A grading method and system based on quantitative statistical analysis of SPECT lung ventilation-perfusion imaging matching.

Abstract

This invention discloses a grading method based on quantitative statistical analysis of SPECT lung ventilation-perfusion imaging matching. The method includes the following steps: Step 1: Pre-collecting ventilation and perfusion plain scan images of the subjects from SPECT V-Q imaging as sample database data for model training; Step 2: Collecting the grading labels of the subjects mentioned in Step 1 as sample database labels for model training; Step 3: Performing tomographic reconstruction, amplitude equalization, and correction on the ventilation and perfusion plain scan images obtained in Step 1, calculating the difference image, and statistically analyzing the difference pixels; classifying the subjects according to the grading labels in Step 2 to obtain a grading standard model; Step 4: Processing the SPECT ventilation and perfusion plain scan images of the subjects to be tested using the grading standard model to provide grading suggestions for the subjects. This invention also discloses a grading system for implementing the above grading method.

Description

Technical Field

[0001] This invention belongs to the field of medical data processing and relates to a grading method and system based on quantitative statistical analysis of matching degree of SPECT lung ventilation and perfusion imaging for non-disease diagnosis or treatment purposes. Background Technology

[0002] Asthma is a common chronic inflammatory disease of the respiratory system. It can be diagnosed through a comprehensive assessment of symptoms, signs, and laboratory tests: 1. Recurrent episodes of wheezing, shortness of breath, chest tightness, and coughing, worsening at night or in the early morning, are often related to exposure to allergens (such as pollen), physical (such as cold air), chemical (such as paint) irritants, viral upper respiratory tract infections, and exercise. Some patients can identify specific triggers for asthma, while others cannot. 2. During an asthma attack, the lungs are in a state of hyperinflation. Patients can clearly feel chest fullness, and auscultation reveals widespread wheezing, predominantly expiratory, in both lungs. In severe cases, a chirping sound can be heard without a stethoscope. Due to difficulty exhaling, patients often unconsciously purse their lips, a phenomenon known as "fish-mouth breathing." This may be accompanied by tachycardia, paradoxical chest and abdominal movements, and cyanosis of the lips. 3. The above symptoms and signs should be symptom-free, excluding other underlying diseases, and should resolve spontaneously or be relieved by treatment. 4. For atypical asthma patients who do not have obvious wheezing or whistling sounds and only present with intractable cough, laboratory tests are required. Asthma can be diagnosed if any one of the following three conditions is met: ① Positive bronchial provocation test (BPT, used to measure airway responsiveness) or exercise test; ② Positive bronchodilator test (BDT, used to measure the reversibility of airway airflow limitation); ③ Diurnal peak expiratory flow (PEF, used to measure changes in airway ventilation function) variability ≥20%.

[0003] In recent years, with the acceleration of my country's industrialization and the aggravation of air pollution, changes in human lung function, respiratory tract reactivity, and the immune system have occurred, increasing the sensitivity of patients with bronchial asthma (hereinafter referred to as asthma) to antigens, increasing the incidence of asthma, and making its symptoms atypical. Currently, there are over 150 million asthma patients worldwide. The pathogenesis of asthma is extremely complex. Environmental allergens, imbalances between nerves and receptors controlling tracheal smooth muscle, abnormal cellular and humoral immune regulation, and genetic factors all participate in its pathogenesis. In recent years, the new theory of asthma pathogenesis—respiratory tract inflammation—has been widely accepted. Therefore, its diagnosis and differential diagnosis rely heavily on laboratory tests, in addition to clinical manifestations, to understand changes in respiratory tract reactivity and lung function. Certain reliable indicators have been established for the diagnosis of asthma. Therefore, a series of examinations are needed clinically for the diagnosis and differential diagnosis of asthma.

[0004] The current problems in the diagnosis of asthma are: 1. Severe asthma can be diagnosed by doctors through auscultation, but the diagnosis of atypical asthma requires many other laboratory tests; 2. Laboratory tests rely on multiple methods for comprehensive judgment, resulting in low diagnostic efficiency and low sensitivity; 3. It is not a quantitative diagnosis, as there are no quantitative evaluation standards for the severity of asthma, and no grading standards and methods based on quantitative indicators.

