A method for processing brain magnetic resonance images of a population exposed to a profession

Abstract

The application discloses a brain magnetic resonance image processing method for a population with occupational exposure, and belongs to the technical field of medical image processing, and comprises the following steps: acquiring brain multi-sequence MRI image data and an exposure type label of a subject; selecting a corresponding target brain region prior template from an exposure type-brain region prior knowledge base; registering the multi-sequence MRI image data to a standard brain space to obtain registration transformation parameters, and mapping back to an individual space to obtain an individualized target brain region mask; further calculating eigenvalues of each preset index in the target brain region, and determining the deviation degree of each eigenvalue; based on the preset weight of the eigenvalue deviation degree of each preset index under the exposure type label, weighting and fusing the deviation degree of each eigenvalue to obtain a comprehensive abnormality index and output an abnormality classification. The application solves the problem of insufficient detection of local and occult damage caused by occupational exposure in the existing MRI analysis method, and can be used for occupational health monitoring and early screening.

Description

Technical Field

[0001] This invention belongs to the field of medical image processing technology, specifically relating to a brain magnetic resonance image processing method for occupationally exposed populations. Background Technology

[0002] In the context of health monitoring and occupational disease prevention for people with occupational exposure to heavy metals (such as manganese, lead, and mercury), brain magnetic resonance imaging (MRI) is increasingly used for imaging assessment of neurological damage. Different heavy metals accumulate in brain tissue and cause different damage patterns. For example, manganese tends to accumulate in the basal ganglia, while lead and mercury are more likely to affect areas such as the white matter and corpus callosum. Therefore, targeted and refined imaging analysis for specific exposure types is of great significance.

[0003] Currently, the analysis of brain MRI (Magnetic Resonance Imaging) images in occupationally exposed populations typically involves a general examination of the entire brain using multiple MRI sequences. Physicians then manually interpret the images, providing qualitative descriptions or a binary positive / negative assessment. Even when imaging indicators are extracted from specific brain regions, the analysis still considers the entire brain. Clearly, these methods are insufficient to accurately assess the localized and occult damage caused by specific exposures. Summary of the Invention

[0004] To address the aforementioned problems in the existing technology, this invention provides a method for processing brain magnetic resonance images in occupationally exposed populations. The technical problem to be solved by this invention is achieved through the following technical solution: This invention provides a method for processing brain magnetic resonance images of occupationally exposed populations, comprising: Acquire multi-sequence MRI images of the subject's brain, along with exposure type labels associated with the subject; Select a target brain region prior template corresponding to the exposure type label from the pre-built exposure type-brain region prior knowledge base. The target brain region is the anatomical region in the brain tissue where the toxin corresponding to the exposure type label is known to accumulate or is vulnerable. The multi-sequence MRI image data are registered to a standard brain space to obtain registration transformation parameters; Using the registration transformation parameters, the prior template of the target brain region is mapped back to the individual space of the subject to obtain an individualized target brain region mask; Based on the individualized target brain region mask, the feature values ​​of each preset index are calculated within the target brain region. The deviation of each feature value is determined according to the feature values ​​and the preset reference values ​​under the corresponding brain region and the corresponding preset index. Based on the preset weights of the deviations of the feature values ​​of each preset indicator under the exposure type label, the deviations of each feature value are weighted and fused to obtain a comprehensive anomaly index, and the anomaly classification result is output based on the comprehensive anomaly index.

[0005] In one embodiment of the present invention, before the step of registering the multi-sequence MRI image data to a standard brain space to obtain registration transformation parameters, the method further includes: The multi-sequence MRI image data is preprocessed to obtain preprocessed MRI image data; Calculate the signal-to-noise ratio (SNR) of the preprocessed MRI image data and compare the SNR with a preset SNR threshold; If the signal-to-noise ratio is less than the preset signal-to-noise ratio threshold, the quality of the preprocessed MRI image data is determined to be substandard, and re-acquisition is triggered. If the signal-to-noise ratio is equal to or greater than the preset signal-to-noise ratio threshold, the edge voxels of the brain parenchyma region in the preprocessed MRI image data are detected to see if they fall within the preset voxel range of the scanning field of view boundary, so as to determine whether the scanning coverage is complete. When the edge voxels of the brain parenchyma region fall within the preset voxel range, the coverage is determined to be incomplete and re-acquisition is triggered.

