Intelligent image detection system for dynamically monitoring vascular calcification of diabetes

Abstract

The invention discloses an intelligent image detection system for dynamically monitoring vascular calcification of diabetes, and belongs to the technical field of medical images and intelligent diagnosis and treatment. Comprising a multi-modal image acquisition module, a cross-modal data processing module, a dynamic modeling and intelligent prediction module and a closed-loop intervention module. According to the invention, through multi-modal data acquisition and processing, integration and analysis of data from different sources are realized, so that characteristics and change rules of diabetic vascular calcification can be better known. And secondly, by utilizing a dynamic modeling and intelligent prediction module, simulation and prediction of the diabetes vascular calcification process can be realized, and making of personalized diagnosis and treatment plans and intervention measures is facilitated. And finally, the closed-loop intervention module can adjust a treatment scheme in real time according to a monitoring result so as to achieve a better treatment effect. Accurate dynamic monitoring and closed-loop intervention of diabetic vascular calcification are achieved, and the method has the advantages of intelligent switching of detection modes, deep fusion of multi-modal data, dynamic optimization of treatment schemes and the like.

Description

Technical Field

[0001] The present invention belongs to the technical field of medical imaging detection and intelligent health management, and in particular relates to an intelligent imaging detection system for dynamic monitoring of diabetic vascular calcification. Background Art

[0002] Diabetic vascular calcification is an important pathological feature of diabetic cardiovascular complications. Its dynamic evolution process involves multi-dimensional interactions between the biochemistry and biomechanics of the vascular wall. In the field of early diagnosis and dynamic monitoring of diabetic vascular calcification, existing technical means such as reference patents CN116386860A and CN119655879A have achieved intelligent assisted prediction and diagnosis based on multimodal imaging, as well as planning and navigation of thyroid nodule surgery. However, these technologies are insufficient in the accurate monitoring of diabetic vascular calcification and real-time optimization of personalized treatment plans.

[0003] Reference patent CN116386860A achieves predictive evaluation of diabetes and its complications through artificial intelligence algorithms, but lacks in-depth interpretation of the dynamic process of vascular calcification and dynamic adjustment of treatment strategies; and reference patent CN119655879A, although improving the real-time and accuracy of surgical planning, does not involve long-term dynamic monitoring of vascular calcification and closed-loop optimization of treatment plans.

[0004] In addition, the prior art also has the following defects: Limitations of single-modality detection: Traditional CT or ultrasound only provides anatomical structural information and cannot simultaneously obtain biochemical microenvironmental parameters related to calcification; Static analysis limitations: Conventional image post-processing techniques (such as the Agatston score) only assess calcification burden and cannot quantify calcium deposition rate and mechanical stability; Intervention lag: Treatment strategy adjustments rely on manual experience and lack a closed-loop feedback mechanism based on dynamic prediction models.

[0005] Therefore, there is an urgent need for an intelligent imaging detection system that can dynamically monitor, intelligently predict, and provide personalized intervention for diabetic vascular calcification. Summary of the Invention

[0006] The intelligent imaging detection system for dynamic monitoring of diabetic vascular calcification in this invention comprises a multimodal image acquisition module, a cross-modal data processing module, a dynamic modeling and intelligent prediction module, and a closed-loop intervention module. These modules work together to achieve comprehensive, dynamic monitoring and accurate prediction of diabetic vascular calcification, while also providing personalized testing solutions and treatment recommendations.

[0007] To solve the above technical problems, the present invention is achieved through the following technical solutions: The intelligent imaging detection system for dynamic monitoring of diabetic vascular calcification of the present invention is characterized by comprising Multimodal image acquisition module, used to obtain images of vascular structure, chemical composition and biomechanical parameters; The cross-modal data processing module realizes spatiotemporal registration and feature extraction of multi-source data based on a hybrid gated attention mechanism, and integrates temporal biochemical features with frequency-domain mechanical features through a dual-stream architecture. The dynamic modeling and intelligent prediction module builds a four-dimensional calcification dynamics model based on fused data and uses a cross-modal graph neural network to output calcification progression prediction results; The closed-loop intervention module dynamically adjusts the detection protocol and treatment strategy based on the model prediction results and generates personalized detection plans and treatment recommendations.

