A multimodal data processing method and system for risk stratification of cognitive function related to white matter lesions.
By generating a cognitive function risk index related to white matter lesions through multimodal data fusion analysis, this technology solves the problem that existing technologies cannot fully reflect the relationship between white matter lesions and peripheral microcirculation status and cognitive function impairment. It provides an auxiliary reference for early identification and hierarchical management, reduces the cost of image assessment, and improves the interpretability of assessment results.
Patent Information
- Application Number
- CN202611228273.0
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2026-08-13
- Publication Date
- 2026-11-17
AI Technical Summary
Existing technologies are insufficient to fully reflect the comprehensive relationship between white matter lesions, peripheral microcirculation status, and cognitive impairment. Furthermore, imaging assessment is costly and complex, and there is a lack of multimodal data processing solutions to unify MRI images, cognitive scales, fundus vascular grading, and vascular risk factors into fusionable structured indicators.
By receiving head MRI image assessment data, cognitive scale data, fundus vascular grading, sublingual vascular grading, and vascular risk factor data, a cognitive function risk index related to white matter lesions is generated, and risk stratification information and indicator contribution information are output for structured processing and fusion analysis.
It enables early identification and stratification without directly providing disease diagnosis or treatment plans, assists in the identification and management of cognitive function risks related to white matter lesions, and provides interpretable assessment results and data integrity prompts.
Smart Images

Figure CN123120293A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the fields of medical image data processing, clinical scale data processing, and peripheral microcirculation information fusion assessment, and particularly to a multimodal data processing method and system for risk stratification of cognitive function related to white matter lesions. Background Technology
[0002] White matter lesions, also known as white matter hyperintensity or leukoaraiosis, are a common imaging manifestation of cerebral small vessel disease. They typically appear as abnormally high signal intensity in the periventricular or deep white matter regions on T2-weighted MRI scans and FLAIR sequences of the head. White matter lesions are closely related to factors such as aging, hypertension, metabolic abnormalities, and microcirculatory disturbances. Increased white matter lesions can affect the efficiency of fronto-subcortical circuits and brain network connectivity, thereby being associated with a decline in attention, executive function, memory function, and overall cognitive function. Therefore, how to conduct early, convenient, and multidimensional assessment of the cognitive function risks associated with white matter lesions is an important issue in the screening and risk management of vascular cognitive impairment.
[0003] In the prior art, there are already solutions for automated analysis of images of white matter lesions. For example, patent document CN113077887B discloses an automated quantitative analysis system and interpretation method for white matter lesions. This system uses artificial intelligence to automatically delineate lesions and calculate their volume. Simultaneously, through human-computer interaction, it determines information such as the anatomical location, morphology, size, time of onset, and presence of hemorrhage of the lesions, thereby quantifying image indicators and standardizing reporting terminology. Such solutions can improve the consistency and quantification of white matter lesion image interpretation and help reduce the subjectivity of manual interpretation. However, their focus is mainly on the segmentation, localization, volume calculation, and interpretation of lesions in central nervous system imaging. They do not fully integrate peripheral microcirculation indicators such as fundus vessels and sublingual vascular tracts, nor do they use cognitive scale scores and vascular risk factors as unified data fusion objects for risk stratification output.
[0004] For example, patent document CN115880287A discloses a method for segmenting and rating high-signal lesion regions in the brain white matter. This method acquires FLAIR image data of the patient's brain, preprocesses the brain image data, performs cranial-brain separation, background separation, high-signal threshold segmentation of the brain white matter, lesion region extraction, and lesion type determination, and inputs the classification results into a preset neural network model for scoring to obtain a rating result for high-signal lesions in the brain white matter. This scheme can segment and rate high-signal lesion regions in the brain white matter based on MRI images and has a certain degree of automation. However, its technical approach still mainly focuses on the image data itself, and it does not adequately utilize the correlation between multiple sources of clinical information such as patient cognitive scale data, fundus vascular grading, sublingual vascular grading, age, history of hypertension, history of heart disease, dyslipidemia, and homocysteine levels.
[0005] In clinical practice, cognitive decline associated with white matter lesions is not solely determined by the size or grade of lesions on imaging. Vascular risk factors such as age, hypertension, history of heart disease, and homocysteine levels can influence the progression of cerebral small vessel disease. Simultaneously, morphological changes in retinal vessels, as a directly observable microcirculatory network, such as arteriosclerosis, arteriovenous crossings, vascular tortuosity, exudation, cotton wool spots, and hemorrhages, can reflect the state of systemic small vessel damage to some extent. Changes in the filling, elongation, tortuosity, branching, and punctate dilated glomus pulvinas of the sublingual venous vessels can also serve as supplementary indicators of peripheral microcirculation. Therefore, relying solely on head MRI image segmentation or single image grading results is insufficient to comprehensively reflect the integrated relationship between white matter lesions, peripheral microcirculation status, and cognitive impairment.
[0006] Furthermore, current imaging assessments of white matter lesions typically rely on radiologists or specific image analysis software. For primary healthcare institutions, physical examinations, or initial screening, MRI examinations are costly and relatively complex. In contrast, fundus photography, tongue imaging, cognitive scale scoring, and routine risk factor collection offer better accessibility and ease of operation. While existing technologies address the quantification of white matter lesion images, the segmentation and rating of high-signal areas in white matter, and the prediction of diseases using fundus images, a multimodal data processing solution for cognitive function risks related to white matter lesions is lacking. This solution should be able to uniformly convert MRI image load, MMSE / MoCA cognitive scores, fundus vascular grading, sublingual choroidal grading, and vascular risk factors into fusionable structured indicators, and further output risk stratification information, data integrity prompts, and indicator contribution information.
[0007] Therefore, it is necessary to provide a multimodal data processing method for risk stratification of cognitive function related to white matter lesions. This method can perform structured processing and fusion analysis of collected image, scale, peripheral microcirculation and risk factor data without directly providing disease diagnosis or treatment plan. This will provide auxiliary reference for early identification, stratification indication and subsequent clinical review of cognitive function risks related to white matter lesions. Summary of the Invention
[0008] The technical objective of this invention is to provide a multimodal data processing method and system for risk stratification of cognitive function related to white matter lesions. This method generates a risk index for cognitive function related to white matter lesions by performing structured processing and fusion analysis on head MRI image assessment data, cognitive scale data, fundus vascular grading, hypolingual choroid grading, and vascular risk factor data. It also outputs risk stratification information and indicator contribution information for clinical reference to help identify individuals at risk of cognitive decline related to white matter lesions.
[0009] Firstly, in order to achieve the above-mentioned objectives, the present invention adopts the following technical solution:
[0010] A multimodal data processing method for risk stratification of cognitive function related to white matter lesions, the method being executed by a computer device, is used to output risk stratification information for clinical reference based on collected clinical examination data, without directly generating disease diagnosis conclusions or treatment plans. The method includes the following steps:
[0011] S1. Receive multimodal data of the subject to be evaluated, including head MRI image evaluation data, cognitive scale data, peripheral microcirculation evaluation data, and vascular risk factor data;
[0012] S2. Obtain the white matter lesion burden level based on the head MRI image assessment data. , Determined based on the Fazekas visual rating scale;
[0013] S3. Obtain a Mini-Mental State Examination score based on the cognitive scale data. and the Montreal Cognitive Assessment Scale score adjusted for education level And generate an index of cognitive impairment. ;
[0014] S4. Obtain fundus vascular grading based on the peripheral microcirculation assessment data. and sublingual vein classification And generate indicators of peripheral microcirculation disorders. ,in, exist The weight in is greater than The weights;
[0015] S5. Generate vascular risk factor indicators based on the vascular risk factor data. The vascular risk factor data include one or more of the following: age, history of hypertension, history of diabetes, history of heart disease, dyslipidemia, and homocysteine levels.
[0016] S6. The level of white matter lesion burden. Cognitive Impairment Indicators Peripheral microcirculation disorder indicators and vascular risk factor indicators By performing fusion, a cognitive function risk index related to white matter lesions was obtained. And according to the risk index Output risk stratification information for low-risk, medium-risk, or high-risk risks, along with corresponding indicator contribution information.
[0017] As a further improvement, in S1, the multimodal data comes from examination data completed within the same hospitalization cycle or the same preset assessment cycle; when the head MRI image assessment data, cognitive scale data, peripheral microcirculation assessment data, or vascular risk factor data are missing, the image quality does not meet the preset clarity requirements, the identity identification is inconsistent, or the acquisition time exceeds the preset assessment cycle, a quality mark is generated for the corresponding data, and a data integrity prompt is output simultaneously when the risk stratification information is output.
[0018] As a further improvement, in S2, the head MRI imaging assessment data includes the grade of periventricular white matter lesions. and deep white matter lesion grade ;
[0019] The grade of periventricular white matter lesions The grading is based on the presence or absence of high signal in the periventricular white matter, cap-shaped high signal, ring-shaped high signal, halo-shaped white matter lesions, degree of irregularity of the margins, and degree of extension into the deep white matter;
[0020] The level of deep white matter lesions The lesions are classified according to the presence or absence of deep white matter lesions, punctate lesions, small patchy lesions, extensive patchy lesions, early fusion tendency, and large fused lesions.
