Thyroid disease multi-omics mass spectrum construction analysis method and system

Abstract

The invention provides a thyroid disease multi-omics mass spectrum construction analysis method and system.The thyroid disease multi-omics mass spectrum construction analysis method comprises the steps that a thyroid ultrasound image and multi-omics mass spectrum data are obtained, and a cross-modal incidence relation is generated based on anatomical structure characteristics of the thyroid ultrasound image and biomolecule space distribution characteristics of the multi-omics mass spectrum data; in combination with the thyroid anatomical structure deformation field and a multi-task optimization target, performing nonlinear coupling calculation on a back propagation error in a multi-task learning iteration process to obtain a gradient contribution weight between tasks; and generating a target multi-omics mass spectrum based on the cross-modal correlation features, the anatomical structure deformation field and the gradient contribution weight, and performing thyroid pathological feature analysis on the target multi-omics mass spectrum to obtain a three-dimensional positioning coordinate and a malignancy probability prediction value of the thyroid nodule. The thyroid disease diagnosis comprehensiveness and accuracy are improved.

Description

Technical Field

[0001] The embodiments of the present application relate to the technical fields of medical image analysis, multi-omics data fusion, and artificial intelligence-assisted diagnosis, and particularly to a method and system for constructing and analyzing multi-omics mass spectrometry maps of thyroid diseases. Background Art

[0002] In the diagnosis and treatment of thyroid diseases, accurately identifying the location and malignancy degree of thyroid nodules is crucial. Currently, there is an urgent need in the medical field for a technology that can integrate multi-modal data (such as ultrasound images and mass spectrometry data) to improve the accuracy and efficiency of diagnosis. This technology needs to be able to process and analyze data from different sources and generate reliable pathological analysis results to assist doctors in making more accurate diagnosis and treatment decisions.

[0003] Currently, in response to this technical need, existing research has proposed a multi-modal data fusion method based on deep learning. This method constructs a deep learning model to extract and fuse features from thyroid ultrasound images and mass spectrometry data, and then realizes the location and malignancy prediction of thyroid nodules. Specifically, this method uses a convolutional neural network (CNN) to extract features from ultrasound images, and at the same time uses a specific algorithm to process mass spectrometry data. Finally, these features are integrated through a multi-task learning framework to complete the location and malignancy prediction of nodules.

[0004] Although the existing solutions have made certain progress in the location and malignancy prediction of thyroid nodules, there are still defects in processing multi-modal data. Specifically, when fusing ultrasound images and mass spectrometry data, the existing solutions fail to fully consider the cross-modal correlation relationship between the two, resulting in a large error between the generated mass spectrometry map and the anatomical structure features in the ultrasound image. This error not only affects the accuracy of nodule location but also reduces the reliability of malignancy prediction, thus limiting the application effect of this solution in clinical practice. Summary of the Invention

[0005] The embodiments of the present application provide a method and system for constructing and analyzing multi-omics mass spectrometry maps of thyroid diseases to solve the problems of poor comprehensiveness and low accuracy in the diagnosis of thyroid diseases in the prior art.

[0006] In a first aspect, the embodiments of the present application provide a method for constructing and analyzing multi-omics mass spectrometry maps of thyroid diseases, including: Obtain a thyroid ultrasound image and multi-omics mass spectrometry data, and generate a cross-modal correlation relationship according to the anatomical structure features of the thyroid ultrasound image and the biomolecular spatial distribution features of the multi-omics mass spectrometry data, where the anatomical structure includes thyroid nodules; Generate a thyroid anatomical structure deformation field based on the anatomical structure features, and construct a multi-task optimization objective according to the preset thyroid disease diagnosis requirements. The multi-task optimization objective includes a thyroid nodule localization task and a thyroid nodule malignancy prediction task; During the multi-task learning iteration process, perform non-linear coupling calculation on the backpropagation error of the thyroid anatomical structure deformation field to obtain the gradient contribution weights of the thyroid nodule localization task and the thyroid nodule malignancy prediction task respectively; Generate cross-modal association features according to the cross-modal association relationship, and based on the thyroid anatomical structure deformation field and the cross-modal association features, combined with the gradient contribution weights, generate a target multi-omics mass spectrum. The error between the spatial coordinates of the metabolomics signal hot spot region in the target multi-omics mass spectrum and the microcalcification cluster in the thyroid ultrasound image is within the preset anatomical structure tolerance range; Perform thyroid pathological feature analysis on the target multi-omics mass spectrum to obtain a thyroid pathological analysis result, which includes the three-dimensional localization coordinates and the malignancy probability prediction value of the thyroid nodule.

[0007] Optionally, the generating cross-modal association features according to the cross-modal association relationship, and based on the thyroid anatomical structure deformation field and the cross-modal association features, combined with the gradient contribution weights, to generate a target multi-omics mass spectrum includes: Based on the cross-modal association relationship, combined with the anatomical structure features and the biomolecular spatial distribution features, construct a three-dimensional spatial weight matrix, and perform multi-level spatial decomposition on the anatomical structure features according to the three-dimensional spatial weight matrix to obtain the texture distribution patterns at different anatomical levels within the thyroid tissue; Fuse the texture distribution patterns and the multi-omics mass spectrometry data layer by layer to generate cross-modal association features; Perform task-aware filtering on the cross-modal association features according to the gradient contribution weights to obtain filtered cross-modal association features; Non-linearly superimpose the filtered cross-modal association features and the thyroid anatomical structure deformation field to generate an initial multi-omics mass spectrum; Optimize the initial multi-omics mass spectrum to generate a target multi-omics mass spectrum.

[0008] Optionally, the fusing the texture distribution patterns and the multi-omics mass spectrometry data layer by layer to generate cross-modal association features includes: Analyze the multi-omics mass spectrometry data to obtain mass spectrometry detection points. Based on the hierarchical resolution of the thyroid ultrasound image, dynamically adjust the mass spectrometry detection points within the coordinate mapping range of each level, where the coordinate mapping range of the global level covers the topological boundary between the overall contour of the thyroid and adjacent organs, and the mapping range of the local level is limited to the local spatial neighborhood of microcalcification clusters in the thyroid ultrasound image; At each hierarchical resolution, perform spatial interpolation compensation processing on the intensity distribution of metabolomics signals in the multi-omics mass spectrometry data according to the spatial distance between the adjusted mass spectrometry detection points and the ultrasound image voxels to obtain the compensated intensity distribution of metabolomics signals. The ultrasound image voxel is the smallest spatial unit of the thyroid ultrasound image; According to the spatial consistency between the spatial distribution path of the expression levels of proteomic markers in the multi-omics mass spectrometry data and the directional characteristics of the texture distribution pattern, perform directional enhancement processing on the texture characteristics of the texture distribution pattern to obtain enhanced texture characteristics; Stitch the compensated intensity distribution of metabolomics signals and the enhanced texture characteristics according to the hierarchical resolution to generate cross-modal correlation features. The stitching channels at the global level include the covariance matrix of the overall deformation trend of the thyroid and the global distribution of metabolomics signals, and the stitching channels at the local level include the texture details of microcalcification clusters and the spatial offset of the peak position of metabolic signals.

[0009] Optionally, the non-linear superposition of the filtered cross-modal correlation features and the thyroid anatomical structure deformation field to generate an initial multi-omics mass spectrometry map includes: Decompose the thyroid anatomical structure deformation field into local deformation parameters and global deformation trends within the thyroid tissue; Based on the local deformation parameters, perform spatial coordinate transformation on the filtered cross-modal correlation features to obtain the cross-modal correlation features after spatial coordinate transformation; Based on the global deformation trend, perform inter-level weight assignment on the cross-modal correlation features after spatial coordinate transformation to obtain the cross-modal correlation features after weight assignment; Non-linearly superpose the cross-modal correlation features after weight assignment and the deformation parameters of the thyroid anatomical structure deformation field, and generate an initial multi-omics mass spectrometry map by iteratively optimizing the spatial coordinates of the metabolomics signal hot spots and the microcalcification clusters.

[0010] Optionally, the thyroid pathological feature analysis of the target multi-omics mass spectrometry map to obtain the thyroid pathological analysis result, where the thyroid pathological analysis result includes the three-dimensional positioning coordinates and malignant probability prediction values of thyroid nodules, includes: Decompose the thyroid anatomical structure in the target multi-omics mass spectrometry map into global hierarchical anatomical deformation features and local hierarchical microcalcification cluster density features. During the decomposition process, the global hierarchical anatomical deformation features are generated by quantifying the displacement gradient direction of the contact surface between the thyroid edge and the carotid artery, and the local hierarchical microcalcification cluster density features are obtained by statistically analyzing the three-dimensional coordinate distribution density of microcalcification clusters in a preset spatial neighborhood; Extract the thyroid nodule candidate region from the overlapping region between the metabolomics signal hotspot region and the local hierarchical microcalcification cluster density features. Based on the thyroid nodule candidate region, combine the consistency coefficient between the metabolomics signal gradient direction and the texture distribution pattern direction to generate an initial boundary contour; Based on the proteomics biomarker expression level, combine the global hierarchical anatomical deformation features, and screen out the set of mutation positions related to the preset malignant risk from the thyroid nodule candidate region. Based on the set of mutation positions, generate protein expression abnormal hotspots; Calculate the confidence of the malignant risk direction of the thyroid nodule candidate region according to the spatial angle between the metabolomics signal gradient direction and the texture distribution pattern; Overlay the spatial distribution path of the protein expression abnormal hotspots and the confidence of the malignant risk direction to generate the three-dimensional positioning coordinates and the malignant probability prediction value of the thyroid nodule.

[0011] Optionally, the overlaying the spatial distribution path of the protein expression abnormal hotspots and the confidence of the malignant risk direction to generate the three-dimensional positioning coordinates and the malignant probability prediction value of the thyroid nodule includes: Generate the spatial distribution path of the protein expression abnormal hotspots by calculating the spatial distance between mutation positions in the set of mutation positions and the similarity of the proteomics biomarker expression level; Overlay the spatial distribution path and the confidence of the malignant risk direction to generate an overlay result; Based on the overlay result, optimize the coupling error between the spatial distribution path and the initial boundary contour to determine the three-dimensional positioning coordinates of the thyroid nodule, and verify the spatial correlation between the spatial distribution path and the texture distribution pattern to generate the malignant probability prediction value of the thyroid nodule.

[0012] Optionally, during the multi-task learning iteration process, perform non-linear coupling calculation on the backpropagation error of the thyroid anatomical structure deformation field to obtain the gradient contribution weights of the thyroid nodule positioning task and the thyroid nodule malignancy prediction task respectively, including: Decompose the backpropagation error of the thyroid anatomical structure deformation field into the backpropagation error component of the thyroid nodule localization task and the backpropagation error component of the thyroid nodule malignancy prediction task; Based on the non-linear relationship between the backpropagation error component of the thyroid nodule localization task and the backpropagation error component of the thyroid nodule malignancy prediction task, construct a non-linear coupling function; Based on the non-linear coupling function, calculate the gradient contribution weights of the thyroid nodule localization task and the thyroid nodule malignancy prediction task respectively.

[0013] In a second aspect, an embodiment of the present application provides a multi-omics mass spectrometry atlas construction and analysis system for thyroid diseases, including: An acquisition module, configured to acquire thyroid ultrasound images and multi-omics mass spectrometry data, and generate a cross-modal association relationship according to the anatomical structure characteristics of the thyroid ultrasound images and the biomolecular spatial distribution characteristics of the multi-omics mass spectrometry data; A construction module, configured to generate a thyroid anatomical structure deformation field based on the anatomical structure characteristics, and construct a multi-task optimization objective according to the preset thyroid disease diagnosis requirements, where the multi-task optimization objective includes a thyroid nodule localization task and a thyroid nodule malignancy prediction task; A calculation module, configured to perform non-linear coupling calculation on the backpropagation error of the thyroid anatomical structure deformation field during the multi-task learning iteration process to obtain the gradient contribution weights of the thyroid nodule localization task and the thyroid nodule malignancy prediction task respectively; A generation module, configured to generate cross-modal association features according to the cross-modal association relationship, and generate a target multi-omics mass spectrometry atlas based on the thyroid anatomical structure deformation field and the cross-modal association features, in combination with the gradient contribution weights, where the error between the metabolomics signal hot spot region in the target multi-omics mass spectrometry atlas and the spatial coordinates of the microcalcification cluster in the thyroid ultrasound image is within the preset anatomical structure tolerance range; An analysis module, configured to perform thyroid pathological feature analysis on the target multi-omics mass spectrometry atlas to obtain a thyroid pathological analysis result, where the thyroid pathological analysis result includes the three-dimensional localization coordinates of the thyroid nodule and the malignant probability prediction value.

