Quantitative water-fat mapping method based on conventional multi-planar MRI sequences and related apparatuses

Abstract

The application provides a quantitative water-fat map generation method based on a conventional multi-plane MRI sequence and related equipment, and the method comprises the following steps: acquiring a target tissue mask of a target object and conventional MRI data comprising a fat suppression sequence, a non-fat suppression sequence and different scanning plane sequences; pre-processing and spatially registering the multiple sequences so that they are mapped to a unified three-dimensional physical coordinate system; extracting the multi-sequence normalized signal intensity of each voxel in the target tissue region, constructing a voxel-level signal vector, and obtaining multiple habitat sub-regions through unsupervised clustering; determining a water-dominant habitat, a fat-dominant habitat and a water-fat mixed habitat according to the multi-sequence signal characteristics of each habitat sub-region, generating a virtual water map and a virtual fat map, and outputting water-fat related indexes. The application does not need to input water-fat separation quantitative sequences, and can represent the water-fat spatial distribution in the tissue by using conventional MRI data.

Description

Technical Field

[0001] This invention relates to the field of medical data processing technology, and more specifically, to a method and related equipment for generating quantitative water-lipid maps based on conventional multiplanar MRI sequences. Background Technology

[0002] In the development of degenerative joint diseases such as knee osteoarthritis, the increase in water content, decrease in fat content, inflammation, edema, and fibrosis in soft tissues such as the infrapatellar fat pad may reflect pathological changes in the local tissue microenvironment. Therefore, obtaining spatial distribution information of water and fat signals within the target tissue will help to more objectively assess the tissue degeneration status, degree of inflammation, and risk of disease progression.

[0003] In existing technologies, specialized quantitative magnetic resonance imaging (MRI) sequences for water-fat separation are typically required, such as multi-echo Dixon or mDIXON-quant sequences, to physically separate water and fat signals. These sequences acquire MRI signals at multiple echo times and utilize the chemical shift differences between water and fat to reconstruct water maps, fat maps, or fat fraction maps. The resulting water and fat maps have a relatively clear physical imaging basis and can directly reflect the distribution of water and fat signals within tissues.

[0004] However, the aforementioned dedicated water-lipid separation and quantification sequences still have significant limitations in practical clinical applications. On the one hand, these sequences typically rely on specific scanning protocols, equipment configurations, or software licenses, and are not essential sequences in primary care hospitals or routine knee joint examination procedures. On the other hand, acquiring additional water-lipid separation sequences increases scanning time and examination costs, reducing the efficiency of MRI equipment. Furthermore, in a large amount of historical imaging data, patients typically only acquired routine knee MRI sequences, without acquiring water-lipid separation sequences such as mDIXON-quant, making it difficult to directly use this real-world historical data for quantitative analysis of tissue water-lipid distribution.

[0005] Unlike dedicated fat-water separation sequences, routine clinical knee MRI examinations typically include fat-suppressed sequences, non-fat-suppressed sequences, and sequences from different scanning planes. For example, fat-suppressed T1-weighted sequences can suppress fat signals, making edema, effusion, or inflammation-related water signals appear as high signals; in non-fat-suppressed T1-weighted sequences, adipose tissue usually appears as a relatively high signal; and sequences from different scanning planes can provide information on tissue spatial distribution from different directions. Although these routine sequences are not specifically designed for fat-water separation, they objectively contain multidimensional information related to water signals, fat signals, and spatial structure.

[0006] Current techniques typically rely solely on visual observation by physicians using standard MRI images, or on extracting overall grayscale, texture, or shape features of the target tissue using traditional radiomics methods. These methods struggle to comprehensively utilize fat-suppressed signals, non-fat-suppressed signals, and cross-planar spatial information at the voxel level, and also find it difficult to convert the water-lipid response differences implicit in standard sequences into spatial distribution results similar to water maps and fat maps. Therefore, without acquiring dedicated water-lipid separation sequences, current techniques cannot utilize standard multiplanar MRI sequences to generate alternative water-lipid atlases that characterize water-dominant and fat-dominant regions within the target tissue.

[0007] Therefore, it is evident that a pressing technical problem needs to be solved: how to fully utilize the fat-suppressed signals, non-fat-suppressed signals, and orthogonal spatial information in conventional multiplanar MRI sequences to obtain the spatial distribution of water-dominant and fat-dominant regions within the target tissue without additional acquisition of specialized water-fat separation sequences such as mDIXON-quant. Summary of the Invention

[0008] In view of the shortcomings of the existing technology, the purpose of this invention is to provide a method, device, medium and computer equipment for generating quantitative water and lipid maps based on conventional multiplanar MRI sequences, so as to overcome the shortcomings of the existing technology.

[0009] The above-mentioned technical objective of the present invention is achieved through the following technical solution: Firstly, a method for generating quantitative hydrolipid maps based on conventional multiplanar MRI sequences, comprising: Acquire a target tissue mask of the target object and conventional multiplanar MRI sequence data, wherein the conventional multiplanar MRI sequence data includes at least one fat-suppressed MRI sequence, at least one non-fat-suppressed MRI sequence, and MRI sequences of at least two different scanning planes; The conventional multiplanar MRI sequence data is preprocessed and spatially registered to map multiple MRI sequences in the conventional multiplanar MRI sequence data to a unified three-dimensional physical coordinate system; Based on the target tissue mask, the target tissue region is determined in the unified three-dimensional physical coordinate system, and the signal intensity of each voxel in the target tissue region in multiple MRI sequences is extracted to construct the multi-sequence signal vector corresponding to each voxel. Unsupervised clustering is performed based on the multi-sequence signal vectors corresponding to multiple voxels within the target tissue region to obtain multiple habitat sub-regions and corresponding three-dimensional habitat maps. Based on the multi-sequence signal characteristics corresponding to each habitat sub-region, water-dominant habitats, lipid-dominant habitats, and / or water-lipid mixed habitats are determined from the multiple habitat sub-regions. Based on the spatial distribution of the water-dominant habitat, the fat-dominant habitat, and / or the mixed water-lipid habitat in the target tissue region, a virtual water-lipid map corresponding to the target tissue is generated. The virtual water-lipid map includes a virtual water map and a virtual lipid map, and quantitative indicators related to water and lipid in the target tissue are output based on the virtual water-lipid map.

[0010] In one embodiment, the conventional multiplanar MRI sequence data includes sagittal proton density-weighted fat-suppressed sequences, sagittal T1-weighted unsuppressed sequences, and coronal proton density-weighted fat-suppressed sequences; the generation process of the virtual water-lipid map does not use water-lipid separation and quantitative sequences as input data.

[0011] In one embodiment, the preprocessing and spatial registration of the conventional multiplanar MRI sequence data, mapping multiple MRI sequences in the conventional multiplanar MRI sequence data to a unified three-dimensional physical coordinate system, includes: The conventional multiplanar MRI sequence data were subjected to format conversion and signal intensity normalization. Based on the voxel spacing, image origin, and scanning direction of each MRI sequence, determine the position information of each MRI sequence in physical space; Using one of the MRI sequences as a reference sequence, rigid or non-rigid registration is performed on the remaining MRI sequences to spatially align them in the unified three-dimensional physical coordinate system.

[0012] In one embodiment, constructing the multi-sequence signal vector corresponding to each voxel includes: For any voxel within the target tissue region, the normalized signal intensity of the voxel in the sagittal proton density weighted lipid-suppressed sequence, the normalized signal intensity in the sagittal T1-weighted unsuppressed sequence, and the normalized signal intensity in the coronal proton density weighted lipid-suppressed sequence are extracted respectively. Based on the extracted multiple normalized signal intensities, construct the multi-sequence signal vector corresponding to any voxel: ; in, Indicates the location in spatial coordinates The multi-sequence signal vector corresponding to the voxel. This represents the normalized signal intensity of the voxel in the sagittal proton density-weighted lipid-suppressed sequence. This represents the normalized signal intensity of the voxel in the sagittal T1-weighted unsuppressed fat sequence. This represents the normalized signal intensity of the voxel in the coronal proton density-weighted lipid-suppressed sequence.

