Children urinary system image and pathological image reconstruction and analysis system based on multi-modal data

Abstract

The invention discloses a children urinary system image and pathological image reconstruction and analysis system based on multi-modal data, which belongs to the technical field of image processing and comprises a multi-modal data input module, a key area positioning module, a dynamic selective preprocessing unit and a reconstruction analysis module. The multi-modal data input module is used for receiving multi-modal data of a child urinary system and performing format standardization; the key area positioning module is used for positioning a key area and a non-key area of the urinary system of the child. According to the method, only high-compression-ratio storage, simplified denoising and other low-intensity processing are executed on a non-key area, and the calculation steps and carrying cost of redundant data are greatly reduced; the key areas such as the pelvis and the ureter opening are preprocessed according to the characteristics of the key areas, and it is ensured that diagnosis core information is not lost. According to the processing mode, the rigid contradiction that the full-amount processing efficiency is low and the full-amount simplification feature is damaged is solved, and the accurate balance between the efficiency and the integrity is realized.

Description

Technical Field

[0001] This invention belongs to the field of image processing technology, and in particular relates to a reconstruction and analysis system for pediatric urinary system images and pathological images based on multimodal data. Background Technology

[0002] Diagnosis of pediatric urinary system diseases, such as ureteropelvic reflux and hydronephrosis, typically requires the integration of imaging data from CT, MRI, and ultrasound with pathological slide images. Current technologies generally employ a full-data preprocessing and fixed-process fusion model for multimodal data processing. Specifically, uniform high-precision preprocessing is performed on all regions of the pediatric urinary system (including critical areas such as the renal pelvis and ureter, and non-critical areas such as surrounding adipose tissue), including multiple noise reduction steps, fine registration, and low-compression storage.

[0003] The core problem with this model is that non-critical areas account for a very high percentage of the pediatric urinary system, usually exceeding 70%. Performing the same high-precision processing on these areas as on critical areas would result in excessive data transfer and lengthy processing times, making it difficult to meet the timeliness requirements of emergency cases that need to produce diagnostic results quickly. This creates a rigid contradiction between processing efficiency and data integrity: if processing is simplified to ensure efficiency, the characteristics of critical areas may be lost; if full processing is performed to ensure integrity, the efficiency is low and cannot be adapted to emergency scenarios. Summary of the Invention

[0004] To address the problems existing in the prior art, this invention provides a reconstruction and analysis system for pediatric urinary system images and pathological images based on multimodal data. This system has the advantage of resolving the rigid contradiction between processing efficiency and data integrity, thus solving the problems of the prior art.

[0005] This invention is implemented as follows: a reconstruction and analysis system for pediatric urinary system images and pathological images based on multimodal data, comprising a multimodal data input module, a key region localization module, a dynamic selective preprocessing unit, and a reconstruction analysis module; the multimodal data input module receives multimodal data of the pediatric urinary system, including at least two of the following: CT, ultrasound, and pathological slide data, and performs format standardization; the key region localization module locates key and non-key regions of the pediatric urinary system, wherein key regions include the renal pelvis and ureteral orifice, and non-key regions include surrounding adipose tissue; and outputs the key regions. The system includes the division results of non-critical regions and the proportion D of critical region data to the total data, where D ranges from 0 to 1. The dynamic selective preprocessing unit is connected to the critical region positioning module and performs differentiated preprocessing based on the region division results: preprocessing adapted to the characteristics of critical regions is performed, while low-intensity preprocessing is performed on non-critical regions to reduce the amount of data transported and the processing time of non-critical regions. The low-intensity preprocessing for non-critical regions includes at least one of high compression ratio storage and simplified noise reduction process. The reconstruction and analysis module is used to reconstruct and analyze the data based on the preprocessed critical region data and output diagnostic results.

[0006] As a preferred embodiment of the present invention, the key area localization module also outputs the clinical importance weight I, noise level N, and structural complexity S of the key area; the clinical importance weight I is set according to the anatomical priority of the pediatric urinary system, wherein the I value of the renal pelvis-ureter junction is 0.9-1.0, the I value of the bladder trigone is 0.8-0.9, and the I value of the surrounding adipose tissue is 0.1-0.2.

[0007] In a preferred embodiment of the present invention, the key region localization module includes a multi-dimensional feature extraction unit, a dynamic key weight calculation unit, a region dynamic partitioning unit, and a linkage signal generation unit. The multi-dimensional feature extraction unit extracts the anatomical features, pathological correlation features, data quality features, and clinical scene features of the region. The anatomical features include location and morphology; the pathological correlation features include spatial overlap P with abnormal pathological sections; the data quality features include noise level and modal integrity Q; and the clinical scene features include case type and chief complaint relevance C. The dynamic key weight calculation unit calculates the dynamic key weight W based on the multi-dimensional features, using the formula W=0. 0.3×A+0.4×P+0.2×Q+0.1×C, where A is the anatomical basis weight, and the initial value is equal to the clinical importance weight I; the region dynamic division unit divides regions based on W using a non-fixed threshold, where W≥0.4 is a critical region for emergency cases, W≥0.6 is a critical region for routine cases, and regions where the W value of the corresponding region of the follow-up case differs from that of the previous child by more than 30% are critical regions; the linkage signal generation unit outputs region type signal, preprocessing priority signal, and feature protection requirement signal to the dynamic selective preprocessing unit, so that the dynamic selective preprocessing unit adjusts the processing intensity based on the signal, forming a dynamic linkage between positioning and preprocessing.

[0008] As a preferred embodiment of the present invention, the key region localization module further includes a feature weight fusion unit, used to resolve potential conflicts between the dynamic criticality weight W and the clinical importance weight I. Specifically, it involves: establishing a baseline mapping between I and A: using the clinical importance weight I as the initial value of the anatomical basis weight A; dynamically adjusting constraints: when the deviation between W and I is ≤20%, maintaining the original calculation formula of W; when the deviation is >20%, adjusting the formula to W=0.5×A+0.3×P+0.15×Q+0.05×C, increasing the weight of A to force W to converge towards I, avoiding excessive degradation of key anatomical regions due to dynamic features; and unified output adaptation: mapping the noise level N to the inverse value of the data quality feature Q, i.e., N=1-Q, fusing the structural complexity S with the pathological association feature P, and simultaneously outputting I, N, S, and the constrained adjusted W, so that the dynamic selective preprocessing unit can both meet the fixed anatomical priority requirements based on I and adapt to dynamic scenarios through W.