[0005] For example, existing V / Q imaging techniques for diagnosing PE (corresponding to V / P in this software, where V stands for ventilation and P for perfusion):

[0006] The diagnostic criteria for PE using V / Q scintigraphy are perfusion defects visible in lobes, segments, or multiple subsegments of the lung, while ventilation imaging remains normal. PIOPED data shows that, in a comparative study of V / Q scintigraphy and pulmonary angiography, V / Q scintigraphy has a sensitivity of 92% and a specificity of 87% in diagnosing PE. To better interpret V / Q imaging results, they are recently categorized into three types: ① High probability: perfusion imaging shows two or more perfusion defects, while ventilation imaging is normal; the probability of diagnosing PE is 88%; ② Normal or near-normal: lung perfusion imaging shows no perfusion defects, ruling out PE; the probability of PE is only 0.2%; ③ Non-diagnostic abnormality: V / Q imaging shows both perfusion and ventilation defects, with signs between high probability and normal, including the previously low and moderate probability. Approximately 50% of suspected PE patients have this non-diagnostic V / Q imaging, with a PE probability of 16-33%. Further investigation is needed for these patients. Routine V / Q imaging for diagnosing PE involves observing the perfusion defect area on plain SPECT images. Figure 3 As shown.

[0007] Therefore, the conventional VQ imaging matching method has the following shortcomings:

[0008] 1. Plain CT scans suffer from volumetric effects; the superposition of information from different areas can lead to unclear visualization and localization of defects. 2. Defects are generally diagnosed by the number of defect areas, but the size of these areas is not clearly defined, resulting in a semi-quantitative diagnosis. 3. Accurate grading is impossible for cases with both perfusion and ventilation defects. 4. Quantitative evaluation indicators are difficult to provide for subtle defects. Therefore, currently only 50%–80% of PE patients can be accurately diagnosed.

[0009] In addition, there are several methods for auxiliary examinations and specific monitoring of asthma:

[0010] 1.1. Blood routine tests show that red blood cell count and hemoglobin are mostly within the normal range. However, both may be elevated in patients with long-term and severe emphysema or pulmonary heart disease. The total white blood cell count and neutrophil count are generally normal, but will increase accordingly if there is an infection. Eosinophil count is generally above 0.06, and can be as high as 0.30.

[0011] 1.2. Sputum is often white and frothy, mostly containing crystal-like asthma beads, and is relatively firm and granular. When complicated by infection, the sputum is yellow or green, thicker and more viscous. During severe coughing, capillaries in the bronchial walls may rupture, resulting in blood-tinged sputum. Microscopic examination can reveal Kushman's spirochetes and Ryd's crystals. If the sputum is stained, a large number of eosinophils can be found, which is helpful in diagnosing asthma. In cases of complicated infection, the number of eosinophils decreases, replaced by an increase in neutrophils. Exfoliative cytology examination reveals a large number of columnar ciliated epithelial cells. Generally, no pathogenic bacteria are found in the sputum of asthma patients; common bacteria are most frequently found in catarrhal bacteria and viridans streptococci.

[0012] 1.3. Chemical Changes in Blood: Electrolytes in the blood of asthma patients are generally within the normal range. Even after long-term use of corticotropin or corticosteroids, there is no significant extracellular electrolyte imbalance. Fasting blood glucose, non-protein nitrogen, sodium, potassium, chloride, calcium, phosphorus, and alkaline phosphatase levels are all within the normal range.

[0013] 1.4 Arterial Blood Gas Analysis: In general, during mild to moderate asthma attacks, PCO2 is low, PO2 is mostly within the normal range, and pH > 7.45. However, when PCO2 is between 7.4 and 53 kPa, PO2 < 8.0 kPa, and pH < 7.45, it indicates a severe attack. When pH < 7.45, PO2 < 7.3 kPa, and PCO2 > 6.7 kPa, it indicates a more severe condition.

[0014] 1.5. X-ray Examination: In asthma patients without complications, chest X-rays usually show no specific findings. Changes on X-ray are more common in children with recurrent exogenous asthma, such as increased lung field translucency, thickened bronchial walls, prominent pulmonary aortic arch, descending diaphragm, narrow and elongated cardiac silhouette, uniform reduction in the diameter of blood vessels in the central and peripheral lung fields, and deepened hilar shadows. Scattered small, dense shadows may be seen in the central and peripheral lung fields; their appearance in a short period suggests secondary localized atelectasis caused by transient mucus plug obstruction in a lung segment.

[0015] 1.6 Fiberoptic Bronchoscopy The purpose of fiberoptic bronchoscopy is to identify or examine lesions within the bronchi to determine the cause of asthma. During asthma remission, inflammatory reactions of the mucosa can be seen under the bronchoscope; during an attack, mucosal edema is observed, and secretions are thick, adhering to the bronchial walls and difficult to remove. When infection is present, the secretions are purulent, and the tracheal and bronchial walls collapse during exhalation. Bronchial wall biopsy via fiberoptic bronchoscopy is a method for studying patients experiencing asthma attacks. Fiberoptic bronchoscopy is easier than rigid bronchoscopy, but the tissue sample obtained is smaller. Pathological findings include thickening of the bronchial basement membrane and eosinophilic infiltration, which can lead to a diagnosis of asthma.