[0021] When the periventricular white matter lesion grade is obtained simultaneously and the aforementioned deep white matter lesion grade At that time, the white matter lesion burden level was determined as follows: :
[0022] ;
[0023] In the formula, The level of white matter lesion burden; Grade of periventricular white matter lesions; Grade of deep white matter lesions; This indicates that the higher of the input levels should be selected.
[0024] As a further improvement, in S3, the education-adjusted Montreal Cognitive Assessment Scale score is... The scores are obtained as follows: when the years of education of the subject to be assessed are less than the preset years, a preset correction score is added to the original Montreal Cognitive Assessment Scale score; when the years of education of the subject to be assessed are not less than the preset years, the original Montreal Cognitive Assessment Scale score is retained.
[0025] The cognitive impairment level index Generate according to the following formula:
[0026] ;
[0027] In the formula, As an indicator of the degree of cognitive impairment; Scoring using the Mini-Mental State Examination; The score is based on the education level-adjusted Montreal Cognitive Assessment Scale. The maximum score on the cognitive scale is 30. Weights corresponding to the scores of the Mini-Mental State Examination; The weights corresponding to the scores of the Montreal Cognitive Assessment Scale; and The sum is 1.
[0028] As a further improvement, in S4, the fundus vascular grading... Determined based on bilateral fundus images acquired by a non-mydriatic color fundus camera;
[0029] Fundus vascular grading It should be determined based on at least one or more of the following fundus vascular morphological features: enhanced retinal artery wall reflex, copper wire-like changes, silver wire-like changes, arteriovenous crossing sign, vascular tortuosity, hard exudates, cotton wool spots, and hemorrhage spots.
[0030] Furthermore, the fundus vascular grading The grading system is progressive, based on the following stages: normal fundus, retinal arteriosclerosis, hard exudates, cotton wool spots, and hemorrhage spots.
[0031] And / or, in S4, the sublingual vascular grading Determined based on sublingual images captured by a tongue image acquisition device;
[0032] The sublingual vein classification It should be determined based on at least one or more morphological features of the main trunk of the sublingual veins, including the degree of fullness, length, tortuosity, branching, punctate dilated vascular bulbs, and grape-bead-like changes.
[0033] Furthermore, the sublingual vascular grading... The classification is progressive, based on whether the main trunk is normal, filled and elongated, significantly filled, elongated and tortuous, or has punctate dilated globules or beaded changes.
[0034] And / or, in S4, the peripheral microcirculation disorder index Generate according to the following formula:
[0035] ;
[0036] In the formula, This is an indicator of peripheral microcirculatory disturbances; Grading of retinal blood vessels; This is the lowest level in the retinal vascular grading system. This is the highest level of retinal vascular grading. The sublingual veins are graded; This is the lowest level in the classification of sublingual veins; This is the highest level of the sublingual vein classification; Weights corresponding to the retinal vascular grading; The weights corresponding to the grading of the sublingual veins; and The sum is 1, and Greater than .
[0037] As a further improvement, in S5, the vascular risk factor index Generate according to the following formula:
[0038] ;
[0039] In the formula, Indicators of vascular risk factors; The normalized age; Marker of a history of hypertension; Markers of diabetes history; Marker of a history of heart disease; This represents the normalized homocysteine level. Markers for dyslipidemia; , , , , and These are the weights of the corresponding risk factors, and the sum of all weights is 1;
[0040] The normalized age and the normalized homocysteine level The following formulas are used to obtain the results:
[0041] In the formula, The normalized age; The age of the person to be evaluated; The minimum age of the sample or a preset lower age limit; The maximum age of the sample or a preset upper age limit;
[0042] ;
[0043] In the formula, This represents the normalized homocysteine level. The homocysteine level of the subjects to be evaluated; This represents the minimum or preset lower limit of homocysteine levels in the sample. This represents the maximum or preset upper limit of homocysteine levels in the sample.
[0044] As a further improvement, in S6, the risk index for cognitive function related to white matter lesions is... Generate according to the following formula:
[0045] ;
[0046] In the formula, A risk index for cognitive function related to white matter lesions; Normalized white matter lesion burden level; As an indicator of the degree of cognitive impairment; This is an indicator of peripheral microcirculatory disturbances; Indicators of vascular risk factors; The weights corresponding to the white matter lesion burden levels; The weights corresponding to the indicators of cognitive impairment degree; The weights corresponding to the peripheral microcirculation disorder indicators; The weights corresponding to the vascular risk factor indicators; , , and The sum is 1;
[0047] The normalized white matter lesion burden level It is obtained using the following formula:
[0048] ;
[0049] In the formula, Normalized white matter lesion burden level; The level of white matter lesion burden; This is the lowest level of the white matter lesion burden gradation; This represents the highest level of white matter lesion burden.
[0050] And / or, in S6, based on the first risk threshold Second risk threshold Risk index for cognitive function related to the aforementioned white matter lesions Conduct risk stratification;
[0051] When the risk index Less than the first risk threshold At that time, output low-risk stratification information;
[0052] When the risk index Not less than the first risk threshold And less than the second risk threshold At that time, output risk stratification information;
[0053] When the risk index Not less than the second risk threshold At that time, output high-risk stratification information;
[0054] in, This serves as the dividing line between low and medium risk. This serves as the dividing line between medium and high risk. Less than ;
[0055] And / or, in S6, the indicator contribution information includes at least the level of white matter lesion burden. Risk index Contribution items, cognitive impairment indicators Risk index Contribution items, peripheral microcirculation disorder indicators Risk index Contribution items and vascular risk factor indicators Risk index Contributions;
[0056] Among them, when the peripheral microcirculation barrier index Mid-retinal vascular classification When the contribution of a factor exceeds a preset contribution threshold, a peripheral microcirculation abnormality alert is output in the risk stratification information; when the white matter lesion burden level... and the aforementioned fundus vascular grading When the risk level exceeds the corresponding preset level, a prompt indicating the coexistence of cerebral white matter lesion burden and peripheral microcirculation abnormality is output in the risk stratification information.
[0057] And / or, in S6, the indicator contribution information is generated in the following manner:
[0058] ;
[0059] In the formula, The contribution of the white matter lesion burden level to the risk index; The weights corresponding to the white matter lesion burden levels; Normalized white matter lesion burden level;
[0060] ;
[0061] In the formula, The contribution of the cognitive impairment level indicator to the risk index; The weights corresponding to the indicators of cognitive impairment degree; As an indicator of the degree of cognitive impairment;
[0062] ;
[0063] In the formula, The contribution of peripheral microcirculation disorder indicators to the risk index; The weights corresponding to the peripheral microcirculation disorder indicators; This is an indicator of peripheral microcirculatory disturbances;
[0064] ;
[0065] In the formula, The contribution of vascular risk factor indicators to the risk index; The weights corresponding to the vascular risk factor indicators; Indicators of vascular risk factors;
[0066] And / or, S6 further includes, based on the said white matter lesion burden level Fundus vascular classification Sublingual vein classification Cognitive Impairment Indicators and vascular risk factor indicators Generate consistency verification results;
[0067] When the white matter lesion burden level With the aforementioned fundus vascular grading All increased, and the indicators of cognitive impairment also increased. When the cognitive impairment threshold is exceeded, a multimodal consistency enhancement marker is generated;
[0068] When the white matter lesion burden level Elevated but the peripheral microcirculation disorder index The preset microcirculation threshold or the peripheral microcirculation barrier index was not exceeded. The level of white matter lesion burden is elevated, but the level of white matter lesion burden is elevated. If the preset image threshold is not exceeded, a multimodal inconsistency verification mark is generated.
[0069] Secondly, the present invention also provides a risk stratification system for cognitive function related to white matter lesions, characterized in that the system is used to implement the method, comprising:
[0070] The data receiving module is used to receive head MRI image assessment data, cognitive scale data, peripheral microcirculation assessment data, and vascular risk factor data of the subject to be evaluated.
[0071] The image burden processing module is used to obtain the white matter lesion burden level based on the head MRI image assessment data. ;
[0072] The cognitive scale processing module is used to score based on the Mini-Mental State Examination. and the Montreal Cognitive Assessment Scale score adjusted for education level Generate cognitive impairment index ;
[0073] The peripheral microcirculation processing module is used to classify retinal blood vessels. and sublingual vein classification Generate peripheral microcirculation disorder indicators ;
[0074] The risk factor processing module is used to generate vascular risk factor indicators based on one or more of the following factors: age, history of hypertension, history of diabetes, history of heart disease, dyslipidemia, and homocysteine level. ;
[0075] The risk fusion module is used to determine the level of white matter lesion burden based on the risk level. Cognitive Impairment Indicators Peripheral microcirculation disorder indicators and vascular risk factor indicators Generate a risk index for cognitive function related to white matter lesions. ;
[0076] Output module, used to output based on the risk index Output risk stratification information, data integrity prompts, indicator contribution information, and multimodal consistency verification results.
[0077] Thirdly, the present invention also provides an electronic device, including a processor, a memory, and a computer program stored in the memory and executable on the processor, wherein when the processor executes the computer program, it implements the method described above.
[0078] Fourthly, the present invention also provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the method described thereon.