[0013] In one embodiment, the step of performing unsupervised clustering based on multiple sequence signal vectors corresponding to multiple voxels within the target tissue region to obtain multiple habitat sub-regions and corresponding three-dimensional habitat maps includes: The multi-sequence signal vectors corresponding to each voxel within the target tissue region are used as clustering samples; The clustering samples are clustered using the K-Means clustering algorithm to obtain multiple cluster categories; Based on the cluster category to which each voxel belongs, a habitat label is assigned to each voxel within the target tissue region. The three-dimensional habitat map is generated based on the habitat labels corresponding to each voxel.

[0014] In one embodiment, determining water-dominant habitats, lipid-dominant habitats, and / or mixed water-lipid habitats from the plurality of habitat sub-regions based on the multi-sequence signal characteristics corresponding to each habitat sub-region includes: Calculate the cluster centers or average signal features corresponding to each habitat sub-region; The habitat subregions with high signal intensity in non-fat-suppressed MRI sequences and low signal intensity in fat-suppressed MRI sequences were identified as fat-dominant habitats. Habitat subregions with high signal intensity in fat-suppressed MRI sequences and relatively low signal intensity in non-fat-suppressed MRI sequences were identified as water-dominant habitats. Habitat subregions whose signal characteristics lie between the lipid-dominant habitat and the water-dominant habitat are defined as water-lipid mixed habitats.

[0015] In one embodiment, generating a virtual water-lipid map corresponding to the target tissue based on the spatial distribution of the water-dominant habitat, the lipid-dominant habitat, and / or the mixed water-lipid habitat in the target tissue region includes: Map the voxels corresponding to the water-dominant habitats to the high-value voxels in the virtual water map; Map the voxels corresponding to the fat-dominant habitat to the high-value voxels in the virtual fat map; The voxels corresponding to the water-lipid mixed habitat are mapped to the transition value voxels in the virtual water map and the virtual lipid map; The water-lipid related quantitative indicators include the volume ratio of water-dominant habitat, the volume ratio of lipid-dominant habitat, the volume ratio of mixed water-lipid habitat, the signal heterogeneity of water-dominant habitat, and / or the spatial integrity of lipid-dominant habitat.

[0016] Secondly, a quantitative lipid atlas generation device based on conventional multiplanar MRI sequences includes: The data acquisition unit is used to acquire the target tissue mask of the target object and conventional multiplanar MRI sequence data, wherein the conventional multiplanar MRI sequence data includes at least one fat-suppressed MRI sequence, at least one non-fat-suppressed MRI sequence, and at least two MRI sequences with different scanning planes. The coordinate mapping unit is used to preprocess and spatially register the conventional multiplanar MRI sequence data, so that multiple MRI sequences in the conventional multiplanar MRI sequence data are mapped to a unified three-dimensional physical coordinate system. The vector construction unit is used to determine the target tissue region in the unified three-dimensional physical coordinate system based on the target tissue mask, and extract the signal intensity of each voxel in the target tissue region in multiple MRI sequences to construct the multi-sequence signal vector corresponding to each voxel. Clustering unit is used to perform unsupervised clustering based on the multi-sequence signal vectors corresponding to multiple voxels within the target tissue region to obtain multiple habitat sub-regions and corresponding three-dimensional habitat maps. The habitat classification unit is used to determine water-dominant habitats, lipid-dominant habitats, and / or mixed water-lipid habitats from the multiple habitat sub-regions based on the multi-sequence signal characteristics corresponding to each habitat sub-region. The quantitative indicator unit is used to generate a virtual water-lipid map corresponding to the target tissue based on the spatial distribution of the water-dominant habitat, the fat-dominant habitat, and / or the water-lipid mixed habitat in the target tissue region. The virtual water-lipid map includes a virtual water map and a virtual lipid map, and outputs water-lipid related quantitative indicators of the target tissue based on the virtual water-lipid map.

[0017] Thirdly, a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of the method described in the first aspect.

[0018] Fourthly, a computer device includes a memory and a processor, the memory storing a computer program, the processor executing the computer program to implement the steps of the method described in the first aspect.

[0019] In summary, the present invention has the following beneficial effects: It eliminates the need for additional acquisition of dedicated water-lipid separation sequences such as mDIXON-quant, enabling the generation of virtual water maps and virtual fat maps of target tissues using routine clinical multiplanar MRI data. By registering fat-suppressed sequences, non-fat-suppressed sequences, and orthogonal plane information into a unified space, and performing habitat clustering and water-lipid dominance mapping at the voxel level, it can objectively characterize the spatial distribution of increased water, decreased fat, and mixed transition regions within the tissue, reducing scanning costs and time, revitalizing existing routine imaging data, and providing quantifiable imaging evidence for disease assessment. Attached Figure Description

[0020] Figure 1 This is a flowchart of the quantitative water-lipid atlas generation method based on conventional multiplanar MRI sequences of the present invention; Figure 2 This is a structural diagram of the quantitative water-lipid atlas generation device based on conventional multiplanar MRI sequences in an embodiment of the present invention; Figure 3 This is an internal structural diagram of the computer device in an embodiment of the present invention; Figure 4 This is a schematic diagram of displaying the habitat VOI generated based on conventional multiplanar MRI sequences superimposed on a sagittal proton density-weighted fat-suppressed sequence image, where the grayscale image represents the sagittal proton density-weighted fat-suppressed sequence image, and the colored area represents the spatial distribution of different habitat sub-regions within the target tissue; Figure 5 for Figure 4 The mDIXON-quant water map corresponding to the target object is used to display the spatial distribution of water signals in the area where the target tissue is located and its neighboring areas; Figure 6 for Figure 4 The mDIXON-quant lipid map corresponding to the target object is used to display the spatial distribution of fat signals in the region where the target tissue is located and its neighboring regions.

[0021] In the diagram: 1. Data acquisition unit; 2. Coordinate mapping unit; 3. Vector construction unit; 4. Clustering unit; 5. Habitat classification unit; 6. Indicator quantification unit. Detailed Implementation

[0022] To make the objectives, features, and advantages of the present invention more apparent and understandable, specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings. Several embodiments of the present invention are shown in the drawings. However, the present invention can be implemented in many different forms and is not limited to the embodiments described herein.

[0023] In this application embodiment, "at least one" refers to one or more, and "more than one" refers to two or more. "And / or" describes the relationship between related objects, indicating that three relationships can exist. For example, A and / or B can represent the existence of A alone, the simultaneous existence of A and B, or the existence of B alone. A and B can be singular or plural. The character " / " generally indicates that the preceding and following related objects are in an "or" relationship. "At least one of the following" and similar expressions refer to any combination of these items, including any combination of single or plural items. For example, at least one of a, b, and c can represent: a, b, c, ab, ac, bc, or abc, where a, b, and c can be single or multiple.

[0024] Those skilled in the art will recognize that the units and algorithm steps described in the embodiments disclosed herein can be implemented using electronic hardware, computer software, or a combination of electronic hardware and software. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this application.

[0025] Those skilled in the art will understand that, for the sake of convenience and brevity, the specific working processes of the systems, devices, and units described above can be referred to the corresponding processes in the foregoing method embodiments, and will not be repeated here.

[0026] In the several embodiments provided in this application, any function, if implemented as a software functional unit and sold or used as an independent product, can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, or a part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of this application. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0027] The above description is merely a specific embodiment of this application. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the protection scope of this application. The protection scope of this application should be determined by the protection scope of the claims.

[0028] The present invention will now be described in detail with reference to the accompanying drawings and embodiments.

