Risk assessment system and method for analyzing abnormal sleep breathing of children based on CBCT (cone beam computed tomography) double channels

Abstract

The invention discloses a risk assessment system and method for analyzing abnormal sleep breathing of children based on CBCT dual-channel, and the method comprises the steps: collecting local three-dimensional image data of a maxillofacial region through CBCT, and extracting the sagittal area, volume and multi-dimensional angle and distance parameters of an upper airway through head correction and anatomical mark positioning; and constructing a logistic regression clinical prediction model combining single-factor and multi-factor logistic regression to realize risk prediction. Standardization and equal-interval sampling are synchronously carried out on image data, a multi-channel three-dimensional image data cube is generated, and deep learning classification is realized by inputting the multi-channel three-dimensional image data cube into a 3D ResNet network based on identical fast connection. And finally, performing weighted collaborative analysis on results of the two diagnosis channels, and outputting children obstructive sleep apnea risk classification. Through the parameter driving and image learning dual-channel fusion design, the accuracy and applicability of early screening and risk assessment of children obstructive sleep apnea are improved.

Description

Technical Field

[0001] The present invention relates to the field of intelligent medical monitoring management, and in particular to a system and method for risk assessment of abnormal sleep breathing in children based on CBCT dual-channel analysis. Background Art

[0002] Obstructive sleep apnea (OSA) is a common sleep disorder in children, characterized by partial or complete upper airway obstruction during sleep, disrupting normal ventilation and sleep patterns. The etiology, clinical manifestations, and diagnostic criteria for OSA in children are nearly independent criteria and screening areas. The prevalence of OSA in children is high, and the harm is significant and persistent, requiring technical assistance for early detection and efficacy assessment.

[0003] The current gold standard for diagnosis is polysomnography (PSG). While simplifying the testing equipment, PSG signal processing needs to be optimized. However, it has the following significant limitations in children: Difficulty in coordination: Multiple sensors need to be connected, which can easily cause discomfort and resistance; Unnatural environment: The hospital laboratory environment is different from daily sleep and is prone to the "first night effect"; Poor resources and accessibility: Sleep centers are concentrated in large cities, with scarce equipment and long waiting times; High cost and time-consuming: The equipment is expensive and requires professional personnel to monitor all night and perform complex analysis.

[0004] Existing PSG interpretation methods and patented diagnostic algorithms require extensive manual preprocessing before they can be used for OSA assessment and modeling. These techniques are highly sensitive and primarily focus on adult OSA or PSG technology optimization, lacking effective alternatives for pediatric OSA. OSA in children differs from that in adults, with etiologies more commonly linked to adenoid / tonsillar hypertrophy and craniofacial structural abnormalities, and diagnostic criteria also differ (an AHI ≥ 1 in children is considered abnormal).

[0005] In view of this, the present invention proposes a risk assessment system and method for children's sleep apnea based on CBCT dual-channel analysis. Summary of the Invention

[0006] The purpose of the present invention is to provide a system and method for risk assessment of sleep apnea disorders in children based on CBCT dual-channel analysis, which solves the problems of complex operation, high cost and poor compliance in the existing diagnosis of obstructive sleep apnea (OSA) in children relying on polysomnography.

[0007] In a first aspect, the present invention provides a method for assessing the risk of sleep apnea in children based on CBCT dual-channel analysis, comprising the following steps: S101: performing head position correction and three-dimensional positioning tracing processing on the CBCT original image data to obtain a maxillofacial image parameter set, wherein the maxillofacial image parameter set includes sagittal area and volume of different upper airway segments, as well as maxillofacial multi-dimensional angle parameters and multi-dimensional distance parameters; S102: Based on the maxillofacial imaging parameter set, a logistic regression clinical prediction model for obstructive sleep apnea in children was constructed using univariate regression screening and multivariate logistic regression modeling; S103: Performing equal-interval sampling and pixel value normalization on the CBCT raw image data to generate an equal-dimensional three-dimensional image data cube for deep learning; S104: Obtaining a deep learning classification result of images of obstructive sleep apnea in children using a three-dimensional convolutional neural network model that uses the three-dimensional image data cube as input, wherein the three-dimensional convolutional neural network model is a 3D ResNet network with an identical shortcut connection structure; S105: Collaboratively analyze the output results of the logistic regression clinical prediction model and the image deep learning classification results of the three-dimensional convolutional neural network model to obtain a risk classification result for obstructive sleep apnea in children.