[0079] This invention uses the brain white matter lesion burden level Cognitive Impairment Indicators Peripheral microcirculation disorder indicators and vascular risk factor indicators Converting them into unified, fusionable quantitative indicators avoids the problem of insufficient information dimensions caused by relying solely on MRI image grading or single cognitive scale scores; among them, peripheral microcirculation disorder indicators Simultaneously, fundus vascular grading was introduced. and sublingual vein classification This allows the system to better utilize the correlation between changes in fundus vascular morphology and white matter lesion burden and cognitive scores, and to give fundus vascular grading a higher weight in this indicator. (Through risk index) By employing fusion calculations and threshold stratification, this invention can output low-risk, medium-risk, or high-risk alerts, and further provide the contribution of each indicator to the risk index, making the assessment results interpretable. Simultaneously, through data integrity alerts and multimodal consistency checks, it can indicate situations such as missing examination data or inconsistencies between imaging and peripheral microcirculation findings, facilitating subsequent manual review. Therefore, this invention can provide auxiliary reference for early screening, stratified management, and clinical review of cognitive function risks related to white matter lesions without directly making disease diagnoses or treatment plans. Attached Figure Description
[0080] Figure 1 This is a schematic diagram of the overall process of the method of the present invention.
[0081] Figure 2 This is a schematic diagram of the system structure of the present invention.
[0082] Figure 3 This is a schematic diagram of the multimodal data quality control and structured processing workflow.
[0083] Figure 4 This is a schematic diagram of the process for generating the white matter lesion burden level.
[0084] Figure 5 This is a schematic diagram illustrating the process of generating indicators for peripheral microcirculation obstruction.
[0085] Figure 6 A schematic diagram illustrating the integration of risk indices and the output of indicators.
[0086] Figure 7 This is a schematic diagram of the multimodal consistency verification process.
[0087] Figure 8 A schematic diagram showing the changes in cognitive scores under different levels of white matter lesion burden.
[0088] Figure 9 This diagram illustrates the changes in peripheral microcirculation indicators under different levels of white matter lesion burden.
[0089] Figure 10This is a diagram illustrating the correlation of multimodal indicators.
[0090] Figure 11 This is a schematic diagram illustrating the risk index fusion and indicator contribution output in an application example of the present invention.
[0091] Figure labeling: 100, Risk stratification system for cognitive function related to white matter lesions; 101, Data receiving module; 102, Image load processing module; 103, Cognitive scale processing module; 104, Peripheral microcirculation processing module; 105, Risk factor processing module; 106, Risk fusion module; 107, Output module; 108, Quality control unit; 109, Consistency verification unit. Detailed Implementation
[0092] The specific embodiments of the present invention will be described below with reference to the accompanying drawings. It should be understood that the following embodiments are used to illustrate the technical solutions of the present invention, and not to limit the scope of protection. Without departing from the technical concept of the present invention, those skilled in the art can make equivalent substitutions or conventional adjustments to the data acquisition equipment, scoring rules, weight setting methods, threshold setting methods, and output interface formats.
[0093] This invention provides a multimodal data processing method and system for risk stratification of cognitive function related to white matter lesions. The method is executed by a computer and processes data obtained through clinical examination, imaging assessment, cognitive scale testing, fundus image acquisition, sublingual image acquisition, and laboratory tests. The output of this invention includes risk stratification information, data integrity indicators, indicator contribution information, and multimodal consistency verification results, used to assist clinical review and risk indication. It does not directly output disease diagnosis conclusions or generate treatment plans.
[0094] I. Terminology Explanation
[0095] In this embodiment, white matter lesions refer to imaging changes that appear as abnormally high signal intensity in the periventricular or deep white matter regions of the brain on head MRI images. The white matter lesion burden level is denoted as... This is used to indicate the degree of imaging burden of white matter lesions in the brain of the subject being evaluated.
[0096] Periventricular white matter lesions are graded as follows: It is determined based on the presence or absence of high signal in the periventricular white matter, cap-shaped high signal, ring-shaped high signal, halo-like lesions, degree of irregularity of the margins, and the extent of extension into the deep white matter.
[0097] The grade of deep white matter lesions is denoted as It is determined based on the presence or absence of deep white matter lesions, punctate lesions, small patchy lesions, extensive patchy lesions, early fusion tendency, and large fused lesions.
[0098] The score of the Mini-Mental State Examination is recorded as follows: The corrected score of the Montreal Cognitive Assessment Scale is recorded as follows: .in, This is a score adjusted for educational attainment based on the original MoCA score. The degree of cognitive impairment is indicated by... It is used to and Converted into a co-directional cognitive risk indicator, i.e. The larger the value, the higher the risk of cognitive decline.
[0099] The classification of fundus vessels is recorded as follows: It is determined based on morphological characteristics such as enhanced retinal artery wall reflection, copper wire-like changes, silver wire-like changes, arteriovenous crossing sign, vascular tortuosity, hard exudates, cotton wool spots, and hemorrhage spots in non-mydriatic color fundus images. The sublingual vein grading is recorded as follows: It is determined based on the morphological characteristics of the sublingual venous vessels, including their fullness, elongation, tortuosity, tree-like branches, punctate dilated vascular bulbs, and grape-like beaded changes. Peripheral microcirculatory disorders are indicated by... It is classified by the retinal blood vessel level. and sublingual vein classification Generate together.
[0100] Vascular risk factors are marked as It is used to comprehensively characterize the impact of factors such as age, history of hypertension, history of diabetes, history of heart disease, dyslipidemia, and homocysteine levels on the risk of cognitive function related to white matter lesions.
[0101] The risk index for cognitive function related to white matter lesions is denoted as It is determined by the leukoencephalopathy burden level. Cognitive Impairment Indicators Peripheral microcirculation disorder indicators and vascular risk factor indicators Generated through fusion. Risk index. Used to output risk stratification information of low risk, medium risk, or high risk.
[0102] Indicator contribution information refers to the contribution of various indicators to the risk index. The contributions include contributions from white matter lesion burden, cognitive impairment, peripheral microcirculatory disturbances, and vascular risk factors. Multimodal consistency verification results refer to the consistency enhancement markers or inconsistency review markers generated by the system based on the unidirectional changes in image burden, peripheral microcirculation, cognitive scales, and risk factors.
[0103] II. System Structure
[0104] like Figure 2 As shown, the cognitive function risk stratification system 100 related to white matter lesions of the present invention includes a data receiving module 101, an image load processing module 102, a cognitive scale processing module 103, a peripheral microcirculation processing module 104, a risk factor processing module 105, a risk fusion module 106, and an output module 107. In one embodiment, the system further includes a quality control unit 108 and a consistency verification unit 109.
[0105] The data receiving module 101 is used to receive multimodal data. The multimodal data includes head MRI image assessment data, cognitive scale data, fundus image data, sublingual image data, and vascular risk factor data. Head MRI image assessment data can come from manual assessment results by radiologists or from the output of existing MRI-assisted analysis software. Cognitive scale data includes... The data includes the original MoCA score, years of education, and the time of scale testing. Fundus images can be acquired using a non-mydriatic color fundus camera, and sublingual images can be acquired using a tongue image acquisition device. Vascular risk factor data include age, history of hypertension, history of diabetes, history of heart disease, dyslipidemia, homocysteine levels, and other optional clinical indicators.
[0106] The quality control unit 108 performs unified verification before data enters each processing module. Verification includes checking for consistency of identification, whether the data collection time falls within the same hospitalization cycle or the same preset assessment cycle, whether the image clarity meets requirements, whether fundus images are obstructed or overexposed, whether sublingual images show tongue displacement or abnormal lighting, whether the cognitive scale is complete, and whether laboratory indicators are missing. Data that passes quality control is then processed in subsequent structured workflows; data with missing information or quality issues generates quality tags and displays the corresponding data integrity status in the final output.
[0107] Image burden processing module 102 is used to generate white matter lesion burden levels. In one implementation, the module receives and and in accordance with Figure 4 The process shown is determined The cognitive scale processing module 103 is used to process... and Generate cognitive impairment index The peripheral microcirculation processing module 104 is used to classify retinal blood vessels according to their grading. and sublingual vein classification Generate peripheral microcirculation disorder indicators The risk factor processing module 105 is used to generate vascular risk factor indicators based on vascular risk factors. Risk fusion module 106 is used to integrate... , , and By merging the data, a risk index can be obtained. The consistency verification unit 109 is used to verify the consistency of the data according to the data. , , , and The relationships between the data are used to generate consistency verification results. Output module 107 is used to output risk stratification information, indicator contribution information, data integrity prompts, and consistency verification results.
[0108] The aforementioned system can be deployed in hospital information systems, scientific research data platforms, physical examination center workstations, image-assisted assessment terminals, or cloud servers. The system can also be implemented by electronic devices, which include a processor, a memory, and a computer program stored in the memory. When the computer program is executed by the processor, the method steps of this invention are implemented.
[0109] III. Overall Method Flow
[0110] like Figure 1 As shown, the method of the present invention includes the following steps:
[0111] S1. Receive multimodal data of the object to be evaluated;
[0112] S2. Obtain the white matter lesion burden level based on head MRI imaging data. ;
[0113] S3. Obtain data from the cognitive scale. and And generate an index of cognitive impairment. ;
[0114] S4. Obtain fundus vascular grading based on peripheral microcirculation assessment data. and sublingual vein classification And generate indicators of peripheral microcirculation disorders. ;
[0115] S5. Generate vascular risk factor indicators based on vascular risk factor data. ;
[0116] S6, to , , and By merging the data, a risk index can be obtained. It outputs risk stratification information, indicator contribution information, and consistency verification results.