[0014] In a third aspect, an embodiment of the present application provides a computing device, including a processor and a memory, where a computer program is stored in the memory, and the processor is configured to run the computer program to execute the method for constructing and analyzing a multi-omics mass spectrometry atlas for thyroid diseases according to any one of the first aspects.

[0015] Fourthly, an embodiment of the present application provides a computer storage medium, on which computer program instructions are stored. When the computer program instructions are executed by a processor, the method for constructing and analyzing a multi-omics mass spectrometry map of a thyroid disease described in any one of the first aspects is implemented.

[0016] In an embodiment of the present application, a method for constructing and analyzing a multi-omics mass spectrometry map of a thyroid disease is provided. The method includes: obtaining a thyroid ultrasound image and multi-omics mass spectrometry data, and generating a cross-modal association relationship according to the anatomical structure features of the thyroid ultrasound image and the biomolecular spatial distribution features of the multi-omics mass spectrometry data, where the anatomical structure includes thyroid nodules; generating a thyroid anatomical structure deformation field based on the anatomical structure features, and constructing a multi-task optimization objective according to the preset thyroid disease diagnosis requirements, where the multi-task optimization objective includes a thyroid nodule localization task and a thyroid nodule malignancy prediction task; in the multi-task learning iteration process, performing non-linear coupling calculation on the backpropagation error of the thyroid anatomical structure deformation field to obtain the gradient contribution weights of the thyroid nodule localization task and the thyroid nodule malignancy prediction task respectively; generating cross-modal association features according to the cross-modal association relationship, and generating a target multi-omics mass spectrometry map based on the thyroid anatomical structure deformation field and the cross-modal association features, in combination with the gradient contribution weights, where the error between the spatial coordinates of the metabolomics signal hotspot region in the target multi-omics mass spectrometry map and the microcalcification cluster in the thyroid ultrasound image is within the preset anatomical structure tolerance range; performing thyroid pathological feature analysis on the target multi-omics mass spectrometry map to obtain a thyroid pathological analysis result, where the thyroid pathological analysis result includes the three-dimensional localization coordinates and the malignancy probability prediction value of the thyroid nodule.

[0017] The technical solution of the present application has the following beneficial effects: The present application realizes the acquisition of multi-modal data, providing a basis for subsequent cross-modal data fusion. An association is established between the anatomical structure features of the ultrasound image and the biomolecular features of the multi-omics mass spectrometry data, solving the problem of isolated multi-modal data. The deformation characteristics of the thyroid tissue are quantified, providing spatial deformation information for multi-task optimization. The optimization objectives of the thyroid nodule localization and malignancy prediction tasks are clarified, supporting the collaborative optimization of multi-task learning. Through non-linear coupling calculation, the gradient contribution weights between tasks are dynamically adjusted, solving the problem of gradient conflict between tasks. The efficient fusion and spatial alignment of multi-modal data are realized, generating a high-precision multi-omics mass spectrometry map. Based on the multi-omics mass spectrometry map, the three-dimensional localization coordinates and the malignancy probability prediction value of the thyroid nodule are generated, improving the comprehensiveness and accuracy of the diagnosis.

[0018] Furthermore, the embodiment of the present application also constructs a three-dimensional spatial weight matrix based on the cross-modal correlation relationship, combines the anatomical structure features and the spatial distribution features of biomolecules, performs multi-level spatial decomposition on the anatomical structure features to obtain the texture distribution patterns at different anatomical levels within the thyroid tissue; layer-by-layer fuses the texture distribution patterns with multi-omics mass spectrometry data to generate cross-modal correlation features; performs task-aware filtering on the cross-modal correlation features according to the gradient contribution weights to obtain the filtered cross-modal correlation features; non-linearly superimposes the filtered cross-modal correlation features with the thyroid anatomical structure deformation field to generate an initial multi-omics mass spectrometry map; and optimizes the initial multi-omics mass spectrometry map to generate a target multi-omics mass spectrometry map.

[0019] By constructing a three-dimensional spatial weight matrix and performing multi-level spatial decomposition, the efficient fusion of thyroid anatomical structure features and multi-omics mass spectrometry data is achieved; by task-aware filtering and non-linear superposition, the spatial alignment accuracy between cross-modal correlation features and the anatomical structure deformation field is optimized; the finally generated target multi-omics mass spectrometry map improves the comprehensiveness and accuracy of thyroid disease diagnosis, and solves the problems of isolated multi-modal data and insufficient spatial alignment accuracy.

[0020] These aspects or other aspects of the present application will be more clearly understood in the following description of the embodiments. BRIEF DESCRIPTION OF THE DRAWINGS

[0021] In order to more clearly illustrate the technical solutions in the embodiments of the present application or the prior art, the following will briefly introduce the drawings required for use in the description of the embodiments or the prior art. Obviously, the drawings in the following description are some embodiments of the present application. For those of ordinary skill in the art, other drawings can be obtained based on these drawings without creative efforts.

[0022] Figure 1 It is a flowchart of a method for constructing and analyzing a multi-omics mass spectrometry map of thyroid diseases provided by an embodiment of the present application; Figure 2 It is a schematic structural diagram of a system for constructing and analyzing a multi-omics mass spectrometry map of thyroid diseases provided by an embodiment of the present application; Figure 3 It is a schematic structural diagram of a computing device provided by an embodiment of the present application. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0023] In order to enable those skilled in the art to better understand the solutions of the present application, the following will clearly and completely describe the technical solutions in the embodiments of the present application with reference to the drawings in the embodiments of the present application.

[0024] In some of the processes described in the specification, claims, and the above-mentioned drawings of the present application, a plurality of operations appear in a specific order. However, it should be clearly understood that these operations may not be executed in the order in which they appear herein or may be executed in parallel. The serial numbers of the operations, such as 101, 102, etc., are only used to distinguish different operations, and the serial numbers themselves do not represent any execution order. Additionally, these processes may include more or fewer operations, and these operations may be executed in sequence or in parallel. It should be noted that the descriptions such as "first", "second", etc. in this document are used to distinguish different messages, devices, modules, etc., and do not represent a sequential order, nor do they limit that "first" and "second" are of different types.

[0025] Next, the technical solutions in the embodiments of the present application will be clearly and completely described in conjunction with the accompanying drawings in the embodiments of the present application. Obviously, the described embodiments are only a part of the embodiments of the present application, rather than all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without making creative efforts belong to the scope of protection of the present application.

[0026] The present application generates a cross-modal association relationship based on the anatomical structure features (including thyroid nodules) of the thyroid ultrasound image and the biomolecular spatial distribution features of the multi-omics mass spectrometry data by acquiring the thyroid ultrasound image and the multi-omics mass spectrometry data; generates a thyroid anatomical structure deformation field based on the anatomical structure features, and constructs a multi-task optimization objective (including the thyroid nodule localization task and the malignancy prediction task) according to the thyroid disease diagnosis requirements; in the multi-task learning iteration process, performs non-linear coupling calculation on the backpropagation error of the thyroid anatomical structure deformation field to obtain the gradient contribution weights of each task; generates cross-modal association features based on the cross-modal association relationship, and combines the thyroid anatomical structure deformation field and the gradient contribution weights to generate a target multi-omics mass spectrometry map, ensuring that the spatial coordinate error between the metabolomics signal hot spot area and the microcalcification cluster in the thyroid ultrasound image is within the preset anatomical tolerance range; performs thyroid pathological feature analysis on the target multi-omics mass spectrometry map to obtain a thyroid pathological analysis result (including the three-dimensional localization coordinates of the thyroid nodule and the malignancy probability prediction value), thereby realizing the accurate diagnosis and prediction of thyroid diseases.

[0027] Figure 1 The following is a flowchart of a method for constructing and analyzing a multi-omics mass spectrometry map for thyroid diseases provided by an embodiment of the present application, as Figure 1 shown. The method includes: Step 101: Acquire a thyroid ultrasound image and multi-omics mass spectrometry data, and generate a cross-modal association relationship according to the anatomical structure features of the thyroid ultrasound image and the biomolecular spatial distribution features of the multi-omics mass spectrometry data.

[0028] In this step, the multi-omics mass spectrometry data includes the metaboliteomics signal intensity distribution and the proteomics biomarker expression levels. The cross-modal association relationships are used to fuse data of different modalities. The anatomical structures include thyroid nodules. The thyroid ultrasound image is an image of the thyroid structure obtained by ultrasonic imaging technology, which is used to display the morphological, size, location, and nodule and other anatomical structure features of the thyroid. For example, the boundary, internal echo, and blood flow conditions of the thyroid nodule can be clearly shown in the image. The biomolecular spatial distribution characteristics refer to the spatial distribution rules of biomolecules such as metaboliteomics and proteomics in the multi-omics mass spectrometry data. For example, the metaboliteomics signals are highly concentrated in the thyroid nodule area.

[0029] In actual operation, the thyroid ultrasound image and the multi-omics mass spectrometry data are obtained, and cross-modal association relationships are generated according to the anatomical structure features of the thyroid ultrasound image and the biomolecular spatial distribution characteristics of the multi-omics mass spectrometry data. Specifically, the anatomical structure features of the ultrasound image are associated with the biomolecular distribution characteristics of the mass spectrometry data through feature alignment technology to generate cross-modal association relationships.

[0030] For example, the thyroid ultrasound image of patient A shows that the nodule is located in the right lobe and its size is 1.5 cm × 1.0 cm; the multi-omics mass spectrometry data shows that the metaboliteomics signals are highly concentrated in the right lobe area. The ultrasound image and the mass spectrometry data are aligned through spatial registration technology to generate cross-modal association relationships.

[0031] Step 102: Generate a thyroid anatomical structure deformation field based on the anatomical structure features, and construct a multi-task optimization objective according to the preset thyroid disease diagnosis requirements.

[0032] In this step, the thyroid anatomical structure deformation field is used to describe the deformation and spatial distribution of the thyroid structure. The multi-task optimization objective includes a thyroid nodule localization task and a thyroid nodule malignancy prediction task, which are used to guide multi-task learning. The thyroid disease diagnosis requirements include the precise localization and malignancy prediction of thyroid nodules. Specifically, the precise localization requirement requires accurately identifying the three-dimensional coordinates of the nodule, and the malignancy prediction requirement requires evaluating the malignancy probability of the nodule. For example, the diagnosis requirements may include a nodule localization error < 0.1 cm and a malignancy prediction accuracy > 90%.

[0033] In actual operation, a thyroid anatomical structure deformation field is generated based on the anatomical structure features, and a multi-task optimization objective is constructed according to the preset thyroid disease diagnosis requirements. Specifically, a thyroid anatomical structure deformation field is generated through a deformation model (such as elastic registration), and a multi-task optimization objective (such as the mean square error of nodule localization and the cross-entropy loss of malignancy prediction) is constructed.

[0034] For example, an anatomical structure deformation field is generated based on the thyroid ultrasound image of patient A to describe the deformation characteristics of the nodule region; a multi-task optimization objective is constructed, including a nodule localization task (target error < 0.1 cm) and a malignancy prediction task (target accuracy rate > 90%).

[0035] Step 103: During the multi-task learning iteration process, perform non-linear coupling calculation on the backpropagation error of the thyroid anatomical structure deformation field to obtain the gradient contribution weights of the thyroid nodule localization task and the thyroid nodule malignancy prediction task respectively.

[0036] In this step, the gradient contribution weight refers to the contribution ratio of the thyroid nodule localization task and the malignancy prediction task to the model training in multi-task learning, and is used to dynamically adjust the optimization objective weights between tasks. In actual operation, during the multi-task learning iteration process, perform non-linear coupling calculation on the backpropagation error of the thyroid anatomical structure deformation field to obtain the gradient contribution weights of the thyroid nodule localization task and the thyroid nodule malignancy prediction task respectively. Specifically, calculate the gradient contribution weights of each task through a non-linear coupling algorithm (such as weighted summation).