[0006] In one embodiment of the present invention, the step of preprocessing the multi-sequence MRI image data to obtain preprocessed MRI image data includes: Intensity normalization is performed on each sequence in the multi-sequence MRI image data to obtain normalized MRI image data; For the multi-sequence MRI image data containing the same sequence acquired at least twice, for the normalized MRI image data obtained from two adjacent repeated acquisitions, a normalized cross-correlation coefficient is calculated between the two; the normalized cross-correlation coefficient is compared with a first preset threshold; when the normalized cross-correlation coefficient is lower than the first preset threshold, it is determined that there is motion artifact between the normalized MRI image data obtained from the two adjacent repeated acquisitions, and the normalized MRI image data obtained from the latter acquisition is removed from the multi-sequence MRI image data; Based on a brain tissue probability model, the scalp, skull, and extracranial tissues are separated and removed from the MRI image data after motion artifact detection, while the brain parenchyma is preserved. The resulting MRI image data after skull removal is used as the preprocessed MRI image data.

[0007] In one embodiment of the present invention, the step of registering the multi-sequence MRI image data to a standard brain space to obtain registration transformation parameters includes: Using a standard brain atlas corresponding to the standard brain space as a fixed image, and structural images in the preprocessed MRI image data as floating images, affine registration is performed on the floating images to obtain registration transformation parameters that map the individual space of the subject to the standard brain space.

[0008] In one embodiment of the present invention, the step of mapping the prior template of the target brain region back to the individual subject space using the registration transformation parameters to obtain an individualized target brain region mask includes: Based on the registration transformation parameters, the prior template of the target brain region is transformed from the standard brain space to the individual space of the subject; The transformed target brain region prior template is resampled to obtain an individualized target brain region mask covering the corresponding anatomical region in the individual space of the subject.

[0009] In one embodiment of the present invention, the step of calculating feature values ​​of each preset index within the target brain region based on the individualized target brain region mask, and determining the deviation of each feature value according to the feature values, the corresponding brain region, and the preset reference value under the corresponding preset index, includes: Using the individualized target brain region mask as a constraint, voxels located in the target brain region are extracted from the preprocessed MRI image data; For each of the preset indices, calculate the feature value of that preset index on the extracted voxel. ; Based on the characteristic values ​​of each preset indicator And the corresponding brain regions, and the preset reference mean under the corresponding preset indicators. Compared with the preset reference standard deviation Calculate the characteristic values ​​of each preset indicator. Deviation: .

[0010] In one embodiment of the present invention, when the target brain region includes a first lateral brain region and a second lateral brain region that are symmetrically arranged, after the step of extracting voxels located in the target brain region from the multi-sequence MRI image data using the individualized target brain region mask as a constraint, the method further includes: For each of the preset indices, calculate the feature value of that preset index on a voxel in the first lateral brain region. And the feature value of the preset index on the voxel in the second lateral brain region. ; Calculate the eigenvalues ​​respectively First side deviation and the characteristic value Second side deviation ; With the first side deviation Deviation from the second side The larger absolute value is taken as the deviation of the feature value of the preset index; Based on the first side deviation Deviation from the second side Calculate the left-right asymmetry index corresponding to the preset index: ; When the left-right asymmetry index When the deviation exceeds a preset asymmetry threshold, the target brain region is determined to have abnormal lateralization, and the affected hemisphere is marked; the affected hemisphere is the brain region on the side with a larger absolute value of deviation.

[0011] In one embodiment of the present invention, the preset index includes at least two of the following: volume, mean signal strength, anisotropy fraction, average diffusivity, and magnetic susceptibility.

[0012] In one embodiment of the present invention, before the step of calculating the feature values ​​of each preset index within the target brain region based on the individualized target brain region mask, the method further includes: Calculate the local variance of the voxel signal values ​​within the mask of the individualized target brain region. When the local variance exceeds a second preset threshold, reduce the preset weight of the deviation of the feature values ​​based on the preset index within the target brain region.

[0013] In one embodiment of the present invention, the exposure type label includes at least: manganese exposure, lead exposure, and mercury exposure; wherein, The target brain regions for manganese exposure include the basal ganglia, while the target brain regions for lead and mercury exposure include the white matter and corpus callosum.

[0014] Compared with the prior art, the beneficial effects of the present invention are as follows: First, this invention narrows the computational scope to the target brain region corresponding to the exposure type label by using an exposure type-brain region prior knowledge base, rather than extracting features from all regions of the whole brain. This avoids the influence of irrelevant brain region signals on the analysis results and improves the detection sensitivity of hidden damage.

[0015] Second, this invention checks the signal-to-noise ratio and scanning coverage of the preprocessed image before registration, and checks the local variance of voxel signal values ​​within the individualized target brain region mask after target brain region extraction. For target brain regions with high local variance, the weight of their deviation in the weighted fusion is reduced. These two checks focus on global image quality and the measurability of the target brain region, respectively, ensuring that the deviation data of the final comprehensive abnormality index has sufficient reliability.