[0009] As a preferred embodiment of the present invention, the dynamic selective preprocessing unit is based on a preprocessing adaptation coefficient. And preprocessing priority signal adjustment processing intensity, preprocessing adaptation coefficient The calculation formula is: ;in, Weighted by clinical importance. Noise level, , For structural complexity, This represents the percentage of data volume in key regions. The dynamic keyness weights are adjusted according to constraints. ;

[0010] Treatment intensity classification:

[0011] when When it is a "highest priority signal", high-intensity preprocessing is used;

[0012] when For "high priority signals", use medium-intensity preprocessing;

[0013] when When there are no high-priority signals, low-intensity preprocessing is used.

[0014] As a preferred embodiment of the present invention, the CT data slice thickness received by the multimodal data input module is 0.5-1mm, the ultrasound data frame rate is 30fps, the pathological slide data magnification is 20× / 40×, and the format is standardized to DICOM format; the data will be synchronously transmitted to the multidimensional feature extraction unit for calculating pathological correlation feature P and data quality feature Q.

[0015] In a preferred embodiment of the present invention, the reconstruction analysis module, based on preprocessed key region data, outputs a 3D urinary system structure and pathological-imaging fusion atlas, automatically identifies abnormal features and outputs quantitative indicators; abnormal feature identification prioritizes and focuses on... or The area.

[0016] As a preferred embodiment of the present invention, it further includes an individual baseline dynamic calibration module, which is linked with the region dynamic division unit and is based on key regions. Value and Values ​​are used to construct an individual baseline for the child; when the baseline deviation is... At that time, Value rise And the upward fluctuation is the same as The values ​​are positively correlated.

[0017] Preferably, the invention also includes a clinical feedback module, which collects correction information from doctors regarding diagnostic results and updates the feature weights of the multi-dimensional feature extraction unit based on the correction information: if the correction is for localization deviation, then the anatomical features of that region are improved. Weights in value calculation; if corrected for feature omission, the pathological association features of that region are improved. exist Weights in value calculation.

[0018] As a preferred embodiment of the present invention, in high-load scenarios, the dynamic selective preprocessing unit automatically triggers a maximum efficiency mode, adjusting based on the region type signal and the preprocessing priority signal:

[0019] For non-critical areas: the processing intensity is further reduced;

[0020] Key areas or The highest priority region: maintain high-intensity preprocessing to ensure that core features are not lost.

[0021] Compared with the prior art, the beneficial effects of the present invention are as follows:

[0022] 1. The core drawback of traditional full-data preprocessing and fixed-process fusion models is that they apply the same high-precision processing to non-critical areas (such as surrounding adipose tissue and normal renal parenchyma) that account for more than 70% of the pediatric urinary system as to critical areas. This leads to a surge in data transfer volume, inefficient use of computing resources, and ultimately low processing efficiency. This system addresses this problem through an innovative logic of dividing critical and non-critical areas and implementing differentiated preprocessing: non-critical areas undergo only low-intensity processing such as high-compression storage and simplified noise reduction, significantly reducing computational steps and data transfer costs for redundant data; critical areas such as the renal pelvis and ureteral orifices undergo preprocessing tailored to their characteristics, ensuring no core diagnostic information is lost. This processing model fundamentally resolves the rigid contradiction between low efficiency in full-data processing and feature loss in full-data simplification, achieving a precise balance between efficiency and completeness.

[0023] 2. Pediatric urinary system emergencies (such as acute exacerbation of hydronephrosis or impacted ureteral stones) are characterized by rapid disease progression and low tolerance. Traditional systems with lengthy processing cycles (often several hours) cannot meet the clinical demand for rapid diagnostic results in emergencies, potentially delaying intervention. This system significantly shortens the overall data processing chain through low-intensity preprocessing of non-critical areas, concentrating core processing resources on feature preservation and optimization in critical areas, ensuring that emergency cases can obtain diagnostic reports in a short time. This advantage buys valuable time for timely diagnosis and treatment of children with emergencies, reducing the risk of complications due to diagnostic delays (such as renal parenchymal compression injury or infection spread). Attached Figure Description

[0024] Figure 1 This is a data input and key area positioning block diagram provided in the embodiments of the present invention;

[0025] Figure 2 This is a block diagram of dynamic preprocessing and diagnostic output provided in an embodiment of the present invention. Detailed Implementation

[0026] To further understand the invention's content, features, and effects, the following embodiments are provided, and detailed descriptions are given in conjunction with the accompanying drawings.

[0027] The structure of the present invention will now be described in detail with reference to the accompanying drawings.

[0028] like Figures 1 to 2 As shown in the figure, the present invention provides a reconstruction and analysis system for pediatric urinary system images and pathological images based on multimodal data, including a multimodal data input module, a key region localization module, a dynamic selective preprocessing unit, and a reconstruction analysis module;

[0029] The multimodal data input module is used to receive multimodal data of the pediatric urinary system, including at least two of the following: CT, ultrasound, and pathological slide data, and to standardize the format.

[0030] The critical area positioning module is used to locate the critical and non-critical areas of the pediatric urinary system. The critical areas include the renal pelvis and ureteral orifice, while the non-critical areas include the surrounding adipose tissue. The module also outputs the division results of the critical and non-critical areas and the proportion D of the data volume of the critical area to the total data volume, where D ranges from 0 to 1.

[0031] The dynamic selective preprocessing unit is signal-connected to the key area positioning module and performs differentiated preprocessing based on the area division results: preprocessing adapted to the characteristics of key areas is performed, and low-intensity preprocessing is performed on non-key areas to reduce the amount of data transported and the processing time of non-key areas; wherein, performing low-intensity preprocessing on non-key areas includes at least one of high compression ratio storage and simplified noise reduction process.

[0032] The reconstruction analysis module is used to reconstruct and analyze key area data based on preprocessed data and output diagnostic results.

[0033] This application breaks away from the traditional indiscriminate processing of all data by focusing on key regions and using differentiated preprocessing. It prioritizes efficiency while maintaining the integrity of key regions, enabling efficient analysis of multimodal data. The specific process is as follows:

[0034] Multimodal data input and standardization: The system receives raw data such as CT, ultrasound, and pathological slides of the pediatric urinary system through the multimodal data input module. First, it converts different formats (such as DICOM for CT, MP4 for ultrasound, and TIFF for pathological slides) into the standardized DICOM format to eliminate compatibility issues caused by data format differences and lay the foundation for subsequent processing.