[0029] Example 1 To address the aforementioned problems, this invention provides a method for generating quantitative hydrolipid maps based on conventional multiplanar MRI sequences, such as... Figure 1 As shown, it includes the following steps: Step S1: Acquire the target tissue mask and conventional multiplanar MRI sequence data of the target object. The target tissue mask is used to define the spatial range of subsequent image analysis; it can represent the position and boundaries of the target tissue in three-dimensional image space. The conventional multiplanar MRI sequence data includes at least one fat-suppressed MRI sequence, at least one non-fat-suppressed MRI sequence, and MRI sequences from at least two different scanning planes. The fat-suppressed MRI sequence provides tissue signal information after fat signal suppression, the non-fat-suppressed MRI sequence provides tissue signal information under conditions without fat suppression, and the MRI sequences from different scanning planes provide imaging information of the target tissue in different spatial directions. Thus, the system can simultaneously acquire the fat-suppressed signal, the non-fat-suppressed signal, and cross-planar spatial information of the target tissue.

[0030] In a preferred embodiment, the target tissue is the infrapatellar fat pad in the knee joint region. Increased water content, decreased fat content, inflammatory edema, or fibrotic changes within the infrapatellar fat pad can reflect changes in the local tissue microenvironment associated with knee osteoarthritis. Therefore, when obtaining the target tissue mask, a three-dimensional mask corresponding to the infrapatellar fat pad can be obtained, and the region defined by this three-dimensional mask can be used as the analysis area for subsequent voxel-level multiple sequence signal extraction, habitat clustering, and virtual water-lipid atlas generation. By limiting the processing range to the interior of the infrapatellar fat pad, interference from signals from bone, cartilage, synovial fluid, and other non-target tissues on the habitat segmentation results can be reduced.

[0031] Step S2 involves preprocessing and spatial registration of conventional multiplanar MRI sequence data, mapping multiple MRI sequences within the data to a unified three-dimensional physical coordinate system. Preprocessing may include data reading from different MRI sequences, image formatting, spatial information parsing, and signal intensity standardization, enabling unified access and comparison of data from different sequences. Since different MRI sequences may have different scanning directions, voxel spacing, image origins, or image matrix sizes, spatial registration is necessary to align multiple MRI sequences to the same three-dimensional physical coordinate system. After spatial registration, voxels at the same spatial location can correspond to signal intensities in multiple MRI sequences, providing a basis for subsequent voxel-level signal analysis. After the above spatial registration, any spatial location within the target tissue region can correspond to the signal intensity in sagittal proton density-weighted fat-suppressed sequences, sagittal T1-weighted unsuppressed sequences, and coronal proton density-weighted fat-suppressed sequences. Thus, each target tissue voxel has a signal description from three conventional MRI sequences, enabling the subsequent construction of a multi-sequence signal vector to simultaneously include water-related signals, fat-related signals, and orthogonal plane supplementary signals.

[0032] In actual clinical acquisition, conventional knee MRI sequences often exhibit issues such as large slice thickness, inconsistent in-plane and inter-slice resolution, and different slice orientations between different scanning planes. For example, sagittal and coronal sequences may not be entirely consistent in image matrix orientation, inter-slice spacing, and scan coverage. Directly extracting voxel signals from their respective original image matrices can easily lead to signal mismatches at the same anatomical location between different sequences. Therefore, this embodiment maps multiple MRI sequences to a unified three-dimensional physical coordinate system through spatial registration, enabling spatial correspondence between sequences from different scanning planes within the target tissue region, thereby ensuring that the same voxel location can obtain matched signal intensities from multiple sequences.

[0033] Step S3: Based on the target tissue mask, the target tissue region is determined in a unified three-dimensional physical coordinate system, and the signal intensity of each voxel within the target tissue region in multiple MRI sequences is extracted to construct a multi-sequence signal vector corresponding to each voxel. Specifically, the target tissue mask can be mapped to a unified three-dimensional physical coordinate system, and the voxels within the mask range are determined as the voxels to be analyzed within the target tissue region. For any voxel within the target tissue region, the signal intensity of that voxel in multiple MRI sequences is read, and the multiple signal intensities are combined in a predetermined order to form the multi-sequence signal vector corresponding to that voxel. Through the above processing, each voxel in the target tissue can be characterized by a combination of signals from multiple MRI sequences, instead of relying solely on the grayscale information of a single sequence.

[0034] Step S4 involves unsupervised clustering based on the multi-sequence signal vectors corresponding to multiple voxels within the target tissue region, resulting in multiple habitat sub-regions and corresponding 3D habitat maps. Specifically, the multi-sequence signal vectors corresponding to each voxel within the target tissue region can be used as clustering samples input into the unsupervised clustering model. This allows voxels with similar signal response patterns to be grouped into the same category, while voxels with different signal response patterns are grouped into different categories. After clustering, each voxel obtains a corresponding category label. Based on the category labels of each voxel, the system forms multiple spatial distribution regions in 3D space. Each spatial distribution region corresponds to a habitat sub-region, and these multiple habitat sub-regions together constitute the 3D habitat map of the target tissue. This 3D habitat map is used to characterize the spatial heterogeneity within the target tissue based on differences in multi-sequence signals.

[0035] Step S5: Based on the multi-sequence signal characteristics corresponding to each habitat sub-region, determine water-dominant habitats, fat-dominant habitats, and / or mixed water-lipid habitats from multiple habitat sub-regions. Specifically, the signal intensity distribution of voxels within each habitat sub-region in multiple MRI sequences can be statistically analyzed, and the corresponding water-lipid dominance type can be determined based on the comprehensive signal performance of the habitat sub-region in fat-suppressed MRI sequences and non-fat-suppressed MRI sequences. When a habitat sub-region exhibits water-related dominant signal characteristics in a fat-suppressed MRI sequence, it can be identified as a water-dominant habitat; when a habitat sub-region exhibits fat-related dominant signal characteristics in a non-fat-suppressed MRI sequence, it can be identified as a fat-dominant habitat; when the multi-sequence signal characteristics of a habitat sub-region are between those of a water-dominant habitat and a fat-dominant habitat, it can be identified as a mixed water-lipid habitat. Thus, the clustered habitat sub-regions not only have spatial division significance but are also endowed with signal interpretations related to the water-lipid distribution of the target tissue.

[0036] In step S6, a virtual water-lipid map is generated for the target tissue based on the spatial distribution of water-dominant, fat-dominant, and / or mixed water-lipid habitats within the target tissue region. This virtual water-lipid map includes a virtual water map and a virtual lipid map, and quantitative water-lipid related indicators of the target tissue are output based on the virtual water-lipid map. Specifically, voxels or regions identified as water-dominant habitats can be mapped onto the virtual water map to characterize the spatial distribution of water-dominant regions within the target tissue; voxels or regions identified as fat-dominant habitats can be mapped onto the virtual lipid map to characterize the spatial distribution of fat-dominant regions within the target tissue; and mixed water-lipid habitats can be represented as transitional regions in both the virtual water map and virtual lipid map. Furthermore, quantitative water-lipid related indicators of the target tissue can be calculated based on the virtual water map and virtual lipid map, such as the volume of water-dominant regions, the volume of fat-dominant regions, the volume of mixed water-lipid regions, and the proportion of each type of region relative to the total volume of the target tissue. These quantitative water-lipid related indicators can reflect the spatial distribution of water dominance and fat dominance within the target tissue.

[0037] In some verification scenarios, the virtual water map and virtual lipid map generated in this embodiment can be compared with the spatial distribution consistency of the water-lipid separation quantitative sequence atlas of the same target object. High-value regions in the virtual water map represent water-dominant regions inferred from conventional multiplanar MRI sequences, and high-value regions in the virtual lipid map represent fat-dominant regions inferred from conventional multiplanar MRI sequences. These virtual atlases are not directly acquired from the water-lipid separation quantitative sequence, but rather are alternative atlases generated based on the combination of multiple sequence signals from conventional MRI sequences. They can be used to characterize the spatial distribution of water-lipid dominance corresponding to the real water map and lipid map.

