Abnormal auricle form data model construction method and system

Abstract

The invention relates to the technical field of intelligent medical treatment, in particular to an abnormal auricle form data model construction method and system, and the method comprises the following steps: obtaining auricle form data, and preprocessing the auricle form data to obtain auricle form preprocessing data; constructing a normal auricle form threshold value, and performing abnormal feature recognition on the auricle form preprocessing data according to the normal auricle form threshold value to obtain abnormal auricle form data; obtaining abnormal auricle form fusion data according to the abnormal auricle form data so as to construct an abnormal auricle form joint feature tensor; performing abnormal space mapping according to the abnormal auricle form joint feature tensor to obtain abnormal auricle form space features; and constructing an abnormal auricle form data model based on the abnormal auricle form spatial features. According to the method, the abnormal auricle form data model is constructed, a quantitative basis is provided for personalized treatment of the abnormal auricle, and the accuracy of abnormal auricle diagnosis is remarkably improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of intelligent medical treatment, and in particular to an abnormal auricle morphology data model construction method and system. BACKGROUND

[0002] Traditional auricle morphology analysis mainly relies on two-dimensional images or manual measurement, and evaluates morphology abnormalities through auricle geometric parameters, and some methods analyze tissue characteristics in combination with mechanical properties such as elastic modulus. In terms of data processing, the conventional process includes registration, noise reduction, feature extraction and threshold comparison, and abnormality recognition is based on a single static threshold or expert experience. For example, the diagnosis of microtia is usually made by comparing the geometric template of a normal auricle and judging the tissue development abnormalities in combination with the difference in elastic modulus; and cup ear is evaluated by measuring the ear roll curling angle and the deviation degree from the normal range.

[0003] The defects of the traditional method include: first, the multi-modal data integration capability is limited, the traditional method is difficult to effectively fuse multi-modal data, leading to insufficient information utilization; second, the noise reduction and feature extraction mechanism lacks self-adaptive ability, and the fixed parameter filtering method cannot adapt to individual differences or dynamic physiological changes; third, abnormality recognition relies on a single threshold or linear model, ignores multi-scale and multi-level abnormal features, and is easy to cause missed diagnosis or misdiagnosis; in addition, the feature fusion strategy is simple, and the non-linear interaction between features and the spatio-temporal evolution law are not considered, the model dynamicity is insufficient, and it is difficult to capture the spatial distribution and time development trend of abnormalities. The traditional method has low sensitivity and specificity in the diagnosis of complex auricle deformities, and is difficult to meet the needs of precision medicine. SUMMARY

[0004] In view of the defects in the prior art, the present application provides an abnormal auricle morphology data model construction method and system.

[0005] In order to achieve the above object, in a first aspect, the present application provides an abnormal auricle morphology data model construction method, the method comprising the following steps: obtaining auricle morphology data, and preprocessing the auricle morphology data to obtain auricle morphology pretreatment data; constructing a normal auricle morphology threshold, identifying abnormal features of the auricle morphology pretreatment data according to the normal auricle morphology threshold to obtain abnormal auricle morphology data; obtaining abnormal auricle morphology fusion data according to the abnormal auricle morphology data to construct an abnormal auricle morphology joint feature tensor; performing abnormal space mapping according to the abnormal auricle morphology joint feature tensor to obtain abnormal auricle morphology space features; and constructing an abnormal auricle morphology data model based on the abnormal auricle morphology space features. The present application realizes high-precision modeling and diagnosis of auricle deformity through multi-modal data fusion and dynamic analysis technology; data preprocessing improves the signal-to-noise ratio and registration accuracy of the original data; the identification sensitivity of complex deformity is improved by combining a progressive abnormality recognition mechanism; the abnormal auricle morphology sensitive features are effectively integrated through a fusion strategy to construct a joint feature tensor to quantify the abnormal space distribution and evolution trend; finally, an abnormal auricle morphology data model is constructed based on a dynamic evolution tensor network, embedding a biomechanical constraint mechanism to provide a quantitative basis for personalized treatment; the limitations of traditional static threshold methods are broken through, and the accuracy of abnormal auricle diagnosis is significantly improved.

[0006] Optionally, the auricle morphology data is obtained, and the auricle morphology data is pretreated to obtain auricle morphology pretreatment data, comprising: obtaining geometric point cloud data, elastic modulus matrix and blood flow thermal map of the auricle to construct multi-modal data, taking the multi-modal data as the auricle morphology data; constructing a biomechanical constraint condition, registering the auricle morphology data based on the biomechanical constraint condition to obtain auricle morphology registration data; establishing an adaptive manifold denoising mechanism, obtaining auricle morphology denoising data in combination with the auricle morphology registration data, and taking the auricle morphology denoising data as the auricle morphology pretreatment data. The present application significantly improves the accuracy of auricle morphology analysis through multi-modal data and biomechanical constraint registration; the data preprocessing process significantly improves the data signal-to-noise ratio and structural consistency, providing high-quality input for subsequent abnormal feature recognition and model construction, which is helpful for fine analysis of complex auricle deformity.

[0007] Optionally, the adaptive manifold denoising mechanism is established by: obtaining a blood flow oscillation index of the auricle based on the blood flow thermal map, obtaining a dynamic denoising intensity according to the blood flow oscillation index; performing multi-modal joint curvature analysis on the auricle morphology data to construct a multi-modal curvature manifold of the auricle; obtaining a biomechanical edge of the auricle according to the biomechanical constraint condition and the multi-modal curvature manifold; and constructing the adaptive manifold denoising mechanism in combination with the dynamic denoising intensity, the multi-modal curvature manifold and the biomechanical edge. The present application significantly improves the signal-to-noise ratio and structural fidelity of the auricle morphology data by dynamically regulating the denoising intensity; and by dynamically regulating the denoising intensity, multi-modal manifold modeling and biomechanical prior fusion, the abnormal sensitive features are retained while the random noise is suppressed, thereby providing a high-quality data basis for subsequent anomaly recognition.