[0035] Key and non-key area division: The key area positioning module is based on the anatomical features of the pediatric urinary system (such as the renal pelvis and ureteral orifice as core diagnostic areas, and surrounding adipose tissue as non-diagnostic core areas), automatically identifies and divides key and non-key areas, and calculates the proportion D of key area data to the total data, thus clarifying the focus of data processing.

[0036] Dynamic selective preprocessing: The dynamic selective preprocessing unit works in conjunction with the key area localization module to execute differentiated processing strategies.

[0037] For critical regions: perform preprocessing adapted to their characteristics (such as basic optimizations like denoising and registration) to ensure that the structural / pathological features required for diagnosis are not lost;

[0038] For non-critical areas: perform low-intensity preprocessing (such as high compression ratio storage and simplified noise reduction process) to significantly reduce the amount of non-core data to be moved and the computational overhead, and avoid excessive processing that consumes resources.

[0039] Reconstruction Analysis and Result Output: The reconstruction analysis module performs 3D structural reconstruction, pathological image fusion, and abnormal feature identification based solely on preprocessed key area data, generating a report containing diagnostic conclusions. The entire process focuses on key data, avoids redundant processing of non-critical areas, and ensures that emergency cases can obtain results quickly.

[0040] Timeliness verification closed loop: The system has built-in emergency timeliness verification logic to ensure that the processing time meets the needs of clinical emergency care, and finally outputs a diagnostic report that meets the timeliness requirements, thus completing the entire data processing and analysis process.

[0041] Furthermore, the key region localization module also outputs the clinical importance weight I (value 0-1), noise level N (value 0-1), and structural complexity S (value 0-1) of the key region.

[0042] The clinical importance weight I is set according to the anatomical priority of the pediatric urinary system, with I values ​​of 0.9-1.0 for the renal pelvis-ureter junction, 0.8-0.9 for the bladder trigone, and 0.1-0.2 for the surrounding adipose tissue. Quantifying the characteristics of key regions provides a basis for adjusting the preprocessing intensity, addressing the issue of feature damage caused by insufficient processing of highly important key regions. Noise level N and structural complexity S are described below.

[0043] Furthermore, the key area localization module includes a multi-dimensional feature extraction unit, a dynamic keyness weight calculation unit, a dynamic area division unit, and a linkage signal generation unit:

[0044] The multi-dimensional feature extraction unit is used to extract the anatomical features (location, morphology), pathological correlation features, data quality features, and clinical scene features of the region, breaking through the traditional logic of relying solely on anatomical location for division; among them, the anatomical features of the region include location and morphology, the pathological correlation features include spatial overlap with abnormal pathological slices P, the data quality features include noise level and modal integrity Q, and the clinical scene features include case type and chief complaint relevance C.

[0045] The dynamic criticality weight calculation unit calculates the dynamic criticality weight W based on multi-dimensional features, with the formula W=0.3×A+0.4×P+0.2×Q+0.1×C, where A is the anatomical basis weight and the initial value is equal to the clinical importance weight I.

[0046] The dynamic regional division unit uses a non-fixed threshold to divide regions based on W. Among them, W≥0.4 is a key region for emergency cases, W≥0.6 is a key region for routine cases, and regions where the difference in W value between follow-up cases and the corresponding regions of the same child in the previous cases is >30% is a key region, breaking the static division mode of "fixed anatomical list".

[0047] The linkage signal generation unit outputs a region type signal (critical / non-critical), a preprocessing priority signal (W≥0.8 is the highest priority, 0.6≤W<0.8 is a high priority, and W<0.6 is a normal priority) and a feature protection requirement signal (marked when P≥0.7 or Q≤0.3) to the dynamic selective preprocessing unit, so that the dynamic selective preprocessing unit adjusts the processing intensity based on the signal to form a dynamic linkage between positioning and preprocessing.

[0048] It should be noted that traditional key area localization relies on a fixed anatomical list (such as presupposing the renal pelvis and ureter as key areas and adipose tissue as non-key areas), which is a static and empirical division that cannot adapt to the dynamic clinical needs of pediatric urology. For example, a certain area may be non-key in routine cases but become a core diagnostic basis in specific pathological scenarios. This solution breaks away from the technical inertia of determining keyness based on anatomical location, and adopts a three-dimensional approach based on the dynamism of clinical needs, the correlation of data characteristics, and the feedback from pathological results to achieve dynamic area division, allowing the boundary between key and non-key areas to be adjusted in real time according to case characteristics.

[0049] The following is a detailed explanation:

[0050] 1. The multi-dimensional feature extraction unit is used to extract non-traditional partitioning features from multimodal data, including:

[0051] Anatomical features: Traditional location information (such as renal pelvis coordinates, ureteral course);

[0052] Pathological association feature P: Spatial overlap with pathological section abnormalities (such as glomerulosclerosis, inflammatory cell infiltration) (0-1, the higher the overlap, the higher the potential criticality).

[0053] Data quality characteristics Q: Noise level and modal integrity of regional data (e.g., ultrasound artifact intensity and CT slice thickness continuity);

[0054] Clinical scenario feature C: Case type (emergency / routine / follow-up), chief complaint (e.g., "lower back pain" suggests higher priority for the renal pelvis area).

[0055] 2. Dynamic Keyness Weight Calculation Unit: Based on multi-dimensional features, calculates the dynamic keyness weight W (0-1, higher values ​​indicate a more key region) for each region. The formula uses weighted logic: W=0.3×A+0.4×P+0.2×Q+0.1×C, where:

[0056] A: Anatomical baseline weights (0-1, initial value of 0.8 for traditional key areas such as the renal pelvis, and initial value of 0.2 for adipose tissue, but these can be adjusted by other features).

[0057] P: Pathological association characteristics (0-1, the higher the overlap with abnormal pathological sections, the larger the value);

[0058] Q: Data quality characteristics (0-1, lower noise and higher integrity, larger values ​​are used to avoid misjudging non-critical data due to poor data).

[0059] C: Clinical scenario characteristics (0-1, emergency / chief complaint related area value is 1, routine area value is 0.5).

[0060] 3. Dynamic Regional Division: Regions are divided based on dynamic criticality weight W (no fixed threshold; the threshold is dynamically adjusted according to case type):

[0061] Emergency cases: Use a lenient threshold (W≥0.4 is considered a critical area) to avoid missing potential abnormal areas;

[0062] For routine cases: standard thresholds are used (W≥0.6 is considered a critical area);

[0063] Follow-up cases: Historical comparison thresholds were used (areas with a difference of >30% from the previous critical area W value of the child were included in the critical area).