[0008] As a preferred technical solution of the present invention, the multimodal image acquisition module includes: Structural imaging unit, integrating coronary CT angiography, optical coherence tomography, and intravascular ultrasound to obtain the three-dimensional anatomical structure of blood vessels and the distribution of plaque components; The molecular targeted imaging unit, through intravenous injection of RGD peptide-modified superparamagnetic iron oxide nanoprobes, synchronously collects near-infrared zone II fluorescence signals and magnetic resonance T2* sequence images to achieve receptor-specific targeted imaging of calcification foci and quantify the amount of iron deposition based on signal intensity.

[0009] As a preferred technical solution of the present invention, the multimodal image acquisition module further includes a non-invasive detection subsystem: The skin glycation end products detection unit uses excitation light to penetrate the epidermis and captures the fluorescence lifetime distribution of AGEs in the dermis through the fluorescence lifetime imaging module to obtain the fluorescence lifetime change rate of epidermal AGEs. The vascular shear wave detection unit uses a multi-frequency focused ultrasound transducer array to collect the shear wave frequency shift of the vascular endothelium, and combines it with the differential equation of elastic mechanics to infer the elastic modulus of the calcification focus.

[0010] As a preferred technical solution of the present invention, the cross-modal dynamic fusion module adopts a dual-stream Transformer architecture, including: Time domain flow, processing AGEs fluorescence lifetime time series data, with a window length of 30-120 minutes; Frequency domain flow, analyzing the short-time Fourier transform spectrum of shear wave frequency shift in the frequency range of 5-30Hz; A hybrid gated attention layer addresses the dimensionality difference between biochemical and mechanical parameters through a learnable weight matrix.

[0011] As a preferred technical solution of the present invention, the weight calculation of the hybrid gated attention layer satisfies The query vector Q comes from the time domain features, the bond vector K comes from the frequency domain features, and the entangled state parameter ψ is optimized by the exponential decay function: , .

[0012] As a preferred technical solution of the present invention, the dynamic modeling and intelligent prediction module includes a four-dimensional calcification phase field equation: Where, μ is the interfacial energy coefficient, μ∈[0.12,0.18], The calcium deposition rate β∈[1.2×10⁻³,1.8×10⁻³], the inhibition factor γ∈[0.07,0.09], and H(AGEs) is a step function driven by the AGEs concentration.

[0013] As a preferred technical solution of the present invention, the dynamic modeling and intelligent prediction module adopts a dual-delay deep deterministic policy gradient algorithm, takes glycated hemoglobin and pulse wave velocity as core reward functions, and outputs a dynamic dosing regimen for metformin and strontium ranelate.

[0014] As a preferred technical solution of the present invention, the closed-loop intervention system includes: Wearable monitoring unit to collect blood glucose levels and vascular mechanics parameters in real time; Adaptive scheduling unit automatically adjusts the imaging detection cycle according to the calcification risk level; The drug optimization unit dynamically adjusts the dosage ratio of hypoglycemic drugs and anti-calcification drugs based on reinforcement learning algorithms.

[0015] As a preferred technical solution of the present invention, the multimodal image acquisition module, cross-modal data processing module, dynamic modeling and intelligent prediction module, and closed-loop intervention module work together through the following process: When the wearable monitoring unit detects blood glucose ≥11.1 mmol / L or the AGEs fluorescence lifetime change rate ≥0.5 ns / h, the molecular targeted imaging program is triggered; The vascular shear wave detection unit synchronously collects circumferential strain and radial pressure data, which are input into the cross-modal feature fusion module together with the image data; The dynamic modeling and prediction module outputs the volume change rate and risk level of calcified plaques, and the closed-loop intervention module generates detection frequency adjustment instructions and individualized treatment recommendations based on this.

[0016] The present invention has the following beneficial effects: The present invention adopts multimodal image acquisition technology, which can obtain information on vascular structure, chemical composition, biomechanics and other aspects, and comprehensively reflect the pathological characteristics of diabetic vascular calcification. Compared with single modality detection, it has higher accuracy and comprehensiveness.

[0017] The data processing method based on the hybrid gated attention mechanism and the dual-stream Transformer architecture effectively solves the problems of spatiotemporal alignment and feature fusion of multimodal data, improves the efficiency and accuracy of data processing, and provides strong support for accurate prediction.