[0016] 1.7 Blood Pressure, Pulse, and Electrocardiogram Examinations: Patients with extremely severe asthma attacks may exhibit low blood pressure and pulsus paradoxus. The electrocardiogram may show tachycardia, right axis deviation, and tall, peaked P waves. Other patients generally have normal results on the above examinations. Summary of the Invention

[0017] To address the shortcomings of existing technologies, the present invention aims to propose a grading method and system based on quantitative statistical analysis of SPECT lung ventilation and perfusion imaging matching for non-disease diagnosis or treatment purposes. This method can be used for quantitative and accurate grading evaluation of atypical asthma and for prognostic assessment.

[0018] The grading method based on quantitative statistical analysis of SPECT lung ventilation-perfusion imaging matching for non-diagnostic purposes, as described in this invention, includes the following steps:

[0019] Step 1: Collect a large number (no less than 20 cases) of SPECT VQ imaging ventilation and perfusion plain scans of subjects in advance as sample library data for model training;

[0020] Step 2: Collect the grading labels of the subjects in Step 1 as the sample library labels for model training; the grading labels include three categories: normal, mild asthma, and asthma, and the grading labels are obtained through existing comprehensive methods or the gold standard of detection;

[0021] Step 3: The ventilation and perfusion plain scan images obtained in Step 1 are subjected to tomographic reconstruction, amplitude equalization and correction, and the difference image is calculated and the difference pixels are statistically analyzed; corresponding to the subject classification labels in Step 2, the subjects are classified to obtain the classification standard model.

[0022] Step 4: Process the SPECT ventilation and perfusion plain scan images of the subjects under test using the grading standard model, and then provide the grading recommendations for the subjects under test.

[0023] In steps one and two, the subjects include three groups: normal individuals, those with mild asthma, and those with asthma. Their ventilation plain scan (V), perfusion plain scan (P), and grading labels (D) are collected. Both the ventilation plain scan (V) and perfusion plain scan (P) are three-dimensional matrices, represented as V[m,n,l] and P[m,n,l], respectively. Here, m and n are the length and width of the SPECT plain scan matrix, with a typical matrix size of 64. 64; l represents the number of 360-degree horizontal sweeps, that is, after each horizontal sweep, the rotation... Rotate 1 degree and then perform the next horizontal sweep. For example, if l is 360 times, it means that after each horizontal sweep, rotate 1 degree and then perform the next horizontal sweep. A typical number of horizontal sweeps can be selected as 120 times.

[0024] In step three, the tomographic reconstruction refers to obtaining ventilation tomographic image V_T and perfusion tomographic image P_T by using a back-projection reconstruction algorithm on the SPECT lung ventilation plain scan image V and perfusion plain scan image P, respectively. Specifically, the tomographic reconstruction process is as follows: the ventilation plain scan image matrix V[m,n,l] is transposed to obtain V'[n,l,m], and m two-dimensional matrices [n,l] are further obtained; the l column vectors of each two-dimensional matrix [n,l] are then processed... The back projections are superimposed, and each two-dimensional matrix [n,l] yields a fault map, resulting in a total of m ventilation fault maps V_T; similarly, a total of m perfusion fault maps P_T are obtained. Since the plain scan map matrix is ​​relatively small, fault reconstruction can be performed after interpolation. One or more interpolation methods can be used, such as bicubic interpolation, nearest neighbor interpolation, bilinear interpolation, and cubic spline interpolation.

[0025] In step three, amplitude equalization refers to using one of the SPECT lung ventilation tomography (V_T) and perfusion tomography (P_T) as a reference to unify the maximum pixel value of the two tomography images; using the ventilation tomography image as a reference, the unified perfusion tomography image P_T' is expressed as max(V_T) / max(P_T). P_T; Taking the perfusion fault map as a reference, the same ventilation fault map V_T' is represented as max(P_T) / max(V_T). V_T.

[0026] In step three, the correction process includes image scanning time correction and slice correction. Image scanning time correction refers to dividing the same perfusion tomographic image P_T' and ventilation tomographic image V_T by the scanning time, or dividing the same ventilation tomographic image V_T' and perfusion tomographic image P_T by the scanning time, to eliminate matching differences caused by different scanning times. Slice correction involves arranging the same anatomical location of the lung in the same perfusion tomographic image P_T' and ventilation tomographic image V_T, or the same anatomical location of the lung in the same ventilation tomographic image V_T' and perfusion tomographic image P_T, on a matching sequence.

[0027] The matching sequence refers to the corresponding positions in the images, that is, in the ventilation and perfusion image regions, the layers with the same anatomical structure are located at the same position.

[0028] In step three, the calculation of the difference map includes matching and subtracting the corrected perfusion map and ventilation map, and classifying and statistically analyzing the pixel differences. Specifically, the pixel amplitude equalization and correction values ​​of V_T and P_T' or V_T' and P_T are subtracted by a matrix and then divided by the mean to obtain the difference map.