[0117] In the above steps, S1 and S2 are mainly used to ensure the reliability of data sources and image load data; S3 is used to convert cognitive scale scores into fusionable risk direction indicators; S4 is an important processing step that distinguishes this invention from simple MRI image analysis schemes, which includes fundus vessels and sublingual vascular vessels as peripheral microcirculation disorder indicators; S5 is used to supplement systemic vascular risk background such as age and hypertension; S6 is used to fuse multi-source information into a risk index and generate interpretable output.
[0118] (a) Data Reception and Quality Control
[0119] like Figure 3 As shown, after receiving multimodal data, the data receiving module 101 first performs a structured check on the data by the quality control unit 108. Specifically, the system reads the unique identifier of the subject to be evaluated, the examination date, the examination items, and the data source. If the head MRI assessment data, cognitive scale data, fundus image data, sublingual image data, and laboratory data originate from the same hospitalization period or from the same assessment period within a preset time range, they are determined to have consistent acquisition periods; otherwise, the system generates an acquisition period abnormality mark for data that exceeds the period.
[0120] For head MRI image evaluation data, a systematic check is performed to determine if it contains... , This may allow us to deduce information about the level of white matter lesion burden. For cognitive scale data, a systematic check should be performed to determine if it contains... MoCA raw score and years of education. For fundus image data, the system checks whether the image is centered on the optic disc or contains evaluable retinal vascular regions, and whether there is significant occlusion, severe blurring, overexposure, or underexposure. For sublingual image data, the system checks whether the sublingual choroidal region is visible, and whether there is significant positional deviation, reflection, color distortion, or acquisition failure. For vascular risk factor data, the system checks whether age, hypertension, diabetes, history of heart disease, dyslipidemia, and homocysteine levels are available records.
[0121] When any modality of data is missing, the system does not directly delete the entire sample; instead, it generates a missing data marker. For cases where core data is missing, such as head MRI image assessment data or cognitive scale data, the system can output a message stating "Core data is incomplete; review recommended." For cases where auxiliary data is missing, such as sublingual images or a specific laboratory indicator, the system can calculate a partial risk index based on the available data and indicate in the output that this risk index was generated based on incomplete data. This approach avoids the failure of the entire assessment due to the absence of a single modality and clarifies the data basis for subsequent users' risk stratification results.
[0122] The purpose of this quality control process is to improve data consistency before multimodal fusion and reduce the risk of bias caused by mixing data from different inspection times, objects, or quality levels. Compared to schemes that only receive single image scores, this implementation uses multimodal data quality control as a pre-risk stratification step, giving the final risk index a better interpretable basis.
[0123] (II) Generation of white matter lesion burden grades
[0124] like Figure 4 As shown, after receiving head MRI image evaluation data, the image burden processing module 102 extracts the grade of periventricular white matter lesions. and deep white matter lesion grade .in, Grading can be based on the extent and morphology of high signal intensity in the periventricular white matter. The lesions in the deep white matter can be classified according to their number, size, fusion trend, and degree of fusion.
[0125] In one implementation, when and When both are available, the system determines the white matter lesion burden level according to the following formula. :
[0126] ;
[0127] In the formula, The level of white matter lesion burden; Grade of periventricular white matter lesions; Grade of deep white matter lesions; This indicates that the higher of the input levels should be selected.
[0128] A higher grade was used as the white matter lesion burden grade. The advantage is that it avoids masking the effects of averaging when the spatial distribution of periventricular lesions and deep white matter lesions is inconsistent. For example, some subjects may have milder periventricular white matter lesions and more severe deep white matter lesions, or vice versa. Simply taking an average might reduce the impact of severe localized lesions on risk stratification. This implementation, by taking a higher level, allows the system to retain sensitivity to more severe imaging burdens.
[0129] In another implementation, the system can also be based on and Weighted combination generation However, the specific calculation method needs to be recorded in the output. For actual deployment scenarios, if the hospital already has the Fazekas total score or white matter lesion grading results from the radiology department, the system can directly read this grading as... The data source is then labeled as either a human evaluation result or a software evaluation result.
[0130] In order to The risk fusion module 106 can be integrated with other indicators in subsequent steps. Normalization was performed. The normalized white matter lesion burden level was denoted as... The calculation method is as follows:
[0131] ;
[0132] In the formula, Normalized white matter lesion burden level; The level of white matter lesion burden; This is the lowest level of the white matter lesion burden gradation; This represents the highest level of white matter lesion burden.
[0133] in, and The classification can be set based on the Fazekas classification range or based on the minimum and maximum classifications in the sample database. After normalization, the white matter lesion burden level can be compared with indicators of cognitive impairment, peripheral microcirculatory disturbance, and vascular risk factors, which is beneficial for subsequent fusion calculations.
[0134] (III) Generation of indicators of cognitive impairment
[0135] Cognitive scale processing module 103 receives The original MoCA score and years of education are considered. For the MoCA score, this implementation method takes into account the impact of education level on the scale results. When the years of education of the subject being assessed are less than a preset number of years, a preset correction score is added to the original MoCA score; when the years of education of the subject being assessed are not less than the preset number of years, the original MoCA score is retained. The corrected MoCA score is denoted as... .
[0136] because and A higher score generally indicates better cognitive function, but risk fusion models require indicators that align with risk direction; therefore, this implementation method will... and Converted into an indicator of cognitive impairment The specific calculation method is as follows:
[0137] ;
[0138] In the formula, As an indicator of the degree of cognitive impairment; Scoring using the Mini-Mental State Examination; The score is based on the education level-adjusted Montreal Cognitive Assessment Scale. The maximum score on the cognitive scale is 30. Weights corresponding to the scores of the Mini-Mental State Examination; The weights corresponding to the scores of the Montreal Cognitive Assessment Scale; and The sum is 1.
[0139] Using the above formula, the scale score is converted into an indicator of the direction of cognitive impairment. When or When the value is lower, the corresponding fraction increases. Increase; when the scale score is high, the corresponding fraction decreases. This approach reduces the risk of cognitive scale data from white matter lesion burden, peripheral microcirculatory disturbances, and vascular risk factors.
[0140] In practical applications, and It can be determined based on sample data training or preset according to clinical application purposes. For example, when the system focuses more on sensitive recognition of mild cognitive decline, it can improve... The weights; when the system is used for general cognitive state screening, it can maintain and The weights are similar. Once the weights are determined, the system should save them in the configuration file or model parameter table and output them along with the risk results when needed for traceability.
[0141] (iv) Generation of indicators of peripheral microcirculation disorders
[0142] like Figure 5 As shown, the peripheral microcirculation processing module 104 is a processing module that contributes significantly to this embodiment. This module does not rely solely on MRI images or cognitive scales, but rather classifies the retinal vessels. and sublingual vein classification Converted into indicators of peripheral microcirculation disorders It is used to characterize the peripheral microcirculation status associated with cerebral small vessel lesions.
[0143] Fundus vascular grading The system determines retinal vascular morphology based on bilateral fundus images acquired by a non-mydriatic color fundus camera. The system can obtain fundus vascular morphological features through manual input, semi-automatic annotation, or image recognition models. These features include increased retinal artery wall reflection, copper wire-like changes, silver wire-like changes, arteriovenous crossings, vascular tortuosity, hard exudates, cotton wool spots, and hemorrhages. Fundus vascular grading can be progressively graded from normal fundus to retinal arteriosclerosis, combined with hard exudates, combined with cotton wool spots, and combined with hemorrhages.
[0144] Sublingual vein classification The sublingual veins are identified based on images captured by a tongue image acquisition device. The system can identify the degree of fullness, elongation, tortuosity, branching, punctate dilated vascular bulbs, and grape-like changes in the main trunk of the sublingual veins through manual input or image recognition. The sublingual vein grading can be progressively graded as follows: normal, full and elongated main trunk, significantly full and elongated main trunk with tortuosity, and presence of punctate dilated vascular bulbs or grape-like changes.
[0145] Because the retinal vessels constitute a directly observable microvascular network, they have a strong correlation with cerebral small vessel disease and cognitive function risk; while the sublingual venous vessels primarily reflect peripheral venous return and local microcirculation status, and are used as supplementary indicators in this approach. Therefore, this implementation method constructs an indicator of peripheral microcirculation disorders. At that time, the retinal blood vessels were graded. The weight is greater than that of the sublingual vein grading. The weights are calculated as follows:
[0146] ;
[0147] In the formula, This is an indicator of peripheral microcirculatory disturbances; Grading of retinal blood vessels; This is the lowest level in the retinal vascular grading system. This is the highest level of retinal vascular grading. The sublingual veins are graded; This is the lowest level in the classification of sublingual veins; This is the highest level of the sublingual vein classification; Weights corresponding to the retinal vascular grading; The weights corresponding to the grading of the sublingual veins; and The sum is 1, and Greater than .