[0037] For example, during the multi-task learning iteration, the calculated gradient contribution weight of the nodule localization task is 0.6, and the gradient contribution weight of the malignancy prediction task is 0.4.

[0038] Step 104: Generate cross-modal association features according to the cross-modal association relationship, and based on the thyroid anatomical structure deformation field and the cross-modal association features, combine the gradient contribution weights to generate a target multi-omics mass spectrometry map, where the error between the spatial coordinates of the metabolomics signal hot spot region in the target multi-omics mass spectrometry map and the microcalcification cluster in the thyroid ultrasound image is within the preset anatomical structure tolerance range.

[0039] In this step, the target multi-omics mass spectrometry map includes the concentration distribution of metabolomics-protein dual-omics biomarkers, the three-dimensional localization coordinates of the nodule, and the malignancy risk probability value, and is used to accurately locate the metabolomics signal hot spot region. The spatial coordinate error is an explicit indicator of cross-modal data alignment, while the backpropagation error is an implicit indicator of model training optimization. The cross-modal association features are used to describe the association relationship between two types of modal data. For example, the association between the boundary features of the nodule in the ultrasound image and the spatial distribution features of the metabolomics signal in the mass spectrometry data. The metabolomics signal hot spot region refers to the high-value region of the metabolomics signal intensity in the thyroid tissue, which reflects the spatial distribution law of metabolite concentration.

[0040] In actual operation, cross-modal association features are generated according to the cross-modal association relationship, and based on the thyroid anatomical structure deformation field and the cross-modal association features, combined with the gradient contribution weight, a target multi-omics mass spectrometry map is generated. Specifically, a target multi-omics mass spectrometry map is generated through a feature fusion technique (such as weighted fusion) to ensure that the spatial coordinate error between the metabolomics signal hotspot region and the microcalcification cluster in the thyroid ultrasound image is within the preset anatomical structure tolerance range.

[0041] For example, based on the cross-modal association features and anatomical structure deformation field of patient A, a target multi-omics mass spectrometry map is generated, where the spatial coordinate error between the metabolomics signal hotspot region and the microcalcification cluster in the ultrasound image is <0.1 cm.

[0042] Step 105: Analyze the thyroid pathological features of the target multi-omics mass spectrometry map to obtain the thyroid pathological analysis result.

[0043] In this step, the thyroid pathological analysis result includes the three-dimensional positioning coordinates and malignant probability prediction value of the thyroid nodule. For example, the analysis result may include the three-dimensional coordinates of the nodule (1.2 cm, 0.8 cm, 0.5 cm) and the malignant probability prediction value of 85%. The three-dimensional positioning coordinates are spatially corrected by fusing the gradient change direction of the metabolomics signal intensity distribution and the texture expansion pattern of the hypoechoic region in the thyroid ultrasound image, and the malignant probability prediction value is generated by coupling the expression mutation position of the proteomics biomarker and the anatomical deformation characteristics at the thyroid-carotid artery contact surface to generate a risk confidence index.

[0044] In actual operation, the thyroid pathological features of the target multi-omics mass spectrometry map are analyzed to obtain the thyroid pathological analysis result. Specifically, the target multi-omics mass spectrometry map is analyzed through a pathological feature extraction technique (such as a deep learning model) to generate the three-dimensional positioning coordinates and malignant probability prediction value of the thyroid nodule.

[0045] For example, by analyzing the pathological features of the target multi-omics mass spectrometry map of patient A, the three-dimensional positioning coordinates of the thyroid nodule are (1.2 cm, 0.8 cm, 0.5 cm), and the malignant probability prediction value is 85%.

[0046] By obtaining thyroid ultrasound images and multi-omics mass spectrometry data, generating cross-modal association relationships and thyroid anatomical structure deformation fields, and constructing a multi-task optimization objective; in the multi-task learning iteration process, generating gradient contribution weights through non-linear coupling calculations and combining cross-modal association features to generate a target multi-omics mass spectrometry map; finally, generating a thyroid pathological analysis result through pathological feature analysis, the accurate diagnosis and prediction of thyroid diseases are realized, and the accuracy and reliability of the diagnosis are improved.

[0047] To solve the problems of isolation of multimodal data and insufficient spatial alignment accuracy in the diagnosis of thyroid diseases, based on cross-modal correlation relationships, this application combines the anatomical structure features of thyroid ultrasound images with the biomolecular spatial distribution features of multi-omics mass spectrometry data to construct a three-dimensional spatial weight matrix, quantifying the feature matching degree between ultrasound image voxels and mass spectrometry detection points; through multi-level spatial decomposition, texture distribution patterns at different anatomical levels within the thyroid tissue are extracted (including the overall contour at the global level, regional nodule distribution at the mesoscopic level, and microcalcification cluster texture at the local level); the texture distribution patterns are fused layer by layer with the multi-omics mass spectrometry data to generate cross-modal correlation features; task-aware filtering is performed on the cross-modal correlation features according to the gradient contribution weights, dynamically adjusting the feature contribution ratios of the thyroid nodule localization task and the malignant prediction task; the filtered cross-modal correlation features are non-linearly superimposed with the thyroid anatomical structure deformation field to generate an initial multi-omics mass spectrometry map; by iteratively optimizing the spatial coordinate error between the metabolomics signal hot spot region and the microcalcification cluster, a target multi-omics mass spectrometry map is generated, realizing the efficient fusion and spatial alignment of multimodal data.

[0048] In some embodiments, step 104 of generating cross-modal correlation features according to the cross-modal correlation relationships, and generating a target multi-omics mass spectrometry map based on the thyroid anatomical structure deformation field and the cross-modal correlation features, in combination with the gradient contribution weights, includes: Step 201: Based on the cross-modal correlation relationships, combining the anatomical structure features with the biomolecular spatial distribution features, construct a three-dimensional spatial weight matrix, and perform multi-level spatial decomposition on the anatomical structure features according to the three-dimensional spatial weight matrix to obtain texture distribution patterns at different anatomical levels within the thyroid tissue.

[0049] In step 201, the three-dimensional spatial weight matrix is a three-dimensional tensor, and each of its elements represents the feature matching degree between the ultrasound image voxel and the mass spectrometry detection point at a specific spatial position (x, y, z), which is used to quantify the spatial consistency of multimodal data. Multilevel spatial decomposition refers to the multi-resolution analysis of the anatomical structure features of the thyroid ultrasound image to extract texture distribution patterns at different levels. Decomposition process: Apply the three-dimensional spatial weight matrix to the anatomical structure features of the ultrasound image for multilevel spatial decomposition. For example, through a convolution operation, the weight matrix is decomposed layer by layer with the anatomical structure features of the ultrasound image to extract texture distribution patterns at different levels. According to the distribution of the weight matrix, extract the texture distribution patterns at different anatomical levels within the thyroid tissue. For example, the overall contour at the global level, the regional nodule distribution at the mesoscopic level, and the microcalcification cluster texture at the local level. Specific numerical example: Assume that the physical space range of the thyroid region is 0 - 100 mm on the x-axis, 0 - 100 mm on the y-axis, and 0 - 50 mm on the z-axis, and construct a three-dimensional spatial weight matrix of 100×100×50, where each element A represents the feature matching degree between the ultrasound image voxel and the mass spectrometry detection point at the spatial position (i, j, k). For example, the weight value at position (50, 60, 20) is 0.12, the weight value at position (55, 65, 22) is 0.15, and the weight value at position (70, 80, 25) is 0.08. Based on this weight matrix, perform multilevel spatial decomposition on the anatomical structure features of the thyroid ultrasound image: At the global level, perform low-pass filtering on the weight matrix and multiply each element by the low-resolution features (64×64×16) of the ultrasound image. For example, the weight value at position (25, 30, 8) is 0.10, and the texture intensity is 0.8. The value at this position in the global-level texture distribution pattern is 0.10×0.8 = 0.08, which reflects the topological relationship between the overall contour of the thyroid and adjacent organs; At the mesoscopic level, perform band-pass filtering on the weight matrix and multiply each element by the mid-resolution features (128×128×32) of the ultrasound image. For example, the weight value at position (50, 60, 16) is 0.12, and the texture intensity is 0.6. The value at this position in the mesoscopic-level texture distribution pattern is 0.12×0.6 = 0.072, which reflects the regional nodule distribution and blood flow signal; At the local level, perform high-pass filtering on the weight matrix and multiply each element by the high-resolution features (256×256×64) of the ultrasound image. For example, the weight value at position (100, 120, 32) is 0.15, and the texture intensity is 0.9. The value at this position in the local-level texture distribution pattern is 0.15×0.9 = 0.135, which reflects the microcalcification cluster texture details.Finally, texture distribution patterns at the global level, meso level, and local level are output, which respectively describe the overall contour of the thyroid gland, the regional nodule distribution, and the texture of microcalcification clusters. The texture distribution pattern refers to the texture features at different anatomical levels in the thyroid ultrasound image, including the overall contour at the global level, the regional nodule distribution at the meso level, and the texture of microcalcification clusters at the local level. Thyroid tissue refers to the physiological tissue of the thyroid gland, including components such as thyroid cells, blood vessels, and nerves. For example, thyroid tissue is composed of thyroid follicular cells and interstitial tissue and is responsible for secreting thyroid hormones. The thyroid anatomical structure refers to the overall and local anatomical features of the thyroid gland, including the edge of the thyroid gland, internal tissues, and topological relationships with adjacent organs (such as the carotid artery and trachea). For example, the overall contour of the thyroid gland, the regional blood flow distribution, and the texture of local microcalcification clusters. A thyroid nodule refers to a locally abnormal proliferation area within the thyroid tissue, usually manifested as a hypoechoic or hyperechoic area in the ultrasound image, which may be benign or malignant. For example, a thyroid nodule may appear as a mass or cyst within the thyroid tissue. The thyroid anatomical structure is the overall and local anatomical features of the thyroid gland, while the thyroid nodule is a local pathological feature within the thyroid anatomical structure. Thyroid tissue is the physiological basis of the thyroid anatomical structure, and the thyroid anatomical structure is a spatial and functional description of the thyroid tissue. The thyroid nodule is a local pathological feature within the thyroid tissue and is also part of the thyroid anatomical structure.

[0050] In the embodiment of the present application, first, based on the cross-modal correlation relationship, combining the anatomical structure features of the thyroid ultrasound image and the biomolecular spatial distribution features of multi-omics mass spectrometry data, a three-dimensional spatial weight matrix is constructed. Specifically, by calculating the spatial distance and feature similarity between the ultrasound image voxels and the mass spectrometry detection points, each element of the weight matrix is generated. Then, according to the three-dimensional spatial weight matrix, the anatomical structure features of the thyroid ultrasound image are decomposed at multiple levels in space, and the overall contour at the global level, the regional nodule distribution at the meso level, and the texture of microcalcification clusters at the local level are extracted. Finally, the texture distribution patterns at different anatomical levels within the thyroid tissue are obtained.

[0051] Step 202: Layer-by-layer fuse the texture distribution pattern with the multi-omics mass spectrometry data to generate cross-modal correlation features.

[0052] In step 202, layer-by-layer fusion means aligning and fusing the texture distribution pattern and the multi-omics mass spectrometry data according to the hierarchical resolution to generate cross-modal correlation features.

[0053] In the embodiments of the present application, the texture distribution pattern is fused with multi-omics mass spectrometry data layer by layer. Specifically, at the global level, the overall contour of the thyroid gland is aligned with the global distribution of metabolomics signals; at the mesoscopic level, the regional nodule distribution is aligned with the regional distribution of metabolomics signals; at the local level, the texture of microcalcification clusters is aligned with the local peaks of metabolomics signals. Through layer-by-layer fusion, cross-modal correlation features are generated to achieve efficient fusion of multi-modal data.

[0054] Step 203: Perform task-aware filtering on the cross-modal correlation features according to the gradient contribution weights to obtain filtered cross-modal correlation features.

[0055] In step 203, task-aware filtering dynamically adjusts the contribution ratio of the feature channels of the cross-modal correlation feature tensor between the nodule localization task and the malignancy prediction task. Among them, the feature channels of the nodule localization task retain the spatial coupling information of the texture expansion pattern inside the thyroid tissue and the gradient change direction of metabolomics signals, and the feature channels of the malignancy prediction task retain the temporal correlation information of the expression mutation positions of proteomics markers and the deformation characteristics of the thyroid-carotid artery contact surface.