[0016] The present invention will be further described in detail below with reference to the accompanying drawings and embodiments. Attached Figure Description

[0017] Figure 1 This is a flowchart of a brain magnetic resonance image processing method for occupationally exposed populations provided in an embodiment of the present invention; Figure 2 This is a schematic diagram of registering multi-sequence MRI image data to a standard brain space, provided by an embodiment of the present invention; Figure 3 This is a schematic diagram showing the relationship between target brain regions and sensitive preset indicators corresponding to different exposure types provided in the embodiments of the present invention. Detailed Implementation

[0018] During their research, the inventors discovered the following shortcomings in existing brain MRI analysis methods when applied to health monitoring of occupationally exposed populations: First, existing brain MRI analysis methods are mostly limited to general detection of the whole brain, uniformly extracting and analyzing parameters from all regions of the brain. They fail to utilize prior knowledge of neurotoxicology to pre-identify the target brain regions that need to be analyzed in detail, resulting in the dispersion of computing resources and insufficient sensitivity in detecting localized and occult damage caused by specific exposures.

[0019] Second, existing methods often rely on qualitative descriptions or binary positive / negative judgments for image analysis results, making it difficult to generate quantifiable and comparable quantitative indicators. Occupational health surveillance requires objective comparisons across time periods (follow-up examinations of the same subject in different years), across equipment (different hospitals and different models of MRI equipment), and across population groups (with baseline values ​​in the normal population), which current qualitative judgments cannot meet.

[0020] Third, existing methods of judgment are mostly based on a single positive / negative assessment, or rely entirely on physicians' manual interpretation of images, which is insufficient to meet the needs of occupational health management for graded early warning and long-term follow-up comparison.

[0021] Fourth, in actual hospital settings, different equipment models and different technician skill levels can lead to inconsistent MRI image quality. Existing methods usually lack image quality control measures, and problems such as motion artifacts and incomplete scan coverage directly affect the reliability of interpretation results.

[0022] In view of this, the present invention provides a method for processing brain magnetic resonance images of people with occupational exposure.

[0023] The present invention will be further described in detail below with reference to specific embodiments, but the implementation of the present invention is not limited thereto.

[0024] Figure 1 This is a flowchart of a brain magnetic resonance imaging processing method for occupationally exposed populations provided in an embodiment of the present invention. Figure 1 As shown, this embodiment of the invention provides a method for processing brain magnetic resonance images of occupationally exposed populations, including: Step S1: Obtain multi-sequence MRI image data of the subject's brain, as well as exposure type labels associated with the subject.

[0025] It should be understood that occupationally exposed populations refer to workers who are exposed to specific hazardous substances for extended periods due to their work. This embodiment particularly focuses on occupational exposure scenarios in medical institutions. For example, medical staff performing MRI scans are exposed to strong magnetic fields for extended periods and may be exposed to substances such as manganese-containing contrast agents; medical staff performing interventional radiology procedures use lead protective equipment for extended periods and are at risk of lead exposure; and medical staff performing dental procedures are exposed to amalgam filling materials for extended periods and are at risk of mercury vapor inhalation. The health risks faced by these populations are characterized by three features: insidiousness, cumulativeity, and localization. In the early stages, there are often no obvious clinical symptoms, but hazardous substances have gradually accumulated in specific brain regions, causing subclinical damage. The degree of damage increases with the cumulative years of exposure, and the accumulation sites of different hazardous substances vary significantly.

[0026] Brain multi-sequence MRI image data is acquired by scanning the subject's head with an MRI scanner. Here, "multi-sequence MRI image data" refers to multiple MRI image data acquired in a single examination using various imaging parameters, where each imaging parameter (i.e., each sequence) is sensitive to different tissue characteristics and is suitable for reflecting different tissue information. In one embodiment, "multi-sequence MRI image data" includes at least structural images (such as T1-weighted images, suitable for clearly displaying the gray-white matter contrast of brain anatomy), and at least one selected from diffusion tensor imaging (DTI, which reflects white matter fiber structure by detecting the diffusion direction and rate of water molecules in tissue), susceptibility-weighted imaging (SWI), or quantitative susceptibility mapping (QSM, which is highly sensitive to the deposition of paramagnetic substances such as iron and manganese).

[0027] Exposure type labels are used to indicate which specific heavy metal a subject has been exposed to, and these labels can be determined based on the subject's occupational information. For example, occupational information includes: the subject's occupation such as MRI technician, interventional radiologist, or dentist; the type of exposure factor such as manganese, lead, or mercury; the duration of exposure such as 8 years; and the frequency of exposure such as once daily, thereby determining the exposure type label, such as manganese exposure, lead exposure, or mercury exposure.

