and corresponds to the minimum distance AS an AS hip joint affected risk labeling result of the target patient and outputting the AS hip joint affected risk labeling result.

2. The ankylosing spondylitis hip joint involvement risk labeling system according to claim 1, wherein the clinical characteristic data comprises demographic information, AS-related clinical characteristic information and hip joint radiological injury information, wherein the demographic information comprises age, sex, body mass index, complications, smoking, drinking and/or vaccination, the AS-related clinical characteristic comprises onset age, disease course, drug treatment regimen comprising a non-steroidal anti-inflammatory drug treatment and/or DMARDs drug treatment, and the hip joint general magnetic resonance examination result comprises joint effusion examination result, bone marrow edema examination result, attachment point inflammation examination result, joint surface sclerosis examination result, joint erosion examination result, fat lesion examination result and/or joint interstitial stenosis examination result, and the hip joint radiological injury information comprises a hip joint structure injury degree score based on the back ankylosing spondylitis radiological index.

3. The ankylosing spondylitis hip involvement risk labeling system of claim 1, wherein for each AS patient in the plurality of AS patients, extracting the corresponding magnetic resonance image histology features from the corresponding hip magnetic resonance images comprises:

for one AS patient in the plurality of AS patients, a corresponding region of interest is intercepted from a corresponding hip joint magnetic resonance image, wherein the region of interest is an image region containing acetabulum, femoral head and constituted hip joint capsule content;

extracting a plurality of T1WI/T2WI sequence features from a region of interest of the certain AS patient;

preprocessing the plurality of T1WI/T2WI sequence features to obtain a plurality of preprocessed features, wherein the preprocessing comprises removing features with zero inter-patient variance, removing features with intra-observer variability and inter-observer variability smaller than a preset threshold and/or eliminating batch effects;

and carrying out data standardization processing on the preprocessed features to obtain a plurality of standardized features which are suitable for distance calculation in a clustering algorithm and correspond to the preprocessed features one by one, and taking the standardized features AS the magnetic resonance image histology features of the AS patient.

4. A ankylosing spondylitis hip involvement risk labeling system according to claim 3, wherein for a certain AS patient among the plurality of AS patients, intercepting the corresponding region of interest from the corresponding hip magnetic resonance image comprises:

importing the hip joint magnetic resonance image of the certain AS patient into a pre-trained region-of-interest recognition model based on a YOLOv4 target detection algorithm, and outputting to obtain a region-of-interest recognition result;

and according to the identification result of the region of interest, cutting out the region of interest of the AS patient from the hip joint magnetic resonance image of the AS patient, wherein the region of interest refers to an image region containing acetabulum, femoral head and constituted hip joint capsule content.

5. A ankylosing spondylitis hip involvement risk labeling system according to claim 3, characterized in that extracting a plurality of T1WI/T2WI sequence features from the region of interest of the certain AS patient comprises:

according to the radiation characteristics defined by the image biomarker standardization initiative, 18 first-order characteristics, 8 shape characteristics and 75 texture characteristics of the T1WI sequence and the T2WI sequence are respectively extracted from the region of interest of the certain AS patient, and different various image filtering algorithms are applied to decompose the original image of the region of interest, so that 1210 high-order characteristics are extracted, and 1412T 1WI/T2WI sequence characteristics are obtained in total.

6. A ankylosing spondylitis hip involvement risk marking system according to claim 3, characterized in that during the pretreatment the batch effect is eliminated by means of combatt compensation.

7. A ankylosing spondylitis hip involvement risk labeling system according to claim 3, characterized in that the data normalization processing is performed on the plurality of preprocessed features to obtain a plurality of normalized features suitable for distance calculation in a clustering algorithm and corresponding one-to-one to the plurality of preprocessed features, comprising:

and carrying out normalization processing on the preprocessed features by adopting a Z score normalization mode to obtain a plurality of normalization features which are suitable for distance calculation in a clustering algorithm and correspond to the preprocessed features one by one.

8. The ankylosing spondylitis hip joint involvement risk labeling system according to claim 1, wherein the hierarchical clustering algorithm adopts an unsupervised clustering algorithm that performs distance calculation based on euclidean distance algorithm and hierarchical clustering based on Ward connection standard.

9. The ankylosing spondylitis hip joint involvement risk labeling system of claim 1, further comprising a clinical statistical analysis module communicatively coupled to the involvement risk assigning module;

the clinical statistics analysis module is used for carrying out descriptive statistics analysis on each patient according to clinical characteristic data of corresponding AS patients by using an SPSS analysis tool to obtain corresponding clinical explanation.

10. The ankylosing spondylitis hip joint involvement risk labeling system of claim 9, further comprising a data display module communicatively coupled to the clinical statistics analysis module;

the data display module is used for displaying the occurrence probability of each clinical feature in the corresponding clinical explanation through a chord chart for each patient class.