CN-121661047-B - Intelligent brain iron deposition space distribution pattern recognition and parting method and system thereof

CN121661047BCN 121661047 BCN121661047 BCN 121661047BCN-121661047-B

Abstract

The invention relates to the technical field of medical image processing and artificial intelligence, and discloses an intelligent brain iron deposition spatial distribution mode identification and parting method and a system thereof, wherein the method comprises the steps of carrying out spatial standardization and brain region segmentation on a whole brain QSM image; the method comprises the steps of generating a spatial distribution feature vector through a multi-scale spatial attention feature extraction network, dividing an iron deposition mode into a leather layer type, a deep nucleus type, a ventricle surrounding type and a diffuse type by adopting a spatial neighborhood constraint fuzzy clustering algorithm, establishing a mapping relation between the iron deposition mode and a vascular cognitive impairment etiology subtype, evaluating bilateral hemispheric iron deposition asymmetry, realizing automatic identification and typing of the brain iron deposition spatial distribution mode, achieving the mode identification accuracy of more than 85%, and providing an auxiliary diagnosis basis for the vascular cognitive impairment etiology typing.

Inventors

MA TAO
GUO ZIMING
Jiao Zunlong
JI CHENGHONG
ZHU XICHEN
LU YINGTONG
LI RONG

Assignees

江南大学

Dates

Publication Date: 20260508
Application Date: 20260205

Claims (10)

