A dual-energy X-ray-based spinal osteoporosis grading diagnosis device and diagnosis method

Abstract

The application discloses a kind of based on dual-energy X-ray spinal osteoporosis grading diagnostic device and diagnostic method, dual-energy X-ray emission module: contain vertical and horizontal two groups of X-ray source (model: M200 microfocus X-ray tube, focal size ≤0.1mm), vertical group is set in the top central of scanning bed (120cm from bed surface), horizontal group is symmetrically set in the scanning bed both sides rack (80cm from bed surface, interval 60cm);The application not only solves the core defects of prior art in diagnostic accuracy, safety, practicality, but also expands the application scenario through portability, intelligent design, provides a new technical scheme for the precise grading diagnosis and standardized treatment of spinal osteoporosis, with significant clinical value and market prospect.

Description

Technical Field

[0001] This invention relates to the field of spinal osteoporosis technology, and in particular to a dual-energy X-ray diffraction (DXD)-based diagnostic device and method for grading spinal osteoporosis. Background Technology

[0002] Osteoporosis is a metabolic bone disease characterized by decreased bone mass and destruction of bone microstructure. The spine is the most commonly affected site, accounting for approximately 70% of osteoporotic fractures. Currently, dual-energy X-ray absorptiometry (DXA) is the core technique for clinical diagnosis of spinal osteoporosis, but existing techniques have the following key limitations:

[0003] (i) The diagnostic dimensions are too limited and the assessment accuracy is insufficient.

[0004] Current DXA devices rely solely on bone mineral density (BMD) as the core diagnostic indicator, using a single threshold of "T-score ≤ -2.5" to determine osteoporosis, neglecting the influence of bone microstructure (such as cortical bone porosity and trabecular morphology) on bone strength. Studies have shown that approximately 20% of patients, although having BMD within the "osteoporosis" range, have an actual fracture risk at osteoporotic levels due to trabecular thinning and cortical bone porosity. Furthermore, current devices cannot simultaneously assess vertebral compression fractures, requiring subjective interpretation by physicians, resulting in a misdiagnosis rate as high as 35% (especially for mild compression fractures), leading to a disconnect between the grading results and actual clinical risk.

[0005] (ii) Fixed scanning mode leads to a prominent contradiction between radiation dose and image quality.

[0006] Current DXA devices mostly use a single vertical scanning direction, and the X-ray beam cannot match the physiological curvature of the spine (Cobb angle 20-40°), making it prone to artifacts due to overlapping of transverse processes and soft tissues. The bone density calculation error can reach ±8%. To reduce artifacts, some devices improve image quality by increasing the number of scans, but the radiation dose per scan is generally 3-5 μSv. Children are more sensitive to radiation, limiting its clinical application. Furthermore, the lack of a dynamic calibration mechanism means that dose drift after 2 hours of operation can increase the error to ±10%, further affecting diagnostic accuracy. (iii) Weak data integration capabilities and insufficient clinical applicability.

[0007] Current devices only output BMD values ​​and cannot correlate with bone metabolism markers such as serum P1NP and CTX, making it difficult to distinguish between different pathological types such as "insufficient bone formation" and "hyperbone resorption," resulting in a lack of targeted treatment plans. Furthermore, the lack of follow-up tracking functionality prevents dynamic monitoring of bone mass changes, requiring physicians to manually compare reports, which is inefficient. In addition, current grading methods do not integrate the FRAX® fracture risk assessment model and cannot directly output 10-year fracture risk values, requiring additional calculation through third-party software, increasing the complexity of clinical operations.

[0008] (iv) Poor equipment compatibility and limited application scenarios

[0009] Existing DXA equipment generally exceeds 0.5m³ in volume and weighs over 100kg, requiring fixed installation in a dedicated radiology room, which cannot meet the immediate diagnostic needs of scenarios such as orthopedic wards and community hospitals; moreover, the protective structure mostly uses a single lead shield with a protective equivalent of only 1mmPb, requiring additional lead screens, resulting in low operational flexibility and a high risk of radiation exposure for medical staff when operating near the table.

[0010] In summary, existing dual-energy X-ray diffraction (DXD)-based diagnostic techniques for spinal osteoporosis have significant shortcomings in terms of diagnostic dimensions, scanning safety, data integration capabilities, and equipment compatibility. There is an urgent need for a multi-dimensional, low-dose, and highly integrated diagnostic device and method to improve diagnostic accuracy and clinical applicability. Summary of the Invention

[0011] In order to overcome the shortcomings of the prior art, one of the objectives of this invention is to provide a dual-energy X-ray-based spinal osteoporosis grading diagnostic device and diagnostic method.

[0012] One of the objectives of this invention is achieved through the following technical solution:

[0013] A dual-energy X-ray diffraction (DXD)-based diagnostic device for grading spinal osteoporosis includes:

[0014] Dual-energy X-ray emission module: Includes two sets of X-ray sources, one vertical and one horizontal (model: M200 microfocus X-ray tube, focal spot size ≤0.1mm). The vertical set is located at the center of the top of the scanning bed (120cm from the bed surface), and the horizontal sets are symmetrically located on both sides of the scanning bed frame (80cm from the bed surface, 60cm apart). Both sets are equipped with a 50kHz high-frequency high-voltage generator, which can simultaneously emit dual-energy X-ray beams: high energy 60-80kV (corresponding current 0.3-0.5mA) and low energy 40-60kV (corresponding current 0.5-0.8mA). The angle between the horizontal X-ray beam and the bed surface is adjusted by a stepper motor. The adjustment trigger condition is: based on the pre-scanned spinal positioning image, calculate the physiological curvature (Cobb angle) of the L1-L4 segments, and then adaptively adjust according to the formula θ=0.3×Cobb angle (°) (range 0-30°).