[0018] The dynamic modeling and intelligent prediction module can dynamically predict the progression of vascular calcification through a four-dimensional calcification dynamics model and a cross-modal graph neural network, and provide personalized drug treatment plans in combination with reinforcement learning algorithms, thus achieving precision medicine.

[0019] The closed-loop intervention module monitors patients' physiological parameters in real time through wearable devices and automatically adjusts detection and treatment strategies based on predicted results, thereby achieving dynamic management of diabetic vascular calcification and improving patients' treatment compliance and efficacy.

[0020] Of course, any product implementing the present invention does not necessarily need to achieve all of the advantages described above at the same time. BRIEF DESCRIPTION OF THE DRAWINGS

[0021] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the following briefly introduces the drawings required for describing the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.

[0022] Figure 1 This is a diagram of the system architecture of the present invention, showing the collaborative process of multimodal acquisition, cross-modal fusion, dynamic modeling and closed-loop intervention; Figure 2 This is the computational flow chart of the hybrid gated attention layer in the present invention, including the time domain and frequency domain feature alignment and weight optimization process; Figure 3 4 is a closed-loop intervention workflow diagram of the present invention. DETAILED DESCRIPTION

[0023] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making any creative efforts shall fall within the scope of protection of the present invention.

[0024] like Figure 1-3 As shown: The present invention is an intelligent imaging detection system for dynamic monitoring of diabetic vascular calcification, including (1) Multimodal image acquisition module: full-domain capture of four-dimensional data Core principle: Through the organic integration of structural imaging, molecular targeted imaging, non-invasive biochemical testing and biomechanical testing, a four-dimensional data chain of "anatomy-molecular-biochemical-mechanical" of vascular calcification is constructed.

[0025] 1. Structural Imaging Unit: Macroscopic and Microscopic Structure Analysis Technology combination: integrated coronary CT angiography CCTA, optical coherence tomography OCT, and intravascular ultrasound IVUS.

[0026] CCTA: Provides the spatial distribution of three-dimensional calcified plaques in blood vessels and quantifies plaque volume and calcification burden index.

[0027] OCT: Identify fibrous cap thickness and lipid pool boundaries with a resolution of 1-10 μm.

[0028] IVUS: measures plaque volume and lipid core load (a lipid core ratio >40% indicates high risk), and simultaneously obtains the vascular remodeling index (RI = vessel area / luminal area, RI>1.1 indicates positive remodeling).

[0029] 2. Molecular Targeted Imaging Unit: Calcification-Specific Markers Probe design: RGD peptide-modified superparamagnetic iron oxide nanoprobe achieves calcification targeting through the specific binding of RGD peptide to integrin αvβ3 on the surface of calcification foci.

[0030] Dual-modality imaging: Near-infrared II (NIR-II) fluorescence imaging: wavelength 1000-1700nm, penetration depth 5-10mm, fluorescence intensity (FI) is positively correlated with probe enrichment.

[0031] Magnetic resonance T2*-weighted imaging: quantifies iron deposition by measuring T2 relaxation time, reflecting the inflammatory activity of calcification foci.

[0032] 3. Non-invasive detection subsystem: monitoring of dynamic body surface indicators Skin glycation end products (AGEs) detection unit: Principle: 375nm excitation light penetrates the epidermis (depth 50-100μm), and AGEs in the dermis produce characteristic fluorescence, and the fluorescence lifetime distribution is captured by the fluorescence lifetime imaging (FLIM) module.

[0033] Quantitative indicators: Based on the double exponential decay model ( ) to calculate the fluorescence lifetime change rate.

[0034] Vascular shear wave detection unit: Technical path: The multi-frequency focused ultrasound transducer array emits shear waves and collects endothelial frequency shift signals.

[0035] Mechanical inversion: Based on the elastic mechanics equation ( ), the elastic modulus of the calcification was inverted using the Levenberg-Marquardt algorithm (E ≥ 100 kPa for hard calcification, and E < 50 kPa for soft calcification).