[0008] Optionally, the normal auricle morphology threshold is constructed, and abnormal auricle morphology data is obtained by performing abnormal feature recognition on the auricle morphology pretreatment data according to the normal auricle morphology threshold, including: constructing a development trajectory function of the normal auricle, obtaining the normal auricle morphology threshold according to the development trajectory function, the normal auricle morphology threshold including a curvature threshold, an elastic modulus threshold and a development state calibration threshold of the normal auricle; establishing a progressive anomaly recognition mechanism based on the normal auricle morphology threshold; and obtaining the abnormal auricle morphology data in combination with the progressive anomaly recognition mechanism and the auricle morphology pretreatment data. The present application constructs a dynamic development trajectory function and a progressive anomaly recognition mechanism, thereby significantly improving the accuracy and adaptability of auricle anomaly detection. The morphology information and biomechanical information are effectively integrated, and auricle malformation is accurately recognized, thereby providing a quantitative basis for early intervention.

[0009] Optionally, the progressive anomaly recognition mechanism is established based on the normal auricle morphology threshold, including: obtaining an auricle confidence level, constructing an auricle multi-level confidence interval in combination with the normal auricle morphology threshold; performing coupling analysis on the auricle morphology pretreatment data in combination with the development trajectory function and the auricle multi-level confidence interval to obtain an auricle development deviation-disease correlation index; and constructing the progressive anomaly recognition mechanism based on the auricle multi-level confidence interval and the auricle development deviation-disease correlation index. The present application screens abnormal regions by multi-level confidence interval layering, quantifies the correlation between morphology deviation and disease in combination with the development trajectory function, and dynamically adapts to individual development stage differences; effectively distinguishes between normal variations and pathological abnormalities, improves the recognition sensitivity of auricle malformation, reduces the missed diagnosis rate, and provides a quantitative basis for accurate diagnosis.

[0010] Optionally, the abnormal auricle morphology data is obtained to obtain abnormal auricle morphology fusion data to construct an abnormal auricle morphology joint feature tensor, comprising: performing multi-scale abnormal sensitive feature extraction on the abnormal auricle morphology data to obtain abnormal auricle morphology sensitive features; establishing a feature fusion strategy, combining the abnormal auricle morphology sensitive features and the feature fusion strategy to obtain the abnormal auricle morphology fusion data; and performing space-time evolution based on the abnormal auricle morphology fusion data to obtain the abnormal auricle morphology joint feature tensor. Through multi-scale feature extraction and space-time evolution analysis, the representation ability of the abnormal auricle morphology is significantly improved; multi-dimensional quantitative basis is provided for accurate diagnosis of the abnormal auricle and analysis of the pathological mechanism, and the ability of the model to describe the dynamic abnormal process is improved.

[0011] Optionally, the feature fusion strategy is established, comprising: obtaining a sensitive feature interaction matrix and a sensitive feature distribution entropy based on the abnormal sensitive features; obtaining an entropy-weighted feature interaction matrix by combining the sensitive feature interaction matrix and the sensitive feature distribution entropy; and establishing the feature fusion strategy according to the entropy-weighted feature interaction matrix. The present application quantifies feature correlation through an interaction matrix, evaluates feature uncertainty through a distribution entropy, optimizes feature weights through entropy weighting, effectively integrates multi-modal abnormal information, and improves the representation robustness of the fusion data for complex auricle deformities.

[0012] Optionally, the abnormal auricle morphology joint feature tensor is used to perform abnormal space mapping to obtain abnormal auricle morphology space features, comprising: constructing an abnormal auricle biomechanics potential field according to the abnormal auricle morphology joint feature tensor; performing abnormal space mapping on the abnormal auricle biomechanics potential field to obtain an abnormal space feature vector; and obtaining the abnormal auricle morphology space features based on the abnormal space feature vector. The present application dynamically quantifies the mechanical distribution and spatial correlation of the abnormal region by constructing an abnormal auricle biomechanics potential field; the potential field is converted into a feature vector through abnormal space mapping, accurately describing the abnormal spatial position and range, improving the spatial positioning accuracy of the auricle deformity, and providing a quantitative spatial basis for individualized correction schemes.

[0013] Optionally, the abnormal auricle morphology data model is constructed based on the abnormal auricle morphology space features, comprising: constructing a dynamic evolution tensor network as a model architecture; establishing a model training mechanism based on the biomechanics constraint condition; and constructing the abnormal auricle morphology data model by combining the abnormal auricle morphology space features, the model architecture and the model training mechanism. The present application quantifies the abnormal space-time evolution law through a dynamic evolution tensor network architecture, combines the biomechanics constraint to ensure that the model conforms to the auricle tissue mechanical properties, effectively integrates multi-modal space features, and constructs an abnormal auricle morphology data model with morphology-mechanics-physiology correlation; provides quantitative dynamic prediction for individualized correction schemes, and improves the clinical intervention effect on complex auricle deformities.