[0015] Multi-dimensional detection module: Composed of vertical and horizontal detector arrays. The vertical array (30cm×40cm) is located at the bottom of the scanning bed, and the horizontal array (20cm×30cm per group) is coaxially set to correspond to the horizontal X-ray source. The detector adopts an amorphous silicon dynamic digital flat panel detector (model: DP500), with a resolution of 1024×1024 pixels (pixel size 100μm), a detection efficiency of ≥75% (30% higher than the traditional CsI detector, tested according to GB / T35410-2023 standard), and a frame rate of ≥30fps, used to synchronously acquire dual-energy X-ray attenuation data passing through the spine (data format: 16-bit RAW format, grayscale range 0-65535).

[0016] Intelligent Data Processing Module: Built-in DXAExpert 3.0 upgraded data processing platform (based on Intel i9 processor + NVIDIA RTX4090 GPU), integrating:

[0017] Image reconstruction unit: An improved ordered subset expectation-maximization algorithm (OS-EM) is used to generate three-dimensional tomographic images of the spine based on dual-energy attenuation data. The number of iterations is 12-16, the voxel resolution is 0.1mm×0.1mm×0.1mm, and the slice thickness is 0.5mm.

[0018] Parameter extraction unit: Extracts parameters using the following algorithm:

[0019] Bone mineral density (BMD): Calculated using dual-energy absorptiometry (DXA) with the formula BMD = (high-energy attenuation value × 0.65 + low-energy attenuation value × 0.35) / vertebral body projection area, in g / cm².

[0020] Cortical bone porosity (Ct.Po): Porous regions were identified using a threshold segmentation method (threshold range 1500-2000 HU), and the calculation formula was Ct.Po = pore volume / total cortical bone volume × 100%;

[0021] Trabecular bone number (Tb.N) and separation (Tb.Sp): Trabecular bone structure was extracted based on a skeletonization algorithm. Tb.N is in mm. -1 Tb.Sp is in mm;

[0022] Biomarker input unit: Supports importing LIS system data via USB interface or manual entry, including serum P1NP (detection range 10-500 ng / mL), CTX (detection range 0.01-5 ng / mL) and 25-hydroxyvitamin D (detection range 5-100 ng / mL) bone metabolism biomarkers data;

[0023] Dynamic calibration module: Includes a standard bone mineral density phantom (material: hydroxyapatite, density 0.2-1.8 g / cm², containing simulated L1-L4 vertebral structures) and a real-time correction algorithm. Calibration is automatically initiated after every 10 scans or after 2 hours of continuous operation of the X-ray source: First, phantom attenuation data is collected, and then dose drift is corrected through a linear regression algorithm (formula: Corrected_Value=Raw_Value×K+B, where K is the slope 0.98-1.02, and B is the intercept -0.02-0.02). The calibration accuracy is ≤±0.5% (based on WHO standard phantom).

[0024] Grading output module: Equipped with a 19-inch touch screen (1920×1080 resolution) and an HL7 standard cloud transmission interface, it outputs grading reports with 3D image annotations. The reports include: basic information, BMDT / Z values, bone microstructure parameters, bone metabolism marker levels, grading results, and FRAX® risk values. It supports integration with HIS / PACS systems (compatible with DICOM 3.0 standard).

[0025] As a further improvement to the above technical solution:

[0026] The dual-energy X-ray emission module adopts burst imaging technology (exposure sequence: high energy and low energy interval ≤1ms, synchronous trigger accuracy ±0.1ms), dual-energy exposure time ≤5ms, single scan radiation dose: adults ≤1μSv, children ≤0.5μSv (compliant with ALARA principles and GBZ130-2020 standards); the child-specific mode has three preset parameter levels: 2-6 years old (low energy 40kV / 0.3mA, high energy 60kV / 0.2mA), 7-12 years old (low energy 45kV / 0.4mA, high energy 65kV / 0.3mA), 13-18 years old (low energy 50kV / 0.5mA, high energy 70kV / 0.4mA), and is linked to the automatic raising and lowering of the lead protective skirt (protection equivalent 0.5mmPb).

[0027] The intelligent data processing module also includes an artifact removal unit, which performs the following steps:

[0028] The original image was processed using adaptive histogram equalization (CLAHE) with a contrast gain coefficient of 2.0-3.0.

[0029] The Otsu thresholding method (threshold 1000HU) was used to distinguish between bone and soft tissue regions, and morphological operations (erosion kernel size 3×3 pixels, expansion kernel size 5×5 pixels) were used to identify overlapping transverse processes.

[0030] Using the average pixel value of the soft tissue region (200-400HU) as a benchmark, Gaussian filtering (standard deviation 1.5) was used to correct the pixel value of the spinal region. The correction formula is: Corrected_Pixel=Raw_Pixel-(Overlap_Pixel-Softtissue_Mean)×0.7. After correction, the bone mineral density calculation error was reduced by ≥15% (compared to before correction, HR-pQCT detection value is the gold standard).

[0031] The device measures 80cm×60cm×30cm (volume <0.144m³) and weighs ≤50kg. It is equipped with a trolley with braked universal wheels (wheel diameter 10cm). The protective structure uses a lead alloy + tungsten alloy composite shield: the outer shell has a protection equivalent of 2mmPb, and the radiation source chamber has a protection equivalent of 3mmPb, which complies with GBZ / T180-2020 standards. The operating modes are: near stage (touch screen distance from the scanning bed ≤1.5m) / far stage (connected to the workstation via wired / wireless (WiFi6) connection, control distance ≥5m). The far stage mode supports real-time image transmission (delay ≤200ms).