[0036] (2) Cross-modal data processing module: deep modeling of spatiotemporal correlation Core architecture: A two-stream Transformer network based on a hybrid gated attention mechanism to achieve spatiotemporal alignment and nonlinear fusion of time-domain biochemical features and frequency-domain mechanical features.

[0037] 1. Two-stream feature extraction Time Stream: Input: AGEs fluorescence lifetime time series; Processing flow: One-dimensional convolution layer, output time domain feature vector after layer normalization .

[0038] Frequency Stream: Input: Short-time Fourier transform (STFT) spectrum of shear wave frequency shift; Processing flow: The two-dimensional convolution layer (kernel=3×3, stride=1) extracts the energy distribution features of the 5-30Hz frequency band, and outputs the frequency domain feature vector after maximum pooling. .

[0039] 2. Hybrid Gated Attention Mechanism Dimension normalization: Normalize (Z-score) the time domain features (dimension ns) and the frequency domain features (dimension Hz) to eliminate the difference in physical units.

[0040] Cross-modal interaction: Weight calculation: The weight calculation of the hybrid gated attention layer satisfies The query vector Q comes from the time domain features, the bond vector K comes from the frequency domain features, and the entangled state parameter ψ is optimized by the exponential decay function: .

[0041] (3) Dynamic Modeling and Intelligent Prediction Module: Accurate Four-Dimensional Evolution Deduction Core framework: A dual-track prediction system of "physical modeling + data-driven" based on calcification phase field model and reinforcement learning.

[0042] 1. Four-dimensional calcification phase field equation Mathematical description: Physical meaning: : calcification phase field variable (0 = non-calcified phase, 1 = calcified phase); : interfacial energy coefficient, which controls the diffusion rate of the calcified boundary; : Calcium deposition rate, related to extracellular calcium concentration ( ) positively correlated; : TGF-β-mediated inhibitory factor, reflecting anti-calcification ability; : AGEs-driven step function (when AGEs fluorescence lifetime τ ≥ 4 ns, H = 1, otherwise H = 0).

[0043] Numerical simulation: The finite element method (FEM) was used to discretize the equations, with a spatial grid resolution of 0.1 mm and a time step of Δt = 1 h, to simulate the dynamic growth of calcified plaques in three-dimensional space (e.g., predicting the rate of change of plaque volume ΔV / Δt within 12 months).

[0044] 2. Reinforcement Learning Intervention Optimization Algorithm selection: Double-delayed deep deterministic policy gradient algorithm (TD3), suitable for optimization in continuous action space.

[0045] State space (State): 12-dimensional feature vector, including blood glucose (Glu), HbA1c, AGEs fluorescence lifetime (τ), elastic modulus (E), plaque volume (V), PWV, etc.

[0046] Action space: Dynamic matching of metformin dose (500-2000mg / d, 250mg increments) and strontium ranelate dose (1-2g / d, 0.5g increments).

[0047] Reward function (Reward): Among them, ΔRFU is the rate of change of NIR-II fluorescence intensity (reflecting the reduction of probe enrichment), and the balance between blood glucose control and calcification reversal is achieved by maximizing the reward function.

[0048] (IV) Closed-loop intervention module: dynamic response and intelligent control Core mechanism: A closed loop of "risk assessment-strategy adjustment-effect feedback" based on real-time monitoring data to achieve adaptive optimization of detection and treatment.

[0049] 1. Wearable monitoring unit Hardware composition: Continuous glucose monitoring (CGM) sensor (accuracy ±0.8mmol / L, sampling frequency 1Hz); Flexible pressure sensor array (fitted to the radial artery, measuring radial pressure of the vessel wall, with a resolution of 0.1 kPa).

[0050] Trigger conditions: When blood glucose ≥ 11.1 mmol / L or the AGEs fluorescence lifetime change rate ≥ 0.5 ns / h, the molecular targeted imaging program is automatically triggered.

[0051] 2. Adaptive Scheduling Unit Risk grading standards: Low risk: ΔV / Δt<0.5mm³ / month, detection cycle 6 months; Medium risk: 0.5≤ΔV / Δt<1.5mm³ / month, detection cycle 3 months; High risk: ΔV / Δt ≥ 1.5 mm³ / month, testing cycle 1 month.