[0064] Output: Dynamic partitioning results (coordinate mask of key / non-key regions) and W value, pathological association feature P, and data quality feature Q of key regions.

[0065] 4. Linkage Signal Generation Unit: Converts the region segmentation results and core features (W, P, Q) into control signals that can be recognized by the dynamic selective preprocessing unit, including:

[0066] Region type signal (critical / non-critical);

[0067] Preprocessing priority signals (based on W value: W≥0.8 is the highest priority, 0.6≤W<0.8 is a high priority, and W<0.6 is a normal priority);

[0068] When the feature protection requirement signal (based on P and Q: P≥0.7 or Q≤0.3) is met, the tag needs enhanced feature protection.

[0069] Through the above settings, the key area localization module and the dynamic selective preprocessing unit form a two-way dynamic feedback mechanism, which has the following effects:

[0070] First, real-time adaptation of regional criticality and preprocessing intensity: The dynamic selective preprocessing unit receives the W value and preprocessing priority signal from the linkage signal and automatically matches the processing intensity.

[0071] Highest priority region (W≥0.8): Enhanced high-intensity preprocessing (6 iterations of noise reduction, sub-pixel level registration + edge enhancement, lossless compression);

[0072] High-priority regions (0.6≤W<0.8): employ conventional high-intensity preprocessing (5 iterations of noise reduction, sub-pixel level registration);

[0073] For critical regions with normal priority (W<0.6): use medium-intensity preprocessing;

[0074] Non-critical areas: Maintain low-intensity pretreatment, but when the W value fluctuates by more than 20% (e.g., W in a certain area increases from 0.3 to 0.5 in a follow-up case), automatically and temporarily increase to medium intensity.

[0075] Second, preprocessing enhancement of pathologically associated regions: When the linkage signal marker requires enhanced feature protection (P≥0.7 or Q≤0.3), the dynamic selective preprocessing unit triggers the pathological association enhancement mode:

[0076] For regions overlapping with abnormal pathological sections (P≥0.7): A pathology-image feature alignment step is added (e.g., forced registration of the cellular texture of the pathological section with the structural features of the image). Specifically, in the pathology-image feature alignment step of this system, forced registration actively corrects the spatial deviation between the microscopic features of the pathological section and the macroscopic structure of the image through an algorithm, ensuring a rigid alignment operation where both correspond precisely in the same coordinate system. Its core is to break down the spatial separation between pathological section and image data, allowing the position, scale, and angle of microscopic pathological details (such as cell texture and lesion infiltration range) and macroscopic anatomical structures (such as the renal pelvis wall and renal parenchyma) to completely match. This alignment is unaffected by data acquisition deviations, prioritizing the accuracy of the corresponding critical diagnostic areas.

[0077] Because pathological sections and imaging data are acquired in completely different ways, spatial biases naturally exist: pathological sections are tissue specimens that have been sliced, stained, and photographed under a microscope, reflecting microscopic local tissues (such as glomeruli within 1 mm²); imaging (CT / ultrasound) is a whole-body scan, reflecting macroscopic anatomical locations (such as the three-dimensional structure of the entire renal pelvis). Ordinary automatic alignment may result in blurred alignment due to deviations in specimen sampling location and scanning angles (for example, sclerotic glomeruli in pathological sections cannot find a precise corresponding anatomical location on imaging). For areas with P ≥ 0.7 (highly overlapping with abnormal pathology), which are core diagnostic areas, the algorithm must be forced to eliminate biases through precise correction, even if the adjustment requires a larger margin, to ensure a one-to-one correspondence between microscopic pathological features and macroscopic imaging structures.

[0078] Taking a case of glomerulosclerosis in a child as an example, the specific process is as follows:

[0079] Extract alignment benchmarks: Extract rigid markers from pathological sections, such as the clustered distribution outline of sclerotic glomeruli and the aggregation area of ​​inflammatory cells (microscopic benchmarks); extract corresponding anatomical markers from images (such as CT scans), such as the specific thickness range of the renal cortex and the direction of vascular branches in the glomerular region (macroscopic benchmarks).

[0080] Algorithm-forced correction: Adjust the coordinate system of the pathological slide through rigid transformations such as translation, rotation, and scaling: for example, align the center position of the sclerotic glomerular clusters in the pathological slide with the center position of the corresponding area of ​​the renal cortex in the image;

[0081] Correcting scale differences: One pixel in a pathological section corresponds to 1 μm in actual tissue, while one pixel in an image corresponds to 0.1 mm. By scaling up, the scales of the two are forced to be consistent (for example, by magnifying the features of the pathological section by 100 times to match the structure of the image).

[0082] Verifying alignment accuracy: The overlap between pathological features and imaging features is calculated (e.g., the overlap rate between the contour of sclerotic glomeruli and low-density areas in the image). If the overlap does not reach a preset threshold (e.g., ≥90%), the algorithm will iterate and adjust again until the accuracy requirement is met. This requirement to meet the accuracy requirement is the core manifestation of the mandatory process.

[0083] After forced registration, when doctors view images, they can click on a certain area of ​​the renal pelvis wall and directly associate it with the corresponding pathological section details (e.g., if the image shows thickening in that area, the corresponding pathological section shows 20% glomerular sclerosis). This avoids the problem of pathological abnormalities not being able to find the corresponding image location or image abnormalities not having pathological support due to alignment deviations, and provides a precise fusion of macroscopic structure and microscopic pathology for diagnosis.

[0084] For low-quality data regions (Q≤0.3): Enable the redundancy information compensation algorithm (e.g., complete the noise region of the current layer based on the thickness data of adjacent CT layers).

[0085] Third, dynamic iteration of regional criticality: If the dynamic selective preprocessing unit finds a feature anomaly in a non-critical region during the processing (such as the presence of high gray values ​​of suspected stones after preprocessing), it will send a feature anomaly signal back to the critical region positioning module.

[0086] After receiving the signal, the region dynamic division unit of the key region positioning module recalculates the W value of the region (automatically temporarily increasing the P weight to 0.5). If the new W is greater than or equal to the threshold, it is reclassified as a key region, and the preprocessing unit is notified to start supplementary processing (upgrading from low intensity to medium intensity).