[0028] In a preferred embodiment, the following rule table can be used to determine the exposure type label: the type of exposure factor is used as the primary key, and the occupation type, exposure years, and exposure frequency are used as auxiliary fields. When the type of exposure factor is clearly a certain heavy metal, the exposure type label is directly determined by that heavy metal. When the type of exposure factor is missing, but the occupation type, exposure years, and exposure frequency exceed the corresponding thresholds, the exposure type label is determined according to the typical heavy metal corresponding to that job position.

[0029] Step S2: Select the target brain region prior template corresponding to the exposure type label from the pre-constructed exposure type-brain region prior knowledge base. The target brain region is the anatomical region in the brain tissue where the toxin corresponding to the exposure type label is known to accumulate or is vulnerable.

[0030] Different heavy metals accumulate in different locations in brain tissue. Specifically, manganese, as a paramagnetic substance, preferentially accumulates in the basal ganglia (mainly involved in motor control and cognitive function regulation) after entering the human body, with typical imaging manifestations of changes in the volume and abnormal magnetization of this area; lead and mercury are more likely to affect the white matter (responsible for information transmission between different brain regions) and the corpus callosum (the fiber fasciculus connecting the left and right cerebral hemispheres), leading to demyelination and axonal damage, with typical imaging manifestations of changes in the diffusion characteristics of water molecules.

[0031] In step S2, a priori template of the target brain region corresponding to the exposure type label determined in step S1 is selected from the pre-constructed exposure type-brain region prior knowledge base. For subjects with an exposure type label of "manganese exposure", a priori template of the basal ganglia is selected from the pre-constructed exposure type-brain region prior knowledge base; while for subjects with an exposure type label of "lead exposure", priori templates of the white matter and corpus callosum are selected from the pre-constructed exposure type-brain region prior knowledge base.

[0032] In one embodiment, before the step of registering multi-sequence MRI image data to a standard brain space to obtain registration transformation parameters, the method further includes: Step a: Preprocess the multi-sequence MRI image data to obtain preprocessed MRI image data.

[0033] First, intensity normalization is performed on each sequence in the multi-sequence MRI image data. Here, "intensity" refers to the numerical value of the strength of the tissue magnetic resonance signal recorded at that location by each voxel in the MRI image data. Through intensity normalization, these values ​​are uniformly mapped to the same scale (e.g., normalized to 0~1), thereby eliminating the overall brightness deviation of the image caused by differences in equipment parameters when acquiring different sequences.

[0034] Next, for the same sequence in the multi-sequence MRI image data that includes at least two repeated acquisitions, the normalized cross-correlation coefficient (NCC) is calculated between the normalized MRI image data obtained from two adjacent repeated acquisitions. The NCC is a value between -1 and 1, used to measure the pixel-level similarity between two images. The closer the NCC is to 1, the more similar the two images are, indicating that the subject moved almost no distance between the two acquisitions; conversely, a significantly lower NCC indicates that head movement by the subject caused image misalignment. The NCC is compared with a first preset threshold, such as 0.85. When the NCC is lower than the first preset threshold, motion artifacts are determined to exist between the normalized MRI image data obtained from two adjacent repeated acquisitions, meaning that the subject's head displacement or rotation during the scan caused image blurring, ghosting, or geometric distortion. In this case, the normalized MRI image data obtained from the latter acquisition in the two adjacent repeated acquisitions is removed from the multi-sequence MRI image data and no longer participates in subsequent calculations.

[0035] Taking a single patient as an example, if the T1-weighted image sequence is acquired twice during an MRI scan, and the normalized cross-correlation coefficient (NCC) between the normalized MRI image data obtained from the first and second acquisitions is 0.92 (>0.85), it indicates that the patient remained stable between the two acquisitions, and both data sets can be retained. If the NCC between the normalized MRI image data obtained from the third acquisition and the previous acquisition is 0.62 (<0.85), then the normalized MRI image data obtained from the third acquisition is marked as containing motion artifacts and discarded. In other words, the normalized MRI image data obtained earlier is retained according to the acquisition order, and the later data that has shifted relative to the earlier acquisition data is discarded.

[0036] Finally, based on the brain tissue probability model, extracranial structures such as the scalp, skull, and soft tissues of the eye socket and face are automatically identified and removed from the MRI image data after motion artifact detection, retaining only the brain parenchyma. This process is called "skull stripping" or "scalp stripping." The resulting MRI image data after skull stripping is used as the preprocessed MRI image data. It should be noted that the brain tissue probability model is a pre-trained statistical model that can give the probability that each voxel in the brain MRI image belongs to gray matter, white matter, or cerebrospinal fluid.

[0037] Step b: Calculate the signal-to-noise ratio (SNR) of the preprocessed MRI image data and compare the SNR with a preset SNR threshold.