[0052] Scheduling logic: Dynamically adjust the testing frequency based on the latest prediction results. For example, if the ΔV / Δt of a medium-risk patient drops to 0.4 mm³ / month for two consecutive times, the patient is downgraded to low risk.

[0053] 3. Drug Optimization Unit Dosage adjustment rules: When HbA1c > 8.5% and PWV > 10 m / s, the metformin dose was increased by 250 mg / d to a maximum dose of 2000 mg / d; When the NIR-II fluorescence intensity continued to increase (ΔFI > 10% / month), strontium ranelate 1 g / d treatment was initiated, and the efficacy was evaluated every 2 weeks and adjusted to a maximum dose of 2 g / d.

[0054] A specific application scenario of this system is as follows: Scene Background A 56-year-old male patient with an 8-year history of type 2 diabetes, with a long-term HbA1c level of 8.0%-8.5%, and a history of hypertension (140 / 90 mmHg), had experienced chest tightness with activity for the past 3 months. Clinically, he was suspected of progressive coronary artery calcification and was enrolled in this system for dynamic monitoring.

[0055] 1. Multimodal Image Acquisition Module: The Whole Process of Data Acquisition 1. Acquisition of Vascular Structural Images Coronary CT angiography (CCTA) Scanning parameters: Third-generation dual-source CT, tube voltage 100 kV, tube current 250 mAs, pitch 0.5, slice thickness 0.625 mm, contrast agent volume 60 ml (iodixanol 320 mgI / ml).

[0056] Data output: Three-dimensional reconstruction showed 40% calcified plaque (volume 18 mm³) in the proximal segment of the left anterior descending artery, calcification burden index (CVI) = 35%, and lumen stenosis of 20%-30%.

[0057] Intravascular ultrasound (IVUS) Operation process: A 40MHz IVUS catheter was inserted through the radial artery with an automatic retraction speed of 0.5mm / s to acquire cross-sectional images of the blood vessels.

[0058] Key indicators: Plaque volume 22 mm³, lipid core 40% (grayscale IVUS shows hypoechoic area), vascular remodeling index (RI) = 1.2 (positive remodeling, indicating plaque instability).

[0059] Optical coherence tomography (OCT) Imaging conditions: 20 ml of normal saline was injected to flush the vascular lumen, 1310 nm near-infrared light was emitted, and the axial resolution was 10 μm.

[0060] Feature identification: fibrous cap thickness 50μm (<65μm indicates fragility), and the continuity of the fibrous cap on the surface of the calcified nodule is interrupted.

[0061] 2. Chemical component detection Molecular targeted imaging: RGD-SPIO probe dual-modality imaging Dosage regimen: RGD-SPIO probe (dose 50 μmol / kg) was injected intravenously, and imaging was started 90 minutes later.

[0062] Near-infrared II (NIR-II) fluorescence imaging: The excitation wavelength was 785 nm, and the fluorescence signal was collected at 1000-1700 nm, showing that the fluorescence intensity (FI) of the calcified focus of the left anterior descending artery was 2200 cps (background FI = 300 cps, signal-to-noise ratio SNR = 7.3:1), indicating high expression of integrin αvβ3.

[0063] Magnetic resonance T2-weighted imaging (MRIT2WI): Using a gradient echo sequence (TE = 20 ms, TR = 300 ms), the T2 relaxation time of the calcification foci was measured to be 16 ms (normal blood vessel T2 = 48 ms), and the iron deposition was calculated to be 80 μg / g tissue (indicating active macrophage infiltration).

[0064] Non-invasive biochemical detection: fluorescence lifetime imaging of skin AGEs Detection parameters: 375 nm excitation light, 80 μm penetration depth, fluorescence lifetime imaging (FLIM) module to collect dermal layer signals.

[0065] Data analysis: The double exponential fitting of fluorescence lifetimes yielded τ1=2.8ns, τ2=4.9ns, and the calculated fluorescence lifetime change rate Δτ / Δt=0.7ns / h (>0.5ns / h threshold, indicating accelerated glycosylation damage).

[0066] 3. Biomechanical Parameter Collection Vascular shear wave detection unit Ultrasound parameters: multi-frequency focused transducer array (center frequency 5 MHz), emitting shear waves and collecting endothelial frequency shift signals.