[0038] Using the above-described implementation method, virtual water and lipid maps within the target tissue can be obtained without acquiring additional dedicated water-lipid separation MRI sequences, utilizing fat-suppressed signals, non-fat-suppressed signals, and cross-planar spatial information from conventional multiplanar MRI sequences. This method improves the utilization rate of different sequence signals through unified spatial registration and voxel-level multi-sequence signal analysis, and can transform the heterogeneity within the target tissue into visualized and quantifiable water-lipid distribution results, thereby reducing examination costs and enhancing the reuse value of previous conventional MRI data.

[0039] In one embodiment, conventional multiplanar MRI sequence data specifically includes sagittal proton density-weighted fat-suppressed sequences, sagittal T1-weighted unsuppressed sequences, and coronal proton density-weighted fat-suppressed sequences. These three MRI sequences are all relatively common basic sequences in clinical knee MRI examinations and can be obtained from the target subject's routine MRI examination data without the need for additional water-fat separation and quantification sequences.

[0040] The three MRI sequences described above constitute a sequence combination of two fat-suppressed sequences and one non-fat-suppressed sequence, and a combination of the same sagittal plane and different coronal planes. Specifically, both the sagittal and coronal proton density-weighted fat-suppressed sequences can reflect changes in water-related signals within the target tissue under conditions where fat signals are suppressed; the sagittal T1-weighted non-fat-suppressed sequence can reflect high fat-related signals within the target tissue; and the coronal proton density-weighted fat-suppressed sequence, being on different scanning planes than the sagittal sequences, can provide supplementary information on the spatial distribution of water-related signals within the target tissue in orthogonal directions. Therefore, this sequence combination can simultaneously provide water-sensitive signals, fat-sensitive signals, and cross-planar spatial constraints, providing a data foundation for subsequent identification of water-dominant and fat-dominant regions based on habitat subregions.

[0041] The sagittal proton density-weighted fat-suppressed sequence provides post-fat-suppression signal information of the target tissue in the sagittal direction. Because this sequence suppresses fat signal, areas of increased water content, inflammatory edema, or other water-related signal changes within the target tissue are more likely to show relatively significant signal differences in this sequence. The sagittal T1-weighted unsuppressed sequence provides signal information of the target tissue under unsuppressed conditions. Since adipose tissue typically exhibits relatively high signal in T1-weighted unsuppressed sequences, this sequence provides a signal basis for subsequent identification of fat-dominant regions. The coronal proton density-weighted fat-suppressed sequence provides post-fat-suppressed signal information from a different scanning plane than the sagittal plane, thus providing cross-planar supplementation to the spatial distribution of water-related signals within the target tissue.

[0042] During the generation of virtual water-lipid maps, the system uses the aforementioned sagittal proton density-weighted lipid-suppressed sequences, sagittal T1-weighted unsuppressed sequences, and coronal proton density-weighted lipid-suppressed sequences as input sequences. In subsequent processing, these input sequences undergo preprocessing, spatial registration, voxel signal extraction, and habitat analysis. Thus, each target tissue voxel can simultaneously acquire lipid-suppressed sequence signals, unsuppressed sequence signals, and spatial signal information from different scanning planes, enabling the subsequent identification of water-dominant and fat-dominant habitats based on the combination of multiple sequence signals.

[0043] In this embodiment, the generation of the virtual water-lipid atlas does not use quantitative water-lipid separation sequences as input data. That is, the system does not require mDIXON-quant, multi-echo Dixon, or other quantitative MRI sequences specifically designed for water-lipid separation when generating virtual water and lipid maps. Quantitative water-lipid separation sequences can be used as validation or control data during the research phase to evaluate the consistency between the virtual and real water-lipid atlases, but they do not participate in the actual generation process of the virtual water-lipid atlas. Therefore, this embodiment can virtually generate the dominant water-lipid distribution of the target tissue based on conventional MRI sequences.

[0044] It should be noted that during the actual generation of the virtual water-lipid atlas, the quantitative water-lipid separation sequence does not participate in the construction of the voxel signal vector within the target tissue mask, nor does it participate in unsupervised clustering or habitat type determination. In other words, the system only processes conventional multiplanar MRI sequence data when inferring and generating virtual water and lipid maps. The quantitative water-lipid separation sequence can only be used as a control atlas during the method validation phase to evaluate the spatial distribution consistency between the generated virtual water and lipid maps and the real water-lipid separation atlas.

[0045] Using the above-described implementation method, by combining sagittal fat-suppressed sequences, sagittal non-fat-suppressed sequences, and coronal fat-suppressed sequences, water-related signals, fat-related signals, and cross-planar spatial information can be obtained simultaneously, providing a basis for virtual water-fat distribution analysis from conventional MRI data. Since the generation process does not rely on water-fat separation and quantification sequences, it reduces the need for additional scanning, lowers examination costs, and allows historical image data from previously unacquired sequences such as mDIXON to still be used for water-fat correlation analysis.

[0046] In one embodiment, after acquiring conventional multiplanar MRI sequence data of the target object, preprocessing can be performed on different MRI sequences to ensure that images from different scanning planes, different scanning parameters, or different data formats can enter the same processing flow. The preprocessing includes format conversion and signal intensity normalization. Format conversion can convert the original MRI examination data from the medical image format output by the device or stored in the hospital system into a unified three-dimensional image data format, enabling each MRI sequence to be read, stored, and retrieved with a unified data structure. Signal intensity normalization is used to reduce intensity differences between different sequences caused by variations in scanning parameters, device gain, or image grayscale range, providing a consistent data basis for subsequent comparisons of signal intensity between different sequences. Signal intensity normalization can be performed separately for each MRI sequence, converting the voxel signals in each sequence to the same or similar numerical scale. Through normalization, numerical offsets caused by differences in scanning parameters, receiving coils, device gain, or grayscale range between different sequences can be reduced, preventing a particular sequence from having excessive weight in subsequent clustering simply because of its larger numerical range. Thus, each MRI sequence can provide a relatively balanced signal contribution in the multi-sequence signal vector.

[0047] After format conversion and signal intensity normalization, the system determines the physical location information of each MRI sequence based on the voxel spacing, image origin, and scan direction. The voxel spacing represents the distance between adjacent voxels in physical space, the image origin represents the starting position of the image coordinate system relative to the physical coordinate system, and the scan direction represents the directional relationship of the rows, columns, and slices of the image matrix in physical space. Based on this information, the voxel index positions in the image matrix can be converted into physical spatial positions.

[0048] For example, for voxel indexing in a certain MRI sequence Its coordinates in physical space can be represented as: ;in, Represents the coordinates of a voxel in physical space. Represents the origin of the image. Represents the scan direction matrix. These represent the voxel spacing in the three directions. Through the above transformation, the corresponding positions of each voxel in different MRI sequences in real three-dimensional physical space can be clearly defined.

[0049] Furthermore, using one MRI sequence from a conventional multiplanar MRI dataset as a reference sequence, rigid or non-rigid registration is performed on the remaining MRI sequences. The reference sequence provides a spatial benchmark for a unified three-dimensional physical coordinate system, and the remaining MRI sequences are mapped to the space containing the reference sequence through registration transformation. Rigid registration can be used to correct overall spatial deviations such as translation and rotation caused by differences in patient position or scanning location between different sequences; non-rigid registration can be used to correct local deformations or local spatial inconsistencies between different sequences. After registration, multiple MRI sequences are spatially aligned in a unified three-dimensional physical coordinate system, allowing the same target tissue location to correspond to the signal intensity in multiple MRI sequences.

[0050] Therefore, when constructing voxel-level multi-sequence signal vectors, the system can extract signal intensities from multiple MRI sequences at the same voxel location in a unified three-dimensional physical coordinate system, instead of directly reading signals from their respective original image matrices. This avoids signal mismatches caused by differences in slice thickness, orientation, origin, or voxel spacing between different sequences, ensuring that subsequent habitat clustering and virtual hydrolipid map generation are based on spatially consistent multi-sequence signals.