[0014] In a second aspect, the present application provides an abnormal auricle morphology data model construction system, which performs the abnormal auricle morphology data model construction method provided by the present application. The system comprises an input device, an output device, a processor and a memory, which are connected with each other. The memory is used for storing a computer program, the computer program comprises program instructions, and the processor is configured to call the program instructions. The present application realizes real-time input of multi-modal data through the input device, and quickly performs dynamic evolution tensor network construction and biomechanical constraint training in combination with the powerful parallel computing capability of the processor. The memory provides large-capacity and high-speed data read-write support to ensure the coherence of data processing. The output device presents abnormal spatial features and model evolution results, and provides stable and efficient hardware support for precise diagnosis and treatment of complex auricle deformities. BRIEF DESCRIPTION OF DRAWINGS

[0015] Figure 1 An abnormal auricle morphology data model construction method flowchart according to an embodiment of the present application; Figure 2 An abnormal auricle morphology data model construction system framework diagram according to an embodiment of the present application. DETAILED DESCRIPTION

[0016] The specific embodiments of the present application will be described in detail below. It should be noted that the embodiments described herein are only used for illustration and do not limit the present application. In the following description, a large number of specific details are set forth in order to provide a thorough understanding of the present application. However, it is obvious to those skilled in the art that the specific details need not be used to implement the present application. In other instances, well-known circuits, software or methods have not been specifically described in order to avoid obscuring the present application.

[0017] Throughout the specification, references to "one embodiment", "an embodiment", "one example" or "an example" mean that a particular feature, structure, or characteristic described in connection with the embodiment or example is included in at least one embodiment of the present application. Therefore, the phrases "in one embodiment", "in an embodiment", "one example" or "an example" appearing throughout the specification do not necessarily refer to the same embodiment or example. In addition, specific features, structures, or characteristics can be combined in any appropriate combination and / or sub-combination in one or more embodiments or examples. In addition, those skilled in the art should understand that the drawings provided herein are for illustrative purposes only and the drawings are not necessarily drawn to scale.

[0018] Please refer to Figure 1 In one embodiment of the present application, an abnormal auricle morphology data model construction method is provided, which comprises the following steps: S1, acquire auricle morphology data, and pre-process the auricle morphology data to obtain auricle morphology pre-processed data.

[0019] The S1 specifically comprises the following steps: S11, acquire geometric point cloud data, elastic modulus matrix and blood flow thermal map of the auricle, and construct multi-modal data by using the multi-modal data as the auricle morphology data.

[0020] In this embodiment, a high-precision handheld three-dimensional scanner is used to obtain original point cloud data by performing auricle surface scanning based on a structured light projection technology. When scanning, a resolution of 0.1 mm and a sampling frequency of 10 Hz are set to ensure that all anatomical structures such as the helix, antihelix and earlobe are covered. During the scanning process, the scanning head needs to be kept perpendicular to the auricle surface, and multi-angle scanning is used to avoid occlusion. The original point cloud data is denoised to remove outliers, and multi-view point cloud alignment is performed, and holes are filled by Poisson surface reconstruction to finally generate geometric point cloud data with uniform density.

[0021] In this embodiment, the auricle tissue mechanical property measurement is performed by using a biomechanical test platform. 20-30 feature points are selected on the auricle surface, a cylindrical probe with a diameter of 2 mm is used to apply a compression force at a speed of 0.5 mm / s, and a compression force-auricle deformation curve is recorded. The elastic modulus is obtained by analyzing the curve, and the anisotropy of the tissue is considered, and the test is repeated in X / Y / Z three directions. The value of the elastic modulus is mapped to the corresponding node of the geometric point cloud data to construct the elastic modulus matrix.

[0022] In this embodiment, a high-resolution infrared thermal imager is used to shoot the temperature distribution of the auricle surface in a constant temperature environment. When shooting, the lens is kept 30 cm away from the auricle, the exposure time is 1 / 60 s, and 30 seconds of continuous collection is performed to obtain stable data. The physiological parameters such as heart rate and blood pressure of the subject are recorded synchronously, and the relationship between blood flow and temperature is calibrated by using a linear regression model. An adaptive threshold algorithm is used to segment the auricle region, and the resolution of the thermal map is improved to be consistent with the geometric point cloud data by interpolation, and finally the blood flow thermal map is obtained.

[0023] Further, the geometric point cloud data is used as a reference coordinate system, and a coordinate conversion method based on feature points is used: 10-15 anatomical landmark points are extracted on the auricle surface, and the spatial mapping relationship between the elastic modulus matrix and the blood flow thermal map is established. The elastic modulus matrix and the blood flow thermal map are mapped to each node of the geometric point cloud by using an interpolation algorithm, and finally a five-dimensional feature vector containing three-dimensional coordinates, elastic modulus and blood flow velocity is generated to construct a multi-modal data set.

[0024] S12, construct biomechanical constraint conditions, and register the auricle morphology data based on the biomechanical constraint conditions to obtain auricle morphology registration data.

[0025] In this embodiment, the biomechanical constraint conditions are constructed in combination with the mechanical properties and anatomical features of the auricle. First, a mechanical parameter library of the normal auricle is established: the elastic modulus of the normal auricle tissue is tested, the anisotropic properties of the regions such as the helix, antihelix, and earlobe are distinguished, and the Poisson's ratio is determined. Second, the surface anatomical landmark points of the auricle are extracted as spatial constraint references. Finally, the biomechanical constraint conditions are constructed based on the continuum mechanics theory, including the mechanical parameter range constraint, the spatial constraint of the anatomical landmark points, and the deformation energy minimization constraint.