[0056] In the embodiments of the present application, task-aware filtering is performed on the cross-modal correlation features according to the gradient contribution weights. Specifically, by calculating the backpropagation errors of the thyroid nodule localization task and the malignancy prediction task, the gradient contribution weights between tasks are dynamically adjusted. Then, the cross-modal correlation features are filtered according to the gradient contribution weights, and the features that contribute the most to the thyroid nodule localization task and the malignancy prediction task are retained to generate filtered cross-modal correlation features.

[0057] Step 204: Nonlinearly superimpose the filtered cross-modal correlation features and the thyroid anatomical structure deformation field to generate an initial multi-omics mass spectrometry map.

[0058] In step 204, nonlinear superimposition means weighted fusion of the filtered cross-modal correlation features and the thyroid anatomical structure deformation field.

[0059] In the embodiments of the present application, the filtered cross-modal correlation features and the thyroid anatomical structure deformation field are nonlinearly superimposed. Specifically, an initial multi-omics mass spectrometry map is generated by weighted fusion of cross-modal correlation features and deformation parameters. Then, by iteratively optimizing the spatial coordinate error between the hot spots of metabolomics signals and microcalcification clusters, the spatial alignment accuracy of the initial map is corrected.

[0060] Step 205: Optimize the initial multi-omics mass spectrometry map to generate a target multi-omics mass spectrometry map.

[0061] In step 205, the optimization is performed by constraining the spatial correlation degree threshold between the expression mutation position of the proteomics biomarker and the texture expansion pattern of the hypoechoic region in the ultrasound image, while restricting the deformation amplitude of the three-dimensional localization coordinates of the nodule and the deformation field of the thyroid anatomical structure within a preset tolerance range.

[0062] In the embodiment of the present application, the initial multi-omics mass spectrometry map is optimized. Specifically, by constraining the spatial correlation degree threshold between the expression mutation position of the proteomics biomarker and the texture expansion pattern of the hypoechoic region, and restricting the deformation amplitude of the three-dimensional localization coordinates of the nodule and the deformation field of the thyroid anatomical structure within a preset tolerance range, a target multi-omics mass spectrometry map is generated.

[0063] The following is a specific example: Assume that the physical space range of the thyroid region is 0 - 100 mm on the x-axis, 0 - 100 mm on the y-axis, and 0 - 50 mm on the z-axis, and a three-dimensional space weight matrix of 100×100×50 is constructed. At the global level, the overall contour of the thyroid is aligned with the global distribution of metabolomics signals to generate a global-level texture distribution pattern; at the mesoscopic level, the regional nodule distribution is aligned with the regional distribution of metabolomics signals to generate a mesoscopic-level texture distribution pattern; at the local level, the texture of the microcalcification cluster is aligned with the local peak of the metabolomics signal to generate a local-level texture distribution pattern. Task-aware filtering is performed on the cross-modal association features according to the gradient contribution weight, and the features that contribute the most to the thyroid nodule localization task and the malignant prediction task are retained. The filtered cross-modal association features are non-linearly superimposed with the deformation field of the thyroid anatomical structure to generate an initial multi-omics mass spectrometry map. By iteratively optimizing the spatial coordinate error between the hot spot region of the metabolomics signal and the microcalcification cluster, a target multi-omics mass spectrometry map is generated.

[0064] Through the construction of a three-dimensional space weight matrix and multi-level space decomposition, the efficient fusion of thyroid anatomical structure features and multi-omics mass spectrometry data is achieved; through task-aware filtering and non-linear superposition, the spatial alignment accuracy of cross-modal association features and the anatomical structure deformation field is optimized; the finally generated target multi-omics mass spectrometry map improves the comprehensiveness and accuracy of thyroid disease diagnosis, and solves the problems of isolation of multi-modal data and insufficient spatial alignment accuracy.

[0065] In order to further improve the accuracy and efficiency of multimodal data fusion, based on the hierarchical resolution of thyroid ultrasound images, the present application dynamically adjusts the spatial positions of mass spectrometry detection points within the coordinate mapping ranges of each level. The coordinate mapping range of the global level covers the topological boundaries of the overall thyroid contour and adjacent organs, and the mapping range of the local level is limited to the local spatial neighborhood of microcalcification clusters; at each hierarchical resolution, according to the spatial distances between the adjusted mass spectrometry detection points and the ultrasound image voxels, spatial interpolation compensation processing is performed on the intensity distribution of metabolomics signals in the multi-omics mass spectrometry data to solve the problem of insufficient spatial resolution of mass spectrometry data; according to the consistency between the spatial distribution path of proteomic biomarker expression levels and the directional characteristics of the texture distribution pattern, directional enhancement processing is performed on the texture features to enhance the ability of the texture features to express pathological information; the compensated intensity distribution of metabolomics signals and the enhanced texture features are channel stitched according to the hierarchical resolution to generate cross-modal association features. The stitching channels of the global level are used to calculate the covariance matrix of the overall thyroid deformation trend and the global distribution of metabolomics signals, and the stitching channels of the local level are used to calculate the spatial offset between the texture details of microcalcification clusters and the peak positions of metabolic signals, realizing the efficient fusion and spatial alignment of multimodal data.

[0066] In some embodiments, the step of layer-by-layer fusing the texture distribution pattern and the multi-omics mass spectrometry data to generate cross-modal association features in step 202 includes: Step 301: Analyze the multi-omics mass spectrometry data to obtain mass spectrometry detection points, and dynamically adjust the mass spectrometry detection points within the coordinate mapping ranges of each level based on the hierarchical resolution of the thyroid ultrasound image. The coordinate mapping range of the global level covers the topological boundaries of the overall thyroid contour and adjacent organs, and the mapping range of the local level is limited to the local spatial neighborhood of microcalcification clusters in the thyroid ultrasound image.

[0067] In step 301, the mass spectrometry detection points refer to the detection points in the multi-omics mass spectrometry data collected by the mass spectrometry detection equipment, including three-dimensional spatial coordinates and biomolecular information (such as metabolomics signal intensity, proteomics biomarker expression level). The hierarchical resolution refers to the multi-scale resolution of the thyroid ultrasound image, including the global level (low resolution), the mesoscopic level (medium resolution), and the local level (high resolution). The coordinate mapping range refers to the spatial distribution range of the mass spectrometry detection points within the thyroid region. The global level covers the topological boundaries of the overall thyroid contour and adjacent organs, and the local level is limited to the local spatial neighborhood of the microcalcification clusters. The adjacent organs of the thyroid include the carotid arteries, which are located on both sides of the thyroid and supply blood to the thyroid; the trachea, which is located behind the thyroid and is in close contact with the thyroid; the esophagus, which is located behind the thyroid and is adjacent to part of the thyroid region; the recurrent laryngeal nerve, which passes through the back of the thyroid and controls the movement of the vocal cords; and the parathyroid glands, which are located on the back of the thyroid and are responsible for regulating calcium and phosphorus metabolism. The adjacent range generally refers to the organs and tissues that are in direct contact with the thyroid or are no more than 10 mm away from the thyroid edge.

[0068] In the embodiments of the present application, first, the multi-omics mass spectrometry data is analyzed to obtain the mass spectrometry detection points. Then, based on the hierarchical resolution of the thyroid ultrasound image, the spatial positions of the mass spectrometry detection points are dynamically adjusted within the coordinate mapping range of each level. Specifically, at the global level, the coordinate mapping range of the mass spectrometry detection points covers the topological boundaries of the overall thyroid contour and adjacent organs (such as the carotid arteries, trachea); at the local level, the coordinate mapping range of the mass spectrometry detection points is limited to the local spatial neighborhood of the microcalcification clusters. By dynamically adjusting the coordinate mapping range of the mass spectrometry detection points, the spatial alignment accuracy of the multi-modal data is ensured.

[0069] Step 302: At each level of resolution, according to the spatial distance between the adjusted mass spectrometry detection points and the ultrasound image voxels, perform spatial interpolation compensation processing on the metabolomics signal intensity distribution in the multi-omics mass spectrometry data to obtain the compensated metabolomics signal intensity distribution.

[0070] In step 302, spatial interpolation compensation means performing interpolation compensation on the metabolomics signal intensity distribution according to the spatial distance between the mass spectrometry detection points and the ultrasound image voxels to solve the problem of insufficient spatial resolution of the mass spectrometry data. An ultrasound image voxel refers to the smallest spatial unit of the thyroid ultrasound image, including three-dimensional coordinates and texture intensity information.

[0071] In the embodiments of the present application, at each level of resolution, according to the spatial distance between the adjusted mass spectrometry detection points and the ultrasound image voxels, spatial interpolation compensation processing is performed on the intensity distribution of metabolomics signals in the multi-omics mass spectrometry data. Specifically, by calculating the texture intensity gradient direction of the ultrasound image voxels within a preset neighborhood of the mass spectrometry detection points, the attenuation coefficient of the metabolomics signals is corrected to generate the compensated intensity distribution of the metabolomics signals.

[0072] Step 303: According to the consistency in space between the spatial distribution path of the expression levels of proteomics markers in the multi-omics mass spectrometry data and the directional characteristics of the texture distribution pattern, perform directional enhancement processing on the texture characteristics of the texture distribution pattern to obtain enhanced texture characteristics.

[0073] In step 303, the directional characteristics of the texture distribution pattern refer to the spatial expansion directions of the texture characteristics (such as hypoechoic regions, microcalcification clusters) in the thyroid ultrasound image. Directional enhancement means enhancing the texture characteristics according to the consistency between the spatial distribution path of the expression levels of proteomics markers and the directional characteristics of the texture distribution pattern, so as to improve the ability of the texture characteristics to express pathological information. The expression level of proteomics markers refers to the expression level of specific proteins in the thyroid tissue, which is usually related to the disease state (such as malignant tumors).

[0074] In the embodiments of the present application, according to the consistency between the spatial distribution path of the expression levels of proteomics markers and the directional characteristics of the texture distribution pattern, directional enhancement processing is performed on the texture characteristics. Specifically, by matching the spatial distribution path of the expression mutation positions of protein markers and the blood flow signals in the ultrasound image, the directional information of the texture characteristics is enhanced to generate enhanced texture characteristics.

[0075] Step 304: Perform channel splicing on the compensated intensity distribution of the metabolomics signals and the enhanced texture characteristics according to the level of resolution to generate cross-modal association features, where the splicing channels at the global level include the covariance matrix of the overall deformation trend of the thyroid and the global distribution of the metabolomics signals, and the splicing channels at the local level include the texture details of the microcalcification clusters and the spatial offset of the peak positions of the metabolic signals.

[0076] In step 304, a specific numerical example of channel concatenation: Assume data, the compensated metabolomics signal intensity distribution (global level): Matrix size: 64×64×32 (x, y, z coordinates); Example value: The signal intensity at coordinate (10, 20, 5) is 1200. The enhanced texture feature (global level): Matrix size: 64×64×32 (x, y, z coordinates); Example value: The texture intensity at coordinate (10, 20, 5) is 0.8. Hierarchical resolution: Global level: 64×64×32; Local level: 128×128×64. Global level concatenation: Concatenate the compensated metabolomics signal intensity distribution and the enhanced texture feature according to the global level resolution. Example: At coordinate (10, 20, 5), the metabolomics signal intensity is 1200 and the texture intensity is 0.8; the cross-modal correlation feature value after concatenation is (1200, 0.8).

[0077] In the embodiments of the present application, the compensated metabolomics signal intensity distribution and the enhanced texture feature are concatenated by channels according to the hierarchical resolution to generate cross-modal correlation features. Specifically, at the global level, the concatenated channel is used to calculate the covariance matrix of the overall deformation trend of the thyroid gland and the global distribution of the metabolomics signal; at the local level, the concatenated channel is used to calculate the spatial offset between the texture details of the microcalcification cluster and the peak position of the metabolic signal.