[0051] The above implementation method improves data consistency across different MRI sequences through format conversion and normalization. It also determines the true spatial location of each sequence using voxel spacing, image origin, and scanning direction, and completes spatial registration using a reference sequence. This allows MRI signals from different scanning planes to correspond to the same tissue location in the same three-dimensional physical coordinate system, reducing signal mismatch and improving the accuracy and stability of voxel-level multi-sequence signal vector construction, habitat delineation, and virtual hydrolipid map generation.

[0052] In one embodiment, after multiple MRI sequences have been mapped to a unified three-dimensional physical coordinate system, the system determines the target tissue region based on a target tissue mask and extracts signal intensity voxel-by-voxel within that region. For any voxel within the target tissue region, the system first determines the voxel's spatial coordinates in the unified three-dimensional physical coordinate system. Then, the normalized signal intensity of the spatial coordinates was read in the sagittal proton density weighted lipid-suppressed sequence, the sagittal T1 weighted unsuppressed sequence, and the coronal proton density weighted lipid-suppressed sequence, respectively.

[0053] The normalized signal intensity in the sagittal proton density-weighted lipid-suppressed sequence is denoted as . It is used to characterize the signal performance of this voxel under sagittal fat-suppressed imaging conditions; the normalized signal intensity in the sagittal T1-weighted unsuppressed sequence is denoted as... It is used to characterize the signal performance of the voxel under sagittal non-fat-suppressed imaging conditions; the normalized signal intensity in the coronal proton density-weighted fat-suppressed sequence is denoted as... It is used to characterize the signal performance of this voxel under coronal fat-suppressed imaging conditions.

[0054] After obtaining the above three normalized signal intensities, the system combines them in a predetermined order to construct the multi-sequence signal vector corresponding to the voxel: ;in, Indicates the location in spatial coordinates The multi-sequence signal vector corresponds to each voxel. Through this multi-sequence signal vector, a voxel is no longer represented solely by the grayscale value of a single MRI sequence, but simultaneously includes sagittal fat-suppressed signal, sagittal non-fat-suppressed signal, and coronal fat-suppressed signal. Therefore, each voxel within the target tissue region can form a corresponding three-dimensional signal representation, providing a data foundation for subsequent unsupervised clustering and habitat segmentation based on voxel signal differences. In the aforementioned multi-sequence signal vector, and It reflects the signal performance of voxels under fat-suppressed imaging conditions and can provide characterization of water-related signal changes; Reflecting the signal performance of voxels under non-fat-suppressed T1 imaging conditions, this data can provide characterization for fat-related signals. Therefore, this multi-sequence signal vector can reflect the differences in water and lipid responses of the same voxel under different imaging conditions in a combined form, providing a basis for subsequent habitat sub-region segmentation based on signal similarity.

[0055] In practical implementation, the above signal extraction and vector construction processes can be sequentially performed on all voxels within the target tissue mask to obtain the set of voxel signal vectors corresponding to the target tissue region. This set can be represented as: ;in, This represents the target tissue region defined by the target tissue mask. This set of voxel signal vectors is used to characterize the differences in multi-sequence signal responses at different spatial locations within the target tissue, and can subsequently be used as data input for unsupervised clustering.

[0056] By employing the above-described implementation method, a multi-sequence signal vector containing sagittal fat suppression, sagittal non-fat suppression, and coronal fat suppression signals is constructed for each target tissue voxel under a unified spatial coordinate system. This fully utilizes the differential responses of different MRI sequences to water and fat signals, avoids the problem of insufficient information from a single sequence, provides a clearer signal basis for subsequent habitat clustering, and improves the spatial consistency and interpretability of virtual water and fat atlas generation.

[0057] In one embodiment, after constructing the multi-sequence signal vectors of each voxel within the target tissue region, the system uses each voxel within the target tissue region as the object to be clustered, and the multi-sequence signal vector corresponding to each voxel as the clustering sample. That is, each voxel is not classified primarily based on its spatial coordinates, but rather on the combination of its signal responses in multiple MRI sequences. For voxels located at spatial coordinates... The clustered samples of voxels can be represented as: ;in, , and These represent the normalized signal intensity of the voxel in the corresponding MRI sequence.

[0058] In this embodiment, the system forms a cluster sample set by combining the multi-sequence signal vectors corresponding to all voxels within the target tissue region, and then clusters this sample set based on the K-Means clustering algorithm. The K-Means clustering algorithm groups voxels with similar signal response patterns into the same cluster category based on the similarity between the signal vectors, and groups voxels with significantly different signal response patterns into different cluster categories. The clustering process can be represented as minimizing the sum of the distances from each sample to its respective cluster center, for example: ;in, Indicates the number of cluster categories. Indicates the first Cluster categories, Indicates the first The multi-sequence signal vector corresponding to the individual element. Indicates the first Each cluster category corresponds to a cluster center. Because different spatial locations within the target tissue may have different water dominance, fat dominance, or water-lipid mixture states, their signal response combinations in fat-suppressed and non-fat-suppressed sequences will also differ. Therefore, in the context of… , and In the constructed multi-sequence signal space, voxels with similar water-lipid response patterns are more likely to cluster into the same cluster category, while voxels with different water-lipid response patterns are classified into different cluster categories. Thus, the clustering results can reflect the habitat partitioning within the target tissue based on differences in water-lipid signals.

[0059] After clustering, the system obtains multiple cluster categories. Each cluster category corresponds to a group of voxels with similar multi-sequence signal characteristics. Since these voxels are all located within the target tissue region, each cluster category can be understood as a habitat sub-region within the target tissue. The system assigns corresponding habitat labels to each voxel within the target tissue region based on its cluster category. For example, voxels belonging to the first cluster category are assigned the first habitat label, voxels belonging to the second cluster category are assigned the second habitat label, and so on.

[0060] After assigning habitat labels to each voxel, the system backfills these labels according to the spatial position of each voxel in a unified three-dimensional physical coordinate system, forming a three-dimensional habitat map corresponding to the target tissue region. This three-dimensional habitat map is used to represent the spatial distribution of different habitat sub-regions within the target tissue. In other words, each voxel position in the three-dimensional habitat map corresponds to a habitat label, voxels with the same habitat label constitute one or more spatial distribution regions, and voxels with different habitat labels represent regions within the target tissue with different multi-sequence signal response characteristics.

[0061] Using the above implementation method, the differences in multi-sequence signals at the voxel level within the target tissue can be transformed into spatially identifiable habitat sub-regions, giving the originally dispersed grayscale signals a three-dimensional regional representation. Unsupervised partitioning of voxel signal vectors using K-Means clustering reduces the need for manual rule intervention and provides a stable data foundation for subsequently determining water dominance, fat dominance, or water-lipid mixture states based on the signal characteristics of different habitat sub-regions.

[0062] In one embodiment, after obtaining multiple habitat sub-regions corresponding to the target tissue region, the system can further perform statistical analysis on the multi-sequence signal characteristics of each habitat sub-region to determine the water-lipid dominance type corresponding to each habitat sub-region. Specifically, the system can determine the set of voxels contained in each habitat sub-region based on the habitat labels of each voxel in the three-dimensional habitat map, and calculate the cluster center or average signal characteristics corresponding to the habitat sub-region based on the multi-sequence signal vectors of each voxel in the voxel set.

[0063] For example, for the first Each habitat subregion, containing a set of voxels, can be represented as... The average signal characteristics corresponding to this habitat sub-region can be expressed as: ;in, Indicates the first Average signal characteristics of each habitat subregion This represents the average normalized signal intensity of the habitat sub-region in the sagittal proton density-weighted lipid-suppressed sequence. This represents the average normalized signal intensity of the habitat subregion in the sagittal T1-weighted unsuppressed lipid sequence. This represents the average normalized signal intensity of the habitat sub-region in the coronal proton density-weighted lipid-suppressed sequence.