[0029] V_P=abs((V_T-P_T') / (V_T+P_T')),

[0030] Or, P_V=abs((P_T-V_T') / (P_T+V_T')),

[0031] The pixel value range of the difference map is [0,1], and the pixels are linearly classified and statistically analyzed. If the pixel classification is binary, [0, 0.5] can be set as high match and (0.5, 1] ​​as low match; if the pixel classification is tri-class, [0, 0.1) can be set as high match, [0.1, 0.9) as medium match and [0.9, 1] as low match; it can also be divided into more categories as needed.

[0032] In step three, the pixel difference classification statistics are used as data features, and the labels are normal, mild asthma, and asthma. The classification line corresponding to the grading standard model is given.

[0033] The method described in this invention further includes re-entering the perfusion-ventilation matching statistics of the subjects in step four (i.e., the analysis and processing data of high matching, medium matching, and low matching of the subjects) and the grading label results of the subjects (labels from existing comprehensive methods or the gold standard of detection) into the sample library to retrain the grading standard model. Multilayer perceptrons, support vector machines, and other methods can be used in the training. If the pixel statistics are divided into two categories, it indicates that the feature space is two-dimensional and the grading standard is a straight line; if the statistical pixels are divided into three categories, it indicates that the feature space is three-dimensional and the grading standard is a plane; if the number of categories is greater than 3, it indicates that the feature space is a hyperspace and the grading standard is a hyperplane.

[0034] The present invention also provides a grading system for implementing the above-mentioned grading method, the system comprising: a grading standard model training module and a grading calculation output and sample storage module.

[0035] The hierarchical standard model training module includes a ventilation-perfusion plain scan image and label import module, a hierarchical standard model processing module, and a classification line display module; the hierarchical standard model processing module includes: a ventilation-perfusion fault reconstruction module, a ventilation-perfusion fault interpolation module, a ventilation-perfusion fault time correction module, a ventilation-perfusion fault amplitude equalization module, a ventilation-perfusion fault layer correction module, a ventilation-perfusion fault difference map calculation module, a ventilation-perfusion difference map pixel classification statistics module, a sample library and label library storage module, a hierarchical standard model training module, and a hierarchical line storage module; such as Figure 21 As shown, the standard model training module moves along the arrow in the figure, trains to obtain the classification standard line, and stores and displays it.

[0036] The grading calculation output and sample storage function module includes a module for importing ventilation-perfusion plain scan images and labels to be graded, a grading calculation module, and a grading result output and display module. The grading calculation module includes: a ventilation-perfusion fault reconstruction module, a ventilation-perfusion fault interpolation module, a ventilation-perfusion fault time correction module, a ventilation-perfusion fault amplitude equalization module, a ventilation-perfusion fault layer correction module, a ventilation-perfusion fault difference map calculation module, a ventilation-perfusion difference map pixel classification statistics module, and a grading result calculation module. The grading calculation module, combined with the grading standard model processing module, realizes the sample storage function.

[0037] like Figure 22 As shown, after obtaining the ventilation-perfusion difference image classification statistics results for the graded image along the direction indicated by the arrow, the graded results are calculated based on the graded standard line in the graded standard model training function module, and then the graded results are output (labels) and displayed.

[0038] The statistical data of the images to be graded and the gold standard test results (labels) of the subjects can be stored in the sample library and label library in the grading standard model training function module; after the samples are updated, the grading standard line can be retrained, making the system an updatable open system.

[0039] The beneficial effects of this invention include: 1. Eliminating the influence of volume effect based on tomographic image matching; 2. Grading based on pixel matching statistics rather than the number of defective regions, thus possessing fully quantitative characteristics; 3. Statistical and grading standards can be trained separately for perfusion defects and ventilation defects; 4. The grading standards are updatable and can be retrained as the sample library expands, forming an open sample library. The training method employs artificial intelligence support vector machine or multilayer perceptron technology.

[0040] The method of this invention can use a single indicator to complete the determination and severity assessment with high sensitivity and specificity, which is especially valuable for the severity assessment, treatment plan selection and treatment effect evaluation of patients with mild or atypical asthma. Attached Figure Description

[0041] Figure 1 This is a general method block diagram of the grading method of the present invention.

[0042] Figure 2 This is a flowchart illustrating the specific processing steps for model training and evaluation in the hierarchical method of this invention.

[0043] Figure 3 This is a schematic diagram of the existing conventional VQ imaging matching method. Conventional perfusion-ventilation matching detection suffers from shortcomings such as non-quantitative matching degree, non-quantitative matching area, and the influence of volume effects.

[0044] Figure 4 The software system interface is based on the method of this invention.