[0087] This setup, through dynamic criticality weights W, allows the same anatomical region to have a variable role in different cases (e.g., adipose tissue is non-critical in routine cases but becomes critical in a pathological scenario of fat infiltration), addressing the insufficient clinical adaptability of traditional fixed classifications. Furthermore, it upgrades the system from a unidirectional output of "localization to preprocessing" to a bidirectional closed loop of "localization to preprocessing and then to localization correction," enabling preprocessing not only to passively execute the classification results but also to actively feed back into the region criticality assessment, enhancing the system's overall adaptability. Additionally, it adjusts the classification threshold using clinical scenario features C, ensuring that potential abnormalities are not overlooked in emergency cases and that resources are not wasted in routine cases, achieving a balance between comprehensiveness and efficiency.

[0088] Example: In a case of "suspected hydronephrosis in an obese child," the multi-dimensional feature extraction unit found that the overlap between the adipose tissue and the pathological section's fat infiltration area was P=0.8, while the clinical scenario was a routine examination with C=0.5. The dynamic criticality weight calculation unit calculated W=0.3×0.2 (initial A value of adipose tissue) + 0.4×0.8 + 0.2×0.9 (good data quality) + 0.1×0.5=0.56. The region dynamic segmentation unit (with a threshold of 0.6 for routine cases) initially classified the adipose tissue as a non-critical region, but because P=0.8≥0.7, the linkage signal marked "feature protection needs to be strengthened." After receiving the signal, the dynamic selective preprocessing unit performed "non-critical region and feature protection" processing on the adipose tissue (the compression ratio was relaxed from 1:20 to 1:10, preserving infiltration details).

[0089] During preprocessing, a suspected blurred infiltration boundary was detected, and a "feature anomaly signal" was reported. The key area localization module recalculated W (the weight of P was temporarily increased to 0.5), and the new W = 0.62 ≥ 0.6. The area was then reclassified as a key area, and supplementary processing (adding an edge sharpening step) was initiated.

[0090] Furthermore, the key region localization module also includes a feature weight fusion unit to resolve potential conflicts between the dynamic criticality weight W and the clinical importance weight I, specifically:

[0091] Establish a baseline mapping between I and A: use the clinical importance weight I as the initial value of the anatomical basis weight A to ensure that A conforms to a fixed anatomical priority;

[0092] Dynamic adjustment constraint: When the deviation between W and I is ≤20%, the original calculation formula of W is maintained; when the deviation is >20%, the formula is adjusted to W=0.5×A+0.3×P+0.15×Q+0.05×C, and the weight of A is increased to force W to converge to I, so as to avoid key anatomical regions being over-degraded due to dynamic features.

[0093] Unified output adaptation: The noise level N is mapped to the inverse value of the data quality feature Q, i.e., N=1-Q. The structural complexity S is fused with the pathological correlation feature P. At the same time, I, N, S and W adjusted by constraints are output, so that the dynamic selective preprocessing unit can meet the fixed anatomical priority requirements based on I and adapt to the dynamic scenario through W.

[0094] Specifically, the calculation logic of the dynamic criticality weight W (weakening the weight of fixed anatomy and strengthening dynamic features such as pathology and scene) may conflict with the clinical importance weight I, which is based on the priority of fixed anatomy.

[0095] Since I is static (e.g., I is fixed at 0.9-1.0 at the renal pelvis-ureter junction), it serves as one of the core bases for adapting pretreatment intensity.

[0096] The calculation of dynamic criticality weight W reduces the weight of anatomical basis weight A to 0.3 (weakening the influence of fixed anatomy), which may lead to inconsistency between the criticality reflected by W and I, and thus cause contradiction between the signal received by the dynamic selective preprocessing unit (based on W) and the preprocessing basis based on I.

[0097] The core logic and operational steps to solve the problem are as follows:

[0098] The baseline mapping between I and A: The clinical importance weight I is used as the initial value for the anatomical basis weight A (ensuring that A conforms to a fixed anatomical priority):

[0099] Renal pelvis-ureter junction: Initial value of A = I = 0.9-1.0;

[0100] Bladder trigone: Initial value of A = I = 0.8-0.9;

[0101] Surrounding adipose tissue: Initial value of A = I = 0.1-0.2.

[0102] Dynamic adjustment constraints: The calculation of dynamic keyness weight W requires I as the baseline constraint.

[0103] When the deviation between the dynamically calculated W and I is ≤20%, the original calculation logic of W is maintained: W=0.3A+0.4P+0.2Q+0.1C to ensure dynamic adaptability.

[0104] When the deviation between W and I is greater than 20% (e.g., W drops to 0.6 in a certain renal pelvis region due to poor data quality, while I = 0.9, resulting in a deviation of 33%), the weight of A in W is automatically increased to 0.5 (adjustment formula: W = 0.5A + 0.3P + 0.15Q + 0.05C), forcing W to converge to I and preventing key anatomical regions from being over-degraded due to dynamic features.

[0105] Unified output adaptation: Two sets of collaborative parameters are output to the dynamic selective preprocessing unit. I, N, and S are retained (N corresponds to the inverse mapping of data quality feature Q, and S integrates pathological association feature P and structural complexity); dynamic criticality weight W and adjustment instructions are added (such as W being calculated based on I constraints), so that the preprocessing unit can both meet the fixed anatomical priority requirements based on I and adapt to dynamic scenarios through W.

[0106] Verification of linkage effect and conflict resolution:

[0107] Taking an example where the initial W=0.6 in the renal pelvis region was I=0.9 due to severe ultrasound artifacts (Q=0.2):

[0108] The feature weight fusion unit detected a deviation of 33% (>20%) between W and I, and initiated a constraint mechanism: adjusted the W calculation formula and increased the weight of A to 0.5;

[0109] Recalculate W = 0.5 × 0.9 (A = I) + 0.3 × 0.7 (P, assuming moderate pathological correlation) + 0.15 × 0.2 (Q) + 0.05 × 0.5 (C, routine case) = 0.45 + 0.21 + 0.03 + 0.025 = 0.715. The deviation from I = 0.9 is reduced to 20.6% (close to the threshold), avoiding excessive downgrading.

[0110] Output to the dynamic selective preprocessing unit: I=0.9, N=0.8 (Q=0.2 reverse mapping, high noise), S=0.7 (fusion pathological association), with W=0.715 and "adjustment based on I constraint" explanation attached;

[0111] The preprocessing unit combines I=0.9 (high anatomical importance) and W=0.715 (moderate dynamic criticality) to employ high-intensity preprocessing, while also specifically enhancing denoising (due to N=0.8), which is both conflict-free and adaptable to dynamic features.