[0148] The above formula first... and Separate normalization is performed, followed by weighted fusion to avoid scale bias in fundus vascular grading and hypolingual vascular grading due to different grading ranges. This is achieved by setting... Greater than The system can make fuller use of the high contribution of fundus vascular grading to the correlation between white matter lesion burden and cognitive score, while preserving the role of the sublingual vascular tract as a supplementary indicator of peripheral microcirculation.
[0149] In one specific implementation, if the fundus vascular grading of the two eyes is different, the fundus vascular grading of the higher eye can be used as the standard. Alternatively, the average value of the binocular grading can be used as... Using a higher ocular grading can improve sensitivity to more severe fundus vascular changes; using an average grading can reduce the impact of random image errors. The system allows selection of specific strategies in parameter configuration and records the adopted strategy in the output. If the sublingual image quality is insufficient, the system can... Mark as missing and process according to the missing data handling strategy. Perform partial calculations or prompt for review.
[0150] (v) Generation of vascular risk factor indicators
[0151] Risk factor processing module 105 is used to generate vascular risk factor indicators based on vascular risk factor data. In this embodiment, vascular risk factors may include age, history of hypertension, history of diabetes, history of heart disease, dyslipidemia, and homocysteine levels. For medical history variables, the system can perform structured processing using a labeled or unlabeled approach; for continuous variables, such as age and homocysteine levels, the system performs normalization.
[0152] Vascular risk factor indicators It can be generated using the following formula:
[0153] ;
[0154] In the formula, Indicators of vascular risk factors; The normalized age; Marker of a history of hypertension; Markers of diabetes history; Marker of a history of heart disease; This represents the normalized homocysteine level. Markers for dyslipidemia; , , , , and These are the weights of the corresponding risk factors, and the sum of the weights is 1.
[0155] Among them, the normalized age It can be obtained using the following formula:
[0156] ;
[0157] In the formula, The normalized age; The age of the person to be evaluated; The minimum age of the sample or a preset lower age limit; This represents the maximum age of the sample or a preset upper age limit.
[0158] Normalized homocysteine levels It can be obtained using the following formula:
[0159] ;
[0160] In the formula, This represents the normalized homocysteine level. The homocysteine level of the subjects to be evaluated; This represents the minimum or preset lower limit of homocysteine levels in the sample. This represents the maximum or preset upper limit of homocysteine levels in the sample.
[0161] In practical applications, , , , , and The risk factor can be determined through correlation analysis or regression models of training samples, or it can be calibrated based on data from within the medical institution. If a certain risk factor is missing, the system can handle it using a missing data flag and indicate in the output that the corresponding risk factor was not included in the calculation.
[0162] (vi) Risk index integration and risk stratification
[0163] like Figure 6 As shown, the risk fusion module 106 will normalize the white matter lesion burden level. Cognitive Impairment Indicators Peripheral microcirculation disorder indicators and vascular risk factor indicators By fusing the data, a risk index for cognitive function related to white matter lesions can be generated. The specific calculation method is as follows:
[0164] ;
[0165] In the formula, A risk index for cognitive function related to white matter lesions; Normalized white matter lesion burden level; As an indicator of the degree of cognitive impairment; This is an indicator of peripheral microcirculatory disturbances; Indicators of vascular risk factors; The weights corresponding to the white matter lesion burden levels; The weights corresponding to the indicators of cognitive impairment degree; The weights corresponding to the peripheral microcirculation disorder indicators; The weights corresponding to the vascular risk factor indicators; , , and The sum is 1.
[0166] The aforementioned fusion method integrates central nervous system imaging load, current cognitive performance, peripheral microcirculation status, and systemic vascular risk factors into a single risk index. Compared to methods that only output MRI grading or only output abnormal scale indications, this implementation can reflect the contribution of multiple dimensions to risk in a single output result.
[0167] Risk stratification can be achieved through a first risk threshold. Second risk threshold Completed. When the risk index... Less than the first risk threshold When the risk index is low, the system outputs low-risk stratification information; when the risk index is high... Not less than the first risk threshold And less than the second risk threshold When the risk index is high, the system outputs risk stratification information; when the risk index is low... Not less than the second risk threshold At that time, the system outputs high-risk stratification information.
[0168] in, This serves as the dividing line between low and medium risk. This serves as the dividing line between medium and high risk. Less than The threshold can be determined based on the risk index distribution of historical samples, subject operating characteristic curves, quantile rules, or pre-set rules of the medical institution. To ensure feasibility, the system can preset a default threshold or recalibrate the threshold after accumulating local data.
[0169] (vii) Generation of indicator contribution information
[0170] To improve the interpretability of risk stratification results, this implementation method outputs a risk index. Simultaneously, indicator contribution information is generated. This information includes contributions from white matter lesion burden levels, cognitive impairment levels, peripheral microcirculatory disturbances, and vascular risk factors.
[0171] The contribution factor for the white matter lesion burden level is generated according to the following formula:
[0172] ;
[0173] In the formula, The contribution of the white matter lesion burden level to the risk index; The weights corresponding to the white matter lesion burden levels; This represents the normalized white matter lesion burden level.
[0174] The contribution factors of the cognitive impairment severity index are generated according to the following formula:
[0175] ;
[0176] In the formula, The contribution of the cognitive impairment level indicator to the risk index; The weights corresponding to the indicators of cognitive impairment degree; This is an indicator of the degree of cognitive impairment.
[0177] The contribution of the peripheral microcirculation disorder index is generated according to the following formula:
[0178] ;
[0179] In the formula, The contribution of peripheral microcirculation disorder indicators to the risk index; The weights corresponding to the peripheral microcirculation disorder indicators; It is an indicator of peripheral microcirculatory disturbance.
[0180] The contribution items of vascular risk factor indicators are generated according to the following formula:
[0181] ;
[0182] In the formula, The contribution of vascular risk factor indicators to the risk index; The weights corresponding to the vascular risk factor indicators; It is an indicator of vascular risk factors.
[0183] Through output , , and The system can not only indicate the risk level of the subject being evaluated, but also suggest whether the risk primarily stems from imaging burden, decreased cognitive scores, peripheral microcirculatory abnormalities, or vascular risk factors. This information helps clinicians determine whether further MRI image review, cognitive scale retesting, supplementary fundus examination, or further assessment of vascular risk factors is necessary.
[0184] In one implementation, when the peripheral microcirculation impairment index Mid-retinal vascular classification When the contribution is high, the system outputs a peripheral microcirculation abnormality indication; when the white matter lesion burden level and fundus vascular classification When the levels exceed a preset threshold, the system outputs a warning indicating the coexistence of white matter lesion burden and peripheral microcirculatory abnormalities. This warning does not constitute a disease diagnosis, but rather indicates that the subject's multimodal data exhibits relatively consistent risk signals.
[0185] (viii) Multimodal consistency verification
[0186] like Figure 7 As shown, the consistency verification unit 109 checks the white matter lesion burden level. Fundus vascular classification Sublingual vein classification Cognitive Impairment Indicators and vascular risk factor indicators Generate multimodal consistency verification results.
[0187] When the leukoencephalopathy burden level Increased, fundus vascular grading Elevated, and indicators of cognitive impairment When the cognitive impairment threshold is exceeded, the system generates a multimodal consistency enhancement marker. This marker indicates that the imaging burden, peripheral microcirculation abnormalities, and cognitive scale results are basically consistent in the risk direction, suggesting that the risk stratification result has high verification value.
[0188] When the leukoencephalopathy burden level Elevated but peripheral microcirculatory disturbance indicators The threshold for microcirculation or the peripheral microcirculation impairment index was not exceeded. Elevated but white matter lesion burden level If the image threshold is not exceeded, the system generates a multimodal inconsistency review marker. This marker indicates potential issues such as inconsistent examination times, insufficient image quality, single-modal abnormalities, scoring bias, or other clinical factors, requiring manual review in conjunction with the original examination data.
[0189] Setting up multimodal consistency checks helps prevent the system from directly interpreting a single abnormal indicator as an increase in overall risk, and also helps to prompt clinicians to pay attention to contradictory information between different modalities. This approach improves the robustness and interpretability of auxiliary assessment results.
[0190] (ix) Methods for determining weights and thresholds
[0191] The weights in this invention include , , , , , , , , , , , , and In one implementation, the weights can be determined based on correlation coefficients, regression coefficients, or model importance metrics from the training samples.
[0192] For example, for several candidate indicators The system can determine the initial weights based on the correlation strength between the target cognitive score or the white matter lesion burden:
[0193] ;
[0194] In the formula, For the first Initial weights for each candidate indicator; For the first The correlation coefficient or regression coefficient between each candidate indicator and the target variable; It is the sum of the absolute values of the correlation coefficients or regression coefficients of all candidate indicators; and The candidate indicator number.
[0195] In another implementation, multiple linear regression, logistic regression, random forest, gradient boosting tree, or other interpretable models can be used to determine the weights of each indicator. If a machine learning model is used to determine the weights, the model version, training sample range, training date, list of input variables, and threshold setting method should be saved in the system configuration for future reference.
[0196] First risk threshold Second risk threshold The threshold can be determined based on the risk index distribution of the training samples, or it can be determined based on sensitivity, specificity, or quantile rules. For initial deployment scenarios, a default threshold can be set using historical sample statistics; for subsequent expansion scenarios, the threshold can be recalibrated based on local samples to adapt the system to different population sources and inspection conditions.