[0026] Further, the registration process is based on geometric point clouds, and multi-modal data registration is achieved in combination with biomechanical constraints. First, the auricle morphology data is preliminarily spatially aligned based on coordinate conversion and interpolation. Then, the biomechanical constraint is introduced to optimize the registration accuracy: the spatial constraint of the anatomical landmark points is used as a hard constraint to obtain the position deviation of the landmark points; at the same time, the deformation process of the auricle is simulated by combining finite element analysis, and the deformation energy minimization constraint is used as a soft constraint to require the elastic modulus distribution after registration to satisfy the force balance equation; finally, the auricle morphology registration data is obtained through iterative optimization, ensuring the consistency of the geometric morphology and the mechanical properties in space.

[0027] S13, an adaptive manifold denoising mechanism is established, and auricle morphology denoising data is obtained in combination with the auricle morphology registration data, and the auricle morphology denoising data is used as the auricle morphology preprocessing data.

[0028] In this embodiment, the blood flow oscillation index of the auricle is obtained based on the blood flow thermogram, and the dynamic denoising intensity is obtained according to the blood flow oscillation index; the multi-modal curvature analysis of the auricle morphology data is performed to construct the multi-modal curvature manifold of the auricle; the biomechanical edge of the auricle is obtained according to the biomechanical constraint conditions and the multi-modal curvature manifold; and the adaptive manifold denoising mechanism is constructed in combination with the dynamic denoising intensity, the multi-modal curvature manifold, and the biomechanical edge.

[0029] Based on the blood flow thermogram, the time series temperature variation curve of each point on the surface of the auricle is extracted. The temperature signal is segmented by using the sliding window method, the frequency spectrum characteristics of each segment signal are analyzed by using the short-time Fourier transform, and the blood flow oscillation index is calculated. The blood flow oscillation index is defined as the energy ratio of the low frequency (0.1 Hz-0.5 Hz) to the high frequency (0.5 Hz-2 Hz) band, which reflects the stability of the local blood flow: the higher the blood flow oscillation index, the more gentle the blood flow fluctuation, and the stronger the noise interference. Then, the denoising intensity is dynamically adjusted according to the blood flow oscillation index value: in the region where the blood flow oscillation index is >1.5 (blood flow stability zone), high-intensity denoising (such as filter kernel radius 3 mm) is set; in the region where the blood flow oscillation index is <1.0 (blood flow fluctuation zone), the denoising intensity is reduced (such as filter kernel radius 1 mm) to retain potential abnormal signals.

[0030] According to the auricle shape registration data, the geometric, mechanical and blood flow three modal information is integrated to construct the curvature manifold. First, the geometric curvature is calculated for the geometric point cloud data, reflecting the surface concave and convex; the mechanical curvature is calculated for the elastic modulus matrix, and the elastic modulus gradient change is analyzed by finite difference method; the temperature curvature is obtained based on the blood flow thermal map, and the blood flow dynamics is evaluated based on the temperature gradient; then, a weighted fusion strategy is adopted: in the mechanical sensitive areas such as the helix and the antihelix, the mechanical curvature weight is increased; in the blood flow sensitive areas such as the earlobe and the triangular fossa, the temperature curvature weight is increased; the geometric curvature is taken as the basic weight; finally, the multi-modal curvature manifold is obtained to quantify the comprehensive shape characteristics of each point on the auricle surface.

[0031] In combination with the biomechanical constraint condition and the multi-modal curvature manifold, the edges of the key structures of the auricle are identified. First, the mechanical active area is extracted by simulating the deformation of the auricle under stress; at the same time, the 2mm range around the anatomical landmark points is marked as the hard constraint area. Then, in the multi-modal curvature manifold, the curvature mutation points are screened as candidate edges, and the intersection of the mechanical active area and the hard constraint area is verified: if the candidate edge is located in the mechanical active area or the hard constraint area, it is retained as the biomechanical edge; otherwise, it is excluded. Finally, the biomechanical edge of the auricle is obtained, and the key structure boundary that needs to be retained is determined.

[0032] In combination with the dynamic noise reduction intensity, the multi-modal curvature manifold and the biomechanical edge, an adaptive manifold noise reduction mechanism is constructed. In the non-biomechanical edge area, anisotropic diffusion filtering is adopted, and the diffusion coefficient is determined by the dynamic noise reduction intensity and the multi-modal curvature, ensuring that the diffusion is stronger in the low curvature and high noise reduction intensity area, effectively smoothing the noise. In the biomechanical edge area, edge-preserving filtering is switched, and spatial domain weight and value domain weight are set to retain edge details. During the filtering process, the consistency of the blood flow signal and the mechanical parameter is monitored in real time, and if the blood flow oscillation index changes by more than 10% after filtering, the parameters are adjusted again to avoid excessive noise reduction.

[0033] In this embodiment, the auricle shape registration data is denoised by the adaptive manifold noise reduction mechanism to obtain the auricle shape denoising data; after filtering, the noise reduction quality is evaluated in three dimensions: the point cloud density in the geometric dimension is checked (target>50 points / mm 2 ), the modulus gradient continuity in the mechanical dimension is verified (gradient variance<0.05kPa / mm), and the registration error between the blood flow thermal map and the geometry is verified (error<0.2mm). If any dimension does not meet the standard, adjust the dynamic noise reduction intensity or filter parameters to reprocess, and finally obtain the auricle shape denoising data.

[0034] S2, construct a normal auricle shape threshold, and identify abnormal auricle shape data from the auricle shape pretreatment data according to the normal auricle shape threshold.

[0035] In the embodiment, a development trajectory function of a normal auricle is constructed, a normal auricle morphology threshold is obtained according to the development trajectory function, the normal auricle morphology threshold includes a curvature threshold, an elastic modulus threshold and a development state calibration threshold of the normal auricle, a progressive abnormality recognition mechanism is established based on the normal auricle morphology threshold, and abnormal auricle morphology data is obtained by combining the progressive abnormality recognition mechanism and auricle morphology pretreatment data.