[0112] In this system, structural complexity S (with a value of 0-1) is a quantitative indicator of the complexity of the anatomical structure of the pediatric urinary system.

[0113] I. Data Foundation: Structural Feature Extraction from Multimodal Data

[0114] The calculation of S is based on the input standardized multimodal data (CT, ultrasound, pathological slides), and the core extraction objects include:

[0115] Anatomical morphological features (from imaging data):

[0116] Irregularity of the regional contour (such as serrated edges of the renal pelvis due to dilation, degree of tortuosity of the ureter);

[0117] Internal structural density (such as the pixel value distribution complexity of dense areas of the glomerulus, and the ambiguity of the boundary between the renal parenchyma and the renal pelvis).

[0118] Spatial topological relationships (such as the angle of the junction between the renal pelvis and ureter, and the number of branches).

[0119] Pathological correlation features (derived from the fusion of pathological sections and images):

[0120] Spatial overlap with pathologically abnormal areas (e.g., the corresponding area of ​​glomerular sclerosis in the image, P-value);

[0121] The complexity of the microstructure shown in pathological sections (such as the disordered tissue texture caused by inflammatory cell infiltration, which is mapped to the features of the imaging area).

[0122] II. Quantization Logic: Weighted Fusion of Multiple Features

[0123] The specific value of S is obtained by weighting the complexity of anatomical morphology with the complexity of pathological association, using the formula: S = 0.6 × M + 0.4 × P, where:

[0124] M(0-1): Anatomical morphological complexity, quantified based on image features (e.g., the higher the irregularity of the contour, the closer M is to 1).

[0125] P(0-1): Pathological association characteristics (the higher the spatial overlap with abnormal pathological sections, the closer P is to 1).

[0126] Among them, the multi-dimensional feature extraction unit provides raw feature data for M and P (such as image contour extraction results and pathology-image overlap calculation results); when the pathology-related feature P is more critical to diagnosis, the weight of P in S calculation can be dynamically increased (e.g., from 0.4 to 0.5), making S more in line with clinical needs.

[0127] Example: In a case of renal pelvis dilation:

[0128] Imaging showed an irregular renal pelvis outline (serrated edges) with internal stone shadows, and an anatomical morphology complexity M=0.8;

[0129] Pathological sections showed that the area corresponded to renal pelvis mucosal hyperplasia, with a pathological correlation characteristic P=0.7;

[0130] Calculate S = 0.6 × 0.8 + 0.4 × 0.7 = 0.48 + 0.28 = 0.76 (High structural complexity requires intensive preprocessing to preserve details).

[0131] In summary, the core value of S is to quantify the complexity of regional structures by integrating macroscopic anatomical morphology with microscopic pathological correlation, providing a basis for calculating the preprocessing adaptation coefficient K, ensuring that highly complex regions (such as structurally disordered lesion areas) receive sufficient preprocessing intensity, and avoiding feature damage.

[0132] Furthermore, the dynamic selective preprocessing unit is based on the preprocessing adaptation coefficient. and the preprocessing priority signal adjusts the processing intensity, the preprocessing adaptation coefficient The calculation formula is: ;in, Weighted by clinical importance. Noise level ( ), For structural complexity (integrating pathological association features) ), This represents the percentage of data volume in key regions. The dynamic keyness weights are adjusted according to constraints. ;

[0133] Treatment intensity classification:

[0134] when When it is a "highest priority signal", high-intensity preprocessing is used;

[0135] when For "high priority signals", use medium-intensity preprocessing;

[0136] when When there are no high-priority signals, low-intensity preprocessing is used.

[0137] By integrating fixed anatomical importance and dynamic criticality, a more precise preprocessing intensity control logic is constructed, overcoming the limitations of single-parameter judgment. The constrained and adjusted dynamic criticality weight W serves as a multiplier, ensuring that K simultaneously reflects three core types of information: fixed clinical importance I, data features (N, S), and dynamic scenario adaptability W, where N = 1 - Q (Q being the data quality feature), and the pathological association feature P fused with S, achieving end-to-end parameter adaptation. A joint decision-making process using K value and preprocessing priority signals is employed, preserving the quantitative judgment of K value while responding to priority signals (e.g., the "highest priority signal" directly triggers high-intensity processing), avoiding conflicts arising from single-parameter judgment logic. This setting balances fixed anatomical priority I and dynamic scenario W, preventing overprocessing of high-I but low-dynamic-criticality regions or underprocessing of low-I but high-dynamic-criticality regions. Furthermore, dynamic features such as W, P, and Q directly influence processing intensity, enhancing the linkage depth between localization and preprocessing.

[0138] For example: in a routine case of a 6-year-old child, the renal pelvis area ( ), moderate ultrasonic noise ( ), pathological correlation features ( After fusion, a value of 0.7 is used; dynamic keyness weight. (Adjusted for constraints), percentage of data volume .

[0139] calculate value: Decision logic: ,and For high-priority signals, high-intensity preprocessing is triggered (5 iterations of denoising, sub-pixel level registration, and lossless compression), which satisfies both high priority and low latency requirements. The region's characteristic protection needs are also adapted to its dynamic criticality.

[0140] Furthermore, the multimodal data input module receives CT data with a slice thickness of 0.5-1mm, ultrasound data with a frame rate of 30fps, and pathological slide data with a magnification of 20× / 40×. The data is standardized to DICOM format. The data will be synchronously transmitted to the multidimensional feature extraction unit for calculating pathological correlation features P and data quality features Q.

[0141] Data Input and Standardization: Specific parameters for CT, ultrasound, and pathological sections are defined (e.g., CT slice thickness 0.5-1mm), and uniformly converted to DICOM format to ensure data compatibility. Input data is not only used for subsequent preprocessing but also simultaneously transmitted to the multi-dimensional feature extraction unit, directly providing raw data for pathological correlation features P (e.g., calculation of overlap between pathological sections and image regions) and data quality features Q (e.g., ultrasound noise level assessment), ensuring the basis for calculating dynamic criticality weights W.

[0142] Furthermore, based on the preprocessed key region data, the reconstruction analysis module outputs a 3D urinary system structure and pathology-image fusion atlas, automatically identifies abnormal features (such as renal pelvis dilation and glomerular sclerosis) and outputs quantitative indicators (such as dilation diameter and stone volume); the abnormal feature identification prioritizes focusing on... or The area.

[0143] To avoid feature extraction errors caused by inconsistent data formats and parameters, and improve system compatibility, synchronous data transmission ensures that multi-dimensional feature extraction units can obtain complete raw data, making the calculation of P and Q more accurate and laying the foundation for the reliability of W and K values.