[0045] Figure 5 It is a three-dimensional rotating display diagram of perfusion and ventilation flat scan (it can be rotated and displayed simultaneously, roughly showing the corresponding defects and layer errors, etc.).

[0046] Figure 6 This is a schematic diagram showing the reconstruction results of perfusion and ventilation faults, as well as the matching difference map. From left to right, they are the ventilation fault map (V), the perfusion fault map (P), and the difference map (VP).

[0047] Figure 7 This is a schematic diagram showing the reconstruction results of perfusion and ventilation faults, as well as the matching difference map effect. From left to right, they are ventilation fault map V, perfusion fault map P, and difference map VP. The difference map is not obvious, indicating that the matching effect of ventilation fault map V and perfusion fault map P is good.

[0048] Figure 8 This is a diagram showing the subject information, grading criteria, and grading recommendations (for normal subjects).

[0049] Figure 9 This is a diagram showing the tomographic images, matching statistics, and grading recommendations (for asthma subjects). The first row, from left to right, shows the ventilation tomographic image (V), the perfusion tomographic image (P), and the difference image (VP). The difference image is significant, indicating a poor match between the ventilation tomographic image (V) and the perfusion tomographic image (P). This suggests a high probability that the subject has asthma, and the grading result is asthma.

[0050] Figure 10 This is a schematic diagram illustrating the process of layer error and correction (A. The ventilation-perfusion plain scan shows a clear layer error between the two locations; B. The ventilation-perfusion fault scan also verifies that there is a clear layer error between the two locations, and the difference image is even more obvious (because of the layer error, the error is obvious); C. The first step of layer correction is to click on a layer image with the mouse crosshair on the ventilation fault scan; D. The second step of layer correction is to move the mouse to the fault area of ​​the perfusion fault region and the same structure as the first mouse click, and click the mouse crosshair again; E. The third step of layer correction is that the system automatically arranges the fault images clicked by the mouse crosshair twice in the corresponding fault sequence; F. Statistical data and classification results after layer correction).

[0051] Figure 11 This is a schematic diagram illustrating the statistical data analysis results of 23 subjects in this embodiment of the invention. From top to bottom and left to right, the diagrams show the statistical results of high-matching pixels for normal and asthmatic patients, medium-matching pixels for normal and asthmatic patients, low-matching pixels for normal and asthmatic patients, the ratio of low-matching to high-matching pixels for normal and asthmatic patients, and the logarithmic display of the ratio of low-matching to high-matching pixels for normal and asthmatic patients.

[0052] Figure 12 This is the first angle projection image displayed by reading the ventilation and perfusion plain scan in a specific embodiment of the present invention.

[0053] Figure 13 This is a schematic diagram of the operation interface for saving a flat scan image as a high-definition vector image in a specific embodiment of the present invention.

[0054] Figure 14 This is a schematic diagram illustrating the display of severe layer errors on plain scan images, tomographic images, and difference images according to the present invention.

[0055] Figure 15 This is a schematic diagram showing the position of the crosshairs on the perfusion map in the software interface of this invention.

[0056] Figure 16This is a schematic diagram showing the position of the crosshairs on the ventilation diagram cross-section in the software interface of this invention.

[0057] Figure 17 This is a schematic diagram showing the layer position after the present invention performs position correction.

[0058] Figure 18 This is a schematic diagram showing a high degree of matching between the perfusion and ventilation diagrams, a slight difference in the diagram, and a normal result in a specific embodiment of the present invention.

[0059] Figure 19 This is a schematic diagram showing the low matching degree of perfusion and ventilation diagrams and obvious discrepancies in the specific embodiments of the present invention, which is a display of the asthma state.

[0060] Figure 20 This is a schematic diagram illustrating the label setting for saving grading results in a sample library in one specific embodiment of the present invention.

[0061] Figure 21 This is a schematic diagram of the hierarchical standard model training functional module structure of the hierarchical system of the present invention.

[0062] Figure 22 This is a schematic diagram of the hierarchical calculation output and sample storage functional module of the present invention. Detailed Implementation

[0063] The invention will be further described in detail below with reference to the specific embodiments and accompanying drawings. Except for the contents specifically mentioned below, the processes, conditions, and experimental methods for implementing the invention are all common knowledge and general knowledge in the art, and the invention does not have any particular limitations.

[0064] This invention, based on the conventional VQ imaging matching method, establishes a quantitative statistical analysis system for matching degree. After tomographic reconstruction and correction of perfusion / ventilation images, a matching difference map is generated. The difference map is then statistically graded into high, medium, and low matching degree pixels, and the ratio of low-matching degree pixels to high-matching degree pixels is calculated as the grading standard. Analysis of a sample of 23 subjects (5 normal, 18 asthmatic) showed that this processing was completely consistent with the label. In the data processing, this invention performs mean filtering to eliminate noise interference; bicubic interpolation to improve resolution; time correction to eliminate differences in scanning time; and slice correction to eliminate slice errors caused by subject movement during image acquisition. Logarithmic operations are performed to reduce the display area of ​​the graded samples.