[0032] A dual-energy X-ray diffraction (DXD)-based method for grading and diagnosing spinal osteoporosis, applied to the device described in any one of claims 1-4, comprises the following steps:

[0033] Scanning preparation:

[0034] Patient positioning: A laser crosshair locator (positioning accuracy ±1mm) is used to align with the midline of the spine, and a high-definition camera (2 million pixels resolution) is used to confirm that the L1-L4 segments are located in the scanning area;

[0035] Parameter settings: Based on patient information, the parameter library is called: Adults (weight < 50 kg: low energy 45kV / 0.6mA, high energy 70kV / 0.4mA; weight 50-80 kg: low energy 50kV / 0.7mA, high energy 75kV / 0.5mA; weight > 80 kg: low energy 55kV / 0.8mA, high energy 80kV / 0.6mA); children use the corresponding age-appropriate parameter settings.

[0036] Angle adjustment: The pre-scan obtains a lateral localization image of the spine, calculates the L1-L4 Cobb angle (normal range 20-40°), and adjusts the horizontal X-ray beam angle according to θ=0.3×Cobb angle;

[0037] Multi-source scanning: The dual-energy X-ray emission module is activated to simultaneously acquire dual-energy attenuation data for segments L1-L4 (scanning range: extending 2cm from the upper and lower edges of the vertebral body). The dynamic calibration module performs real-time calibration once for each vertebral segment scanned, and the calibration data is fed back to the detection module in real time; Parameter analysis:

[0038] Image reconstruction: The OS-EM algorithm iterates 14 times to generate a three-dimensional tomographic image and automatically removes blurry layers with a signal-to-noise ratio of <3;

[0039] Parameter calculation: BMD, Ct.Po, Tb.N, and Tb.Sp are extracted according to the algorithm described in claim 1. Each parameter is calculated three times and the average value is taken. The error is controlled within ±1% (verified with standard Phantom).

[0040] Fracture assessment: Measure the anterior vertebral body height (H1) and central height (H2), calculate the ratio R = H1 / H2, and mark the fracture as a compression fracture when R < 0.8. Record the fracture segment and degree of compression (mild R = 0.7-0.8, moderate R = 0.6-0.7, severe R < 0.6).

[0041] Data fusion:

[0042] Data preprocessing: parameter standardization (BMD converted to T values, Ct.Po, Tb.N, and Tb.Sp converted to Z values, and markers normalized to reference ranges);

[0043] Model Import: A ResNet-50 deep learning model is used, with a dual input layer: an image feature channel (extracting 1024-dimensional convolutional features from 3D images) and a numerical feature channel (fusing 4 core parameters + 3 biomarker data). The model is trained with 5200 clinical samples (sample composition: 1700 normal cases, 1800 cases of osteopenia, 1700 cases of osteoporosis, including 800 cases of HR-pQCT validation data). The training set:validation set:test set ratio is 7:2:1. The cross-entropy function is used as the loss function. The model is trained for 200 epochs, achieving a test set accuracy of ≥92%. Hierarchical Output:

[0044] Hierarchical determination: Based on the fusion results and the following standard output:

[0045] Normal: BMDT value ≥ -1.0, Ct.Po < 15%, Tb.Sp < 0.2mm, no compression fracture;

[0046] Osteopenia: -2.5 < T value < -1.0, 15% ≤ Ct.Po < 20%, 0.2mm ≤ Tb.Sp < 0.3mm, no compression fracture or mild fracture;

[0047] Osteoporosis: T value ≤ -2.5, Ct.Po ≥ 20%, Tb.Sp ≥ 0.3 mm, accompanied by moderate or greater compression fractures or multiple minor fractures;

[0048] Risk assessment: Automatically import 12 FRAX® parameters (age, sex, height, weight, history of fracture, history of hip fracture in parents, smoking status, history of glucocorticoid use, history of rheumatoid arthritis, history of secondary osteoporosis, history of excessive alcohol consumption, femoral neck BMD value) to generate 10-year hip fracture risk and major osteoporotic fracture risk values. High risk is marked when the risk is ≥3% or ≥20%, and intervention recommendations are given.

[0049] Report generation: Outputs a color report with 3D image annotations, parameter trend charts, and grading criteria within 1 minute, supporting local printing and cloud storage.

[0050] As a further improvement to the above technical solution:

[0051] The deep learning model optimization details are as follows: The Adam optimizer is used (initial learning rate 0.001, decaying by 10% every 50 rounds), batch size 32, and L2 penalty (coefficient 0.001) is used for regularization; spatial features of the input layer image feature channels are extracted through 3×3×3 convolutional kernels, and the numerical feature channels are transformed in dimension through a fully connected layer (64 neurons). The two are then weighted and fused in the fusion layer (128 neurons) through an attention mechanism (image feature weight 0.6, numerical feature weight 0.4). The output layer uses the softmax activation function, and the confusion matrix F1 value is ≥0.91.

[0052] Step 3 parameter analysis also includes bone turnover status assessment: calculate the P1NP / CTX ratio. A ratio >1.5 indicates bone formation-dominant type, a ratio <0.8 indicates bone resorption-dominant type, and a ratio between 0.8 and 1.5 indicates balanced type. This result is included in the grading reference factor: bone loss in patients with bone resorption-dominant type can be upgraded to early warning of osteoporosis.

[0053] The method also includes a follow-up tracking module, which performs the following steps:

[0054] Automatically associates patient ID (supports ID card / medical insurance card recognition and input), stores scan data from each scan (including original images, parameter values, and grading results), and saves data in a format that conforms to the DICOM 3.0 standard;

[0055] Generate trend analysis charts: plot the change curves of BMDT value, Ct.Po, Tb.N, and Tb.Sp with time as the horizontal axis, and mark the normal reference range threshold line;

[0056] Treatment assessment: Calculate the rate of change of parameters between two consecutive scans (e.g., BMD change rate = (current BMD - previous BMD) / previous BMD × 100%). A change rate ≥ 3% is considered effective treatment, -3% ≤ change rate < 3% is considered stable treatment, and a change rate < -3% is considered progression treatment.