[0038] For preprocessed MRI image data, the signal-to-noise ratio (SNR) is calculated by dividing the average signal value of the brain parenchyma region by the standard deviation of the signal in the background air region. The brain parenchyma region includes gray matter and white matter. SNR is an objective indicator of image quality. A higher SNR means clearer brain tissue signals, lower background noise, and sharper tissue structure edges. A lower SNR indicates blurred brain tissue boundaries, leading to reduced reliability of subsequent quantitative measurements.

[0039] Optionally, the signal-to-noise ratio (SNR) of the preprocessed MRI image data is compared with a preset SNR threshold. If the SNR is less than the preset threshold, it indicates that the quality of the preprocessed MRI image data is too poor, triggering a re-acquisition. If the SNR is equal to or greater than the preset threshold, the integrity of the scanning field of view is further checked. Specifically, it is checked whether the edge voxels of the brain parenchyma region (i.e., voxels at the edge of brain tissue) in the preprocessed MRI image data fall within a preset voxel range of the scanning field of view boundary (the physical area actually covered by the MRI scanner). The preset voxel range refers to the range measured from the scanning field of view boundary towards the interior of the field of view. The boundary region formed by individual elements It is a positive integer. When the distance between the edge voxel of the brain parenchyma region and the boundary of the scanning field of view is less than... When acquiring a voxel, if the edge voxel falls within the preset voxel range, a re-acquisition is triggered. For example, if the edge voxel of the brain parenchyma is less than 3 voxels away from the boundary of the scanning field of view, it indicates that the subject's head position is too high, and part of the top of the brain may be cut out of the field of view. In this case, the coverage is determined to be incomplete and a re-acquisition is triggered. This ensures that the entire brain of the subject is completely within the imaging range.

[0040] Step S3: Register the multi-sequence MRI image data to the standard brain space to obtain the registration transformation parameters.

[0041] Figure 2 This is a schematic diagram illustrating the registration of multi-sequence MRI image data to a standard brain space, as provided in an embodiment of the present invention. Please refer to... Figure 2 A standard brain atlas (such as the MNI152 atlas) corresponding to the standard brain space was used as the fixed image, and structural images (such as T1-weighted images) from preprocessed MRI image data were used as floating images. Affine registration was performed on the floating images to align the transformed floating images with the fixed images in terms of overall anatomical structure. After alignment, a set of registration transformation parameters was obtained. The unique registration transformation parameters for each subject reflected the differences between their individual brain morphology and the standard brain.

[0042] Step S4: Using registration transformation parameters, the prior template of the target brain region is mapped back to the individual space of the subject to obtain an individualized target brain region mask.

[0043] Using the obtained registration transformation parameters, the prior template of the target brain region defined in the standard brain space is mapped back to the subject's individual space, that is, the three-dimensional image coordinate system output by the scanner and belonging to the subject. The specific steps are as follows: The prior template of the target brain region is transformed from the standard brain space coordinates to the subject's individual space coordinates using the inverse transformation of the registration transformation parameters; subsequently, the transformed prior template of the target brain region is resampled to ensure a one-to-one correspondence between the prior template and the voxels in the subject's individual space, and it is determined whether each voxel position in the subject's MRI image data belongs to the target brain region; finally, a personalized target brain region mask is obtained, which is a three-dimensional binary image of the same size as the subject's MRI image data, where voxels belonging to the target brain region are marked as 1, and other voxels are marked as 0. By overlaying this mask onto the subject's MRI image data, the signal belonging only to the target brain region can be accurately extracted.

[0044] Taking a subject exposed to manganese as an example, the target brain region template of the basal ganglia is retrieved from the exposure type-brain region prior knowledge base. The registration transformation parameters of the subject's multi-sequence MRI image data to the standard brain space are calculated. Using the inverse transformation of the registration transformation parameters, the basal ganglia template in the standard brain space is attached to the subject's own MRI image data, thereby obtaining an individualized target brain region mask belonging to the subject. The individualized target brain region mask can accurately show the range of the globus pallidus, putamen, and caudate nucleus belonging to the subject.

[0045] Step S5: Based on the individualized target brain region mask, calculate the feature values ​​of each preset index in the target brain region, and determine the deviation of each feature value by comparing the feature values ​​with the preset reference values ​​under the corresponding brain region and corresponding parameters.

[0046] In one embodiment, the preset indicators include at least two of the following: volume, mean signal strength, anisotropy fraction, average diffusivity, and magnetic susceptibility.

[0047] Firstly, long-term accumulation of manganese in the basal ganglia can lead to tissue proliferation or atrophy in this region, thus volume changes are one of its early imaging markers. The volume can be calculated based on structural images (such as T1-weighted images) by multiplying the number of voxels in the target brain region within the individualized target brain region mask by the volume of a single voxel (determined by scanning parameters) to obtain the volume of the target brain region.