[0144] Furthermore, it also includes an individual baseline dynamic calibration module, which is linked with the region dynamic division unit and is based on key regions. Value and Values ​​are used to construct an individual baseline for the child; when the baseline deviation is... At that time, the aforementioned Value rise And the upward fluctuation is the same as Positive correlation ( value) Time rise Time rise ).

[0145] This setting binds individual baseline construction to I and W features, and the K value correction is positively correlated with W, enabling personalized preprocessing intensity adjustment.

[0146] Individual baseline construction: Based on the I value (ensuring that the core anatomical regions are included in the baseline) and the W value (ensuring that dynamic key regions are included in the baseline), individual baselines (such as renal pelvis volume / age ratio, ureteral course curvature) are constructed.

[0147] Baseline deviation correction: When the baseline deviation is greater than 20%, the K value is increased by 10%-15%, and the increase is positively correlated with W—15% for W≥0.8 (highest dynamic criticality) and 10% for W<0.8, so that the dynamic critical region receives more sufficient preprocessing enhancement.

[0148] This setting, combining I and W to construct a baseline, covers both fixed core regions and dynamic critical regions, adapting to individual developmental differences in children. Furthermore, the increase in the K value is linked to W, avoiding insufficient correction for dynamic critical regions or excessive correction for low-dynamic critical regions.

[0149] For example, in the following scenario: a follow-up case of a 1-year-old child, the renal pelvis region I=0.95, the current W=0.82 (highest dynamic criticality), the individual baseline was a renal pelvis volume of 4.2 cm³, the current measurement of the renal pelvis volume was 5.3 cm³, and the baseline deviation was 26.2% > 20%.

[0150] Correction operation: Since W=0.82≥0.8, the K value is increased by 15%; assuming the original K=1.3, the corrected K=1.3×1.15=1.495. Based on the corrected K=1.495, the preprocessing unit adopts enhanced high-intensity preprocessing (6 iterations of noise reduction and edge sharpening) to accurately capture potential abnormal features caused by the increase in renal pelvis volume.

[0151] Furthermore, it also includes a clinical feedback module, which collects doctors' correction information on diagnostic results (such as misdiagnosis of renal pelvis dilation) and updates the feature weights of the multi-dimensional feature extraction unit based on the correction information: if the correction is to a localization error, then the anatomical features of that area are improved. The weight in the value calculation (e.g., increasing from 0.3 to 0.4); if corrected for feature omission, the pathological association features of that region are improved. exist Weights in value calculation (e.g., increasing from 0.4 to 0.5); optimization of the value by updating the weights. Value and the stated The accuracy of the value calculation.

[0152] This setup directly links clinical feedback with the multi-dimensional feature extraction unit, optimizing W and K values ​​through feature weight adjustments to create a closed-loop iteration. Specifically, it collects doctors' correction information on diagnostic results, identifying error types (such as localization bias or feature omission). For different error types, the feature weights of the multi-dimensional feature extraction unit are adjusted—for localization bias, the proportion of anatomical features in the A value (anatomical basis weight); for feature omission, the proportion of pathologically relevant features P in the W value calculation is increased. After the weights are updated, the W value calculation for subsequent cases is more accurate, thereby optimizing the K value and making the preprocessing intensity more aligned with clinical reality.

[0153] This setting transforms physician experience into quantifiable feature weight adjustments, addressing the disconnect between system output and clinical practice. As feedback data accumulates, the calculation accuracy of W and K values ​​gradually improves, making dynamic region partitioning and preprocessing intensity determination more aligned with clinical needs.

[0154] For example, when reviewing a case of an obese child, a doctor missed a ureteral stricture encased in adipose tissue. This region had an I=0.85, but due to fat artifacts, the pathological association feature P weight in the original multi-dimensional feature extraction unit was only 0.4, resulting in a W=0.55, and it was not classified as a critical region. Because the doctor missed labeling features, the clinical feedback module increased the weight of the pathological association feature P in the W value calculation for this region from 0.4 to 0.5. For similar obese child cases, the P value weight for this ureteral region was increased, resulting in a calculated W=0.65 (meeting the standard case threshold W≥0.6), and it was classified as a critical region. The K value was simultaneously increased, and the preprocessing unit used medium-intensity preprocessing, successfully identifying the ureteral stricture feature.

[0155] Furthermore, in high-load scenarios ( (For example, in the case of simultaneous input in the emergency department), the dynamic selective preprocessing unit automatically triggers the maximum efficiency mode, adjusting based on the region type signal and the preprocessing priority signal:

[0156] For non-critical areas Processing intensity was further reduced (compression ratio increased to 1:30, denoising iteration 1); for key areas... or Highest priority area: Maintain high-intensity preprocessing to ensure no core features are lost; balance efficiency under high load with the integrity of critical areas. High load trigger: When ≥10 emergency cases are entered simultaneously, the extreme efficiency mode is automatically triggered.

[0157] The beneficial effects are as follows: processing of non-critical areas is further simplified, significantly reducing data transfer volume and time consumption, ensuring the timeliness of emergency cases. High-priority critical areas are selected using W and I values ​​to maintain high-intensity preprocessing and avoid missed diagnoses due to the pursuit of efficiency.

[0158] For example: A hospital suddenly had 12 emergency cases of pediatric urinary system (high-load scenario), among which 1 child had a renal pelvis region I=0.95, W=0.88 (highest priority), and surrounding adipose tissue W=0.3 (non-critical area).

[0159] Extreme efficiency mode operation:

[0160] Non-critical regions (adipose tissue): The processing intensity is further reduced, with one denoising iteration and a compression ratio of 1:30, to quickly complete data processing and reduce resource consumption;

[0161] Highest priority region (renal pelvis): Maintain high-intensity preprocessing (5 iterations of noise reduction, sub-pixel level registration) to ensure that core abnormal features such as renal pelvis dilation and stones are not lost.

[0162] It should be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article, or apparatus.