[0057] Warning prompt: When the parameter change rate reaches the progression threshold or the grade is upgraded, an automatic reminder will be pushed to the clinical workstation.

[0058] The grading output module is also equipped with a voice broadcast unit (supporting Chinese / English bilingual), which can broadcast the core information of the grading results (grading type, fracture risk level, abnormal values ​​of key parameters), and has a built-in patient education knowledge base (including osteoporosis prevention, dietary advice, and exercise guidance), and can print popular science manuals as needed.

[0059] Compared with the prior art, the beneficial effects of the present invention are as follows:

[0060] (i) Breaking through the limitations of a single dimension, diagnostic accuracy is significantly improved.

[0061] Multi-parameter collaborative assessment: This invention is the first to fuse four-dimensional data: bone mineral density (BMD), bone microstructure (Ct.Po, Tb.N, Tb.Sp), bone metabolism markers, and vertebral fracture status. The data is graded using a ResNet-50 deep learning model, achieving an accuracy of ≥92% on the test set, which is 17 percentage points higher than the existing DXA device (which relies solely on BMD and has an accuracy of approximately 75%). Among these features, the early warning function for osteoporosis in patients with bone resorption-dominant osteoporosis can identify high-risk individuals 6-12 months in advance, avoiding missed diagnoses.

[0062] Precise artifact removal: By combining CLAHE histogram equalization, Otsu threshold segmentation and Gaussian filtering algorithms, the cross process overlap artifacts are corrected, reducing the bone density calculation error by ≥15% (from ±8% to below ±6.8%). Furthermore, blurry layers with a signal-to-noise ratio <3 are automatically removed, ensuring the accuracy of parameter extraction (error ≤ ±1%) and providing reliable data support for grading.

[0063] (ii) Optimize scanning security and expand the applicable population.

[0064] Ultra-low radiation dose: Using burst imaging technology (dual-energy exposure time ≤5ms) and a child-specific parameter setting, the radiation dose per scan for adults is ≤1μSv and for children is ≤0.5μSv, which is 60%-90% lower than existing devices (3-5μSv). It fully complies with ALARA radiation protection principles and can be safely used for long-term follow-up of sensitive populations such as children and pregnant women (non-pregnant).

[0065] Dynamic real-time calibration: Standard phantom calibration is automatically initiated every 10 scans or 2 hours. Dose drift is corrected through a linear regression algorithm, with a calibration accuracy of ≤±0.5%. This ensures the stability of parameters during continuous operation of the equipment and avoids diagnostic deviations caused by aging of the radiation source.

[0066] (III) Enhance clinical applicability and improve diagnostic and treatment efficiency

[0067] One-stop data integration: Built-in LIS system interface can automatically import bone metabolism marker data and simultaneously integrate the FRAX® risk assessment model. It can output a color report with 3D image annotation, parameter trend graph and intervention suggestions within 1 minute. No third-party software assistance is required. The diagnosis process time is reduced from the current 30 minutes to 5 minutes, and the efficiency is improved by 83%.

[0068] Intelligent follow-up tracking: Automatically associates patient IDs and stores historical data, generates change curves for parameters such as BMD and Ct.Po, quantifies treatment effectiveness through the rate of change (≥3% for effective, <-3% for progress), and automatically pushes alerts when the treatment is graded or when parameters progress, helping doctors to adjust treatment plans in a timely manner and reduce delays in treatment.

[0069] (iv) Innovate equipment design and expand application scenarios

[0070] Portability and full protection: The overall volume is <0.144m³ (80cm×60cm×30cm), and the weight is ≤50kg. It is equipped with a swivel trolley and can be flexibly moved to orthopedic wards and community hospitals. It adopts lead alloy + tungsten alloy composite shielding (shell 2mmPb, radiation source chamber 3mmPb), which increases the protection equivalent by 100% compared with existing devices. It supports near-stage / far-stage dual-mode operation (far-stage delay ≤200ms), and the radiation exposure risk of medical staff is close to zero.

[0071] Adaptive scanning adjustment: The Cobb angle of the spine is calculated by pre-scanning, and the horizontal X-ray beam angle is adaptively adjusted (0-30°) according to θ=0.3×Cobb angle to match the physiological curvature of the spine and avoid missed or repeated scans caused by fixed scanning angle. It is especially suitable for patients with scoliosis and kyphosis (the success rate of scanning such patients with existing devices is only 60%, while this invention improves it to more than 98%).

[0072] (v) Standardization and compatibility to lower the threshold for clinical implementation

[0073] Standardized interface: Supports DICOM3.0 and HL7 standards, enabling seamless integration with hospital HIS / PACS systems without data format conversion, reducing information silos; Voice broadcast unit (Chinese and English bilingual) and patient education knowledge base can improve patients' understanding of their condition and enhance treatment adherence.

[0074] Simplified operation: 19-inch touch screen + laser crosshair positioning (accuracy ±1mm), no professional radiologist is required to operate, orthopedic doctors can get started after simple training, lowering the threshold for use in primary hospitals and helping to promote early screening and hierarchical diagnosis and treatment of osteoporosis.

[0075] In summary, this invention not only solves the core deficiencies of existing technologies in terms of diagnostic accuracy, safety, and practicality, but also expands application scenarios through portable and intelligent design, providing a brand-new technical solution for the accurate grading diagnosis and standardized treatment of spinal osteoporosis, and has significant clinical value and market prospects.