[0036] Based on large-scale healthy population data, a development trajectory function of a normal auricle is constructed, first, geometric, mechanical and blood flow parameters of normal auricles in different age groups are collected, then, age correlation analysis is performed on each index, and the development trajectory function is established, based on the trajectory function, threshold values are determined by statistical normal auricle data: the curvature threshold value is the upper limit of the confidence of the geometric curvature, the elastic modulus threshold value distinguishes the regions (the helix E<1.2kPa, the lobe E<0.6kPa), and the development state calibration threshold value is analyzed by the residual of geometric size and age, if the development data deviates from the trajectory function evaluation value by more than 2 standard deviations, it is marked as development abnormality. Finally, the curvature threshold value, the elastic modulus threshold value and the development state calibration threshold value are integrated to form a multi-dimensional normal auricle morphology threshold system.

[0037] The above development trajectory function satisfies the following relationship: wherein, is the age-calibrated feature mean value, is the reference feature mean value, is the development peak coefficient, is the growth function, is the growth rate factor, is the individual actual age, is the growth inflection point age, is the gender correction coefficient, is the individual gender marker.

[0038] In the embodiment, auricle confidence levels are obtained, auricle multi-level confidence intervals are constructed in combination with the normal auricle morphology threshold, auricle development deviation-disease correlation indexes are obtained by coupling analysis of auricle morphology pretreatment data in combination with the development trajectory function and the auricle multi-level confidence intervals, and a progressive abnormality recognition mechanism is constructed based on the auricle multi-level confidence intervals and the auricle development deviation-disease correlation indexes.

[0039] The confidence level of the auricle is obtained in combination with the anatomical features. First, the auricle is divided into three levels of high, medium and low confidence areas according to the auricle anatomy: the high confidence area (such as the helix, the antihelix) has a regular shape and high data acquisition stability, and the confidence level is set to level 1; the medium confidence area (such as the triangular fossa, the concha cavity) has a complex structure but controllable data quality, and is set to level 2; the low confidence area (such as the edge of the ear lobe) is easily affected by motion artifacts, and is set to level 3. Then, in combination with the normal auricle shape threshold, the threshold range is defined for each confidence level: the strict threshold is used for the level 1 area (such as curvature <0.7mm⁻ 1 , elastic modulus <1.0kPa); the 10% relaxation is used for the level 2 area (curvature <0.8mm⁻ 1 , elastic modulus <1.1kPa); and the 15% relaxation is used for the level 3 area (curvature <0.9mm⁻ 1 , elastic modulus <1.2kPa). Finally, the auricle multi-level confidence interval is obtained, and the abnormal detection sensitivity of different areas is determined.

[0040] Further, first, the expected feature mean of each age group is calculated based on the development trajectory function; at the same time, the actual value of the corresponding point in the auricle shape pre-processing data is extracted. Then, in each confidence interval, the deviation of the actual value from the expected value is calculated; finally, the auricle development deviation-disease correlation index is constructed to quantify the correlation strength between development deviation and potential disease.

[0041] The auricle development deviation-disease correlation index satisfies the following relationship: wherein, is the auricle development deviation-disease correlation index, is the development trajectory deviation degree, is the primary abnormality strength, is the development deviation population standard deviation, is the abnormality strength population standard deviation, is the nearest boundary distance of the anatomical functional area, is the anatomical attenuation coefficient, is the second-order gradient of the elastic modulus, is the elastic modulus reference value.

[0042] Progressive anomaly recognition mechanism: the first-level screening is for the high confidence area, if the actual value exceeds the first-level threshold and the ear development deviation-disease correlation index is greater than 1.5, it is marked as mild abnormality; the second-level screening is for the medium confidence area, if the actual value exceeds the second-level threshold and the ear development deviation-disease correlation index is greater than 2.0, it is marked as moderate abnormality; the third-level screening is for the low confidence area, if the actual value exceeds the third-level threshold and the ear development deviation-disease correlation index is greater than 2.5, it is marked as severe abnormality. At the same time, dynamic feedback is introduced: if a certain area is marked in the low-level screening, its confidence level is automatically promoted, and the threshold and correlation index are re-evaluated to avoid missed detection.

[0043] In this embodiment, the ear morphology pre-processing data is subjected to anomaly recognition by using a progressive anomaly recognition mechanism to obtain abnormal ear morphology data. First, the ear morphology pre-processing data is input into the progressive anomaly recognition mechanism, and the ear multi-level confidence interval and development trajectory function are automatically loaded; three-level screening is performed based on the ear development deviation-disease correlation index combined with the progressive anomaly recognition mechanism; and finally, the abnormal ear morphology data containing the abnormal type, severity and spatial positioning are output, providing accurate input for subsequent feature fusion.

[0044] S3, obtaining abnormal ear morphology fusion data from the abnormal ear morphology data to construct an abnormal ear morphology joint feature tensor.

[0045] In this embodiment, multi-scale anomaly sensitive features are extracted from the abnormal ear morphology data to obtain abnormal ear morphology sensitive features; a feature fusion strategy is established, and the abnormal ear morphology sensitive features and the feature fusion strategy are combined to obtain abnormal ear morphology fusion data; and based on the abnormal ear morphology fusion data, spatio-temporal evolution is performed to obtain an abnormal ear morphology joint feature tensor.