[0067] Signal processing: Short-time Fourier transform (STFT, 0.2s window length) was performed on the frequency-shift data. The energy peak in the 18Hz frequency band increased significantly (Δf = 12Hz). The elastic modulus of the calcification focus was inverted based on the elastic mechanics equation to obtain E = 150kPa (>100kPa indicates hard calcification).

[0068] 2. Intelligent Image Detection System: Data Analysis and Processing 1. Cross-modal data processing module: dual-stream Transformer feature fusion Time domain stream processing (AGEs fluorescence lifetime time series data) Input: τ time series data collected continuously for 2 hours (sampling interval is 10 minutes), and a 30-minute sliding window (T=30min) is constructed.

[0069] Feature extraction: one-dimensional convolutional layer ( )extract The upward trend of the time domain feature vector (dimension=32).

[0070] Frequency domain stream processing (shear wave frequency shift STFT spectrum) Input: 18Hz frequency band energy share = 28% (15% increase compared to the baseline), 2D convolution layer extracts spatial frequency distribution features and outputs frequency domain feature vector (dimension=32).

[0071] Hybrid Gated Attention Mechanism Dimensional normalization: Perform Z-score standardization.

[0072] Weight calculation: by optimizing , the time-domain-frequency domain interaction weight matrix is ​​obtained, in which the correlation weight between the AGEs change rate and the 18 Hz frequency shift is 0.82 ( Figure 2 ).

[0073] 3. Prediction results of intelligent image detection system 1. Quantification of Calcification Progression Risk Four-dimensional phase field model output: 12-month predicted value: The volume of calcified plaque increased from 18 mm³ to 39.6 mm³, with a volume change rate of ΔV / Δt = 1.8 mm³ / month (higher than the high-risk threshold of 1.5 mm³ / month).

[0074] Spatial evolution characteristics: The plaque mainly grows toward the vascular lumen (dominated by positive remodeling), which is expected to cause the degree of lumen stenosis to progress from 20%-30% to 45%-55%, and the volume of calcium nodules in the plaque shoulder (vulnerable area) increases by 60%.

[0075] Risk level determination: Based on the comprehensive ΔV / Δt, AGEs fluorescence lifetime (τ=4.9ns), and elastic modulus (E=150kPa), the system automatically marks the patient as having a high risk of calcification progression (risk score 92 / 100).

[0076] 2. Preliminary evaluation of intervention effects (reinforcement learning simulation) TD3 algorithm simulation path: Baseline prediction: If the current treatment (metformin 1000 mg / d) is maintained, HbA1c = 8.1%, PWV = 11.5 m / s, and NIR-II fluorescence intensity FI = 2500 cps after 12 months (indicating continued progression of calcification).

[0077] Post-intervention prediction: Initiation of metformin 1500 mg / d + strontium ranelate 2 g / d treatment is expected to result in HbA1c = 7.2%, PWV = 9.8 m / s, FI = 1800 cps (probe enrichment decreased by 27%), and ΔV / Δt decreased to 0.9 mm³ / month (medium risk) after 6 months.

[0078] IV. Implementation and Dynamic Adjustment of Closed-Loop Intervention Measures 1. Adjustment of detection protocol Adaptive scheduling unit trigger logic: Because the risk level was high, the system automatically shortened the imaging detection cycle from 6 months to 1 month, and the first expedited detection was completed within 3 days.

[0079] Expedited testing content: Repeated skin AGEs testing (monitoring changes in Δτ / Δt) and vascular shear wave testing (assessing dynamic changes in E), and invasive IVUS / OCT will not be repeated for the time being.

[0080] 2. Refined implementation of drug treatment plans Metformin dose titration: Week 1: The dose was increased from 1000 mg / d to 1500 mg / d (taken in three divided doses after meals), and blood glucose was monitored to avoid hypoglycemia (target fasting blood glucose 4.4-7.0 mmol / L, <10.0 mmol / L 2 hours after meal).

[0081] Week 2: Blood sugar levels did not meet the standard (8.2 mmol / L fasting, 11.5 mmol / L postprandial), and the system recommended the addition of a GLP-1 receptor agonist (semaglutide 0.5 mg / week subcutaneous injection).