[0048] Secondly, the arithmetic mean of all voxel signal values ​​within the mask of an individualized target brain region can be calculated in structural images to obtain the mean signal intensity, the magnitude of which is affected by tissue relaxation time characteristics (T1, T2). For example, manganese deposition in the basal ganglia shortens the local T1 relaxation time, resulting in increased signal in that region on T1-weighted images; this is the manifestation of manganese accumulation in MRI image data.

[0049] Thirdly, the anisotropy score is calculated based on the DTI sequence and ranges from 0 to 1. The higher the anisotropy score, the more intact the white matter fiber structure. Lead and mercury can cause demyelination and axonal damage, making the fiber structure loose and reducing preferential diffusion along the fiber direction, which is specifically manifested as a decrease in the anisotropy score. Therefore, the anisotropy score is an important indicator for assessing white matter integrity.

[0050] Fourthly, the average diffusivity, also calculated based on DTI sequences, reflects the average diffusion rate of water molecules in tissues. An increased average diffusivity usually indicates an expansion of the extracellular space, commonly seen in white matter damage caused by lead and mercury exposure.

[0051] Fifthly, magnetic susceptibility values ​​are calculated based on susceptibility-weighted imaging sequences or quantitative magnetic susceptibility mapping sequences to reflect the degree of tissue magnetization in an external magnetic field. Paramagnetic substances such as iron and manganese can enhance the local magnetic field, increasing the magnetic susceptibility value. Quantitative magnetic susceptibility mapping sequences can provide voxel-level quantitative magnetic susceptibility maps and are highly sensitive to manganese deposition in the basal ganglia. In subjects exposed to manganese, elevated magnetic susceptibility values ​​in the globus pallidus are an earlier imaging marker than changes in volume.

[0052] In step S5, using an individualized target brain region mask as a constraint, voxels located within the target brain region are extracted from the preprocessed MRI image data; for each preset index, the feature value of that preset index on the extracted voxel is calculated. Based on the characteristic values ​​of each preset indicator And the corresponding brain regions, and the preset reference mean under the corresponding preset indicators. Compared with the preset reference standard deviation Calculate the characteristic values ​​of each preset indicator. Deviation: It should be noted that the preset reference value includes the preset reference average. and preset reference standard deviation This can be derived statistically from a large number of multi-sequence MRI images of the brains of healthy control individuals.

[0053] Many structures in the human brain are naturally symmetrically distributed, such as the basal ganglia, thalamus, and hippocampus, each with a basically symmetrical structure in both hemispheres. Under normal circumstances, quantitative parameters of the same brain region in healthy individuals are usually highly similar on both sides; significant asymmetry often indicates a unilateral or asymmetrical pathological process. Therefore, in one embodiment, this anatomical characteristic is fully utilized. When the target brain region comprises two symmetrical parts, for each preset indicator, the characteristic value of that preset indicator on a voxel in the first hemisphere is calculated. And the characteristic values ​​of the preset index on voxels in the second lateral brain region. ; Calculate the eigenvalues ​​respectively First side deviation and eigenvalues Second side deviation ; with the first side deviation Deviation from the second side The deviation of the feature value with the larger absolute value is taken as the characteristic value of the preset index; based on the first deviation... Deviation from the second side Calculate the left-right asymmetry index corresponding to the preset index: ; When the left-right asymmetry index When the deviation exceeds the preset asymmetry threshold, the target brain region is determined to have abnormal lateralization, and the affected hemisphere is marked; the affected hemisphere is the brain region on the side with the larger absolute value of deviation.

[0054] Optionally, before performing the step of calculating the feature values ​​of each preset index within the target brain region based on the individualized target brain region mask, the method further includes: calculating the local variance of the voxel signal values ​​within the individualized target brain region mask; when the local variance exceeds a second preset threshold, reducing the preset weight of the deviation of the feature values ​​based on the preset indexes within the target brain region.

[0055] Step S6: Based on the preset weights of the deviations of the feature values ​​of each preset indicator under the exposure type label, the deviations of each feature value are weighted and fused to obtain a comprehensive anomaly index, and the anomaly classification result is output based on the comprehensive anomaly index.

[0056] In one embodiment, based on the known neuropathological mechanisms of each exposure type, a preset weight is determined for the deviation of each preset indicator feature value in the target brain region, so that the deviation of the preset indicator feature value most directly related to the typical damage caused by the exposure type receives a higher weight. Figure 3 This is a schematic diagram illustrating the relationship between target brain regions and sensitive preset indicators corresponding to different exposure types, as provided in embodiments of the present invention. For example... Figure 3 As shown, taking manganese exposure as an example, manganese, as a paramagnetic substance, preferentially accumulates in the basal ganglia, causing changes in volume and magnetic susceptibility in this region. Therefore, the deviation of the target brain region (basal ganglia) from its volume and magnetic susceptibility values ​​is assigned a relatively higher pre-defined weight, while the pre-defined weights corresponding to the deviations in anisotropy fraction and mean diffusivity are relatively lower. Taking lead or mercury exposure as further examples, it is known that they primarily damage the integrity of the myelin sheath in the white matter and fiber bundle structures such as the corpus callosum. Therefore, the deviations in the target brain region from its anisotropy fraction and mean diffusivity are assigned a relatively higher weight, while the pre-defined weights corresponding to the deviations in volume and magnetic susceptibility values ​​are relatively lower.