[0065] The specific implementation of this invention is carried out according to the following steps:

[0066] 1. Loading Perfusion Images: After clicking the "Load Perfusion" button in the software interface based on the method of this invention, a file selection path for the perfusion images will appear. After selecting the correct perfusion image file, the first plain scan of the perfusion image will be displayed, along with its color bars. If the selected file is not a perfusion image, a prompt message will appear, instructing you to reselect the correct file.

[0067] 2. Load Ventilation Images: After clicking the "Load Ventilation" button in the software interface, a file selection path for the ventilation images will appear. Selecting the correct ventilation image file will display the first plain scan image of the ventilation system, along with its color bars. If the selected file is not a ventilation image, a prompt message will appear, instructing you to select the correct file again.

[0068] 3. Stereoscopic Rotation Display: Click the "Image Rotate" button in the software interface to simultaneously rotate and display the perfusion and ventilation scan images in stereoscopic mode. For better observation, select the checkbox next to the button to open a separate page for stereoscopic display. Alternatively, you can save the image as a vector graphic by clicking the "Export Settings" option in the "File" menu within this window. Figure 13 As shown.

[0069] 4. Fault Reconstruction: After clicking the VP tomography button in the software interface, the system performs perfusion / ventilation fault reconstruction and displays the perfusion / ventilation fault reconstruction map and matching difference map in the lower half of the system. For better observation, selecting the checkbox next to the button will open a separate page for fault display, or you can save it as a vector image through the export settings in this window.

[0070] 5. Horizontal Correction: Click the Horizon Correction button in the software interface to perform horizontal correction. During perfusion and ventilation scans, subject movement may cause differences in lateral alignment. This difference can lead to significant mismatches, resulting in false positives or increased severity assessments. Therefore, horizontal correction is necessary. The horizontal correction function is performed manually. After clicking the horizontal correction button, a crosshair will appear. Click on a section of the perfusion map with the crosshair, then drag it to the ventilation map area. Observe the section that most closely resembles the previous section and click it again. This will correct the ventilation map section to the exact same position as the perfusion map section. Then recalculate the data and grading results.

[0071] like Figure 14A patient exhibited severe slice displacement, clearly visible in both plain and tomographic images. In the plain image, there was a significant difference in height between the two. In the tomographic image, the shapes of the corresponding slices were clearly mismatched. The poorly matched image showed a large number of low-match pixels (white), with grading values ​​exceeding the normal display range.

[0072] 6. To better display the horizontal correction crosshairs, change the colorbar to HSV format. Click the "Horizon Correction" button. A black crosshair will appear. First, observe the planes where the fault shapes of the grouting and ventilation maps match well. Click the crosshair at the center of a fault plane in the grouting map (e.g., ...). Figure 15 (As shown).

[0073] 7. Move the crosshair to the center of another layer with the same shape on the ventilation diagram and click the left mouse button (e.g., ...). Figure 16 (As shown).

[0074] 8. The software will automatically correct the position of the layers and then display the following: Figure 17 At this point, the difference map error decreased, and the grading values ​​also changed. Although still outside the asthma red line, the values ​​decreased significantly.

[0075] 9. Graded recommendation results - normal case; the specific operation procedure is the same as steps 1-8 above.

[0076] Model calculations can be performed using matching data from perfusion and ventilation charts of different subjects to provide recommendations for their classification results.

[0077] like Figure 18 As shown, the locations on the plain scan images are basically consistent, the perfusion and ventilation images have a high degree of matching, and the difference in the images is slight. The final result is recommended as normal.

[0078] 10. Classification Recommendation Results - Asthma Case: The specific operation procedure is the same as steps 1-8 above.

[0079] like Figure 19 As shown, another subject had a low match between perfusion and ventilation tomographic images, and the final recommended outcome was asthma.

[0080] Table 1 shows the statistical grading effect of actual subjects' labels and matching data (5 normal, 18 asthma). Label 1 is normal, and label 2 is asthma.