[0082] Strontium Ranelate Start-up and Optimization: Week 1-2: 1 g / d (orally divided into two doses), and renal function test (estimated glomerular filtration rate eGFR = 85 ml / min / 1.73 m², meeting the medication conditions).

[0083] Week 3: Shear wave elastic modulus E = 140 kPa (6.7% decrease from baseline). The system determined that the therapeutic effect was effective and the dose was maintained. If E did not change, the dose was increased to 1.5 g / d.

[0084] 3. Digital management of lifestyle interventions Wearable device linkage: Patients wear a wristband with integrated CGM and pressure sensor to upload blood sugar (every 5 minutes), pulse rate, and exercise intensity data to the system cloud in real time.

[0085] Smart reminder rules: When blood sugar levels are >11.1mmol / L for three consecutive times after meals, a dietary adjustment reminder will be pushed (e.g., reducing staple food by 50g and increasing vegetable intake); If daily exercise is less than 30 minutes, personalized exercise plans (such as brisk walking + resistance training combination) will be pushed.

[0086] 4. Efficacy evaluation and solution iteration First follow-up after 1 month: Multimodal data update: The fluorescence lifetime change rate of AGEs was Δτ / Δt = 0.55 ns / h (a decrease of 21% compared with the baseline), indicating that the rate of glycation damage slowed down; The shear wave elastic modulus E = 135 kPa (decreased by 10%), and the NIR-II fluorescence intensity FI = 2000 cps (decreased by 9%).

[0087] Dynamic modeling update: The four-dimensional phase field model recalculates ΔV / Δt = 1.2 mm³ / month (has the risk level been adjusted from medium risk to low risk? No, the original high risk prediction was 1.8, and the current 1.2 is still medium risk).

[0088] System Decision: Maintain metformin 1500 mg / d + semaglutide 0.5 mg / week, and do not adjust the strontium ranelate dose for the time being; The inspection cycle is maintained at once every 3 months (the risk level is adjusted from medium risk to medium risk? The original high risk → current medium risk, according to the rules, the medium risk is once every 3 months).

[0089] Second assessment after 3 months: Changes in key indicators: HbA1c=7.0%, PWV=10.2m / s, plaque volume ΔV=3.5mm³ (ΔV / Δt=1.17mm³ / month, close to the low-risk threshold).

[0090] Intervention upgrade recommendation: If PWV does not drop to <9 m / s, the system automatically recommends adding statins (such as atorvastatin 20 mg / d). The cross-modal model predicts that statins can prolong T2* relaxation time by 5 ms (reducing iron deposition by 15%).

[0091] Throughout this specification, references to terms such as "one embodiment," "example," or "specific example" indicate that the specific features, structures, materials, or characteristics described in conjunction with that embodiment or example are included in at least one embodiment or example of the present invention. In this specification, schematic representations of these terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in any one or more embodiments or examples.

[0092] The preferred embodiments of the present invention disclosed above are intended only to help illustrate the present invention. These preferred embodiments do not exhaustively describe all details, nor do they limit the present invention to the specific embodiments described. Obviously, many modifications and variations are possible based on the content of this specification. These embodiments are selected and described in detail in this specification to better explain the principles and practical applications of the present invention, thereby enabling those skilled in the art to better understand and utilize the present invention. The present invention is limited only by the claims and their full scope and equivalents.

Claims

1. Intelligent imaging detection system for dynamic monitoring of diabetic vascular calcification, characterized by: include Multimodal image acquisition module, used to obtain images of vascular structure, chemical composition and biomechanical parameters; The cross-modal data processing module realizes spatiotemporal registration and feature extraction of multi-source data based on a hybrid gated attention mechanism, and integrates temporal biochemical features with frequency-domain mechanical features through a dual-stream architecture. The dynamic modeling and intelligent prediction module builds a four-dimensional calcification dynamics model based on fused data and uses a cross-modal graph neural network to output calcification progression prediction results; The closed-loop intervention module dynamically adjusts the detection protocol and treatment strategy based on the model prediction results and generates personalized detection plans and treatment recommendations.