[0064] After obtaining the cluster centers or average signal characteristics corresponding to each habitat sub-region, the system can determine the type of different habitat sub-regions based on the signal intensity relationship between non-fat-suppressed MRI sequences and fat-suppressed MRI sequences. For habitat sub-regions with high signal intensity in non-fat-suppressed MRI sequences and low signal intensity in fat-suppressed MRI sequences, they can be identified as fat-dominant habitats. That is, when a certain habitat sub-region exhibits a strong signal under non-fat-suppressed imaging conditions but a relatively weak signal under fat-suppressed imaging conditions, it indicates that the signal performance of this region is more consistent with the characteristics of a fat-dominant region, and therefore, this habitat sub-region can be marked as a fat-dominant habitat. The "higher" or "lower" signal intensity can be determined based on the relative ranking of the cluster center signals among multiple habitat sub-regions, or it can be determined based on a preset threshold. In the embodiment using the above three MRI sequences as input, a fat-dominant habitat can be characterized by a high average normalized signal intensity in the sagittal T1-weighted non-fat-suppressed sequence and a low average normalized signal intensity in the sagittal proton density-weighted fat-suppressed sequence and the coronal proton density-weighted fat-suppressed sequence. Correspondingly, water-dominant habitats are characterized by higher average normalized signal in sagittal and coronal proton density-weighted lipid-suppressed sequences, and relatively lower average normalized signal in sagittal T1-weighted unsuppressed sequences. Habitat subregions exhibiting signal characteristics between these two categories can be identified as water-lipid mixed habitats.

[0065] For habitat subregions exhibiting high signal intensity on fat-suppressed MRI sequences but relatively low signal intensity on non-fat-suppressed MRI sequences, these can be identified as water-dominant habitats. Specifically, fat-suppressed MRI sequences suppress fat signals; regions that still show high signal intensity in these sequences typically better reflect increased water content, inflammatory changes, or water-related signal enhancement. Therefore, when a habitat subregion shows high signal intensity on fat-suppressed MRI sequences but relatively low signal intensity on non-fat-suppressed MRI sequences, it can be considered a water-dominant habitat.

[0066] Habitat subregions whose signal characteristics fall between those of fat-dominant and water-dominant habitats can be identified as mixed water-lipid habitats. These subregions do not exhibit a clear bias towards either a single water-dominant or fat-dominant signal pattern in either fat-suppressed or non-fat-suppressed MRI sequences; instead, they display multi-sequence signal characteristics intermediate between the two. Therefore, these regions can be considered as spatial areas of mixed or transitional water-lipid states, used to express the transitional distribution of water and lipid components within the target tissue when generating virtual water-lipid maps.

[0067] Through the above processing, the system can further assign water-lipid-related signal meanings to multiple habitat sub-regions obtained from unsupervised clustering. This transforms the 3D habitat map from merely representing the spatial division of signal-similar regions into a tissue atlas foundation that reflects water dominance, lipid dominance, and the water-lipid mixing state. Subsequently, when generating virtual water maps and virtual lipid maps, water-lipid atlas mapping can be performed on different voxels or regions based on the spatial distribution of the aforementioned habitat types.

[0068] Using the above implementation method, the physiological meaning of water and lipid in unsupervised clustering results can be interpreted based on the cluster centers or average signal characteristics of each habitat sub-region, so that the spatial regions obtained by clustering correspond to the signal response relationship of fat-suppressed and non-fat-suppressed MRI. This method avoids relying solely on the overall tissue mean for judgment, and can highlight the local water dominance, fat dominance, and transition regions within the target tissue, thereby improving the interpretability and spatial resolution of the virtual water and lipid atlas generation results.

[0069] In one embodiment, after determining the water-dominant habitat, fat-dominant habitat, and mixed water-fat habitat, the system further generates a virtual water-fat map corresponding to the target tissue based on the spatial distribution of these different habitat types in the target tissue region. The virtual water-fat map includes a virtual water map and a virtual fat map, both established in a unified three-dimensional physical coordinate system, thus ensuring that the spatial positions of each voxel within the target tissue remain consistent across different maps.

[0070] Specifically, for voxels identified as water-dominant habitats, the system maps them to high-value voxels in the virtual water map. Here, a high-value voxel indicates that it is assigned a relatively high spectral value in the virtual water map, representing a region with strong water dominance. Correspondingly, these voxels form high-value regions in the virtual water map, thus reflecting the spatial distribution of water-dominant regions within the target tissue.

[0071] For voxels identified as adipose-dominant habitats, the system maps them to high-value voxels in the virtual fat map. In other words, in the virtual fat map, these voxels are assigned relatively high map values ​​to characterize the region where the voxel is located as having strong fat-dominant characteristics. Thus, adipose-dominant habitats form high-value regions in the virtual fat map, reflecting the spatial distribution of fat-dominant regions within the target tissue.

[0072] For voxels identified as belonging to mixed water-lipid habitats, the system maps them to transitional voxels in both virtual water and virtual lipid maps. Transitional voxels are those that do not correspond to the highest value region in either the virtual water or lipid map, but are represented by values ​​between high and low, reflecting that the region possesses both water and fat characteristics, or is in a transitional state from fat dominance to water dominance. Through this mapping method, the system can express not only water-dominant regions in the virtual water map but also fat-dominant regions in the virtual lipid map, and further utilize transitional voxels to express the mixed transitional regions between the two.

[0073] After the map is generated, the system can further output quantitative indicators related to water and lipids in the target tissue based on the virtual water-lipid map. Specifically, the volume percentage of water-dominant habitat represents the proportion of water-dominant habitat volume in the entire target tissue; the volume percentage of fat-dominant habitat volume represents the proportion of fat-dominant habitat volume in the entire target tissue; and the volume percentage of mixed water-lipid habitat volume represents the proportion of the mixed transition region volume in the entire target tissue. If the total prime number of the target tissue region is denoted as... The number of voxels corresponding to the water-dominant habitat is denoted as The number of voxels corresponding to the fat-dominant habitat is denoted as The number of voxels corresponding to the water-lipid mixed habitat is denoted as The corresponding volume percentages can be expressed as follows: ; ; ;in, This indicates the proportion of water-dominant habitat volume. This indicates the volume percentage of fat-dominant habitats. This indicates the volume percentage of a water-lipid mixed habitat.

[0074] Furthermore, the signal heterogeneity of water-dominant habitats reflects the dispersion of voxel signal distribution within these habitats, and can be characterized by statistically analyzing the fluctuations in voxel signal values. The spatial integrity of fat-dominant habitats reflects their spatial continuity and completeness, and can be characterized by their connectivity in three-dimensional space. The signal heterogeneity of water-dominant habitats can be expressed using the standard deviation, variance, or coefficient of variation of voxel signal intensity within the habitat. The spatial integrity of fat-dominant habitats can be expressed as the ratio of the largest connected region volume to the total volume of the fat-dominant habitat. Through these indicators, the system can quantitatively describe the water-lipid distribution characteristics of target tissues from aspects such as volume ratio, internal signal variation, and spatial structure.

[0075] Therefore, in this embodiment, the system first assigns voxel-level map values ​​based on different habitat types, then constructs virtual water maps and virtual lipid maps based on the assignment results, and finally extracts water-lipid related quantitative indicators from the maps. This not only allows for a visual display of the water-lipid dominance status in different regions within the target tissue in an image-based manner, but also outputs data results for further analysis in a quantitative indicator format.

[0076] Using the above implementation method, the spatial distribution of water-dominant, fat-dominant, and mixed water-lipid habitats can be directly converted into different value regions in virtual water maps and virtual lipid maps. This transforms the water-lipid differences within the target tissue from abstract clustering results into visualized and quantifiable atlas results. Furthermore, by using indicators such as volume percentage, signal heterogeneity, and spatial integrity, the method can further reflect local water anomalies, fat retention levels, and transitional zone states within the target tissue, thereby improving the intuitiveness, interpretability, and clinical analytical value of the results.