[0076] The above description is merely an overview of the technical solution of the present invention. In order to better understand the technical means of the present invention and to implement it in accordance with the contents of the specification, and to make the above and other objects, features and advantages of the present invention more apparent and understandable, preferred embodiments are described in detail below with reference to the accompanying drawings. Attached Figure Description

[0077] Figure 1 This is a flowchart illustrating the overall structure of this embodiment;

[0078] Figure 2 This is the main flowchart of the diagnostic method in this embodiment.

[0079] Figure 3 This is a flowchart of the intelligent data processing in this embodiment;

[0080] Figure 4 This is a flowchart of the dynamic calibration process in this embodiment;

[0081] Figure 5 This is a flowchart of the deep learning model in this embodiment;

[0082] Figure 6 This is a flowchart of the follow-up tracking process in this embodiment;

[0083] Figure 7 This is a flowchart of the radiation safety control process in this embodiment. Detailed Implementation

[0084] The present invention will now be further described in conjunction with the accompanying drawings and specific embodiments. It should be noted that, without conflict, the various embodiments or technical features described below can be arbitrarily combined to form new embodiments.

[0085] It should be noted that when a component is described as "fixed to" another component, it can be directly on the other component or may have a component in between. When a component is considered "connected to" another component, it can be directly connected to the other component or may have a component in between. When a component is considered "set on" another component, it can be directly set on the other component or may have a component in between. The terms "vertical," "horizontal," "left," "right," and similar expressions used in this document are for illustrative purposes only.

[0086] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention pertains. The terminology used herein in the description of the invention is for the purpose of describing particular embodiments only and is not intended to be limiting of the invention. The term "and / or" as used herein includes any and all combinations of one or more of the associated listed items. I. Specific Implementation Methods

[0088] (a) Physical structure of the device

[0089] The dual-energy X-ray-based spinal osteoporosis grading diagnostic device disclosed in this embodiment has the following specific configuration:

[0090] 1. Dual-energy X-ray emission module: It adopts two sets of M200 microfocus X-ray tubes (focal size 0.08mm). The vertical set is fixed to the top gantry of the scanning bed (120cm from the bed surface, with the central axis coinciding with the center line of the bed surface). The horizontal set is symmetrically installed on both sides of the scanning bed frame (80cm from the bed surface, 60cm spacing, horizontal error ≤±2mm). Both sets are equipped with a 50kHz high-frequency high-voltage generator (model: HG-5000), supporting high-energy 60-80kV (0.3-0.5mA) and low-energy 40-60kV (0.5-0.8mA) synchronous output. The beam angle adjustment of the horizontal set is driven by a stepper motor (model: 42HS40-1684), with an adjustment accuracy of ±0.1°. The trigger signal source is the Cobb angle calculation result of the pre-scanned image (automatically measured by ImageJ software, error ≤±1°).

[0091] 2. Multi-dimensional detection module: The vertical detector array is a DP500 amorphous silicon flat plate (30cm×40cm, pixel size 100μm, response time <1ms, quantum detection efficiency 78%), which is mounted on the lead plate (2mm thick Pb) at the bottom of the scanning bed; the horizontal detector array is 20cm×30cm in size and is arranged coaxially with the horizontal X-ray source (coaxiality error ≤±1mm). The detector output is connected to the data processing module through a 16-bit AD converter (sampling rate 100MHz).

[0092] 3. Intelligent Data Processing Module: The hardware utilizes an Intel Core i9-13900K processor (5.8GHz) + NVIDIA RTX 4090 GPU (24GB VRAM) and is equipped with 1TB SSD storage. The software integrates the DXAExpert 3.0 platform. The image reconstruction unit employs an improved OS-EM algorithm (iteration step size 0.05, relaxation factor 0.9). The parameter extraction unit has a built-in threshold database (Ct.Po segmentation threshold 1600HU, soft tissue reference threshold 300HU). The marker input unit supports the HL7FHIR standard data interface (transmission rate 100Mbps).

[0093] 4. Dynamic calibration module: The standard phantom is a cylindrical hydroxyapatite prosthesis (15cm in diameter, 20cm in height, containing simulated L1-L4 vertebrae with density gradients of 0.5, 1.0, and 1.5g / cm²), placed in the calibration area on the right side of the scanning bed (30cm from the scanning center); the calibration algorithm is embedded in the FPGA chip, and the scan-calibration cycle is automatically started after every 10 scans or after the X-ray source has worked for a cumulative 2 hours, with calibration time <30s.

[0094] 5. Tiered Output Module: 19-inch IPS touchscreen (1920×1080 resolution, 5ms response time), equipped with an RJ45 network port (supporting DICOM 3.0 transmission) and a USB 3.0 interface. Built-in voice module (TTS chip, 99% accuracy in Chinese / English pronunciation), patient education knowledge base containing 12 types of popular science documents (PDF format, supports one-click printing).

[0095] (II) Diagnostic Method Implementation Steps

[0096] Taking a 45-year-old female patient (weighing 65kg, chief complaint of "dull pain in the lower back for 3 months") as an example, the specific implementation process is as follows:

[0097] 1. Scanning Preparation

[0098] Patient positioning: Lie prone on the scanning bed (bed surface hardness 70D, adjustable height range 50-80cm), align the laser crosshair locator with the midline of the spine (the line connecting the umbilicus 3cm above the sacral promontory), and use a high-definition camera (2 megapixels, 25fps) to confirm that the L1-L4 segments are located in the scanning area (the upper and lower edges each extend 2cm beyond the scanning area).

[0099] Parameter settings: Select the parameters for weight 50-80kg, low energy 50kV / 0.7mA, high energy 75kV / 0.5mA, exposure time 4ms.

[0100] Angle adjustment: Pre-scan (low dose mode: 30kV / 0.2mA, scan time 1s) to acquire lateral positioning image, ImageJ software calculates L1-L4 Cobb angle 30°, adjust the horizontal beam angle according to θ=0.3×30°=9°, stepper motor drive takes 1.2s.