[0046] A multi-resolution analysis framework was constructed based on preprocessed geometric point cloud data. A Gaussian pyramid model was used to perform three-level downsampling on the original data, forming auricular surface representations at different scales (original scale, 1 / 2 scale, and 1 / 4 scale). At each scale level, local feature analysis based on geometric measures was performed: at the original scale level, the distribution density of principal curvature extrema was calculated to identify micro-protrusions or depressions on the surface; at the 1 / 2 scale level, anisotropic diffusion equations were used for surface smoothing to suppress noise while preserving edge information, followed by calculation of the normal direction consistency index to capture morphological variations over a larger range; at the 1 / 4 scale level, wavelet transform was applied to extract frequency domain features, focusing on analyzing anomalous fluctuation patterns in low-frequency components. A cross-scale correlation mechanism was simultaneously established, mapping and aligning anomalous indicators at different levels by constructing a feature pyramid, and using a dynamic time warping algorithm to quantify the correlation of morphological variations between scales. For the screening of abnormal sensitive features, confidence evaluation is performed on features at each scale: the coefficient of variation of each feature point between adjacent scales is calculated, and when the coefficient of variation exceeds a preset threshold (such as 0.8), it is determined to be a high-sensitivity region. At the same time, the candidate features are classified and verified by support vector machine, and features with a classification accuracy of more than 90% are retained as sensitive features of abnormal auricle morphology, providing sensitive feature input with spatiotemporal correlation for subsequent feature fusion.

[0047] The above coefficients of variation satisfy the following relationship: in, The coefficient of variation of the feature points. For the number of scale levels, For the scale level index variable, For feature points In scale eigenvectors, For feature points The mean of multi-scale features For the Euclidean norm, It is the stability constant. Weights for anatomical locations.

[0048] In the present embodiment, a sensitive feature interaction matrix is constructed, the correlation between features of each scale is quantified, mutual information values between each pair of features are calculated to form an N-order symmetric matrix, and the matrix element value range [0, 1] represents the proportion of the amount of information shared by the feature pair; at the same time, the sensitive feature distribution entropy is calculated, the Shannon entropy formula is applied to each feature dimension, the feature distribution curve is fitted by the kernel density estimation method, and the entropy value vector of each feature is obtained as the sensitive feature distribution entropy; an entropy weighting strategy is designed to weight and fuse the entropy weighted feature interaction matrix and the sensitive feature distribution entropy to obtain the entropy weighted feature interaction matrix, so that the interaction intensity of high-entropy features (rich information) and low-entropy features (high certainty) is nonlinearly enhanced, and the noise interaction of low-information features is inhibited.

[0049] The above entropy weighting strategy satisfies the following relationship: wherein, is the element of the entropy weighted feature interaction matrix, is the element of the entropy weighted feature interaction matrix, is the entropy value of the feature , and is the entropy value of the feature .

[0050] A feature fusion strategy is established based on the entropy weighted feature interaction matrix: a threshold (such as 0.6) is set, when , it is determined that the feature and the feature have strong correlation, a weighted sum fusion strategy is adopted, and the fusion weight is determined by entropy value normalization; when , a maximum value retention strategy is adopted to prevent information loss of weakly correlated features.

[0051] In an optional embodiment, the entropy weighted feature interaction matrix is used to weight and fuse the sensitive features of the abnormal auricle shape to obtain the abnormal auricle shape fusion data. First, the feature space is reconstructed by matrix multiplication to strengthen the interaction intensity of high-entropy features and low-entropy features, then the weighted feature sequence is cut by time window sliding, and the dynamic evolution law of the shape feature is captured through the time sequence transmission of the hidden state, and finally the effective aggregation of the multi-scale abnormal features is realized to obtain the fusion data.

[0052] Further, the spatiotemporal evolution constraint is introduced, the abnormal auricle shape fusion data is input into the long short-term memory network, the dynamic weight adjustment of the cross-scale features is realized through the hidden state transmission of the time step, and finally the third-order feature tensor is output as the abnormal auricle shape joint feature tensor.

[0053] S4, according to the abnormal auricle shape joint feature tensor, an abnormal spatial mapping is performed to obtain an abnormal auricle shape spatial feature.​

[0054] S4, obtaining an abnormal auricle morphology spatial feature based on the abnormal auricle morphology joint feature tensor and the abnormal auricle biomechanics potential field. S41, constructing an abnormal auricle biomechanics potential field according to the abnormal auricle morphology joint feature tensor.

[0055] In this embodiment, first, the multiscale spatiotemporal feature tensor is mapped to the physical space domain, the finite element analysis framework is adopted, the auricle surface is discretized into triangular grid elements, and each node is assigned with attribute parameters based on the feature tensor. By applying the boundary conditions of the simulated sound wave excitation, the displacement field distribution of the grid nodes under steady-state vibration is calculated, and the principal stress direction diagram and the strain energy density cloud diagram are generated as the abnormal auricle biomechanics potential field. The energy aggregation areas are located in the potential field, which correspond to the mechanical imbalance points caused by the abnormal auricle morphology. For example, the biomechanics simulation of patients with congenital microtia shows that there are 3-5 high strain energy concentration areas on the abnormal auricle in the sound wave conduction path, and the distribution pattern is significantly statistically different from that of the normal auricle.

[0056] S42, obtaining an abnormal spatial feature vector by performing abnormal spatial mapping on the abnormal auricle biomechanics potential field.

[0057] In this embodiment, the wavelet threshold denoising algorithm is used to preprocess the potential field data, the soft threshold function is selected according to the noise level, and the 0.1mm-0.5mm level of microscopic deformation characteristics is retained while suppressing the Gaussian white noise; the field line tracking algorithm is introduced, the energy flow path is extracted in the triangular grid element by the gradient descent method with the potential energy gradient direction as the guide, and the abnormal spatial feature points are defined as the breakpoints of the continuous field lines; the topological features are modally fused with the geometric parameters such as the curvature and the circumference of the auricle surface by the non-negative matrix factorization, and the abnormal spatial feature vector with spatial structure information and biomechanics characteristics is generated.