[0078] The following is a specific example: Assume that the physical space range of the thyroid region is 0~100 mm on the x-axis, 0~100 mm on the y-axis, and 0~50 mm on the z-axis, and a three-dimensional spatial weight matrix of 100×100×50 is constructed. At the global level, the coordinate mapping range of the mass spectrometry detection points covers the overall contour of the thyroid gland and the topological boundaries of adjacent organs (such as the carotid artery and trachea); at the local level, the coordinate mapping range of the mass spectrometry detection points is limited to the local spatial neighborhood of the microcalcification cluster. At each hierarchical resolution, according to the spatial distance between the mass spectrometry detection points and the ultrasound image voxels, spatial interpolation compensation processing is performed on the metabolomics signal intensity distribution to generate the compensated metabolomics signal intensity distribution. According to the consistency between the spatial distribution path of the proteomics biomarker expression level and the directional characteristics of the texture distribution pattern, directional enhancement processing is performed on the texture feature to generate the enhanced texture feature. The compensated metabolomics signal intensity distribution and the enhanced texture feature are concatenated by channels according to the hierarchical resolution to generate cross-modal correlation features.

[0079] By dynamically adjusting the coordinate mapping range of mass spectrometry detection points, the spatial alignment accuracy of multi-modal data is ensured; through spatial interpolation compensation and directional enhancement processing, the pathological information expression ability of metabolomics signals and texture features is improved; through channel splicing to generate cross-modal correlation features, efficient fusion and spatial alignment of multi-modal data are achieved, enhancing the comprehensiveness and accuracy of thyroid disease diagnosis.

[0080] To address the problems of insufficient deformation characteristics of the thyroid anatomical structure and insufficient fusion accuracy of multi-modal data, this application decomposes the thyroid anatomical structure deformation field into local deformation parameters within the thyroid tissue and the global deformation trend. Among them, the local deformation parameters are generated by quantifying the relative displacement of the contact surface between the thyroid edge and the carotid artery, and the global deformation trend is generated by extracting the spatial offset direction of the overall thyroid contour; based on the local deformation parameters, spatial coordinate transformation is performed on the filtered cross-modal correlation features to dynamically adjust the three-dimensional coordinate mapping relationship of the mass spectrometry detection points; based on the global deformation trend, hierarchical weight allocation is performed on the cross-modal correlation features after spatial coordinate transformation, and according to the overlap degree between the spatial density distribution of microcalcification clusters and the hot spots of metabolomics signals, the feature contribution ratio between the global level and the local level is dynamically adjusted; the cross-modal correlation features after weight allocation are non-linearly superimposed with the deformation parameters of the thyroid anatomical structure deformation field to generate an initial multi-omics mass spectrometry map, and by iteratively optimizing the spatial coordinate error between the hot spots of metabolomics signals and microcalcification clusters, the spatial alignment accuracy of the initial map is corrected, realizing the efficient fusion and spatial alignment of multi-modal data.

[0081] In some embodiments, the non-linear superposition of the filtered cross-modal correlation features and the thyroid anatomical structure deformation field in step 204 to generate an initial multi-omics mass spectrometry map includes: Step 401: Decompose the thyroid anatomical structure deformation field into local deformation parameters within the thyroid tissue and the global deformation trend.

[0082] In step 401, the local deformation parameter refers to the deformation amount of a local area within the thyroid tissue, which is generated by quantifying the relative displacement of the contact surface between the thyroid edge and the carotid artery. The global deformation trend refers to the spatial offset direction of the overall thyroid contour, which is generated by extracting the deformation characteristics of the overall thyroid contour.

[0083] In the embodiments of this application, the thyroid anatomical structure deformation field is decomposed into local deformation parameters and the global deformation trend. Specifically, the local deformation parameters are generated by quantifying the relative displacement of the contact surface between the thyroid edge and the carotid artery, reflecting the deformation characteristics of the local area; the global deformation trend is generated by extracting the spatial offset direction of the overall thyroid contour, reflecting the overall deformation trend of the thyroid.

[0084] Step 402: Based on the local deformation parameters, perform a spatial coordinate transformation on the filtered cross-modal correlation features to obtain the cross-modal correlation features after spatial coordinate transformation.

[0085] In step 402, the spatial coordinate transformation refers to, according to the local deformation parameters, performing a coordinate mapping on the filtered cross-modal correlation features to adjust the three-dimensional coordinate positions of the mass spectrometry detection points.

[0086] In the embodiment of the present application, based on the local deformation parameters, a spatial coordinate transformation is performed on the filtered cross-modal correlation features. Specifically, by matching the texture extension direction of the hypoechoic region in the thyroid ultrasound image with the metabolicomics signal gradient direction of the cross-modal correlation feature tensor, the three-dimensional coordinate mapping relationship of the mass spectrometry detection points is dynamically adjusted to generate the cross-modal correlation features after spatial coordinate transformation.

[0087] Step 403: Based on the global deformation trend, perform an inter-level weight assignment on the cross-modal correlation features after spatial coordinate transformation to obtain the cross-modal correlation features after weight assignment.

[0088] In step 403, the inter-level weight assignment is to adjust the feature contribution ratio between the global level and the local level according to the overlap degree between the spatial density distribution of the microcalcification clusters and the hot spot region of the metabolicomics signal. The spatial coordinates of the microcalcification clusters are the basic data for the spatial density distribution of the microcalcification clusters.

[0089] In the embodiment of the present application, based on the global deformation trend, an inter-level weight assignment is performed on the cross-modal correlation features after spatial coordinate transformation. Specifically, according to the overlap degree between the spatial density distribution of the microcalcification clusters in the thyroid tissue and the hot spot region of the metabolicomics signal, the feature contribution ratio between the global level and the local level is dynamically adjusted to generate the cross-modal correlation features after weight assignment.

[0090] Step 404: Non-linearly superimpose the cross-modal correlation features after weight assignment with the deformation parameters of the thyroid anatomical structure deformation field, and generate an initial multi-omics mass spectrometry map by iteratively optimizing the spatial coordinates of the hot spot region of the metabolicomics signal and the microcalcification clusters.

[0091] In step 404, the superposition formula example: Superposition result = Cross-modal correlation feature × Deformation parameter + Deformation parameter^2. 1. Input data: Cross-modal correlation feature: Global level: Deformation trend of the overall thyroid contour (eigenvalue: 0.8); Local level: Texture details of microcalcification clusters (eigenvalue: 0.6). Deformation parameters of the thyroid anatomical structure deformation field: Global deformation trend: 0.7; Local deformation amplitude: 0.5. Spatial coordinate error between the hot spot region of metabolomics signal and the microcalcification cluster: Initial error: 5 mm. 2. Nonlinear superposition: Global level superposition result: 0.8×0.7 + 0.7^2 = 0.56 + 0.49 = 1.05, Local level superposition result: 0.6×0.5 + 0.5^2 = 0.30 + 0.25 = 0.55. 3. Iterative optimization: First iteration: Spatial coordinate error: 5 mm; According to the error gradient direction, adjust the coordinate offset of the mass spectrometry detection point, and the error is reduced to 3 mm. Second iteration: Spatial coordinate error: 3 mm; Further adjust the coordinate offset, and the error is reduced to 1 mm. Third iteration: Spatial coordinate error: 1 mm; The error meets the preset tolerance range (<2 mm), and the iteration stops. Generate the initial multi-omics mass spectrometry map. Global level: Superposition result 1.05; Local level: Superposition result 0.55; Spatial coordinate error: 1 mm (meets the preset tolerance range). The initial multi-omics mass spectrometry map includes the covariance matrix of the deformation trend of the overall thyroid contour at the global level and the global distribution of metabolomics signals, the spatial offset between the texture details of microcalcification clusters and the peak positions of metabolomics signals at the local level, and the spatial coordinate error between the hot spot region of metabolomics signals and the microcalcification clusters. The deformation parameters include the deformation trend of the overall thyroid contour (global deformation) and the deformation amplitude of the local region (local deformation).

[0092] In the embodiment of the present application, the cross-modal correlation feature after weight assignment and the deformation parameters of the thyroid anatomical structure deformation field are nonlinearly superimposed to generate the initial multi-omics mass spectrometry map. Specifically, the initial map is generated by weighted fusion of the cross-modal correlation feature and the deformation parameters; then, the spatial alignment accuracy of the initial map is corrected by iteratively optimizing the spatial coordinate error between the hot spot region of the metabolomics signal and the microcalcification cluster.

[0093] The following is a specific example: Assume that the physical space range of the thyroid region is 0 - 100 mm on the x-axis, 0 - 100 mm on the y-axis, and 0 - 50 mm on the z-axis, and construct a three-dimensional space weight matrix of 100×100×50. At the global level, align the overall thyroid contour with the global distribution of metabolomics signals to generate a global-level texture distribution pattern; at the local level, align the texture of microcalcification clusters with the local peaks of metabolomics signals to generate a local-level texture distribution pattern. Perform task-aware filtering on cross-modal association features according to the gradient contribution weights, and retain the features that contribute the most to the thyroid nodule localization task and the malignant prediction task. Nonlinearly superimpose the filtered cross-modal association features with the thyroid anatomical structure deformation field to generate an initial multi-omics mass spectrum. Generate the target multi-omics mass spectrum by iteratively optimizing the spatial coordinate error between the hot spots of metabolomics signals and microcalcification clusters.

[0094] By decomposing the thyroid anatomical structure deformation field into local deformation parameters and global deformation trends, the quantitative description of thyroid tissue deformation characteristics is realized; through spatial coordinate transformation and inter-level weight assignment, the spatial alignment accuracy of cross-modal association features is optimized; through nonlinear superposition and iterative optimization, an initial multi-omics mass spectrum with high precision is generated, improving the comprehensiveness and accuracy of thyroid disease diagnosis.

[0095] To further improve the accuracy and efficiency of thyroid disease diagnosis, this application decomposes the thyroid anatomical structure in the target multi-omics mass spectrum into global-level anatomical deformation characteristics and local-level microcalcification cluster density characteristics. The global-level anatomical deformation characteristics are generated by quantifying the displacement gradient direction of the contact surface between the thyroid edge and the carotid artery, and the local-level microcalcification cluster density characteristics are obtained by statistically analyzing the three-dimensional coordinate distribution density of microcalcification clusters in a preset spatial neighborhood; extract the thyroid nodule candidate regions from the overlapping regions between the hot spots of metabolomics signals and the local-level microcalcification cluster density characteristics, and generate an initial boundary contour by combining the consistency coefficient between the metabolomics signal gradient direction and the texture distribution pattern direction; based on the expression levels of proteomic markers, combined with the global-level anatomical deformation characteristics, screen out the set of mutation positions related to the preset malignant risk from the thyroid nodule candidate regions to generate protein expression abnormal hot spots; calculate the confidence of the malignant risk direction of the thyroid nodule candidate regions according to the spatial angle between the metabolomics signal gradient direction and the corresponding direction of the texture distribution pattern; superimpose the spatial distribution path of the protein expression abnormal hot spots and the confidence of the malignant risk direction to generate the three-dimensional localization coordinates and malignant probability prediction values of thyroid nodules, realizing the high-precision diagnosis and risk assessment of thyroid diseases.

[0096] In some embodiments, in step 105, parsing the target multi-omics mass spectrometry map for thyroid pathological features to obtain a thyroid pathological analysis result, where the thyroid pathological analysis result includes the three-dimensional positioning coordinates and malignant probability prediction value of a thyroid nodule, including: Step 501: Decompose the thyroid anatomical structure in the target multi-omics mass spectrometry map into global hierarchical anatomical deformation features and local hierarchical microcalcification cluster density features. During the decomposition process, the global hierarchical anatomical deformation features are generated by quantifying the displacement gradient direction of the contact surface between the thyroid edge and the carotid artery, and the local hierarchical microcalcification cluster density features are obtained by statistically analyzing the three-dimensional coordinate distribution density of microcalcification clusters within a preset spatial neighborhood.

[0097] In step 501, the global hierarchical anatomical deformation features refer to the deformation trend of the overall thyroid contour, which is generated by quantifying the displacement gradient direction of the contact surface between the thyroid edge and the carotid artery and is used to describe the overall deformation characteristics of the thyroid. The local hierarchical microcalcification cluster density features refer to the density distribution of microcalcification clusters within the thyroid tissue, which is obtained by statistically analyzing the three-dimensional coordinate distribution density of microcalcification clusters within a preset spatial neighborhood and is used to describe local pathological features.