[0101] 2. Multi-source scanning: Start the dual-energy emission module to synchronously acquire attenuation data (RAW format, 16-bit grayscale, data size 800MB). Dynamic calibration is triggered for each vertebral segment (L1, L2, L3, L4) scanned. After calibration, the data is updated to the detection module in real time. The total scanning time is 8 seconds.

[0102] 3. Parameter Analysis

[0103] Image reconstruction: The OS-EM algorithm iterates 14 times to generate a 3D tomographic image (0.1mm voxel, 0.5mm layer thickness), automatically removes the blurry layers with a signal-to-noise ratio of 2.8 in layers 3 and 12, and retains 24 effective layers.

[0104] Parameter calculations: BMD = (high-energy attenuation value 680 × 0.65 + low-energy attenuation value 1250 × 0.35) / vertebral body projection area 12cm² = 0.82g / cm² (T value - 1.8); Ct.Po = pore volume 0.8cm³ / total cortical bone volume 5cm³ × 100% = 16%; Tb.N = 2.1mm⁻¹, Tb.Sp = 0.25mm; each of the three parameters was calculated 3 times, with an average error of ±0.8%.

[0105] Fracture assessment: The anterior height of the L3 vertebral body was measured as H1=28mm, the central height as H2=36mm, and the radius of curvature was R=28 / 36=0.78, which was marked as a mild compression fracture.

[0106] 4. Data fusion

[0107] Data preprocessing: BMDT value -1.8 (normalized -0.9), Ct.Po 16% (Z value 0.3), Tb.Sp 0.25mm (Z value 0.5), serum P1NP 180ng / mL (normalized 0.4), CTX 0.8ng / mL (normalized 0.6), 25-hydroxyvitamin D 35ng / mL (normalized 0.5). Model inference: The ResNet-50 model inputs 1024-dimensional image features and 7 numerical features, using an attention mechanism for weighted fusion (image weight 0.6, numerical weight 0.4), inference time 0.8s, outputting a bone loss probability of 91%.

[0108] 5. Tiered output: Generate a report within 1 minute, showing "Osteopenia (mild compression fracture)", FRAX® risk value: hip fracture 2.1%, major fracture 12.3%, and push intervention suggestion of "supplementing vitamin D + calcium"; voice broadcast "Diagnosis result: osteoopenia, moderate fracture risk", and support printing reports with 3D annotations and popular science booklets on "Osteoporosis Diet Guide".

[0109] II. Experimental Records

[0110] (I) Experimental Design

[0111] 1. Subjects: 520 patients from the orthopedic outpatient department of a tertiary hospital between January and June 2024 were selected, including 180 males and 340 females, aged 22-85 years (mean 62.3±11.5 years); including 170 normal individuals, 180 with osteopenia, 170 with osteoporosis, 80 patients with scoliosis (Cobb angle 15-45°), and 50 children (6-18 years old). 2. Control Equipment: HologicDiscoveryDXA device (current technology representative, radiation dose 3-5 μSv, measuring only BMD, without dynamic calibration).

[0112] 3. Evaluation indicators: diagnostic accuracy, bone density calculation error, radiation dose, scan success rate, diagnostic process time, and calibration accuracy.

[0113] (II) Experimental Data and Analysis

[0114] 1. Diagnostic accuracy experiment

[0115]

[0116] Note: Accuracy is based on HR-pQCT test results as the gold standard; patients with bone resorption-dominant type are defined as P1NP / CTX < 0.8; mild fracture refers to vertebral body compression rate of 10%-20%.

[0117] 2. Scanning security experiment

[0118]

[0119] Note: Radiation dose was measured using a dosimeter (model: FLUKE451P); calibration error was based on the WHO standard phantom.

[0120] 3. Clinical applicability trials

[0121]

[0122] 4. Equipment compatibility test

[0123]

[0124] The above embodiments are merely preferred embodiments of the present invention and should not be construed as limiting the scope of protection of the present invention. Any non-substantial changes and substitutions made by those skilled in the art based on the present invention shall fall within the scope of protection claimed by the present invention.