[0057] Based on the preset weights of the deviations of the feature values ​​of each preset indicator under the exposure type label, a comprehensive abnormality index is obtained by weighted fusion of the deviations of each feature value. An abnormality classification result is then output based on the comprehensive abnormality index. In a preferred embodiment, a detection report can also be output based on the target brain region name, the feature value of the preset indicator, the corresponding deviation, the comprehensive abnormality index, and the abnormality classification result.

[0058] As can be seen from the above embodiments, the beneficial effects of the present invention are as follows: First, this invention narrows the computational scope to the target brain region corresponding to the exposure type label by using an exposure type-brain region prior knowledge base, rather than extracting features from all regions of the whole brain. This avoids the influence of irrelevant brain region signals on the analysis results and improves the detection sensitivity of hidden damage.

[0059] Second, this invention checks the signal-to-noise ratio and scanning coverage of the preprocessed image before registration, and checks the local variance of voxel signal values ​​within the individualized target brain region mask after target brain region extraction. For target brain regions with high local variance, the weight of their deviation in the weighted fusion is reduced. These two checks focus on global image quality and the measurability of the target brain region, respectively, ensuring that the deviation data of the final comprehensive abnormality index has sufficient reliability.

[0060] In the description of this invention, the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of indicated technical features. Thus, a feature defined as "first" or "second" may explicitly or implicitly include one or more of that feature. In the description of this invention, "a plurality of" means two or more, unless otherwise explicitly specified.

[0061] The use of terms such as "one embodiment," "some embodiments," "example," "specific example," or "some examples" indicates that a specific feature, structure, material, or characteristic described in connection with that embodiment or example is included in at least one embodiment or example of the present invention. In this specification, the illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples. In addition, those skilled in the art can combine and integrate the different embodiments or examples described in this specification.

[0062] The above description, in conjunction with specific preferred embodiments, provides a further detailed explanation of the present invention. It should not be construed that the specific implementation of the present invention is limited to these descriptions. For those skilled in the art, various simple deductions or substitutions can be made without departing from the concept of the present invention, and all such modifications and substitutions should be considered within the scope of protection of the present invention.

Claims

1. A method for processing brain magnetic resonance images of occupationally exposed populations, characterized in that, include: Acquire multi-sequence MRI images of the subject's brain, along with exposure type labels associated with the subject; Select a target brain region prior template corresponding to the exposure type label from the pre-built exposure type-brain region prior knowledge base. The target brain region is the anatomical region in the brain tissue where the toxin corresponding to the exposure type label is known to accumulate or is vulnerable. The multi-sequence MRI image data are registered to a standard brain space to obtain registration transformation parameters; Using the registration transformation parameters, the prior template of the target brain region is mapped back to the individual space of the subject to obtain an individualized target brain region mask; Based on the individualized target brain region mask, the feature values ​​of each preset index are calculated within the target brain region. The deviation of each feature value is determined according to the feature values ​​and the preset reference values ​​under the corresponding brain region and the corresponding preset index. Based on the preset weights of the deviations of the feature values ​​of each preset indicator under the exposure type label, the deviations of each feature value are weighted and fused to obtain a comprehensive anomaly index, and the anomaly classification result is output based on the comprehensive anomaly index.

2. The brain magnetic resonance imaging processing method for occupationally exposed populations according to claim 1, characterized in that, Before the step of registering the multi-sequence MRI image data to a standard brain space to obtain registration transformation parameters, the method further includes: The multi-sequence MRI image data is preprocessed to obtain preprocessed MRI image data; Calculate the signal-to-noise ratio (SNR) of the preprocessed MRI image data and compare the SNR with a preset SNR threshold; If the signal-to-noise ratio is less than the preset signal-to-noise ratio threshold, the quality of the preprocessed MRI image data is determined to be substandard, and re-acquisition is triggered. If the signal-to-noise ratio is equal to or greater than the preset signal-to-noise ratio threshold, the edge voxels of the brain parenchyma region in the preprocessed MRI image data are detected to see if they fall within the preset voxel range of the scanning field of view boundary, so as to determine whether the scanning coverage is complete. When the edge voxels of the brain parenchyma region fall within the preset voxel range, the coverage is determined to be incomplete and re-acquisition is triggered.