1. The brain iron deposition space distribution pattern intelligent identification and parting method is characterized by comprising the following steps of: Step S1, QSM image preprocessing, namely acquiring a full brain quantitative magnetic sensitive image of a subject, performing spatial standardization processing on the full brain quantitative magnetic sensitive image, registering the full brain quantitative magnetic sensitive image to a standard brain template space, and performing brain segmentation on the registered image to generate a brain segmentation mask; S2, a spatial distribution feature extraction step, namely inputting the preprocessed quantitative magnetic susceptibility graph image and the brain regions into a multi-scale spatial attention feature extraction network, extracting iron deposition spatial distribution features in different receptive field ranges through a multi-scale convolution layer, carrying out weighted fusion on the features of different brain regions through a spatial attention mechanism to generate spatial distribution feature vectors, and counting magnetic susceptibility value distribution features of each region based on the brain regions, so as to generate regional magnetic susceptibility statistical features; s3, an iron deposition mode clustering and typing step, namely inputting the spatial distribution feature vector and the partitioned magnetic sensitivity statistical features into a self-adaptive iron deposition mode clustering model, dividing the iron deposition spatial distribution mode into a leather layer type, a deep nucleus type, a ventricle surrounding type and a diffuse type based on a fuzzy clustering algorithm of spatial neighborhood constraint, and outputting an iron deposition mode label and a mode confidence score; Step S4, an asymmetry index analysis step, namely dividing a quantitative magnetic sensitivity map image into a left hemisphere region and a right hemisphere region based on the brain region division mask, respectively counting magnetic sensitivity value distribution characteristics of the left hemisphere region and the right hemisphere region, and calculating asymmetry indexes of double hemispheres to quantify the iron deposition difference degree of the left hemisphere and the right hemisphere, wherein the asymmetry indexes are used for identifying asymmetric iron metabolism abnormality related to single vascular lesions; And S5, a causal subtype association mapping step, namely establishing association relation between the iron deposition mode and the causal subtype of the vascular cognitive disorder through a causal subtype probability mapping network based on the iron deposition mode label, the mode confidence score and the asymmetry index output in the step S4, outputting a causal subtype probability vector, and generating causal subtype auxiliary prompt information based on the causal subtype probability vector.
2. The method for intelligently identifying and typing a brain iron deposition spatial distribution pattern according to claim 1, wherein in the step S1, the brain region division mask comprises a cortex region, a deep nucleus region and a ventricle surrounding region, and the spatial normalization process comprises performing rigid registration on the whole brain quantitative magnetic susceptibility map image and a high resolution structural image to obtain a first registration transformation matrix, performing nonlinear registration on the high resolution structural image and a standard brain template to obtain a second registration transformation matrix, and transforming the whole brain quantitative magnetic susceptibility map image into the standard brain template space based on the first registration transformation matrix and the second registration transformation matrix.
3. The method for intelligently identifying and typing a brain iron deposition spatial distribution pattern according to claim 1, wherein in step S2, the multi-scale spatial attention feature extraction network includes a first convolution branch, a second convolution branch, and a third convolution branch in parallel, the convolution kernel of the first convolution branch having a size of 3 x 3, the convolution kernel size of the second convolution branch is 5 multiplied by 5, the convolution kernel size of the third convolution branch is 7 multiplied by 7, the outputs of the three convolution branches are input into the spatial attention module after being spliced by the channel, the spatial attention module generates a spatial attention weighting map based on the brain region segmentation mask to weight the stitching features.
4. The intelligent brain iron deposition spatial distribution pattern recognition and typing method according to claim 1, wherein in the step S3, the iron deposition pattern clustering typing step classifies the iron deposition spatial distribution pattern into four types, namely cortical type characterization iron deposition is intensively distributed in a brain cortical region, deep nucleolus type characterization iron deposition is intensively distributed in a basal ganglion region and a thalamus region, periventricular type characterization iron deposition is intensively distributed in a white matter region around a lateral ventricle, and diffuse type characterization iron deposition is distributed in the cortex, deep nucleolus and periventricular region.
5. The method according to claim 1, wherein in the step S2, the partitioned magnetic susceptibility statistical features include a mean value of magnetic susceptibility, a standard deviation of magnetic susceptibility, a bias of magnetic susceptibility, and kurtosis of magnetic susceptibility of each brain region, wherein each brain region includes frontal cortex, parietal cortex, temporal cortex, occipital cortex, caudate nucleus, putamen, globus pallidus, thalamus, hippocampus, and periventricular white matter.
6. The brain iron deposition spatial distribution pattern intelligent recognition and typing method according to claim 1, wherein in the step S3, the adaptive iron deposition pattern clustering model is implemented based on a spatial neighborhood constraint fuzzy clustering algorithm, the algorithm introduces a spatial neighborhood regularization term in an optimization objective function, so that spatial neighboring voxels tend to belong to the same iron deposition pattern, and the pattern confidence score indicates membership of the sample to each iron deposition pattern.
7. The method for intelligently identifying and typing the brain iron deposition spatial distribution pattern according to claim 1, wherein in the step S4, the calculation of the asymmetry index comprises calculating a magnetic sensitivity value average value of a left hemisphere area and a magnetic sensitivity value average value of a right hemisphere area, obtaining a whole brain asymmetry index based on the ratio of the difference value of the magnetic sensitivity value average value of the left hemisphere area and the magnetic sensitivity value average value of the right hemisphere area to the average value of the two values, and calculating the ratio of the difference value of the magnetic sensitivity value average value of the left brain area and the magnetic sensitivity value average value of the corresponding right brain area to the average value of the two values for the corresponding brain areas respectively, so as to obtain a partition asymmetry index vector.
8. The brain iron deposition spatial distribution pattern intelligent recognition and typing method according to claim 1, wherein in the step S5, the causative subtype of the vascular cognitive impairment includes multiple infarct type, small vessel disease type, low perfusion type and mixed type, the causative subtype probability mapping network receives the independent heat coding vector of the iron deposition pattern label, the pattern confidence score and the asymmetry index as input, outputs the causative subtype probability vector through a normalized index function after being processed by a plurality of fully connected layers, and the causative subtype corresponding to the largest component of the causative subtype probability vector as a prediction result.
9. The method for intelligently identifying and typing a brain iron deposition spatial distribution pattern according to claim 1, further comprising a closed-loop feedback optimization step of calculating a feedback error signal based on cross entropy loss between the etiology subtype probability vector and labeled real etiology subtype labels, and back-propagating the feedback error signal to the multi-scale spatial attention feature extraction network and the adaptive iron deposition pattern cluster model, and updating network parameters of the multi-scale spatial attention feature extraction network and cluster center vectors of the adaptive iron deposition pattern cluster model.
10. An intelligent brain iron deposition spatial distribution pattern recognition and typing system for implementing the brain iron deposition spatial distribution pattern intelligent recognition and typing method of any one of claims 1-9, comprising: The QSM image preprocessing module is used for acquiring a full brain quantitative magnetic sensitive image of a subject, carrying out space standardization processing on the full brain quantitative magnetic sensitive image and registering the full brain quantitative magnetic sensitive image to a standard brain template space, and carrying out brain region division on the registered image to generate a brain region division mask; The spatial distribution feature extraction module is used for inputting the preprocessed quantitative magnetic sensitivity map image and the brain region division mask into a multi-scale spatial attention feature extraction network to generate a spatial distribution feature vector and a regional magnetic sensitivity statistical feature; The iron deposition mode clustering and parting module is used for inputting the spatial distribution feature vector and the partitioned magnetic sensitivity statistical features into an adaptive iron deposition mode clustering model and outputting an iron deposition mode label and a mode confidence score; the etiology subtype association mapping module is used for outputting an etiology subtype probability vector and an etiology subtype auxiliary prompt message through an etiology subtype probability mapping network based on the iron deposition mode label, the mode confidence score and the asymmetry index; And the asymmetry index analysis module is used for calculating the asymmetry index of the bilateral hemispheres based on the distribution characteristics of the magnetic sensitivity values of the left hemispheric region and the right hemispheric region of the brain region division mask respectively.