[0058] S43, obtaining the abnormal auricle morphology spatial feature based on the abnormal spatial feature vector.

[0059] In this embodiment, the non-negative matrix factorization is used to modally fuse the topological features and the geometric features. The abnormal spatial feature vector, the curvature, the circumference and other geometric parameters of the auricle surface are jointly input into the multilayer perception machine, the feature weights are optimized through cross-validation, the mapping relationship between the abnormal spatial feature and the biological attribute is established, and finally the abnormal auricle morphology spatial feature is obtained, including a 128-dimensional feature vector, of which the first 32 dimensions represent the topological abnormalities, the middle 64 dimensions reflect the geometric deformation, and the last 32 dimensions correspond to the biomechanics misalignment.

[0060] S5, constructing an abnormal auricle morphology data model based on the abnormal auricle morphology spatial feature.

[0061] In this embodiment, a dynamic evolution tensor network is constructed as a model architecture; a model training mechanism is established based on biomechanical constraint conditions; and an abnormal auricle morphology data model is constructed by combining the abnormal auricle morphology spatial features, the model architecture, and the model training mechanism.

[0062] Specifically, a dynamic evolution tensor network is used as the model architecture, and the input is the abnormal auricle morphology spatial features. The first layer of the network constructs a space-time graph structure, discretizes the auricle surface into a node graph structure, associates each node with 5 nearest neighbor nodes to form an edge, and determines the graph weight by calculating the Euclidean distance between nodes through a Gaussian kernel function. The second layer introduces a dynamic evolution mechanism, uses a temporal convolution network to capture the temporal dependence of features, sets the receptive field to 8 time steps, and extracts multi-scale time patterns through dilated convolution. The third layer fuses biomechanical potential field information, uses the strain energy density field obtained by finite element analysis as prior knowledge, dynamically adjusts the node weight through a graph attention network, and strengthens the feature expression of abnormal areas. The final output layer uses a fully connected structure to generate an abnormal representation vector.

[0063] Further, a double-constraint loss function is introduced in the training process: a data-driven term and a physics-driven term. The data-driven term uses a cross-entropy loss, which calculates the classification error based on the clinical labeled deformity level. The physics-driven term constructs a biomechanical regularization term, which includes two aspects: first, by comparing the auricle deformation field predicted by the model with the true value field simulated by finite element analysis, the mean square error is calculated to constrain the physical reasonableness of the deformation pattern; second, a strain energy density threshold constraint is introduced, which triggers a penalty term when the local strain energy output by the model exceeds the 95% confidence interval of the normal auricle. The training uses a phased optimization strategy, with only the data-driven term parameters updated in the previous iteration rounds, and the double-constraint terms optimized jointly in the subsequent iteration rounds. To enhance the model's generalization, a transfer learning mechanism is introduced, which initializes the parameters on a pre-training set containing 1000 normal auricle data, and then fine-tunes on 200 abnormal data.

[0064] The abnormal auricle morphology spatial features are encoded; they are processed through 3 layers in the dynamic evolution tensor network; they are mapped to the biomechanical parameter space through a fully connected layer; the total loss is calculated; the optimizer is used for gradient update with dynamic learning rate decay; after training on 200 abnormal auricle data, the morphology features are preserved, and the quantitative evaluation report is obtained based on the abnormal auricle morphology data model, including quantitative indicators such as anatomical variation degree, mechanical stability index, and surgical correction priority.

[0065] See Figure 2In an optional embodiment, the present application provides an abnormal auricle morphology data model construction system, which comprises an input device, an output device, a processor and a memory, and is connected among the hardware facilities, wherein the memory is used to store a computer program comprising program instructions, and the processor is configured to invoke the program instructions to execute the specific steps of the abnormal auricle morphology data model construction method provided by the present application. The abnormal auricle morphology data model construction system provided by the present application has complete structure and objective stability, and improves the overall applicability and practical application ability of the present application.

[0066] In summary, the abnormal auricle morphology data model construction method and system provided by the present application are constructed through multi-modal data acquisition and adaptive noise reduction preprocessing; abnormal features are accurately extracted based on a development trajectory function and a progressive identification mechanism; combined multi-scale sensitive feature extraction, entropy weighted fusion strategy and spatio-temporal evolution generate joint feature tensors; spatial features are obtained through abnormal space mapping, and then an abnormal auricle morphology data model is constructed based on a dynamic tensor network and biomechanical constraints; multi-dimensional quantitative representation of complex auricle deformities is realized, the abnormal detection sensitivity and spatial positioning accuracy are improved, and support is provided for personalized correction schemes; the method of the present application is easy to understand, simple to calculate, and has less workload, and is convenient for practical application, and provides a theoretical basis and technical support for further development of intelligent medical technology field.

[0067] Finally, it should be noted that: the above embodiments are only used to illustrate the technical solutions of the present application, and not to limit them; although the present application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that: it can still modify the technical solutions recorded in the foregoing embodiments, or make equivalent replacement to part or all of the technical features; and these modifications or replacements do not make the essence of the corresponding technical solutions deviate from the scope of the technical solutions of the embodiments of the present application, and they should be covered in the scope of the claims and the specification of the present application.