3. The brain magnetic resonance image processing method for occupationally exposed populations according to claim 2, characterized in that, The step of preprocessing the multi-sequence MRI image data to obtain preprocessed MRI image data includes: Intensity normalization is performed on each sequence in the multi-sequence MRI image data to obtain normalized MRI image data; For the multi-sequence MRI image data containing the same sequence acquired at least twice, for the normalized MRI image data obtained from two adjacent repeated acquisitions, a normalized cross-correlation coefficient is calculated between the two; the normalized cross-correlation coefficient is compared with a first preset threshold; when the normalized cross-correlation coefficient is lower than the first preset threshold, it is determined that there is motion artifact between the normalized MRI image data obtained from the two adjacent repeated acquisitions, and the normalized MRI image data obtained from the latter acquisition is removed from the multi-sequence MRI image data; Based on a brain tissue probability model, the scalp, skull, and extracranial tissues are separated and removed from the MRI image data after motion artifact detection, while the brain parenchyma is preserved. The resulting MRI image data after skull removal is used as the preprocessed MRI image data.

4. The brain magnetic resonance imaging processing method for occupationally exposed populations according to claim 3, characterized in that, The step of registering the multi-sequence MRI image data to a standard brain space to obtain registration transformation parameters includes: Using a standard brain atlas corresponding to the standard brain space as a fixed image, and structural images in the preprocessed MRI image data as floating images, affine registration is performed on the floating images to obtain registration transformation parameters that map the individual space of the subject to the standard brain space.

5. The brain magnetic resonance imaging processing method for occupationally exposed populations according to claim 1, characterized in that, The step of mapping the prior template of the target brain region back to the individual subject space using the registration transformation parameters to obtain an individualized target brain region mask includes: Based on the registration transformation parameters, the prior template of the target brain region is transformed from the standard brain space to the individual space of the subject; The transformed target brain region prior template is resampled to obtain an individualized target brain region mask covering the corresponding anatomical region in the individual space of the subject.

6. The brain magnetic resonance image processing method for occupationally exposed populations according to claim 3, characterized in that, Based on the individualized target brain region mask, the steps of calculating feature values ​​of each preset index within the target brain region, and determining the deviation of each feature value according to the feature values, the corresponding brain region, and the preset reference value under the corresponding preset index, include: Using the individualized target brain region mask as a constraint, voxels located in the target brain region are extracted from the preprocessed MRI image data; For each of the preset indices, calculate the feature value of that preset index on the extracted voxel. ; Based on the characteristic values ​​of each preset indicator And the corresponding brain regions, and the preset reference mean under the corresponding preset indicators. Compared with the preset reference standard deviation Calculate the characteristic values ​​of each preset indicator. Deviation: 。 7. The brain magnetic resonance image processing method for occupationally exposed populations according to claim 6, characterized in that, When the target brain region includes a bilaterally symmetrical first and second lateral brain regions, after the step of extracting voxels within the target brain region from the multi-sequence MRI image data using the individualized target brain region mask as a constraint, the method further includes: For each of the preset indices, calculate the feature value of that preset index on a voxel in the first lateral brain region. And the feature value of the preset index on the voxel in the second lateral brain region. ; Calculate the eigenvalues ​​respectively First side deviation and the characteristic value Second side deviation ; With the first side deviation Deviation from the second side The larger absolute value is taken as the deviation of the feature value of the preset index; Based on the first side deviation Deviation from the second side Calculate the left-right asymmetry index corresponding to the preset index: ； When the left-right asymmetry index When the deviation exceeds a preset asymmetry threshold, the target brain region is determined to have abnormal lateralization, and the affected hemisphere is marked; the affected hemisphere is the brain region on the side with a larger absolute value of deviation.

8. The brain magnetic resonance image processing method for occupationally exposed populations according to claim 1, characterized in that, The preset indicators include at least two of the following: volume, mean signal strength, anisotropy fraction, average diffusivity, and magnetic susceptibility.

9. The brain magnetic resonance imaging processing method for occupationally exposed populations according to claim 1, characterized in that, Before the step of calculating the feature values ​​of each preset index within the target brain region based on the individualized target brain region mask, the method further includes: Calculate the local variance of the voxel signal values ​​within the mask of the individualized target brain region. When the local variance exceeds a second preset threshold, reduce the preset weight of the deviation of the feature values ​​based on the preset index within the target brain region.

10. The brain magnetic resonance imaging processing method for occupationally exposed populations according to claim 1, characterized in that, The exposure type labels include at least: manganese exposure, lead exposure, and mercury exposure; wherein... The target brain regions for manganese exposure include the basal ganglia, while the target brain regions for lead and mercury exposure include the white matter and corpus callosum.

Images