Description

Intelligent brain iron deposition space distribution pattern recognition and parting method and system thereof Technical Field The invention relates to the technical field of medical image processing and artificial intelligence, in particular to an intelligent brain iron deposition spatial distribution pattern identification and parting method and a system thereof. Background Iron acts as an important metal element in the brain and plays a key role in axonal myelination, energy metabolism and neurotransmitter synthesis, while the iron content of specific brain regions gradually changes with age. When the iron homeostasis system is disrupted, a range of age-related neurodegenerative diseases may result. Previous studies have shown that abnormal brain iron deposition occurs in diseases such as Alzheimer's disease, parkinson's disease, multiple sclerosis and vascular cognitive dysfunction. Therefore, quantitative assessment and detection of spatially distributed features of brain iron deposition are of great importance for understanding the disease occurrence mechanism and guiding clinical interventions. Quantitative magnetic susceptibility mapping (Quantitative Susceptibility Mapping, QSM) is an advanced magnetic resonance imaging technique capable of non-invasively quantitatively assessing susceptibility distribution of in vivo tissues through complex image post-processing procedures and inversion algorithms. Since paramagnetic iron is the main source of gray matter magnetic susceptibility, the QSM technique becomes an ideal tool for noninvasive quantitative analysis of iron deposition in living bodies at present, and is widely applied to the research of nervous system diseases. Chinese patent publication No. CN116934662a discloses a data processing method and apparatus for whole brain quantitative magnetic sensitive magnetic resonance imaging. The method comprises the steps of performing QSM reconstruction on original gradient echo data, constructing a standardized brain template of an individual space by utilizing a structural image, performing registration processing on a tested QSM image and the brain template to complete space standardization, and performing voxel-based or region-of-interest-based data analysis. The method can automatically divide QSM images, realizes quantitative analysis of the magnetic sensitivity value, and improves repeatability and comparability. However, the technical scheme has the following defects that firstly, the method only realizes standardized processing and basic statistical analysis of QSM images, depth feature mining and pattern recognition cannot be carried out on spatial distribution patterns of brain iron deposition, different iron deposition spatial distribution types cannot be effectively distinguished, secondly, the method lacks a correlation analysis mechanism between the iron deposition patterns and disease etiology subtypes, auxiliary diagnosis support cannot be provided for the etiology typing of vascular cognitive impairment, thirdly, the method does not consider asymmetry evaluation of bilateral hemispherical iron deposition, asymmetric iron metabolism abnormality related to unilateral vascular lesions cannot be identified, and finally, the method lacks a cooperative optimization mechanism among processing modules, and the whole analysis performance cannot be improved through end-to-end learning. Therefore, an intelligent analysis method capable of automatically extracting brain iron deposition spatial distribution characteristics, identifying different iron deposition modes, establishing association between the iron deposition modes and etiology subtypes of vascular cognitive impairment and evaluating bilateral hemispheric iron deposition asymmetry is needed. Disclosure of Invention Aiming at the technical problems of strong heterogeneity of brain iron deposition spatial distribution patterns and difficult correlation analysis of the etiology types of vascular cognitive impairment in the prior art, the invention provides an intelligent brain iron deposition spatial distribution pattern identification and typing method and a system thereof. The first aspect of the invention provides a brain iron deposition spatial distribution pattern intelligent identification and typing method, which comprises the following steps: Step S1, a QSM image preprocessing step, namely acquiring a full brain quantitative magnetic susceptibility map image of a subject, performing spatial standardization processing on the full brain quantitative magnetic susceptibility map image, registering the full brain quantitative magnetic susceptibility map image to a standard brain template space, and performing brain segmentation on the registered image to generate a brain segmentation mask, wherein the brain segmentation mask comprises a cortex region, a deep nucleus region and a ventricle surrounding region. And S2, spatial distribution feature extraction, namely inputting the preprocesse