[0163] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A system for reconstructing and analyzing pediatric urinary system images and pathological images based on multimodal data, characterized in that, It includes a multimodal data input module, a key area localization module, a dynamic selective preprocessing unit, and a reconstruction analysis module; The multimodal data input module is used to receive multimodal data of the pediatric urinary system, including at least two of the following: CT, ultrasound, and pathological slide data, and to standardize the format. The critical area positioning module is used to locate the critical and non-critical areas of the pediatric urinary system. The critical areas include the renal pelvis and ureteral orifice, while the non-critical areas include the surrounding adipose tissue. The module also outputs the division results of the critical and non-critical areas and the proportion D of the data volume of the critical area to the total data volume, where D ranges from 0 to 1. The dynamic selective preprocessing unit is signal-connected to the key area positioning module and performs differentiated preprocessing based on the area division results: preprocessing adapted to the characteristics of key areas is performed, and low-intensity preprocessing is performed on non-key areas to reduce the amount of data transported and the processing time of non-key areas; wherein, performing low-intensity preprocessing on non-key areas includes at least one of high compression ratio storage and simplified noise reduction process. The reconstruction analysis module is used to reconstruct and analyze key area data based on preprocessed data and output diagnostic results.

2. The system for reconstructing and analyzing pediatric urinary system images and pathological images based on multimodal data according to claim 1, characterized in that, The critical region localization module also outputs the clinical importance weight I, noise level N, and structural complexity S of the critical region. The clinical importance weight I is set according to the anatomical priority of the pediatric urinary system, wherein the I value of the renal pelvis-ureter junction is 0.9-1.0, the I value of the bladder trigone is 0.8-0.9, and the I value of the surrounding adipose tissue is 0.1-0.

2.

3. The system for reconstructing and analyzing pediatric urinary system images and pathological images based on multimodal data according to claim 2, characterized in that, The key region localization module includes a multi-dimensional feature extraction unit, a dynamic keyness weight calculation unit, a region dynamic division unit, and a linkage signal generation unit. The multi-dimensional feature extraction unit is used to extract the anatomical features, pathological correlation features, data quality features, and clinical scene features of the region. The anatomical features of the region include location and morphology; the pathological correlation features include spatial overlap with abnormal pathological slices (P); the data quality features include noise level and modal integrity (Q); and the clinical scene features include case type and chief complaint relevance (C). The dynamic criticality weight calculation unit calculates the dynamic criticality weight W based on multi-dimensional features, with the formula W=0.3×A+0.4×P+0.2×Q+0.1×C, where A is the anatomical basis weight and the initial value is equal to the clinical importance weight I. The dynamic region division unit uses a non-fixed threshold to divide regions based on W. Among them, W≥0.4 is a key region for emergency cases, W≥0.6 is a key region for routine cases, and regions where the difference in W value between follow-up cases and the corresponding regions of the same child in the previous period is >30% is a key region. The linkage signal generation unit outputs a region type signal, a preprocessing priority signal, and a feature protection requirement signal to the dynamic selective preprocessing unit, enabling the dynamic selective preprocessing unit to adjust the processing intensity based on the signals, thus forming a dynamic linkage between positioning and preprocessing.

4. The system for reconstructing and analyzing pediatric urinary system images and pathological images based on multimodal data according to claim 3, characterized in that, The key region localization module also includes a feature weight fusion unit to resolve potential conflicts between the dynamic criticality weight W and the clinical importance weight I, specifically: Establish a baseline mapping between I and A: use the clinical importance weight I as the initial value of the anatomical basis weight A; Dynamic adjustment constraint: When the deviation between W and I is ≤20%, the original calculation formula of W is maintained; when the deviation is >20%, the formula is adjusted to W=0.5×A+0.3×P+0.15×Q+0.05×C, and the weight of A is increased to force W to converge to I, so as to avoid key anatomical regions being over-degraded due to dynamic features. Unified output adaptation: The noise level N is mapped to the inverse value of the data quality feature Q, i.e., N=1-Q. The structural complexity S is fused with the pathological correlation feature P. At the same time, I, N, S and W adjusted by constraints are output, so that the dynamic selective preprocessing unit can meet the fixed anatomical priority requirements based on I and adapt to the dynamic scenario through W.

5. The system for reconstructing and analyzing pediatric urinary system images and pathological images based on multimodal data according to claim 4, characterized in that, The dynamic selective preprocessing unit is based on a preprocessing adaptation coefficient. and the preprocessing priority signal adjusts the processing intensity, the preprocessing adaptation coefficient The calculation formula is: ;in, Weighted by clinical importance. Noise level, , For structural complexity, This represents the percentage of data volume in key regions. The dynamic keyness weights are adjusted according to constraints. ; Treatment intensity classification: when When it is a "highest priority signal", high-intensity preprocessing is used; when For "high priority signals", use medium-intensity preprocessing; when When there are no high-priority signals, low-intensity preprocessing is used.

6. The system for reconstructing and analyzing pediatric urinary system images and pathological images based on multimodal data according to claim 5, characterized in that, The multimodal data input module receives CT data with a slice thickness of 0.5-1mm, ultrasound data with a frame rate of 30fps, and pathological slide data with a magnification of 20× / 40×. The data is standardized to DICOM format. The data will be synchronously transmitted to the multidimensional feature extraction unit for calculating pathological correlation features P and data quality features Q.

7. The system for reconstructing and analyzing pediatric urinary system images and pathological images based on multimodal data according to claim 6, characterized in that, The reconstruction and analysis module, based on preprocessed key region data, outputs a 3D urinary system structure and pathological-imaging fusion atlas, automatically identifies abnormal features and outputs quantitative indicators; the abnormal feature identification prioritizes and focuses on... or The area.

8. The system for reconstructing and analyzing pediatric urinary system images and pathological images based on multimodal data according to claim 7, characterized in that, It also includes an individual baseline dynamic calibration module, which is linked with the region dynamic division unit and is based on key regions. Value and Values ​​are used to construct an individual baseline for the child; when the baseline deviation is... At that time, the aforementioned Value rise And the upward fluctuation is the same as The values ​​are positively correlated.

9. The system for reconstructing and analyzing pediatric urinary system images and pathological images based on multimodal data according to claim 8, characterized in that, It also includes a clinical feedback module, which collects doctors' correction information on diagnostic results and updates the feature weights of the multi-dimensional feature extraction unit based on the correction information: if the correction is for localization deviation, then the anatomical features of that area are improved. Weights in value calculation; if corrected for feature omission, the pathological association features of that region are improved. exist Weights in value calculation.

10. The system for reconstructing and analyzing pediatric urinary system images and pathological images based on multimodal data according to claim 9, characterized in that, Under high-load scenarios, the dynamic selective preprocessing unit automatically triggers the extreme efficiency mode, adjusting based on the region type signal and the preprocessing priority signal: For non-critical areas: the processing intensity is further reduced; Key areas or The highest priority area: maintain high-intensity preprocessing.

Images