Claims

1. A dual-energy X-ray diffraction (DXD)-based diagnostic device for spinal osteoporosis, characterized in that, include: Dual-energy X-ray emission module: Includes two sets of X-ray sources, one vertical and one horizontal (model: M200 microfocus X-ray tube, focal spot size ≤0.1mm). The vertical set is located at the center of the top of the scanning bed (120cm from the bed surface), and the horizontal sets are symmetrically located on both sides of the scanning bed frame (80cm from the bed surface, 60cm apart). Both sets are equipped with a 50kHz high-frequency high-voltage generator, which can simultaneously emit dual-energy X-ray beams: high energy 60-80kV (corresponding current 0.3-0.5mA) and low energy 40-60kV (corresponding current 0.5-0.8mA). The angle between the horizontal X-ray beam and the bed surface is adjusted by a stepper motor. The adjustment trigger condition is: based on the pre-scanned spinal positioning image, calculate the physiological curvature (Cobb angle) of the L1-L4 segments, and then adaptively adjust according to the formula θ=0.3×Cobb angle (°) (range 0-30°). Multi-dimensional detection module: Composed of vertical and horizontal detector arrays. The vertical array (30cm×40cm) is located at the bottom of the scanning bed, and the horizontal array (20cm×30cm per group) is coaxially set to correspond to the horizontal X-ray source. The detector adopts an amorphous silicon dynamic digital flat panel detector (model: DP500), with a resolution of 1024×1024 pixels (pixel size 100μm), a detection efficiency of ≥75% (30% higher than the traditional CsI detector, tested according to GB / T35410-2023 standard), and a frame rate of ≥30fps, used to synchronously acquire dual-energy X-ray attenuation data passing through the spine (data format: 16-bit RAW format, grayscale range 0-65535). Intelligent Data Processing Module: Built-in DXAExpert 3.0 upgraded data processing platform (based on Intel i9 processor + NVIDIA RTX4090 GPU), integrating: Image reconstruction unit: An improved ordered subset expectation-maximization algorithm (OS-EM) is used to generate three-dimensional tomographic images of the spine based on dual-energy attenuation data. The number of iterations is 12-16, the voxel resolution is 0.1mm×0.1mm×0.1mm, and the slice thickness is 0.5mm. Parameter extraction unit: Extracts parameters using the following algorithm: Bone mineral density (BMD): Calculated using dual-energy X-ray absorptiometry (DXA) as BMD = (high-energy attenuation value × 0.65 + low-energy attenuation value × 0.35) / vertebral body projection area, in g / cm²; Cortical bone porosity (Ct.Po): Porous regions were identified using a threshold segmentation method (threshold range 1500-2000 HU), calculated as Ct.Po = pore volume / total cortical bone volume × 100%; Trabecular bone number (Tb.N) and separation (Tb.Sp): Trabecular bone structure was extracted based on a skeletonization algorithm. Tb.N is in mm. -1 Tb.Sp is in mm; Biomarker input unit: Supports importing LIS system data via USB interface or manual entry, including serum P1NP (detection range 10-500 ng / mL), CTX (detection range 0.01-5 ng / mL) and 25-hydroxyvitamin D (detection range 5-100 ng / mL) bone metabolism biomarkers data; Dynamic calibration module: Includes a standard bone mineral density phantom (material: hydroxyapatite, density 0.2-1.8 g / cm², containing simulated L1-L4 vertebral structures) and a real-time correction algorithm. Calibration is automatically initiated after every 10 scans or after 2 hours of continuous operation of the X-ray source: First, phantom attenuation data is collected, and then dose drift is corrected through a linear regression algorithm (formula: Corrected_Value=Raw_Value×K+B, where K is the slope 0.98-1.02, and B is the intercept -0.02-0.02). The calibration accuracy is ≤±0.5% (based on WHO standard phantom). Grading output module: Equipped with a 19-inch touch screen (1920×1080 resolution) and an HL7 standard cloud transmission interface, it outputs grading reports with 3D image annotations. The reports include: basic information, BMDT / Z values, bone microstructure parameters, bone metabolism marker levels, grading results, and FRAX® risk values. It supports integration with HIS / PACS systems (compatible with DICOM 3.0 standard).

2. The apparatus according to claim 1, characterized in that, The dual-energy X-ray emission module adopts burst imaging technology (exposure sequence: high energy and low energy interval ≤1ms, synchronous trigger accuracy ±0.1ms), dual-energy exposure time ≤5ms, single scan radiation dose: adults ≤1μSv, children ≤0.5μSv (compliant with ALARA principles and GBZ130-2020 standards); the child-specific mode has three preset parameter levels: 2-6 years old (low energy 40kV / 0.3mA, high energy 60kV / 0.2mA), 7-12 years old (low energy 45kV / 0.4mA, high energy 65kV / 0.3mA), 13-18 years old (low energy 50kV / 0.5mA, high energy 70kV / 0.4mA), and is linked to the automatic raising and lowering of the lead protective skirt (protection equivalent 0.5mmPb).

3. The apparatus according to claim 1, characterized in that, The intelligent data processing module also includes an artifact removal unit, which performs the following steps: The original image was processed using adaptive histogram equalization (CLAHE) with a contrast gain coefficient of 2.0-3.

0. The Otsu thresholding method (threshold 1000HU) was used to distinguish between bone and soft tissue regions, and morphological operations (erosion kernel size 3×3 pixels, expansion kernel size 5×5 pixels) were used to identify overlapping transverse processes. Using the average pixel value of the soft tissue region (200-400HU) as a benchmark, Gaussian filtering (standard deviation 1.5) was applied to correct the pixel value of the spinal region. The correction formula is: Corrected_Pixel = Raw_Pixel - (Overlap_Pixel - Softtissue_Mean) × 0.

7. The error in bone mineral density calculation was reduced by ≥15% after correction (compared to the uncorrected value, with HR-pQCT test value as the gold standard).

4. The apparatus according to claim 1, characterized in that, The device measures 80cm×60cm×30cm (volume <0.144m³) and weighs ≤50kg. It is equipped with a trolley with braked universal wheels (wheel diameter 10cm). The protective structure uses a lead alloy + tungsten alloy composite shield: the outer shell has a protection equivalent of 2mmPb, and the radiation source chamber has a protection equivalent of 3mmPb, which complies with GBZ / T180-2020 standards. The operating modes are: near stage (touch screen distance from the scanning bed ≤1.5m) / far stage (connected to the workstation via wired / wireless (WiFi6) connection, control distance ≥5m). The far stage mode supports real-time image transmission (delay ≤200ms).