[0081] Table 1

[0082] Label High matching degree Medium matching degree low matching degree Low match / high match Low match / high match log ratio 1 18494 13246 3194 0.172705 1.237304059 1 20972 451 3774 0.179954 1.255162047 1 37640 3541 1073 0.028507 0.454950107 1 13402 17907 1449 0.108118 1.033898772 1 10735 9988 1913 0.178202 1.250912921 2 7379 7536 2901 0.393143 1.594550218 2 7600 11928 8755 1.151974 2.061442558 2 3598 11350 8883 2.468872 2.392498503 2 4389 6758 14556 3.316473 2.520676466 2 14363 12721 7333 0.510548 1.708036525 2 6825 16262 15695 2.299634 2.361658664 2 13234 19629 14939 1.128835 2.052630397 2 11600 15564 13276 1.144483 2.058609255 2 6003 11371 12983 2.162752 2.335006714 2 24638 7232 8949 0.363219 1.560169057 2 21918 10668 6264 0.285792 1.456050826 2 15911 15723 20047 1.259946 2.100351915 2 6914 16678 17106 2.474111 2.393419092 2 16405 15520 10342 0.630418 1.799628299 2 13986 5788 14739 1.05384 2.022774495 2 15136 12944 14246 0.9412 1.973681821 2 5076 12849 13294 2.618991 2.418134061 2 3222 2630 20297 6.299503 2.799306316

[0083] Statistical analysis of 23 subjects revealed the following: High and medium match regions showed very low correlation with the labels, exhibiting no categorical regularity; low match regions showed strong correlation, but still included one outlier. Considering the varying sizes of lung regions among subjects, absolute values ​​were not entirely reasonable. Displaying a relative ratio of low to high match regions resulted in high correlation. However, the data resolution was poor, with small values ​​compressed together and large values ​​showing significant dispersion. Logarithmic display yielded the best resolution. Figure 11 As shown.

[0084] The first image shows the statistical results of highly matched pixels in normal and asthmatic patients, without any graded differences.

[0085] The second figure shows the statistical results of the matched pixels in normal and asthmatic patients, without grading differences.

[0086] The third figure shows the statistical results of low-match pixels between normal and asthmatic patients, which basically have statistical significance for graded differences; however, there are still outlier data points.

[0087] The fourth figure shows the statistical pixel ratio of low-match to high-match between normal and asthmatic patients, which has statistical significance for graded differences; however, the display is not good.

[0088] The fifth figure shows the logarithmic display of the low / high match ratio of pixels in normal and asthmatic patients, demonstrating statistical significance of graded differences; at the same time, the display is very good.

[0089] The software system based on the method of this invention also has the following functional operations:

[0090] i) Save flat scan 3D rotation image function: Click the save rotation image button to save the flat scan image as a GIF animation.

[0091] ii) Save V_P Tomography Image Function: Click the save V_P Tomography button to save the perfusion, ventilation, and perfusion-ventilation difference images in png or jpg format.

[0092] iii) Save Statistical Data Function: Clicking the "Save Statistical Data" button will save the area of ​​high-matching regions, the area of ​​medium-matching regions, and the area of ​​low-matching regions in the difference plot, as well as the logarithmic result of the ratio of low-matching region area to high-matching region area, as an Excel spreadsheet.

[0093] iv) Save data to the database: Click the "send to database" button to save the statistical data of the subjects calculated in this study to the database. Before saving, you need to enter the sample label, i.e., the subject's classification: 1 indicates normal, 2 indicates mild, and 3 indicates asthma. An illustration of one of the classification results is shown below. Figure 20 As shown.

[0094] v) Retraining the Classification Criterion Line: After adding new samples, you can retrain the classification criteria. Clicking "retrain classifier" may adjust the classification line. This software uses a support vector machine (SVM) model for classification training. Because the classification criteria can be changed with the added samples, adding new samples requires great care. If an incorrect sample or label is mistakenly added, it can be deleted from the backend database.

[0095] The scope of protection of this invention is not limited to the above embodiments. Any variations and advantages that can be conceived by those skilled in the art without departing from the spirit and scope of the inventive concept are included in this invention and are protected by the appended claims.