[0077] In some embodiments, the quantitative water-lipid related indicators may further include the proportion of adipose-dominant habitat shrinkage. The proportion of adipose-dominant habitat shrinkage can be characterized by the degree of reduction in the volume of adipose-dominant habitat relative to the total volume of the target tissue, or by the difference between the current proportion of adipose-dominant habitat volume in the target tissue and a preset reference range. The proportion of adipose-dominant habitat shrinkage reflects the degree of reduction or fragmentation of adipose-dominant regions within the target tissue, thereby providing a quantitative basis for evaluating reduced fat content, fibrosis, or abnormal fat metabolism in the target tissue.

[0078] In one optional embodiment, quantitative water-lipid related indicators obtained based on the virtual water-lipid atlas can be used as imaging input data for subsequent disease assessment or efficacy evaluation. For example, based on the volume ratio of water-dominant habitat, fat-dominant habitat, mixed water-lipid habitat, signal heterogeneity of water-dominant habitat, and / or spatial integrity of fat-dominant habitat, the changes in water increase, fat decrease, or mixed transition areas within the target tissue can be characterized. This application does not change the generation process of the virtual water-lipid atlas, but rather utilizes the atlas output results for subsequent analysis.

[0079] Example 2 Please see Figure 2 A quantitative lipid atlas generation device based on conventional multiplanar MRI sequences, the device comprising: Data acquisition unit 1 is used to acquire the target tissue mask of the target object and conventional multiplanar MRI sequence data, wherein the conventional multiplanar MRI sequence data includes at least one fat-suppressed MRI sequence, at least one non-fat-suppressed MRI sequence, and at least two MRI sequences with different scanning planes. Coordinate mapping unit 2 is used to preprocess and spatially register the conventional multiplanar MRI sequence data, so that multiple MRI sequences in the conventional multiplanar MRI sequence data are mapped to a unified three-dimensional physical coordinate system. Vector construction unit 3 is used to determine the target tissue region in the unified three-dimensional physical coordinate system based on the target tissue mask, and extract the signal intensity of each voxel in the target tissue region in multiple MRI sequences to construct a multi-sequence signal vector corresponding to each voxel. Clustering unit 4 is used to perform unsupervised clustering based on the multi-sequence signal vectors corresponding to multiple voxels within the target tissue region to obtain multiple habitat sub-regions and corresponding three-dimensional habitat maps. Habitat classification unit 5 is used to determine water-dominant habitats, lipid-dominant habitats, and / or mixed water-lipid habitats from the multiple habitat sub-regions based on the multi-sequence signal characteristics corresponding to each habitat sub-region. The quantitative indicator unit 6 is used to generate a virtual water-lipid map corresponding to the target tissue based on the spatial distribution of the water-dominant habitat, the fat-dominant habitat, and / or the water-lipid mixed habitat in the target tissue region. The virtual water-lipid map includes a virtual water map and a virtual lipid map, and outputs water-lipid related quantitative indicators of the target tissue based on the virtual water-lipid map.

[0080] Specific limitations regarding the quantitative lipid atlas generation device based on conventional multiplanar MRI sequences can be found in the limitations of the quantitative lipid atlas generation method based on conventional multiplanar MRI sequences described above, and will not be repeated here. Each module in the aforementioned quantitative lipid atlas generation device based on conventional multiplanar MRI sequences can be implemented entirely or partially through software, hardware, or a combination thereof. These modules can be embedded in or independent of the processor in a computer device, or stored in the memory of a computer device as software, so that the processor can call and execute the corresponding operations of each module.

[0081] Those skilled in the art will understand that Figure 2 The structure shown is merely a block diagram of a portion of the structure related to the present application and does not constitute a limitation on the present application. The specific quantitative lipid mapping device based on conventional multiplanar MRI sequences may include more or fewer components than those shown in the figure, or combine certain components, or have different component arrangements.

[0082] Example 3 A computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the quantitative water-lipid atlas generation method based on conventional multiplanar MRI sequences as described in Example 1.

[0083] Example 4 In one embodiment, a computer device is provided, which may be a server, and its internal structure diagram may be as follows: Figure 3 As shown, the computer device includes a processor, memory, network interface, and database connected via a system bus. The processor provides computational and control capabilities. The memory includes a non-volatile storage medium and internal memory. The non-volatile storage medium stores the operating system, computer programs, and the database. The internal memory provides an environment for the operation of the operating system and computer programs in the non-volatile storage medium. When executed by the processor, the computer program implements a method for generating quantitative lipid maps based on conventional multiplanar MRI sequences.

[0084] Those skilled in the art will understand that Figure 3 The structure shown is merely a block diagram of a portion of the structure related to the present application and does not constitute a limitation on the computer device to which the present application is applied. Specific computer devices may include more or fewer components than those shown in the figure, or combine certain components, or have different component arrangements.

[0085] In one embodiment, a computer device is provided, including a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to perform the following steps: including: S1. Obtain the target tissue mask of the target object and conventional multiplanar MRI sequence data, wherein the conventional multiplanar MRI sequence data includes at least one fat-suppressed MRI sequence, at least one non-fat-suppressed MRI sequence, and at least two MRI sequences with different scanning planes; S2. Preprocess and spatially register the conventional multiplanar MRI sequence data to map multiple MRI sequences in the conventional multiplanar MRI sequence data to a unified three-dimensional physical coordinate system; S3. Based on the target tissue mask, determine the target tissue region in the unified three-dimensional physical coordinate system, and extract the signal intensity of each voxel in the target tissue region in multiple MRI sequences to construct a multi-sequence signal vector corresponding to each voxel. S4. Perform unsupervised clustering based on the multi-sequence signal vectors corresponding to multiple voxels within the target tissue region to obtain multiple habitat sub-regions and corresponding three-dimensional habitat maps. S5. Based on the multi-sequence signal characteristics corresponding to each habitat sub-region, determine the water-dominant habitat, lipid-dominant habitat, and / or water-lipid mixed habitat from the multiple habitat sub-regions. S6. Based on the spatial distribution of the water-dominant habitat, the fat-dominant habitat, and / or the water-lipid mixed habitat in the target tissue region, generate a virtual water-lipid map corresponding to the target tissue. The virtual water-lipid map includes a virtual water map and a virtual lipid map. Based on the virtual water-lipid map, output the water-lipid related quantitative indicators of the target tissue.

[0086] Those skilled in the art will understand that all or part of the processes in the methods of the above embodiments can be implemented by a computer program instructing related hardware. The computer program can be stored in a non-volatile computer-readable storage medium, and when executed, it can include the processes of the embodiments of the above methods. Any references to memory, storage, databases, or other media used in the embodiments provided in this application can include non-volatile and / or volatile memory. Non-volatile memory can include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM), or flash memory. Volatile memory can include random access memory (RAM) or external cache memory. By way of illustration and not limitation, RAM is available in various forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), dual data rate SDRAM (DDRSDRAM), enhanced SDRAM (ESDRAM), synchronous link DRAM (SLDRAM), Rambus direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM (RDRAM), etc.

[0087] The technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0088] The above description is merely a preferred embodiment of the present invention. The scope of protection of the present invention is not limited to the above embodiments. All technical solutions falling within the scope of the present invention's concept are within the scope of protection of the present invention. It should be noted that for those skilled in the art, any improvements and modifications made without departing from the principle of the present invention should also be considered within the scope of protection of the present invention.