2. The intelligent imaging detection system for dynamic monitoring of diabetic vascular calcification according to claim 1, characterized in that: The multimodal image acquisition module includes: Structural imaging unit, integrating coronary CT angiography, optical coherence tomography, and intravascular ultrasound to obtain the three-dimensional anatomical structure of blood vessels and the distribution of plaque components; The molecular targeted imaging unit, through intravenous injection of RGD peptide-modified superparamagnetic iron oxide nanoprobes, synchronously collects near-infrared zone II fluorescence signals and magnetic resonance T2* sequence images to achieve receptor-specific targeted imaging of calcification foci and quantify the amount of iron deposition based on signal intensity.

3. The intelligent imaging detection system for dynamic monitoring of diabetic vascular calcification according to claim 2, characterized in that: The multimodal image acquisition module also includes a non-invasive detection subsystem: The skin glycation end products detection unit uses excitation light to penetrate the epidermis and captures the fluorescence lifetime distribution of AGEs in the dermis through the fluorescence lifetime imaging module to obtain the fluorescence lifetime change rate of epidermal AGEs. The vascular shear wave detection unit uses a multi-frequency focused ultrasound transducer array to collect the shear wave frequency shift of the vascular endothelium, and combines it with the differential equation of elastic mechanics to infer the elastic modulus of the calcification focus.

4. The intelligent imaging detection system for dynamic monitoring of diabetic vascular calcification according to claim 3, characterized in that: The cross-modal dynamic fusion module adopts a dual-stream Transformer architecture, including: Time domain flow, processing AGEs fluorescence lifetime time series data, with a window length of 30-120 minutes; Frequency domain flow, analyzing the short-time Fourier transform spectrum of shear wave frequency shift in the frequency range of 5-30Hz; A hybrid gated attention layer addresses the dimensionality difference between biochemical and mechanical parameters through a learnable weight matrix.

5. The intelligent imaging detection system for dynamic monitoring of diabetic vascular calcification according to claim 4, characterized in that: The weight calculation of the hybrid gated attention layer satisfies The query vector Q comes from the time domain features, the bond vector K comes from the frequency domain features, and the entangled state parameter ψ is optimized by the exponential decay function: λ∈[0.05,0.15],ω∈[2π×0.1,2π×1]Hz.

6. The intelligent imaging detection system for dynamic monitoring of diabetic vascular calcification according to claim 1, characterized in that: The dynamic modeling and intelligent prediction module includes the four-dimensional calcification phase field equation: Where, μ is the interfacial energy coefficient, μ∈[0.12,0.18], The calcium deposition rate β∈[1.2×10⁻³,1.8×10⁻³], the inhibition factor γ∈[0.07,0.09], and H(AGEs) is a step function driven by the AGEs concentration.

7. The intelligent imaging detection system for dynamic monitoring of diabetic vascular calcification according to claim 1, characterized in that: The dynamic modeling and intelligent prediction module adopts a double-delayed deep deterministic policy gradient algorithm, takes glycated hemoglobin and pulse wave velocity as core reward functions, and outputs a dynamic dosing regimen for metformin and strontium ranelate.

8. The intelligent imaging detection system for dynamic monitoring of diabetic vascular calcification according to claim 5, characterized in that: The closed-loop intervention system includes: Wearable monitoring unit to collect blood glucose levels and vascular mechanics parameters in real time; Adaptive scheduling unit automatically adjusts the imaging detection cycle according to the calcification risk level; The drug optimization unit dynamically adjusts the dosage ratio of hypoglycemic drugs and anti-calcification drugs based on reinforcement learning algorithms.

9. The intelligent imaging detection system for dynamic monitoring of diabetic vascular calcification according to claim 8, characterized in that: The multimodal image acquisition module, cross-modal data processing module, dynamic modeling and intelligent prediction module, and closed-loop intervention module work together through the following process: When the wearable monitoring unit detects blood glucose ≥11.1 mmol / L or the AGEs fluorescence lifetime change rate ≥0.5 ns / h, the molecular targeted imaging program is triggered; The vascular shear wave detection unit synchronously collects circumferential strain and radial pressure data, which are input into the cross-modal feature fusion module together with the image data; The dynamic modeling and prediction module outputs the volume change rate and risk level of calcified plaques, and the closed-loop intervention module generates detection frequency adjustment instructions and individualized treatment recommendations based on this.

Images