Claims

1. A grading method based on quantitative statistical analysis of SPECT lung ventilation-perfusion imaging matching, characterized in that, The method includes the following steps: Step 1: Pre-collect ventilation and perfusion plain scan images of the subjects from SPECT VQ imaging as sample database data for model training; Step 2: Collect the graded labels of the subjects mentioned in Step 1 as the sample library labels for model training; Step 3: The ventilation and perfusion plain scan images obtained in Step 1 are processed by tomographic reconstruction, amplitude equalization, and correction. The difference image is calculated and the difference pixels are statistically analyzed. The subjects are classified according to the classification labels of the subjects in Step 2 to obtain the classification standard model. In step three, the tomographic reconstruction refers to obtaining ventilation tomographic image V_T and perfusion tomographic image P_T by using a back-projection reconstruction algorithm on the SPECT lung ventilation plain scan image V and perfusion plain scan image P, respectively. Specifically, the tomographic reconstruction process is as follows: the ventilation plain scan image matrix V[m,n,l] is transposed to obtain V'[n,l,m], and m two-dimensional matrices [n,l] are further obtained; the l column vectors of each two-dimensional matrix [n,l] are then processed... The reverse projection superposition yields a tomographic image for each two-dimensional matrix [n, l], resulting in a total of m ventilation tomographic images V_T; similarly, a total of m perfusion tomographic images P_T are obtained; where m and n are the matrix length and width of the SPECT plain scan image, respectively; l is the number of 360-degree plain scans, i.e., after each plain scan, the image is rotated... Then sweep the surface again. Step three, the calculation of the difference map, includes matching and subtracting the corrected V_T and P_T', or V_T' and P_T, and classifying and statistically analyzing the pixel differences; specifically, the pixel amplitude equalization and correction of V_T and P_T', or P_T and V_T', are subtracted by a matrix and then divided by the mean to obtain the difference map. V_P=abs((V_T-P_T') / (V_T+P_T')), Or, P_V=abs((P_T-V_T') / (P_T+V_T')), The pixel value range of the difference map is [0,1], and the pixels are linearly classified and statistically analyzed. In step three, amplitude equalization refers to using one of the SPECT lung ventilation tomography (V_T) and perfusion tomography (P_T) as a reference to unify the maximum pixel value of the two tomography images; using the ventilation tomography image as a reference, the unified perfusion tomography image P_T' is expressed as max(V_T) / max(P_T). P_T; Taking the perfusion fault map as a reference, the same ventilation fault map V_T' is represented as max(P_T) / max(V_T). V_T; In step three, the correction process includes image scanning time correction and slice correction. Image scanning time correction refers to dividing the same perfusion tomographic image P_T' and ventilation tomographic image V_T by the scanning time, or dividing the same ventilation tomographic image V_T' and perfusion tomographic image P_T by the scanning time, to eliminate matching differences caused by different scanning times. Slice correction involves arranging the same anatomical location of the lung in the same perfusion tomographic image P_T' and ventilation tomographic image V_T, or the same anatomical location of the lung in the same ventilation tomographic image V_T' and perfusion tomographic image P_T, on a matching sequence. In step three, the pixel difference classification statistics are used as data features, and the labels are normal, mild asthma, and asthma. The classification line corresponding to the grading standard model is given. Step 4: Process the SPECT ventilation and perfusion plain scan images of the subjects under test using the grading standard model, and then provide the grading recommendations for the subjects under test. The method also includes re-entering the perfusion-ventilation matching statistics of the subjects in step four and the classification label results of the subjects into the sample library to retrain the classification standard model.

2. The grading method based on quantitative statistical analysis of SPECT lung ventilation-perfusion imaging matching degree according to claim 1, characterized in that, The acquired SPECT ventilation plain scan (V), perfusion plain scan (P), and grading label (D) include SPECT ventilation plain scan (V), perfusion plain scan (P), and grading label (D) for three groups of people: normal, mild asthma, and asthma.

3. The grading method based on quantitative statistical analysis of SPECT lung ventilation-perfusion imaging matching degree according to claim 2, characterized in that, The grading label D includes three categories: normal, mild asthma, and asthma; the ventilation plain scan V and the perfusion plain scan P are both three-dimensional matrices, which are represented as V[m,n,l] and P[m,n,l], respectively.

4. The grading method based on quantitative statistical analysis of SPECT lung ventilation-perfusion imaging matching degree according to claim 1, characterized in that, Before fault reconstruction, the process also includes interpolating the scan matrix using one or more interpolation methods such as bicubic interpolation, nearest neighbor interpolation, bilinear interpolation, and cubic spline interpolation.

5. A hierarchical system implementing the hierarchical method as described in any one of claims 1-4, characterized in that, The system includes: a hierarchical standard model training module and a hierarchical calculation output and sample storage module; The hierarchical standard model training module includes a ventilation-perfusion plain scan image and label import module, a hierarchical standard model processing module, and a classification line display module; the hierarchical standard model processing module includes: a ventilation-perfusion fault reconstruction module, a ventilation-perfusion fault interpolation module, a ventilation-perfusion fault time correction module, a ventilation-perfusion fault amplitude equalization module, a ventilation-perfusion fault layer correction module, a ventilation-perfusion fault difference map calculation module, a ventilation-perfusion difference map pixel classification statistics module, a sample library and label library storage module, a hierarchical standard model training module, and a hierarchical line storage module; The hierarchical calculation output and sample storage function module includes a module for importing ventilation-perfusion plain scan images and labels to be classified, a hierarchical calculation module, and a hierarchical result output and display module. The hierarchical calculation module includes: a ventilation-perfusion fault reconstruction module, a ventilation-perfusion fault interpolation module, a ventilation-perfusion fault time correction module, a ventilation-perfusion fault amplitude equalization module, a ventilation-perfusion fault layer correction module, a ventilation-perfusion fault difference map calculation module, a ventilation-perfusion difference map pixel classification statistics module, and a hierarchical result calculation module. The hierarchical calculation module, combined with the hierarchical standard model processing module, realizes the sample storage function.

Images