[0197] IV. Application Examples and Experimental Data
[0198] (I) Sources and application scenarios of experimental data
[0199] This application example is based on 99 patients with white matter lesions who were hospitalized in the Department of Neurology of our hospital and completed head MRI examination, cognitive function assessment, and peripheral microcirculation evaluation. All subjects underwent head MRI examination, and the image quality met the requirements for the Fazekas visual rating scale; they also completed the Mini-Mental State Examination. Montreal Cognitive Assessment Scale score The assessment includes fundus vascular evaluation, sublingual vascular evaluation, and collection of data on vascular risk factors. These vascular risk factors include age, sex, body mass index, smoking history, alcohol consumption history, history of hypertension, history of diabetes, history of coronary heart disease, uric acid, dyslipidemia, C-reactive protein, and homocysteine levels.
[0200] This application example verifies the feasibility and technical effectiveness of the cognitive function risk stratification method related to white matter lesions described in this invention in actual clinical data. This method does not directly output disease diagnostic conclusions. Instead, it converts head MRI image load, cognitive scale scores, fundus vascular grading, hypolingual vascular grading, and vascular risk factors into structured indicators. Through a risk fusion module, it outputs risk stratification information, indicator contribution information, and multimodal consistency verification results, providing auxiliary reference for clinical review.
[0201] (II) Structured processing of multimodal data based on 99 samples
[0202] In this application example, 99 subjects were graded according to their white matter lesion burden based on their head MRI imaging findings. Group 13 cases, Group 55 cases, Group 23 cases, The group consisted of 8 cases. The system first read the head MRI grading data, cognitive scale data, fundus vascular grading data, sublingual vein grading data, and clinical risk factor data of each study subject through the data receiving module.
[0203] For head MRI image data, the system follows... Figure 4 The procedure shown reads the grade of periventricular white matter lesions. and deep white matter lesion grade And generate a white matter lesion burden level. At the same time, and When, it is generated according to the following formula :
[0204] ;
[0205] In the formula, The level of white matter lesion burden; Grade of periventricular white matter lesions; Grade of deep white matter lesions; This indicates that the higher of the input levels should be selected.
[0206] For cognitive scale data, the system reads... and Scoring, and assigning grades based on years of education. After correction, the following results were obtained. Then, the system will and Converted into an indicator of cognitive impairment :
[0207] ;
[0208] In the formula, As an indicator of the degree of cognitive impairment; Scoring using the Mini-Mental State Examination; The score is based on the education level-adjusted Montreal Cognitive Assessment Scale. The maximum score on the cognitive scale is 30. for The corresponding weights; for The corresponding weights; and The sum is 1.
[0209] For peripheral microcirculation data, the system reads fundus vascular grading. and sublingual vein classification and in accordance with Figure 5 The process shown generates peripheral microcirculation barrier indicators. :
[0210] ;
[0211] In the formula, This is an indicator of peripheral microcirculatory disturbances; Grading of retinal blood vessels; This is the lowest level in the retinal vascular grading system. This is the highest level of retinal vascular grading. The sublingual veins are graded; This is the lowest level in the classification of sublingual veins; This is the highest level of the sublingual vein classification; Weights corresponding to the retinal vascular grading; The weights corresponding to the grading of the sublingual veins; and The sum is 1, and Greater than .
[0212] After adopting the above processing method, the heterogeneous data originally scattered in MRI images, cognitive scales, fundus images, sublingual images, and clinical history were uniformly transformed into structured indicators that can be calculated, fused, and output contribution items, providing a basis for subsequent risk indexing. This provides the data foundation for the generation.
[0213] (III) Verification of the relationship between white matter lesion burden and cognitive function scores
[0214] To verify the invention's classification of white matter lesion burden levels Inclusion in risk index The necessity of different grading groups for white matter lesions and The scores were compared, and the results are shown in Table 1. This can be combined with... Figure 8 Please provide an explanation.
[0215] Table 1 Comparison of cognitive scores among different white matter lesion grading groups
[0216]
[0217] As shown in Table 1, with the increase in the grade of white matter lesions, and The ratings are all trending downwards. Especially... Group, The median is 16.50. The median score was 8.00, significantly lower than that of the low-grade group. This result indicates that the white matter lesion burden level... There is a clear correlation with cognitive function scores. As a risk index The core input indicators are based on clinical data.
[0218] Further and Merge as a low-load group, and The results, shown in Table 2, were obtained by merging them into a high-load group.
[0219] Table 2 Comparison of cognitive scores between the low-load and high-load groups of leukoencephalopathy.
[0220]
[0221] As can be seen from Table 2, the high-load group and The scores of all groups were lower than those of the low-burden group, and the differences were statistically significant. This result indicates that, in the risk fusion process of this invention, the white matter lesion burden level is considered... As a risk index The components of this study can effectively reflect the contribution of central imaging load to cognitive function risk.
[0222] Further Spearman correlation analysis revealed that the burden of white matter lesions... and A negative correlation was observed, with a correlation coefficient of -0.458 (p < 0.001); white matter lesion burden. and A negative correlation was observed, with a correlation coefficient of -0.460 (p < 0.001). This result indicates a relatively stable correlation between increased white matter lesion burden and decreased cognitive scores, supporting the inclusion of image burden processing and cognitive scale processing modules in this invention from a data perspective.
[0223] (iv) Verification of the relationship between peripheral microcirculation indicators and cerebral white matter lesion burden
[0224] To verify the present invention's classification of fundus blood vessels and sublingual vein classification Including indicators of peripheral microcirculation disorders The necessity of this study was demonstrated by comparing the fundus vascular grading and sublingual vascular grading of different white matter lesion groups. The results are shown in Table 3 and can be combined with... Figure 9 Please provide an explanation.
[0225] Table 3 Comparison of peripheral microcirculation indicators among different grading groups of white matter lesions
[0226]
[0227] As shown in Table 3, the grade of fundus vessels showed a significant increasing trend with the increase in the grade of white matter lesions, and the differences between groups were statistically significant. Specifically, The fundus vascular grading for this group was 1.00±0.00. The group showed an increase to 2.90±0.55, suggesting a strong correlation between changes in fundus vascular morphology and increased white matter lesion burden. Although the sublingual vascular grading showed a certain upward trend with the increase in white matter lesion grading, the difference between groups did not reach a statistically significant level, and its function is more suitable as a supplementary indicator of peripheral microcirculation status.
[0228] The results of further correlation analysis are shown in Table 4.
[0229] Table 4. Correlation analysis of white matter lesion burden, cognitive score, and peripheral microcirculation indicators.
[0230]
[0231] As shown in Table 4, fundus vascular grading is related to the burden of white matter lesions. The correlation coefficient was 0.486, and it was consistent with... and Both showed a negative correlation. In contrast, the sublingual vein grading and The correlation coefficient was 0.182, indicating a weak correlation with cognitive scores. This result suggests that the correlation with cognitive scores is relatively weak when constructing an index for peripheral microcirculatory disturbances. At that time, the retinal blood vessels were graded. weight Greater than the sublingual vein classification weight It is supported by experimental data.
[0232] Therefore, this invention does not simply use the fundus vessels and sublingual vascular grading as two peripheral indicators in parallel. Instead, based on the correlation strength between these two indicators and the burden of white matter lesions and cognitive scores, it adopts a fusion approach, using fundus vessel grading as the primary peripheral microcirculation indicator and sublingual vascular grading as a supplementary peripheral microcirculation indicator. This design avoids the negative impact of low-correlation indicators on the risk index. It can produce excessive effects, while retaining the value of the sublingual veins as an auxiliary indicator for observing the state of the body's microcirculation.
[0233] (v) Auxiliary verification of the role of vascular risk factors in risk stratification
[0234] This application example also analyzes the relationship between vascular risk factors and white matter lesion burden and cognitive scores. The results show that, with... As the severity grade increased, age gradually increased, and the differences between groups were significant (statistic 32.75, p < 0.001). The proportion of hypertension was significantly higher in the high-grade group (statistic 11.79, p = 0.008). Uric acid and dyslipidemia varied among different groups. The differences between the grades were also statistically significant, with p-values of 0.046 and 0.006, respectively; homocysteine levels varied with... The grading showed an increasing trend, p=0.068.
[0235] Correlation analysis showed that age and The correlation coefficient was 0.521, and... The correlation coefficient was -0.532, and... The correlation coefficient was -0.540; history of hypertension and The correlation coefficient was 0.314; homocysteine and The correlation coefficient was 0.267. These results indicate an association between vascular risk factors such as age, history of hypertension, and homocysteine levels and the burden of white matter lesions in the brain. Therefore, these factors should be included in the vascular risk factor index. It can supplement systemic risk background information beyond simple imaging grading and peripheral microcirculation grading.
[0236] In this invention, vascular risk factor indicators It can be generated using the following formula:
[0237] ;
[0238] In the formula, Indicators of vascular risk factors; The normalized age; Marker of a history of hypertension; Markers of diabetes history; Marker of a history of heart disease; This represents the normalized homocysteine level. Markers for dyslipidemia; , , , , and These are the weights of the corresponding risk factors, and the sum of the weights is 1.