[0098] In the embodiments of the present application, the thyroid anatomical structure in the target multi-omics mass spectrometry map is decomposed into global hierarchical anatomical deformation features and local hierarchical microcalcification cluster density features. Specifically, the global hierarchical anatomical deformation features are generated by quantifying the displacement gradient direction of the contact surface between the thyroid edge and the carotid artery, reflecting the overall deformation trend of the thyroid; the local hierarchical microcalcification cluster density features are obtained by statistically analyzing the three-dimensional coordinate distribution density of microcalcification clusters within a preset spatial neighborhood, reflecting local pathological features.

[0099] Step 502: Extract a thyroid nodule candidate region from the overlapping region between the metabolomics signal hot spot region and the local hierarchical microcalcification cluster density features, and generate an initial boundary contour based on the thyroid nodule candidate region in combination with the consistency coefficient between the metabolomics signal gradient direction and the texture distribution pattern direction.

[0100] In step 502, the thyroid nodule candidate region refers to a potential nodule region extracted from the overlapping region between the metabolomics signal hot spot region and the local hierarchical microcalcification cluster density features. The initial boundary contour refers to the preliminary boundary shape extracted from the thyroid nodule candidate region and is used to describe the preliminary spatial range of the nodule.

[0101] In the embodiments of the present application, a thyroid nodule candidate region is extracted from the overlapping region between the metabolomics signal hotspot region and the local hierarchical microcalcification cluster density feature. Then, an initial boundary contour is generated by combining the consistency coefficient between the metabolomics signal gradient direction and the texture distribution pattern direction. Specifically, the position and shape of the nodule boundary are determined by calculating the spatial angle between the metabolomics signal gradient direction and the texture extension direction.

[0102] Step 503: Based on the expression levels of the proteomics markers, in combination with the global hierarchical anatomical deformation features, a set of mutation positions related to the preset malignant risk is screened out from the thyroid nodule candidate region, and based on the set of mutation positions, a proteome expression abnormality hotspot is generated.

[0103] In step 503, the set of mutation positions refers to the proteomics marker expression mutation positions related to the malignant risk screened out from the thyroid nodule candidate region. For example, when the correlation coefficient between the protein marker expression level and the anatomical deformation amplitude is greater than a preset threshold (such as 0.7), this position is considered a mutation position.

[0104] In the embodiments of the present application, based on the expression levels of the proteomics markers, in combination with the global hierarchical anatomical deformation features, a set of mutation positions related to the preset malignant risk is screened out from the thyroid nodule candidate region. Then, spatial clustering analysis is performed on the set of mutation positions to generate a proteome expression abnormality hotspot. Specifically, by calculating the spatial distance and expression level similarity between the mutation positions, the mutation positions that meet the conditions are clustered into an abnormality hotspot.

[0105] Step 504: Calculate the confidence level of the malignant risk direction of the thyroid nodule candidate region according to the spatial angle between the metabolomics signal gradient direction and the texture distribution pattern.

[0106] In step 504, the confidence level of the malignant risk direction is a confidence index for evaluating the malignant risk direction of the thyroid nodule candidate region based on the consistency degree between the metabolomics signal gradient direction and the texture distribution pattern direction, and is used to evaluate the malignant risk direction of the thyroid nodule candidate region to support the generation of the malignant probability prediction value.

[0107] In the embodiments of the present application, the confidence level of the malignant risk direction of the thyroid nodule candidate region is calculated according to the spatial angle between the metabolomics signal gradient direction and the corresponding direction of the texture distribution pattern. Specifically, the confidence level of the malignant risk direction is generated by calculating the cosine value of the angle between the two.

[0108] Step 505: Superimpose the spatial distribution path of the proteome expression abnormality hotspot and the confidence level of the malignant risk direction to generate the three-dimensional positioning coordinates and the malignant probability prediction value of the thyroid nodule.

[0109] In the embodiments of the present application, the spatial distribution path of protein expression abnormal hotspots is superimposed with the confidence of the malignant risk direction to generate the three-dimensional positioning coordinates and the predicted value of the malignant probability of the thyroid nodule. Specifically, by weighted fusion of the spatial distribution path of protein expression abnormal hotspots and the confidence of the malignant risk direction, a superimposed result is generated; then, based on the superimposed result, the three-dimensional positioning coordinates and the predicted value of the malignant probability of the thyroid nodule are generated.

[0110] The following is a specific example: Assume that the physical space range of the thyroid region is 0 - 100 mm on the x-axis, 0 - 100 mm on the y-axis, and 0 - 50 mm on the z-axis, and a three-dimensional space weight matrix of 100×100×50 is constructed. At the global level, the overall thyroid contour is aligned with the global distribution of metabolomics signals to generate the global-level texture distribution pattern; at the local level, the texture of the microcalcification cluster is aligned with the local peak of the metabolomics signal to generate the local-level texture distribution pattern. From the overlapping area between the metabolomics signal hotspot region and the density characteristics of the local-level microcalcification cluster, candidate regions of the thyroid nodule are extracted, and combined with the consistency coefficient between the metabolomics signal gradient direction and the texture distribution pattern direction, an initial boundary contour is generated. Based on the expression level of proteomics markers, combined with the global-level anatomical deformation characteristics, a set of mutation positions related to the preset malignant risk is screened from the candidate regions of the thyroid nodule to generate protein expression abnormal hotspots. According to the spatial angle between the metabolomics signal gradient direction and the corresponding direction of the texture distribution pattern, the confidence of the malignant risk direction of the candidate region of the thyroid nodule is calculated. The spatial distribution path of protein expression abnormal hotspots is superimposed with the confidence of the malignant risk direction to generate the three-dimensional positioning coordinates and the predicted value of the malignant probability of the thyroid nodule.

[0111] By decomposing the thyroid anatomical structure into global-level and local-level features, the quantitative description of thyroid tissue deformation and pathological features is realized; by extracting candidate regions of the thyroid nodule and generating an initial boundary contour, the accuracy of nodule positioning is optimized; by screening a set of mutation positions and generating protein expression abnormal hotspots, the accuracy of malignant risk assessment is improved; by calculating the confidence of the malignant risk direction and superimposing the spatial distribution path, the three-dimensional positioning coordinates and the predicted value of the malignant probability of the thyroid nodule are generated, improving the comprehensiveness and accuracy of thyroid disease diagnosis.

[0112] To solve the problem of the accuracy of precise positioning of thyroid nodules and prediction of malignant probability, first, in this application, by calculating the similarity between the spatial distances between mutation positions in the mutation position set and the expression levels of proteomic markers, a spatial distribution path of protein expression abnormal hotspots is generated. In this process, a spatial clustering method is used to aggregate mutation positions with similar expression levels and relatively close spatial distances into abnormal hotspots; then, the protein expression abnormal hotspot intensity value at each point on the spatial distribution path is superimposed with the confidence level of the malignant risk direction, and a superimposed result is generated, where the confidence level of the malignant risk direction is calculated based on the consistency between the direction of the metabolomic signal gradient and the direction of the texture distribution pattern; finally, based on the superimposed result, by iteratively optimizing the coupling error between the spatial distribution path and the initial boundary contour, the three-dimensional positioning coordinates of the thyroid nodule are determined, and at the same time, the spatial correlation degree between the spatial distribution path and the texture distribution pattern is verified to generate the malignant probability prediction value of the thyroid nodule, thereby realizing the precise positioning and malignant risk assessment of the thyroid nodule.

[0113] In some embodiments, the step of superimposing the spatial distribution path of the protein expression abnormal hotspots with the confidence level of the malignant risk direction to generate the three-dimensional positioning coordinates and the malignant probability prediction value of the thyroid nodule in step 505 includes: Step 601: By calculating the similarity between the spatial distances between mutation positions in the mutation position set and the expression levels of the proteomic markers, a spatial distribution path of the protein expression abnormal hotspots is generated.

[0114] In step 601, the spatial distance refers to the Euclidean distance between mutation positions and is used to describe the spatial distribution of mutation positions. The spatial distribution path of protein expression abnormal hotspots refers to the spatial distribution trajectory of protein expression abnormal hotspots generated through spatial clustering analysis, reflecting the aggregation law of protein marker expression mutation positions in space. A specific example is given: For each pair of mutation positions in the mutation position set, calculate the spatial distance and expression level similarity between them. For example, the spatial distance between mutation positions A and B is 7.35 mm, and the expression level similarity is 0.92. According to the spatial distance and expression level similarity, spatial clustering analysis is performed on the mutation positions to generate protein expression abnormal hotspots. For example, mutation positions with a spatial distance less than 10 mm and an expression level similarity greater than 0.9 are clustered into an abnormal hotspot: A, B, and C are clustered into an abnormal hotspot; D and E are clustered into another abnormal hotspot. By connecting multiple abnormal hotspots, a spatial distribution path of protein expression abnormal hotspots is generated. For example, calculate the spatial distance and expression level similarity between abnormal hotspot A - B - C and abnormal hotspot D - E. If the preset conditions are met (such as a spatial distance less than 20 mm and an expression level similarity greater than 0.8), then connect the two to generate a spatial distribution path.

[0115] In the embodiments of the present application, first, the spatial distances between the mutation positions in the mutation position set are calculated. For example, the Euclidean distance formula is used to calculate the distance between each pair of mutation positions. Then, the similarity of the proteomic biomarker expression levels between the mutation positions is calculated. For example, the similarity coefficient is calculated by normalizing the expression level difference. Then, based on the spatial distance and the expression level similarity, a spatial clustering algorithm is used to cluster the mutation positions into protein expression abnormal hotspots, generating a spatial distribution path. Finally, the spatial distribution path reflects the spatial aggregation law of the protein biomarker expression mutation positions in the thyroid tissue.

[0116] Step 602: Superimpose the spatial distribution path and the confidence of the malignant risk direction to generate a superimposed result.

[0117] In step 602, an example is given for the superimposing process: The spatial distribution path is represented as a series of spatial coordinate points and their corresponding protein expression abnormal hotspot intensities. For example, the spatial distribution path includes the following coordinate points: Point 1: (50, 60, 20), intensity 1200; Point 2: (55, 65, 22), intensity 1100; Point 3: (52, 63, 21), intensity 1150; Point 4: (70, 80, 25), intensity 900; Point 5: (75, 85, 27), intensity 950. The confidence of the malignant risk direction is represented as the confidence value of each point on the spatial distribution path. For example, the confidence of the malignant risk direction of each point on the spatial distribution path is 0.8. The protein expression abnormal hotspot intensity of each point on the spatial distribution path is weighted and fused with the confidence of the malignant risk direction to generate a superimposed result. For example, the value of each point in the superimposed result is: Superimposed result = Protein expression abnormal hotspot intensity × Confidence of the malignant risk direction.

[0118] In the embodiments of the present application, first, the intensity value of each point on the spatial distribution path of the protein expression abnormal hotspot is obtained. For example, it is obtained by normalizing the protein biomarker expression level. Then, the confidence of the malignant risk direction corresponding to each point is obtained. For example, it is obtained by calculating the spatial angle between the metabolomic signal gradient direction and the texture distribution pattern direction. Then, the intensity value of each point on the spatial distribution path is weighted and superimposed with the confidence of the malignant risk direction to generate a superimposed result. The superimposed result reflects the spatial consistency between the protein expression abnormal hotspot and the malignant risk direction.

[0119] Step 603: Based on the superimposed result, optimize the coupling error between the spatial distribution path and the initial boundary contour to determine the three-dimensional positioning coordinates of the thyroid nodule, and verify the spatial correlation degree between the spatial distribution path and the texture distribution pattern to generate a malignant probability prediction value of the thyroid nodule.

[0120] In step 603, the coupling error refers to the spatial alignment error between the spatial distribution path and the initial boundary contour, which is used to optimize the three-dimensional positioning coordinates of the thyroid nodule. The spatial correlation degree refers to the degree of spatial consistency between the spatial distribution path of the abnormal hot spots of protein expression and the texture distribution pattern, which is used to generate the malignant probability prediction value of the thyroid nodule.

[0121] In the embodiment of the present application, first, based on the superposition result, the coupling error between the spatial distribution path and the initial boundary contour is iteratively optimized, for example, the gradient descent method is used to minimize the spatial alignment error to determine the three-dimensional positioning coordinates of the thyroid nodule; then, the spatial correlation degree between the spatial distribution path and the texture distribution pattern is verified, for example, by calculating the spatial angle or overlap degree between the two, to generate the malignant probability prediction value of the thyroid nodule. Finally, the three-dimensional positioning coordinates and the malignant probability prediction value respectively reflect the spatial position and the malignant risk degree of the thyroid nodule.