Claims

1. A method for generating quantitative water-lipid maps based on conventional multiplanar MRI sequences, characterized in that, include: Acquire a target tissue mask of the target object and conventional multiplanar MRI sequence data, wherein the conventional multiplanar MRI sequence data includes at least one fat-suppressed MRI sequence, at least one non-fat-suppressed MRI sequence, and MRI sequences of at least two different scanning planes; The conventional multiplanar MRI sequence data is preprocessed and spatially registered to map multiple MRI sequences in the conventional multiplanar MRI sequence data to a unified three-dimensional physical coordinate system; Based on the target tissue mask, the target tissue region is determined in the unified three-dimensional physical coordinate system, and the signal intensity of each voxel in the target tissue region in multiple MRI sequences is extracted to construct the multi-sequence signal vector corresponding to each voxel. Unsupervised clustering is performed based on the multi-sequence signal vectors corresponding to multiple voxels within the target tissue region to obtain multiple habitat sub-regions and corresponding three-dimensional habitat maps. Based on the multi-sequence signal characteristics corresponding to each habitat sub-region, water-dominant habitats, lipid-dominant habitats, and / or water-lipid mixed habitats are determined from the multiple habitat sub-regions. Based on the spatial distribution of the water-dominant habitat, the fat-dominant habitat, and / or the mixed water-lipid habitat in the target tissue region, a virtual water-lipid map corresponding to the target tissue is generated. The virtual water-lipid map includes a virtual water map and a virtual lipid map, and quantitative indicators related to water and lipid in the target tissue are output based on the virtual water-lipid map.

2. The method for generating quantitative water-lipid maps based on conventional multiplanar MRI sequences according to claim 1, characterized in that, The conventional multiplanar MRI sequence data includes sagittal proton density-weighted fat-suppressed sequences, sagittal T1-weighted unsuppressed sequences, and coronal proton density-weighted fat-suppressed sequences; the generation process of the virtual water-lipid map does not use water-lipid separation and quantitative sequences as input data.

3. The method for generating quantitative water-lipid maps based on conventional multiplanar MRI sequences according to claim 1, characterized in that, The preprocessing and spatial registration of the conventional multiplanar MRI sequence data, mapping multiple MRI sequences in the conventional multiplanar MRI sequence data to a unified three-dimensional physical coordinate system, includes: The conventional multiplanar MRI sequence data were subjected to format conversion and signal intensity normalization. Based on the voxel spacing, image origin, and scanning direction of each MRI sequence, determine the position information of each MRI sequence in physical space; Using one of the MRI sequences as a reference sequence, rigid or non-rigid registration is performed on the remaining MRI sequences to spatially align them in the unified three-dimensional physical coordinate system.

4. The method for generating quantitative water-lipid maps based on conventional multiplanar MRI sequences according to claim 1, characterized in that, The construction of the multi-sequence signal vector corresponding to each voxel includes: For any voxel within the target tissue region, the normalized signal intensity of the voxel in the sagittal proton density weighted lipid-suppressed sequence, the normalized signal intensity in the sagittal T1-weighted unsuppressed sequence, and the normalized signal intensity in the coronal proton density weighted lipid-suppressed sequence are extracted respectively. Based on the extracted multiple normalized signal intensities, construct the multi-sequence signal vector corresponding to any voxel: ; in, Indicates the location in spatial coordinates The multi-sequence signal vector corresponding to the voxel. This represents the normalized signal intensity of the voxel in the sagittal proton density-weighted lipid-suppressed sequence. This represents the normalized signal intensity of the voxel in the sagittal T1-weighted unsuppressed fat sequence. This represents the normalized signal intensity of the voxel in the coronal proton density-weighted lipid-suppressed sequence.

5. The method for generating quantitative water-lipid maps based on conventional multiplanar MRI sequences according to claim 1, characterized in that, The unsupervised clustering based on the multi-sequence signal vectors corresponding to multiple voxels within the target tissue region yields multiple habitat sub-regions and corresponding three-dimensional habitat maps, including: The multi-sequence signal vectors corresponding to each voxel within the target tissue region are used as clustering samples; The clustering samples are clustered using the K-Means clustering algorithm to obtain multiple cluster categories; Based on the cluster category to which each voxel belongs, a habitat label is assigned to each voxel within the target tissue region. The three-dimensional habitat map is generated based on the habitat labels corresponding to each voxel.

6. The method for generating quantitative water-lipid maps based on conventional multiplanar MRI sequences according to claim 1, characterized in that, The step of determining water-dominant habitats, lipid-dominant habitats, and / or mixed water-lipid habitats from the multiple habitat sub-regions based on the multi-sequence signal characteristics corresponding to each habitat sub-region includes: Calculate the cluster centers or average signal features corresponding to each habitat sub-region; The habitat subregions with high signal intensity in non-fat-suppressed MRI sequences and low signal intensity in fat-suppressed MRI sequences were identified as fat-dominant habitats. Habitat subregions with high signal intensity in fat-suppressed MRI sequences and relatively low signal intensity in non-fat-suppressed MRI sequences were identified as water-dominant habitats. Habitat subregions whose signal characteristics lie between the lipid-dominant habitat and the water-dominant habitat are defined as water-lipid mixed habitats.

7. The method for generating quantitative water-lipid maps based on conventional multiplanar MRI sequences according to claim 1, characterized in that, The step of generating a virtual water-lipid map corresponding to the target tissue based on the spatial distribution of the water-dominant habitat, the fat-dominant habitat, and / or the mixed water-lipid habitat in the target tissue region includes: Map the voxels corresponding to the water-dominant habitats to the high-value voxels in the virtual water map; Map the voxels corresponding to the fat-dominant habitat to the high-value voxels in the virtual fat map; The voxels corresponding to the water-lipid mixed habitat are mapped to the transition value voxels in the virtual water map and the virtual lipid map; The water-lipid related quantitative indicators include the volume ratio of water-dominant habitat, the volume ratio of lipid-dominant habitat, the volume ratio of mixed water-lipid habitat, the signal heterogeneity of water-dominant habitat, and / or the spatial integrity of lipid-dominant habitat.

8. A quantitative water-lipid atlas generation device based on conventional multiplanar MRI sequences, characterized in that, The device includes: The data acquisition unit is used to acquire the target tissue mask of the target object and conventional multiplanar MRI sequence data, wherein the conventional multiplanar MRI sequence data includes at least one fat-suppressed MRI sequence, at least one non-fat-suppressed MRI sequence, and at least two MRI sequences with different scanning planes. The coordinate mapping unit is used to preprocess and spatially register the conventional multiplanar MRI sequence data, so that multiple MRI sequences in the conventional multiplanar MRI sequence data are mapped to a unified three-dimensional physical coordinate system. The vector construction unit is used to determine the target tissue region in the unified three-dimensional physical coordinate system based on the target tissue mask, and extract the signal intensity of each voxel in the target tissue region in multiple MRI sequences to construct the multi-sequence signal vector corresponding to each voxel. Clustering unit is used to perform unsupervised clustering based on the multi-sequence signal vectors corresponding to multiple voxels within the target tissue region to obtain multiple habitat sub-regions and corresponding three-dimensional habitat maps. The habitat classification unit is used to determine water-dominant habitats, lipid-dominant habitats, and / or mixed water-lipid habitats from the multiple habitat sub-regions based on the multi-sequence signal characteristics corresponding to each habitat sub-region. The quantitative indicator unit is used to generate a virtual water-lipid map corresponding to the target tissue based on the spatial distribution of the water-dominant habitat, the fat-dominant habitat, and / or the water-lipid mixed habitat in the target tissue region. The virtual water-lipid map includes a virtual water map and a virtual lipid map, and outputs water-lipid related quantitative indicators of the target tissue based on the virtual water-lipid map.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, it implements the quantitative water-lipid atlas generation method based on conventional multiplanar MRI sequences as described in any one of claims 1-7.

10. A computer device comprising a memory and a processor, wherein the memory stores a computer program, characterized in that, When the processor executes the computer program, it implements the quantitative water-lipid atlas generation method based on conventional multiplanar MRI sequences as described in any one of claims 1-7.

Images