5. A method for grading and diagnosing spinal osteoporosis based on dual-energy X-ray diffraction, applied to the device described in any one of claims 1-4, characterized in that, Includes the following steps: Scanning preparation: Patient positioning: A laser crosshair locator (positioning accuracy ±1mm) is used to align with the midline of the spine, and a high-definition camera (2 million pixels resolution) is used to confirm that the L1-L4 segments are located in the scanning area; Parameter settings: Based on patient information, the parameter library is called: Adults (weight < 50 kg: low energy 45kV / 0.6mA, high energy 70kV / 0.4mA; weight 50-80 kg: low energy 50kV / 0.7mA, high energy 75kV / 0.5mA; weight > 80 kg: low energy 55kV / 0.8mA, high energy 80kV / 0.6mA); children use the corresponding age-appropriate parameter settings. Angle adjustment: The pre-scan obtains a lateral localization image of the spine, calculates the L1-L4 Cobb angle (normal range 20-40°), and adjusts the horizontal X-ray beam angle according to θ=0.3×Cobb angle; Multi-source scanning: The dual-energy X-ray emission module is activated to simultaneously acquire dual-energy attenuation data of L1-L4 segments (scanning range: extending 2cm from the upper and lower edges of the vertebral body). The dynamic calibration module performs real-time calibration once for each vertebral segment scanned, and the calibration data is fed back to the detection module in real time. Parameter analysis: Image reconstruction: The OS-EM algorithm iterates 14 times to generate a three-dimensional tomographic image and automatically removes blurry layers with a signal-to-noise ratio of <3; Parameter calculation: BMD, Ct.Po, Tb.N, and Tb.Sp are extracted according to the algorithm described in claim 1. Each parameter is calculated three times and the average value is taken. The error is controlled within ±1% (verified with standard Phantom). Fracture assessment: Measure the anterior vertebral body height (H1) and central height (H2), calculate the ratio R = H1 / H2, and mark the fracture as a compression fracture when R < 0.

8. Record the fracture segment and degree of compression (mild R = 0.7-0.8, moderate R = 0.6-0.7, severe R < 0.6). Data fusion: Data preprocessing: parameter standardization (BMD converted to T values, Ct.Po, Tb.N, and Tb.Sp converted to Z values, and markers normalized to reference ranges); Model Import: A ResNet-50 deep learning model was adopted, with a dual-channel input layer: an image feature channel (extracting 1024-dimensional convolutional features from 3D images) and a numerical feature channel (fusing 4 core parameters + 3 biomarker data). The model was trained with 5200 clinical samples (sample composition: 1700 normal cases, 1800 cases of osteopenia, 1700 cases of osteoporosis, including 800 cases of HR-pQCT validation data). The training set:validation set:test set ratio was 7:2:1, the cross-entropy function was used as the loss function, and the model was trained iteratively for 200 epochs. The accuracy on the test set was ≥92%. Hierarchical output: Hierarchical determination: Based on the fusion results and the following standard output: Normal: BMDT value ≥ -1.0, Ct.Po < 15%, Tb.Sp < 0.2mm, no compression fracture; Osteopenia: -2.5 < T value < -1.0, 15% ≤ Ct.Po < 20%, 0.2mm ≤ Tb.Sp < 0.3mm, no compression fracture or mild fracture; Osteoporosis: T value ≤ -2.5, Ct.Po ≥ 20%, Tb.Sp ≥ 0.3 mm, accompanied by moderate or greater compression fractures or multiple minor fractures; Risk assessment: Automatically import 12 FRAX® parameters (age, sex, height, weight, history of fracture, history of hip fracture in parents, smoking status, history of glucocorticoid use, history of rheumatoid arthritis, history of secondary osteoporosis, history of excessive alcohol consumption, femoral neck BMD value) to generate 10-year hip fracture risk and major osteoporotic fracture risk values. High risk is marked when the risk is ≥3% or ≥20%, and intervention recommendations are given. Report generation: Outputs a color report with 3D image annotations, parameter trend charts, and grading criteria within 1 minute, supporting local printing and cloud storage.

6. The method according to claim 5, characterized in that, The deep learning model optimization details are as follows: The Adam optimizer is used (initial learning rate 0.001, decaying by 10% every 50 rounds), batch size 32, and L2 penalty (coefficient 0.001) is used for regularization; spatial features of the input layer image feature channels are extracted through 3×3×3 convolutional kernels, and the numerical feature channels are transformed in dimension through a fully connected layer (64 neurons). The two are then weighted and fused in the fusion layer (128 neurons) through an attention mechanism (image feature weight 0.6, numerical feature weight 0.4). The output layer uses the softmax activation function, and the confusion matrix F1 value is ≥0.

91.

7. The method according to claim 5, characterized in that, Step 3 parameter analysis also includes bone turnover status assessment: calculate the P1NP / CTX ratio. A ratio >1.5 indicates bone formation-dominant type, a ratio <0.8 indicates bone resorption-dominant type, and a ratio between 0.8 and 1.5 indicates balanced type. This result is included in the grading reference factor: bone loss in patients with bone resorption-dominant type can be upgraded to early warning of osteoporosis.

8. The method according to claim 5, characterized in that, The method also includes a follow-up tracking module, which performs the following steps: Automatically associates patient ID (supports ID card / medical insurance card recognition and input), stores scan data from each scan (including original images, parameter values, and grading results), and saves data in a format that conforms to the DICOM 3.0 standard; Generate trend analysis charts: plot the change curves of BMDT value, Ct.Po, Tb.N, and Tb.Sp with time as the horizontal axis, and mark the normal reference range threshold line; Treatment assessment: Calculate the rate of change of parameters between two consecutive scans (e.g., BMD change rate = (current BMD - previous BMD) / previous BMD × 100%). A change rate ≥ 3% is considered effective treatment, -3% ≤ change rate < 3% is considered stable treatment, and a change rate < -3% is considered progression treatment. Warning prompt: When the parameter change rate reaches the progression threshold or the grade is upgraded, an automatic reminder will be pushed to the clinical workstation.

9. The apparatus according to claim 1, characterized in that, The grading output module is also equipped with a voice broadcast unit (supporting Chinese / English bilingual), which can broadcast the core information of the grading results (grading type, fracture risk level, abnormal values ​​of key parameters), and has a built-in patient education knowledge base (including osteoporosis prevention, dietary advice, and exercise guidance), and can print popular science manuals as needed.

Images