[0122] The following is a specific example: Suppose the candidate region of the thyroid nodule contains 5 mutation positions. Among them, the three-dimensional coordinates of mutation position A are (50, 60, 20), and the expression level of the proteomic marker is 1200; the three-dimensional coordinates of mutation position B are (55, 65, 22), and the expression level is 1100; the three-dimensional coordinates of mutation position C are (52, 63, 21), and the expression level is 1150; the three-dimensional coordinates of mutation position D are (70, 80, 25), and the expression level is 900; the three-dimensional coordinates of mutation position E are (75, 85, 27), and the expression level is 950. In step 601, first, the spatial distance between the mutation positions is calculated. For example, the distance between A and B is 7.35 mm. Then, the expression similarity is calculated. For example, the similarity between A and B is 0.92. Then, the spatial clustering algorithm is used to cluster A, B, and C into one abnormal hot spot, and D and E into another abnormal hot spot to generate the spatial distribution path. In step 602, the intensity value of each point on the spatial distribution path is obtained. For example, the intensity value of point A is 1.0, and the confidence degree of the malignant risk direction corresponding to each point is obtained. For example, the confidence degree of point A is 0.85. Finally, the intensity value and the confidence degree are weighted and superimposed to generate the superposition result. In step 603, based on the superposition result, the coupling error between the spatial distribution path and the initial boundary contour is iteratively optimized to determine that the three-dimensional positioning coordinates of the thyroid nodule are (53, 64, 21), and the spatial correlation degree between the spatial distribution path and the texture distribution pattern is verified to generate the malignant probability prediction value of the thyroid nodule as 0.78.

[0123] By calculating the similarity between the spatial distances between mutation positions and the expression levels of proteomic markers, a spatial distribution path of abnormal protein expression hotspots is generated; the spatial distribution path is superimposed with the confidence degree of the malignant risk direction to generate a superimposed result; based on the superimposed result, the coupling error between the spatial distribution path and the initial boundary contour is optimized to determine the three-dimensional positioning coordinates of the thyroid nodule, and the spatial correlation degree between the spatial distribution path and the texture distribution pattern is verified to generate a malignant probability prediction value of the thyroid nodule. This method realizes the precise positioning and malignant probability prediction of thyroid nodules, and improves the accuracy and reliability of thyroid disease diagnosis.

[0124] To solve the gradient conflict problem between the thyroid nodule localization task and the malignant prediction task in multi-task learning, first, the backpropagation error of the thyroid anatomical structure deformation field is decomposed into the backpropagation error component of the thyroid nodule localization task and the backpropagation error component of the thyroid nodule malignant prediction task. The error component of the localization task is generated by quantifying the displacement gradient direction of the contact surface between the thyroid edge and the carotid artery, while the error component of the malignant prediction task is generated by quantifying the correlation between the expression mutation positions of proteomic markers and the anatomical deformation amplitude at the thyroid-carotid artery contact surface; then, based on the non-linear relationship between the backpropagation error components of the two tasks, a non-linear coupling function is constructed, which is generated by fusing the contribution degree of the localization task error component to the deformation trend of the overall thyroid contour and the contribution degree of the malignant prediction task error component to the spatial distribution path of the expression mutation positions of proteomic markers; finally, the non-linear coupling function is applied to the backpropagation error of the thyroid anatomical structure deformation field, and the coupling error of the error components of the two tasks is iteratively optimized to calculate the respective gradient contribution weights of the thyroid nodule localization task and the thyroid nodule malignant prediction task, so as to dynamically adjust the optimization target weights of the two tasks in the multi-task learning iteration process and achieve the collaborative optimization between tasks.

[0125] In some embodiments, in the multi-task learning iteration process of step 103, performing non-linear coupling calculation on the backpropagation error of the thyroid anatomical structure deformation field to obtain the respective gradient contribution weights of the thyroid nodule localization task and the thyroid nodule malignant prediction task includes: Step 701: Decompose the backpropagation error of the thyroid anatomical structure deformation field into the backpropagation error component of the thyroid nodule localization task and the backpropagation error component of the thyroid nodule malignant prediction task.

[0126] In step 701, the backpropagation error refers to the error between the predicted value and the true value of the thyroid anatomical structure deformation field during the multi-task learning iteration, which is calculated by the backpropagation algorithm. The backpropagation error component of the thyroid nodule localization task refers to the part of the backpropagation error related to the thyroid nodule localization task, which is generated by quantifying the displacement gradient direction of the contact surface between the thyroid edge and the carotid artery. The backpropagation error component of the thyroid nodule malignancy prediction task refers to the part of the backpropagation error related to the thyroid nodule malignancy prediction task, which is generated by quantifying the correlation between the expression mutation position of the proteomic biomarker and the anatomical deformation amplitude at the thyroid-carotid artery contact surface.

[0127] In the embodiment of the present application, first, the backpropagation error of the thyroid anatomical structure deformation field is obtained, for example, by calculating the mean square error between the predicted value and the true value; then the backpropagation error is decomposed into the backpropagation error component of the thyroid nodule localization task and the backpropagation error component of the thyroid nodule malignancy prediction task, where the error component of the localization task is generated by calculating the displacement gradient direction of the contact surface between the thyroid edge and the carotid artery, and the error component of the malignancy prediction task is generated by calculating the correlation between the expression mutation position of the proteomic biomarker and the anatomical deformation amplitude. Finally, the decomposed error components respectively reflect the contributions of the localization task and the malignancy prediction task to the backpropagation error.

[0128] Step 702: Construct a non-linear coupling function based on the non-linear relationship between the backpropagation error component of the thyroid nodule localization task and the backpropagation error component of the thyroid nodule malignancy prediction task.

[0129] In step 702, the non-linear coupling function refers to a function constructed based on the non-linear relationship between the backpropagation error components of the two tasks, which is used to calculate the gradient contribution weights of the two tasks.

[0130] In the embodiment of the present application, first, analyze the non-linear relationship between the backpropagation error component of the thyroid nodule localization task and the backpropagation error component of the thyroid nodule malignancy prediction task, for example, by plotting a scatter diagram of the error components or calculating the correlation coefficient; then construct a non-linear coupling function based on the non-linear relationship, for example, using an exponential function or a logarithmic function to describe the interaction relationship between the two error components; finally, generate a non-linear coupling function by fusing the contribution degree of the localization task error component to the deformation trend of the overall thyroid contour and the contribution degree of the malignancy prediction task error component to the spatial distribution path of the expression mutation position of the proteomic biomarker.

[0131] Step 703: Calculate the respective gradient contribution weights of the thyroid nodule localization task and the thyroid nodule malignancy prediction task based on the non-linear coupling function.

[0132] In the embodiments of the present application, first, a non - linear coupling function is applied to the back - propagation error of the thyroid anatomical structure deformation field to calculate the gradient contribution weights of the thyroid nodule localization task and the thyroid nodule malignancy prediction task respectively. For example, the weights are determined by iteratively optimizing the coupling error of the error components of the two tasks. Then, based on the gradient contribution weights, the optimization target weights of the two tasks in the multi - task learning iteration process are dynamically adjusted. For example, the optimization target weights are generated by matching the contribution ratios of the error components of the two tasks. Finally, the gradient contribution weights achieve the collaborative optimization between the two tasks.

[0133] The following is a specific example: Suppose the back - propagation error of the thyroid anatomical structure deformation field is 0.5, where the back - propagation error component of the thyroid nodule localization task is 0.3 and the back - propagation error component of the thyroid nodule malignancy prediction task is 0.2. In step 701, first, the back - propagation error is decomposed into the error component of the localization task (0.3) and the error component of the malignancy prediction task (0.2). The error component of the localization task is generated by calculating the displacement gradient direction of the contact surface between the thyroid edge and the carotid artery, while the error component of the malignancy prediction task is generated by calculating the correlation between the expression mutation position of the proteomic biomarker and the anatomical deformation amplitude. In step 702, based on the non - linear relationship between the two error components, a non - linear coupling function is constructed. For example, an exponential function is used to describe the interaction relationship between the two error components. In step 703, the non - linear coupling function is applied to the back - propagation error. By iteratively optimizing the coupling error of the error components of the two tasks, the gradient contribution weight of the localization task is calculated as 0.6, and the gradient contribution weight of the malignancy prediction task is 0.4. And the optimization target weights of the two tasks are dynamically adjusted based on the weights.

[0134] By decomposing the back - propagation error of the thyroid anatomical structure deformation field into the back - propagation error components of the thyroid nodule localization task and the malignancy prediction task, constructing a non - linear coupling function based on the non - linear relationship between the two error components, and calculating the gradient contribution weights of the two tasks respectively, the collaborative optimization between the thyroid nodule localization task and the malignancy prediction task in multi - task learning is achieved, improving the accuracy and reliability of thyroid disease diagnosis.

[0135] Figure 2 The following is a schematic structural diagram of a multi - omics mass spectrometry map construction and analysis system for thyroid diseases provided by the embodiments of the present application, as Figure 2 shown, the system includes: An acquisition module 21, configured to acquire thyroid ultrasound images and multi - omics mass spectrometry data, and generate a cross - modal association relationship according to the anatomical structure features of the thyroid ultrasound images and the biomolecular spatial distribution features of the multi - omics mass spectrometry data; A construction module 22 is configured to generate a thyroid anatomical structure deformation field based on the anatomical structure features, and construct a multi-task optimization objective according to the preset thyroid disease diagnosis requirements. The multi-task optimization objective includes a thyroid nodule localization task and a thyroid nodule malignancy prediction task. A calculation module 23 is configured to perform non-linear coupling calculation on the backpropagation error of the thyroid anatomical structure deformation field during the multi-task learning iteration process to obtain the gradient contribution weights of the thyroid nodule localization task and the thyroid nodule malignancy prediction task respectively. A generation module 24 is configured to generate cross-modal association features according to the cross-modal association relationship, and generate a target multi-omics mass spectrum based on the thyroid anatomical structure deformation field and the cross-modal association features, in combination with the gradient contribution weights. The error between the spatial coordinates of the metabolomics signal hot spot region in the target multi-omics mass spectrum and the microcalcification cluster in the thyroid ultrasound image is within the preset anatomical structure tolerance range. An analysis module 25 is configured to perform thyroid pathological feature analysis on the target multi-omics mass spectrum to obtain a thyroid pathological analysis result, which includes the three-dimensional localization coordinates and the malignancy probability prediction value of the thyroid nodule.

[0136] Figure 2 The described thyroid disease multi-omics mass spectrum construction and analysis system can execute Figure 1 The thyroid disease multi-omics mass spectrum construction and analysis method described in the illustrated embodiment. Its implementation principle and technical effects will not be elaborated further. For the thyroid disease multi-omics mass spectrum construction and analysis system in the above embodiment, the specific manners in which each module and unit perform operations have been described in detail in the embodiment related to the method, and will not be elaborated here.

[0137] In a possible design, Figure 2 The thyroid disease multi-omics mass spectrum construction and analysis system of the illustrated embodiment can be implemented as a computing device, as Figure 3 shown. The computing device can include a storage component 31 and a processing component 32. The storage component 31 stores one or more computer instructions, and the one or more computer instructions are called and executed by the processing component 32.

[0138] The processing component 32 is the above Figure 1 The thyroid disease multi-omics mass spectrum construction and analysis method of the illustrated embodiment.

[0139] Among them, the processing component 32 may include one or more processors to execute computer instructions to complete all or part of the steps in the above methods. Of course, the processing component may also be implemented by one or more application-specific integrated circuits (ASICs), digital signal processors (DSPs), digital signal processing devices (DSPDs), programmable logic devices (PLDs), field-programmable gate arrays (FPGAs), controllers, microcontrollers, microprocessors or other electronic components for executing the above methods.

[0140] The storage component 31 is configured to store various types of data to support the operation of the terminal. The storage component may be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as static random access memory (SRAM), electrically erasable programmable read-only memory (EEPROM), erasable programmable read-only memory (EPROM), programmable read-only memory (PROM), read-only memory (ROM), magnetic memory, flash memory, magnetic disk or optical disk.