[0239] This formula enables the system to transform discrete medical history variables and continuous laboratory variables into a unified risk background indicator, thus making the risk index... It not only reflects the current imaging and cognitive status, but also the background of systemic vascular risk associated with cerebral small vessel disease.
[0240] (vi) Risk index integration and indicator contribution output
[0241] Based on the above experimental data, the system follows... Figure 6 The flowchart shown classifies the leukoencephalopathy burden level. Cognitive Impairment Indicators Peripheral microcirculation disorder indicators and vascular risk factor indicators Integration into a risk index :
[0242] ;
[0243] In the formula, A risk index for cognitive function related to white matter lesions; Normalized white matter lesion burden level; As an indicator of the degree of cognitive impairment; This is an indicator of peripheral microcirculatory disturbances; Indicators of vascular risk factors; The weights corresponding to the white matter lesion burden levels; The weights corresponding to the indicators of cognitive impairment degree; The weights corresponding to the peripheral microcirculation disorder indicators; The weights corresponding to the vascular risk factor indicators; , , and The sum is 1.
[0244] To facilitate clinical review, the system further outputs the contribution items of each indicator:
[0245] ;
[0246] In the formula, The contribution of the white matter lesion burden level to the risk index; The weights corresponding to the white matter lesion burden levels; This represents the normalized white matter lesion burden level.
[0247] ;
[0248] In the formula, The contribution of the cognitive impairment level indicator to the risk index; The weights corresponding to the indicators of cognitive impairment degree; This is an indicator of the degree of cognitive impairment.
[0249] ;
[0250] In the formula, The contribution of peripheral microcirculation disorder indicators to the risk index; The weights corresponding to the peripheral microcirculation disorder indicators; It is an indicator of peripheral microcirculatory disturbance.
[0251] ;
[0252] In the formula, The contribution of vascular risk factor indicators to the risk index; The weights corresponding to the vascular risk factor indicators; It is an indicator of vascular risk factors.
[0253] In actual output, the system can , , and Displaying the data in a bar chart or sorted by contribution is recommended; it is suggested that the output be presented in this format. Figure 11 .pass Figure 11 Doctors or researchers can visually see the risk index. The main source. For example, when and When both are high, it suggests that the subject has both high imaging load and significant peripheral microcirculatory abnormalities; when Higher but A low level suggests that the abnormality in the cognitive scale is not entirely consistent with the imaging load, and further verification by considering other factors is needed; when A higher level suggests that vascular risk factors such as age, hypertension, or homocysteine levels contribute significantly to the risk index.
[0254] This contribution output method avoids the problem that traditional scoring results only provide a single risk level without explanatory information, making the risk stratification results of this invention traceable, interpretable, and easy for manual review.
[0255] (vii) Verification of multimodal consistency
[0256] In this application example, the system further... Figure 7 The illustrated process performs multimodal consistency verification. When the white matter lesion burden level... Increased, fundus vascular grading Elevated, and indicators of cognitive impairment During synchronous elevation, the system generates a multimodal consistency enhancement marker. Based on the experimental data, The correlation coefficient with fundus vascular grading was 0.486. and The correlation coefficient was -0.458. and The correlation coefficient was -0.460, and the fundus vascular grading was related to... The correlation coefficient was -0.420, indicating a positive risk relationship between image burden, abnormal fundus microcirculation, and cognitive score decline. Therefore, for those who simultaneously possess... rise, Increase and The enhanced multimodal consistency markers for the elevated objects are supported by experimental data.
[0257] Conversely, when an object behaves as rise but No increase, or rise but When the risk level is not elevated, the system generates a multimodal inconsistency check marker. This marker is used to indicate potential differences in examination time, image quality issues, local fundus lesions, the influence of educational level on cognitive scales, or other clinical factors. Therefore, this invention does not directly equate a single modality abnormality with high risk, but rather improves the robustness of the output results through a consistency check mechanism.
[0258] As can be seen from the above application examples and experimental data, the present invention achieves at least the following technical effects.
[0259] First, this invention can transform the relationship between the imaging burden of white matter lesions and cognitive scores into a calculable indicator. Experimental data show that, with... Graded increase, and All scores decreased significantly, especially in the high-load group. and Significantly lower than the low-load group, and and , All showed a significant negative correlation. Therefore, and Inclusion in risk index It can reflect the association between central imaging load and cognitive decline.
[0260] Secondly, this invention can incorporate peripheral microcirculatory abnormalities into the risk stratification of cognitive function related to white matter lesions. Experimental data show that fundus vascular grading increases with... The grade of the fundus vessels increased with increasing vascularity, and the difference between groups was statistically significant; fundus vascular grading and... Positively correlated with and A negative correlation was observed. This result indicates that fundus vascular grading... It can serve as an important input indicator of the peripheral microcirculation status. Sublingual vein grading and... It shows a weak correlation, making it suitable as a supplementary indicator for peripheral microcirculation. Therefore, this invention sets... Greater than It can reasonably allocate the internal weights of peripheral microcirculation indicators based on experimental data.
[0261] Furthermore, this invention can incorporate vascular risk factors such as age, hypertension, and homocysteine into a unified risk context. Experimental data show that age, history of hypertension, and homocysteine are associated with... There is a correlation between them, indicating that The settings can supplement systemic risk information beyond imaging, cognition, and peripheral microcirculation.
[0262] Finally, this invention uses a risk index. Indicator Contribution Items , , , Furthermore, the multimodal consistency verification results ensure that the output not only has risk stratification capabilities but also interpretability and verification prompts. Compared to simple MRI grading, simple cognitive scale scoring, or single fundus assessment, this invention can integrate multi-source clinical information through a unified data processing workflow, providing more complete data support for the auxiliary identification of cognitive function risks related to white matter lesions.
[0263] The foregoing description of embodiments of the present invention, through which those skilled in the art are able to implement or use the present invention, will be readily apparent to those skilled in the art. Various modifications to these embodiments will be readily apparent to those skilled in the art. The general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the invention. Therefore, the present invention is not to be limited to the embodiments shown herein, but is to be accorded the widest scope consistent with the principles and novelty disclosed herein.
[0264] Those skilled in the art will understand that embodiments of this application can be provided as methods, systems, or computer program products. Therefore, this application can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, this application can take the form of a computer program product embodied on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.
[0265] This application is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of this application. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart... Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.
[0266] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1The function specified in one or more boxes.
[0267] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.
[0268] In a typical configuration, a computing device includes one or more processors (CPU), input / output interfaces, network interfaces, and memory.
[0269] Memory may include non-persistent memory in computer-readable media, such as random access memory (RAM) and / or non-volatile memory, such as read-only memory (ROM) or flash RAM. Memory is an example of computer-readable media.
[0270] Computer-readable media includes both permanent and non-permanent, removable and non-removable media that can store information using any method or technology. Information can be computer-readable instructions, data structures, modules of programs, or other data. Examples of computer storage media include, but are not limited to, phase-change memory (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), flash memory or other memory technologies, CD-ROM, digital versatile optical disc (DVD) or other optical storage, magnetic tape, magnetic disk storage or other magnetic storage devices, or any other non-transferable medium that can be used to store information accessible by a computing device. As defined herein, computer-readable media does not include transient computer-readable media, such as modulated data signals and carrier waves.
Claims
1. A multimodal data processing method for risk stratification of cognitive function related to white matter lesions, characterized in that, The method, executed by a computer device, is used to output risk stratification information for clinical reference based on collected clinical examination data, without directly generating disease diagnosis conclusions or treatment plans. The method includes the following steps: S1. Receive multimodal data of the subject to be evaluated, including head MRI image evaluation data, cognitive scale data, peripheral microcirculation evaluation data, and vascular risk factor data; S2. Assessing a white matter lesion load grade from the head MRI image data , Determined according to the Fazekas visual rating scale S3, obtaining a mini-mental state examination scale score from the cognitive scale data and a montreal cognitive assessment scale score corrected for education level and generating a cognitive impairment severity indicator ; S4. Obtaining an ocular fundus blood vessel grading from the peripheral microcirculation assessment data and sublingual collateral grading and generating a peripheral microcirculation disorder indicator wherein, the weight of is greater than the weight of ; S5, generating a vascular risk factor indicator from the vascular risk factor data , the vascular risk factor data comprising one or more of age, history of hypertension, history of diabetes, history of heart disease, dyslipidemia, and homocysteine level; S6. The level of white matter lesion burden. Cognitive Impairment Indicators Peripheral microcirculation disorder indicators and vascular risk factor indicators By performing fusion, a cognitive function risk index related to white matter lesions was obtained. And according to the risk index Output risk stratification information for low-risk, medium-risk, or high-risk risks, along with corresponding indicator contribution information.
2. The method according to claim 1, characterized in that, In S1, the multimodal data comes from examination data completed within the same hospitalization cycle or the same preset assessment cycle. When the head MRI image assessment data, cognitive scale data, peripheral microcirculation assessment data, or vascular risk factor data are missing, the image quality does not meet the preset clarity requirements, the identity identification is inconsistent, or the acquisition time exceeds the preset assessment cycle, a quality mark is generated for the corresponding data, and a data integrity prompt is output simultaneously when the risk stratification information is output.