Claims

1. A method for constructing a data model of abnormal auricle morphology, characterized in that, Includes the following steps: Acquire auricular morphology data and preprocess the auricular morphology data to obtain auricular morphology preprocessed data; A normal auricle shape threshold is constructed, and abnormal auricle shape data is obtained by performing abnormal feature identification on the auricle shape preprocessing data based on the normal auricle shape threshold. Abnormal auricle morphology fusion data is obtained based on the abnormal auricle morphology data to construct the abnormal auricle morphology joint feature tensor. The spatial features of the abnormal auricle morphology are obtained by performing anomaly space mapping based on the joint feature tensor of the abnormal auricle morphology. An abnormal auricle morphology data model is constructed based on the spatial features of the abnormal auricle morphology.

2. The method for constructing an abnormal auricle morphology data model according to claim 1, characterized in that, The process of acquiring auricle morphology data and preprocessing the auricle morphology data to obtain auricle morphology preprocessed data includes: Geometric point cloud data, elastic modulus matrix and blood flow heat map of the auricle are acquired to construct multimodal data, and the multimodal data is used as the auricle morphology data; Biomechanical constraints are constructed, and the auricular morphology data is registered based on the biomechanical constraints to obtain auricular morphology registration data. An adaptive manifold noise reduction mechanism is established, and auricular morphology noise reduction data is obtained by combining the auricular morphology registration data. The auricular morphology noise reduction data is used as the auricular morphology preprocessing data.

3. The method for constructing an abnormal auricle morphology data model according to claim 2, characterized in that, The establishment of the adaptive manifold noise reduction mechanism includes: The blood flow oscillation index of the auricle is obtained based on the blood flow heat map, and the dynamic noise reduction intensity is obtained based on the blood flow oscillation index. Multimodal joint curvature analysis was performed on the auricle morphology data to construct a multimodal curvature manifold of the auricle; The biomechanical edge of the auricle is obtained based on the biomechanical constraints and the multimodal curvature manifold. The adaptive manifold noise reduction mechanism is constructed by combining the dynamic noise reduction intensity, the multimodal curvature manifold, and the biomechanical edge.

4. The method for constructing an abnormal auricle morphology data model according to claim 1, characterized in that, The process of constructing a normal auricle morphology threshold and then performing abnormal feature identification on the preprocessed auricle morphology data based on the normal auricle morphology threshold to obtain abnormal auricle morphology data includes: A developmental trajectory function for a normal auricle is constructed, and a normal auricle morphology threshold is obtained based on the developmental trajectory function. The normal auricle morphology threshold includes the curvature threshold, elastic modulus threshold, and developmental state calibration threshold of the normal auricle. A progressive anomaly recognition mechanism is established based on the normal auricle morphology threshold. The abnormal auricle morphology data is obtained by combining the progressive anomaly recognition mechanism and the auricle morphology preprocessing data.

5. The method for constructing an abnormal auricle morphology data model according to claim 4, characterized in that, The progressive anomaly recognition mechanism based on the normal auricle morphology threshold includes: Obtain the confidence level of the auricle and construct a multi-level confidence interval for the auricle by combining it with the normal auricle morphology threshold. By combining the developmental trajectory function and the multi-level confidence interval of the auricle, the auricle morphology preprocessing data is coupled and analyzed to obtain the auricle developmental deviation-lesion correlation index. The progressive abnormality identification mechanism is constructed based on the multi-level confidence interval of the auricle and the auricle developmental deviation-lesion correlation index.

6. The method for constructing an abnormal auricle morphology data model according to claim 1, characterized in that, The step of obtaining abnormal auricle morphology fusion data based on the abnormal auricle morphology data to construct an abnormal auricle morphology joint feature tensor includes: Multi-scale abnormality sensitive feature extraction is performed on the abnormal auricle morphology data to obtain abnormal auricle morphology sensitive features; A feature fusion strategy is established, and the abnormal auricle morphology fusion data is obtained by combining the abnormal auricle morphology sensitive features and the feature fusion strategy. The joint feature tensor of the abnormal auricle morphology is obtained by spatiotemporal evolution based on the fusion data of the abnormal auricle morphology.

7. The method for constructing an abnormal auricle morphology data model according to claim 6, characterized in that, The established feature fusion strategy includes: Based on the aforementioned abnormal sensitive features, obtain the sensitive feature interaction matrix and the sensitive feature distribution entropy; The entropy-weighted feature interaction matrix is ​​obtained by combining the sensitive feature interaction matrix and the sensitive feature distribution entropy; The feature fusion strategy is established based on the entropy-weighted feature interaction matrix.

8. The method for constructing an abnormal auricle morphology data model according to claim 1, characterized in that, The step of obtaining spatial features of abnormal auricle morphology by performing abnormal space mapping based on the joint feature tensor of the abnormal auricle morphology includes: A biomechanical potential field of the abnormal auricle is constructed based on the joint feature tensor of the abnormal auricle morphology. Anomaly space feature vectors are obtained by performing anomaly space mapping on the biomechanical potential field of the abnormal auricle. The spatial features of the abnormal auricle shape are obtained based on the abnormal spatial feature vector.

9. The method for constructing an abnormal auricle morphology data model according to claim 2, characterized in that, The construction of the abnormal auricle morphology data model based on the spatial features of the abnormal auricle morphology includes: A dynamically evolving tensor network is constructed as the model architecture; A model training mechanism is established based on the aforementioned biomechanical constraints; The abnormal auricle morphology data model is constructed by combining the spatial features of the abnormal auricle morphology, the model architecture, and the model training mechanism.

10. A system for constructing a data model of abnormal auricle morphology, characterized in that, The system includes an input device, an output device, a processor, and a memory, which are interconnected. The memory is used to store a computer program, which includes program instructions. The processor is configured to call the program instructions to execute the abnormal auricle morphology data model construction method as described in any one of claims 1-9.

Images