[0141] Of course, the computing device may also necessarily include other components, such as input / output interfaces, display components, communication components, etc.

[0142] The input / output interface provides an interface between the processing component and the peripheral interface module, and the above peripheral interface module may be an output device, an input device, etc.

[0143] The communication component is configured to facilitate communication between the computing device and other devices in a wired or wireless manner, etc.

[0144] Among them, the computing device may be a physical device or an elastic computing host provided by a cloud computing platform, etc. At this time, the computing device may refer to a cloud server, and the above processing component, storage component, etc. may be basic server resources leased or purchased from a cloud computing platform.

[0145] The embodiment of the present application also provides a computer storage medium storing a computer program, and when the computer program is executed by a computer, it can implement the above Figure 1 A method for constructing and analyzing a multi-omics mass spectrometry map of thyroid diseases shown in the embodiment.

[0146] Those skilled in the art can clearly understand that for the convenience and brevity of description, the specific working processes of the above-described systems, devices, and units can refer to the corresponding processes in the foregoing method embodiments and will not be elaborated herein.

[0147] The device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed to multiple network units. Some or all of the modules can be selected according to actual needs to achieve the purpose of the solution of this embodiment. Those of ordinary skill in the art can understand and implement it without creative efforts.

[0148] Through the description of the above embodiments, those skilled in the art can clearly understand that each embodiment can be implemented by means of software plus a necessary general hardware platform, and of course, it can also be implemented by hardware. Based on such an understanding, the essence of the above technical solution, or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, magnetic disk, optical disk, etc., and includes several instructions for causing a computer device (which can be a personal computer, a server, or a network device, etc.) to execute the methods described in each embodiment or some parts of the embodiments.

[0149] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present application and are not intended to limit them. Although the present application has been described in detail with reference to the foregoing embodiments, those of ordinary skill in the art should understand that they can still modify the technical solutions recorded in the foregoing embodiments or perform equivalent replacements for some of the technical features. These modifications or replacements do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present application.

Claims

1. A multi-omics mass spectrometry map construction and analysis method for thyroid diseases, characterized in that Including: Obtaining thyroid ultrasound images and multi-omics mass spectrometry data, and generating cross-modal association relationships according to the anatomical structure features of the thyroid ultrasound images and the biomolecular spatial distribution features of the multi-omics mass spectrometry data, where the anatomical structures include thyroid nodules; Generating a thyroid anatomical structure deformation field based on the anatomical structure features, and constructing a multi-task optimization objective according to the preset thyroid disease diagnosis requirements, where the multi-task optimization objective includes a thyroid nodule localization task and a thyroid nodule malignancy prediction task; During the multi-task learning iteration process, performing non-linear coupling calculation on the backpropagation error of the thyroid anatomical structure deformation field to obtain the gradient contribution weights of the thyroid nodule localization task and the thyroid nodule malignancy prediction task respectively; Generating cross-modal association features according to the cross-modal association relationships, and generating a target multi-omics mass spectrometry map based on the thyroid anatomical structure deformation field and the cross-modal association features, in combination with the gradient contribution weights, where the error between the metabolomics signal hot spot region in the target multi-omics mass spectrometry map and the spatial coordinates of the microcalcification clusters in the thyroid ultrasound image is within the preset anatomical structure tolerance range; Performing thyroid pathological feature analysis on the target multi-omics mass spectrometry map to obtain a thyroid pathological analysis result, where the thyroid pathological analysis result includes the three-dimensional localization coordinates and the malignancy probability prediction value of the thyroid nodule.

2. The method according to claim 1, wherein The generating cross-modal association features according to the cross-modal association relationships, and generating a target multi-omics mass spectrometry map based on the thyroid anatomical structure deformation field and the cross-modal association features, in combination with the gradient contribution weights, includes: Based on the cross-modal association relationships, in combination with the anatomical structure features and the biomolecular spatial distribution features, constructing a three-dimensional spatial weight matrix, and performing multi-level spatial decomposition on the anatomical structure features according to the three-dimensional spatial weight matrix to obtain the texture distribution patterns of different anatomical levels within the thyroid tissue; Fusing the texture distribution patterns and the multi-omics mass spectrometry data layer by layer to generate cross-modal association features; Performing task-aware filtering on the cross-modal association features according to the gradient contribution weights to obtain filtered cross-modal association features; Non-linearly superimposing the filtered cross-modal association features and the thyroid anatomical structure deformation field to generate an initial multi-omics mass spectrometry map; Optimizing the initial multi-omics mass spectrometry map to generate a target multi-omics mass spectrometry map.

3. The method according to claim 2, wherein The fusing the texture distribution patterns and the multi-omics mass spectrometry data layer by layer to generate cross-modal association features includes: Analyzing the multi-omics mass spectrometry data to obtain mass spectrometry detection points, and dynamically adjusting the mass spectrometry detection points within the coordinate mapping ranges of each level based on the hierarchical resolution of the thyroid ultrasound image, where the coordinate mapping range of the global level covers the overall contour of the thyroid and the topological boundaries of adjacent organs, and the mapping range of the local level is limited to the local spatial neighborhood of the microcalcification clusters in the thyroid ultrasound image; At each level of resolution, perform spatial interpolation compensation processing on the intensity distribution of metabolomics signals in the multi-omics mass spectrometry data according to the spatial distance between the adjusted mass spectrometry detection points and the ultrasound image voxels, where the ultrasound image voxels are the smallest spatial units of the thyroid ultrasound image, to obtain the compensated intensity distribution of metabolomics signals; According to the spatial consistency between the spatial distribution path of the expression levels of proteomics markers in the multi-omics mass spectrometry data and the directional characteristics of the texture distribution pattern, perform directional enhancement processing on the texture characteristics of the texture distribution pattern to obtain enhanced texture characteristics; Stitch the compensated intensity distribution of metabolomics signals and the enhanced texture characteristics by channels according to the level of resolution to generate cross-modal correlation features, where the stitching channels at the global level are used to calculate the covariance matrix of the overall deformation trend of the thyroid and the global distribution of metabolomics signals, and the stitching channels at the local level are used to calculate the spatial offset between the texture details of the microcalcification clusters and the peak positions of the metabolic signals.

4. The method according to claim 2, wherein The non-linear superposition of the filtered cross-modal correlation features and the thyroid anatomical structure deformation field to generate an initial multi-omics mass spectrometry map includes: Decompose the thyroid anatomical structure deformation field into local deformation parameters and global deformation trends within the thyroid tissue; Based on the local deformation parameters, perform spatial coordinate transformation on the filtered cross-modal correlation features to obtain the cross-modal correlation features after spatial coordinate transformation; Based on the global deformation trend, perform inter-level weight assignment on the cross-modal correlation features after spatial coordinate transformation to obtain the cross-modal correlation features after weight assignment; Non-linearly superpose the cross-modal correlation features after weight assignment and the deformation parameters of the thyroid anatomical structure deformation field to generate an initial multi-omics mass spectrometry map, where the initial multi-omics mass spectrometry map is corrected by iteratively optimizing the spatial coordinates of the metabolomics signal hotspots and the microcalcification clusters.

5. The method according to claim 1, characterized in that, The analysis of the thyroid pathological features of the target multi-omics mass spectrometry map to obtain the thyroid pathological analysis results, where the thyroid pathological analysis results include the three-dimensional positioning coordinates and malignant probability prediction values of the thyroid nodules, includes: Decompose the thyroid anatomical structure in the target multi-omics mass spectrometry map into global-level anatomical deformation features and local-level microcalcification cluster density features. During the decomposition process, the global-level anatomical deformation features are generated by quantifying the displacement gradient direction of the contact surface between the thyroid edge and the carotid artery, and the local-level microcalcification cluster density features are obtained by statistically analyzing the three-dimensional coordinate distribution density of the microcalcification clusters in a preset spatial neighborhood; Extract the thyroid nodule candidate regions from the overlapping regions between the metabolomics signal hotspots and the local-level microcalcification cluster density features, and generate an initial boundary contour based on the thyroid nodule candidate regions in combination with the consistency coefficient between the metabolomics signal gradient direction and the corresponding direction of the texture distribution pattern. Based on the expression levels of proteomic markers, combined with the global hierarchical anatomical deformation features, a set of mutation positions related to the preset malignant risk is screened from the thyroid nodule candidate regions. Based on the set of mutation positions, protein expression abnormal hotspots are generated; According to the spatial angle between the metabolic signal gradient direction and the corresponding direction of the texture distribution pattern, the malignant risk direction confidence of the thyroid nodule candidate region is calculated; The spatial distribution path of the protein expression abnormal hotspots is superimposed with the malignant risk direction confidence to generate the three-dimensional positioning coordinates and the malignant probability prediction value of the thyroid nodule.

6. The method according to claim 5, wherein The superimposing the spatial distribution path of the protein expression abnormal hotspots with the malignant risk direction confidence to generate the three-dimensional positioning coordinates and the malignant probability prediction value of the thyroid nodule includes: By calculating the spatial distance between the mutation positions in the set of mutation positions and the similarity of the proteomic marker expression levels, the spatial distribution path of the protein expression abnormal hotspots is generated; The protein expression abnormal hotspot intensity value of each point on the spatial distribution path is superimposed with the malignant risk direction confidence to generate a superimposed result; Based on the superimposed result, the coupling error between the spatial distribution path and the initial boundary contour is optimized to determine the three-dimensional positioning coordinates of the thyroid nodule, and the spatial correlation degree between the spatial distribution path and the texture distribution pattern is verified to generate the malignant probability prediction value of the thyroid nodule.

7. The method according to claim 1, wherein The non-linearly coupling the backpropagation error of the thyroid anatomical structure deformation field during the multi-task learning iteration process to obtain the gradient contribution weights of the thyroid nodule localization task and the thyroid nodule malignancy prediction task respectively includes: Decomposing the backpropagation error of the thyroid anatomical structure deformation field into the backpropagation error component of the thyroid nodule localization task and the backpropagation error component of the thyroid nodule malignancy prediction task; Based on the non-linear relationship between the backpropagation error component of the thyroid nodule localization task and the backpropagation error component of the thyroid nodule malignancy prediction task, a non-linear coupling function is constructed; Based on the non-linear coupling function, the gradient contribution weights of the thyroid nodule localization task and the thyroid nodule malignancy prediction task are calculated respectively.

8. A multi-omics mass spectrometry atlas construction and analysis system for thyroid diseases, characterized in that, Including: An acquisition module for acquiring thyroid ultrasound images and multi-omics mass spectrometry data, and generating a cross-modal association relationship according to the anatomical structure features of the thyroid ultrasound images and the biomolecular spatial distribution features of the multi-omics mass spectrometry data; A construction module for generating a thyroid anatomical structure deformation field based on the anatomical structure features and constructing a multi-task optimization objective according to the preset thyroid disease diagnosis requirements, the multi-task optimization objective including a thyroid nodule localization task and a thyroid nodule malignancy prediction task; A calculation module for non-linearly coupling the backpropagation error of the thyroid anatomical structure deformation field during the multi-task learning iteration process to obtain the gradient contribution weights of the thyroid nodule localization task and the thyroid nodule malignancy prediction task respectively; A generation module, configured to generate cross-modal association features according to the cross-modal association relationship, and generate a target multi-omics mass spectrometry map based on the thyroid anatomical structure deformation field and the cross-modal association features, in combination with the gradient contribution weight, wherein the error between the metabolomics signal hot spot region in the target multi-omics mass spectrometry map and the spatial coordinates of the microcalcification cluster in the thyroid ultrasound image is within a preset anatomical structure tolerance range; An analysis module, configured to perform thyroid pathological feature analysis on the target multi-omics mass spectrometry map to obtain a thyroid pathological analysis result, where the thyroid pathological analysis result includes the three-dimensional positioning coordinates and the malignant probability prediction value of the thyroid nodule.

9. A computing device, characterized in that, It includes a processing component and a storage component; the storage component stores one or more computer instructions; the one or more computer instructions are used to be called and executed by the processing component to implement a method for constructing and analyzing a multi-omics mass spectrometry map of thyroid diseases as described in any one of claims 1 to 7.

10. A computer storage medium, characterized in that, A computer program is stored, and when the computer program is executed by a computer, it implements a method for constructing and analyzing a multi-omics mass spectrometry map of thyroid diseases as described in any one of claims 1 to 7.

Images