3. The method according to claim 1, characterized in that, In S2, the head MRI imaging assessment data includes the grade of periventricular white matter lesions. and deep white matter lesion grade ; The grade of periventricular white matter lesions The grading is based on the presence or absence of high signal in the periventricular white matter, cap-shaped high signal, ring-shaped high signal, halo-shaped white matter lesions, degree of irregularity of the margins, and degree of extension into the deep white matter; The level of deep white matter lesions The lesions are classified according to the presence or absence of deep white matter lesions, punctate lesions, small patchy lesions, extensive patchy lesions, early fusion tendency, and large fused lesions. When the periventricular white matter lesion grade is obtained simultaneously and the aforementioned deep white matter lesion grade At that time, the white matter lesion burden level was determined as follows: : In the formula, The level of white matter lesion burden; Grade of periventricular white matter lesions; Grade of deep white matter lesions; This indicates that the higher of the input levels should be selected.
4. The method according to claim 1, characterized in that, In S3, the education-adjusted Montreal Cognitive Assessment Scale score The scores are obtained as follows: when the years of education of the subject to be assessed are less than the preset years, a preset correction score is added to the original Montreal Cognitive Assessment Scale score; when the years of education of the subject to be assessed are not less than the preset years, the original Montreal Cognitive Assessment Scale score is retained. The cognitive impairment level index Generate according to the following formula: ; In the formula, As an indicator of the degree of cognitive impairment; Scoring using the Mini-Mental State Examination; The score is based on the education level-adjusted Montreal Cognitive Assessment Scale. The maximum score on the cognitive scale is 30. Weights corresponding to the scores of the Mini-Mental State Examination; The weights corresponding to the scores of the Montreal Cognitive Assessment Scale; and The sum is 1.
5. The method according to claim 1, characterized in that, In S4, the fundus vascular grading Determined based on bilateral fundus images acquired by a non-mydriatic color fundus camera; Fundus vascular grading It should be determined based on at least one or more of the following fundus vascular morphological features: enhanced retinal artery wall reflex, copper wire-like changes, silver wire-like changes, arteriovenous crossing sign, vascular tortuosity, hard exudates, cotton wool spots, and hemorrhage spots. Furthermore, the fundus vascular grading The grading system is progressive, based on the following stages: normal fundus, retinal arteriosclerosis, hard exudates, cotton wool spots, and hemorrhage spots. And / or, in S4, the sublingual vascular grading Determined based on sublingual images captured by a tongue image acquisition device; The sublingual vein classification It should be determined based on at least one or more morphological features of the main trunk of the sublingual veins, including the degree of fullness, length, tortuosity, branching, punctate dilated vascular bulbs, and grape-bead-like changes. Furthermore, the sublingual vascular grading... The classification is progressive, based on the following criteria: normal, main trunk full and elongated, main trunk significantly full, elongated and tortuous, and presence of punctate dilated globules or beaded changes. And / or, in S4, the peripheral microcirculation disorder index Generate according to the following formula: ; In the formula, This is an indicator of peripheral microcirculatory disturbances; Grading of retinal blood vessels; This is the lowest level in the retinal vascular grading system. This is the highest level of retinal vascular grading. The sublingual veins are graded; This is the lowest level in the classification of sublingual veins; This is the highest level of the sublingual vein classification; Weights corresponding to the retinal vascular grading; The weights corresponding to the grading of the sublingual veins; and The sum is 1, and Greater than .
6. The method according to claim 1, characterized in that, In S5, the vascular risk factor indicators Generate according to the following formula: ; In the formula, Indicators of vascular risk factors; The normalized age; Marker of a history of hypertension; Markers of diabetes history; Marker of a history of heart disease; This represents the normalized homocysteine level. Markers for dyslipidemia; , , , , and These are the weights of the corresponding risk factors, and the sum of all weights is 1; The normalized age and the normalized homocysteine level The following formulas are used to obtain the results: In the formula, The normalized age; The age of the person to be evaluated; The minimum age of the sample or a preset lower age limit; The maximum age of the sample or a preset upper age limit; ; In the formula, This represents the normalized homocysteine level. The homocysteine level of the subjects to be evaluated; This represents the minimum or preset lower limit of homocysteine levels in the sample. This represents the maximum or preset upper limit of homocysteine levels in the sample.
7. The method according to claim 1, characterized in that, In S6, the risk index for cognitive function related to white matter lesions. Generate according to the following formula: ; In the formula, A risk index for cognitive function related to white matter lesions; Normalized white matter lesion burden level; As an indicator of the degree of cognitive impairment; This is an indicator of peripheral microcirculatory disturbances; Indicators of vascular risk factors; The weights corresponding to the white matter lesion burden levels; The weights corresponding to the indicators of cognitive impairment degree; The weights corresponding to the peripheral microcirculation disorder indicators; The weights corresponding to the vascular risk factor indicators; , , and The sum is 1; The normalized white matter lesion burden level It is obtained using the following formula: ; In the formula, Normalized white matter lesion burden level; The level of white matter lesion burden; This is the lowest level of the white matter lesion burden gradation; This represents the highest level of white matter lesion burden. And / or, in S6, based on the first risk threshold Second risk threshold Risk index for cognitive function related to the aforementioned white matter lesions Conduct risk stratification; When the risk index Less than the first risk threshold At that time, output low-risk stratification information; When the risk index Not less than the first risk threshold And less than the second risk threshold At that time, output risk stratification information; When the risk index Not less than the second risk threshold At that time, output high-risk stratification information; in, This serves as the dividing line between low and medium risk. This serves as the dividing line between medium and high risk. Less than ; And / or, in S6, the indicator contribution information includes at least the level of white matter lesion burden. Risk index Contribution items, cognitive impairment indicators Risk index Contribution items, peripheral microcirculation disorder indicators Risk index Contribution items and vascular risk factor indicators Risk index Contributions; Among them, when the peripheral microcirculation barrier index Mid-retinal vascular classification When the contribution of a factor exceeds a preset contribution threshold, a peripheral microcirculation abnormality alert is output in the risk stratification information; when the white matter lesion burden level... and the aforementioned fundus vascular grading When the risk level exceeds the corresponding preset level, a prompt indicating the coexistence of cerebral white matter lesion burden and peripheral microcirculation abnormality is output in the risk stratification information. And / or, in S6, the indicator contribution information is generated in the following manner: ; In the formula, The contribution of the white matter lesion burden level to the risk index; The weights corresponding to the white matter lesion burden levels; Normalized white matter lesion burden level; In the formula, The contribution of the cognitive impairment level indicator to the risk index; The weights corresponding to the indicators of cognitive impairment degree; As an indicator of the degree of cognitive impairment; ; In the formula, The contribution of peripheral microcirculation disorder indicators to the risk index; The weights corresponding to the peripheral microcirculation disorder indicators; This is an indicator of peripheral microcirculatory disturbances; ; In the formula, The contribution of vascular risk factor indicators to the risk index; The weights corresponding to the vascular risk factor indicators; Indicators of vascular risk factors; And / or, S6 further includes, based on the said white matter lesion burden level Fundus vascular classification Sublingual vein classification Cognitive Impairment Indicators and vascular risk factor indicators Generate consistency verification results; When the white matter lesion burden level With the aforementioned fundus vascular grading All increased, and the indicators of cognitive impairment also increased. When the cognitive impairment threshold is exceeded, a multimodal consistency enhancement marker is generated; When the white matter lesion burden level Elevated but the peripheral microcirculation disorder index The preset microcirculation threshold or the peripheral microcirculation barrier index was not exceeded. The level of white matter lesion burden is elevated, but the level of white matter lesion burden is elevated. If the preset image threshold is not exceeded, a multimodal inconsistency verification mark is generated.
8. A risk stratification system for cognitive function related to white matter lesions, characterized in that, The system is used to implement the method according to any one of claims 1-7, comprising: The data receiving module is used to receive head MRI image assessment data, cognitive scale data, peripheral microcirculation assessment data, and vascular risk factor data of the subject to be evaluated. The image burden processing module is used to obtain the white matter lesion burden level based on the head MRI image assessment data. ; The cognitive scale processing module is used to score based on the Mini-Mental State Examination. and the Montreal Cognitive Assessment Scale score adjusted for education level Generate cognitive impairment index ; The peripheral microcirculation processing module is used to classify retinal blood vessels. and sublingual vein classification Generate peripheral microcirculation disorder indicators ; The risk factor processing module is used to generate vascular risk factor indicators based on one or more of the following factors: age, history of hypertension, history of diabetes, history of heart disease, dyslipidemia, and homocysteine level. ; The risk fusion module is used to determine the level of white matter lesion burden based on the risk level. Cognitive Impairment Indicators Peripheral microcirculation disorder indicators and vascular risk factor indicators Generate a risk index for cognitive function related to white matter lesions. ; Output module, used to output based on the risk index Output risk stratification information, data integrity prompts, indicator contribution information, and multimodal consistency verification results.
9. An electronic device, characterized in that, The method includes a processor, a memory, and a computer program stored in the memory and executable on the processor, wherein when the processor executes the computer program, it implements the method according to any one of claims 1 to 7.
10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by a processor, it implements the method described in any one of claims 1 to 7.
Citation Information
Patent Citations
Automated quantitative analysis system and interpretation method for brain white matter lesions
CN113077887B
White matter high-signal focus area segmentation and rating method
CN115880287A