[0008] As a preferred technical solution of the present invention, the step of obtaining the maxillofacial image parameter set includes: Correct the head position of the CBCT original image data based on standard anatomical landmarks; Based on the corrected CBCT original image data, the sagittal areas and volumes of the nasopharynx, velopharynx, glossopharynx, and hypopharynx were extracted. Measure the multi-dimensional angle parameters and multi-dimensional distance parameters of the mandible, hyoid bone and craniomaxillofacial skeleton in a three-dimensional coordinate system; The above area, volume, angle and distance parameters are combined to form the maxillofacial imaging parameter set.

[0009] As a preferred technical solution of the present invention, the steps of constructing the logistic regression clinical prediction model include: The maxillofacial imaging parameter set was randomly divided into a training set and a validation set; The univariate regression method was used to select parameters with P values ​​less than or equal to 0.1 as modeling variables; Multivariate logistic regression modeling was performed based on the screened modeling variables to construct an obstructive sleep apnea risk prediction model; A nomogram score chart is drawn based on the model regression coefficient to achieve individual risk scoring and prediction for children.

[0010] As a preferred technical solution of the present invention, the step of generating an equal-dimensional three-dimensional image data cube for deep learning includes: The CBCT original image data is sampled at equal intervals to generate a three-dimensional data cube of preset size; According to the set window width and window position, the image pixel value is normalized to the preset grayscale value range in the preset pixel interval; The normalized pixel matrix is ​​combined into a three-dimensional image data cube in the form of a multi-channel input, where each channel of the three-dimensional image data cube corresponds to normalized two-dimensional slice data in axial, sagittal and coronal dimensions.

[0011] As a preferred technical solution of the present invention, the step of obtaining the image deep learning classification result includes: Building a 3D ResNet network based on an identical shortcut connection structure, wherein the 3D ResNet network includes a residual module and a global average pooling layer; Input the three-dimensional image data cube and its corresponding obstructive sleep apnea label, train the 3DResNet network, and output corresponding classification features after multi-layer convolution and residual learning; Generating obstructive sleep apnea classification categories using the multi-classification probability results output by the 3D ResNet network, performing supervised training based on the classification features and preset classification labels, and optimizing weight parameters of the 3D ResNet network; The classification accuracy, precision, and recall of the 3D ResNet network were evaluated based on the training set and the validation set.

[0012] As a preferred technical solution of the present invention, the steps of training the 3D ResNet network include: Two network architectures, 3D ResNet50 and 3D ResNet101, were constructed respectively, and the cross entropy loss function and Adam optimizer were used to jointly control the training process of the 3D ResNet network; The 3D ResNet50 and 3D ResNet101 are trained respectively using the training set; The classification performance of the two networks was compared on the validation set, and the network with the highest classification accuracy on the validation set was selected as the final three-dimensional convolutional neural network model.

[0013] As a preferred technical solution of the present invention, the logic for generating risk classification results is: The risk prediction result of the logistic regression clinical prediction model is used as the first risk factor, and the image deep learning classification result is used as the second risk factor; generating a risk classification result of the obstructive sleep apnea in children by fusing the first risk factor and the second risk factor based on a weighted collaborative decision-making mechanism; Among them: the weights of the first risk factor and the second risk factor are obtained by dynamic adjustment based on the classification accuracy of historical clinical samples.

[0014] In the second aspect, the present invention provides a children's sleep breathing abnormality risk assessment system based on CBCT dual-channel analysis, which is based on the implementation of the first aspect and includes a head position correction and image parameter extraction module, a logistic regression prediction modeling module, an image data cube generation module, an image deep learning analysis module and a collaborative analysis module. The modules are connected by wired and / or wireless connections.

[0015] The head position correction and image parameter extraction module performs head position correction and three-dimensional positioning tracing on the original CBCT image data to obtain a maxillofacial image parameter set, which includes sagittal area and volume for different sections of the upper airway, as well as maxillofacial multi-dimensional angle parameters and multi-dimensional distance parameters; A logistic regression prediction modeling module is used to construct a logistic regression clinical prediction model for obstructive sleep apnea in children based on the maxillofacial imaging parameter set. The logistic regression clinical prediction model is generated based on univariate regression screening and multivariate logistic regression modeling, and is used to output a risk prediction result for obstructive sleep apnea in children. An image data cube generation module is used to perform equal-interval sampling and pixel value normalization processing on the CBCT raw image data to generate an equal-dimensional multi-channel three-dimensional image data cube; an image deep learning analysis module, which uses a three-dimensional convolutional neural network model with the three-dimensional image data cube as input to obtain a deep learning classification result of images of obstructive sleep apnea in children, wherein the three-dimensional convolutional neural network model is a 3D ResNet network with an identical shortcut connection structure; The collaborative analysis module collaboratively analyzes the output results of the logistic regression clinical prediction model with the image deep learning classification results of the three-dimensional convolutional neural network model to obtain the risk classification results of obstructive sleep apnea in children.

[0016] In the above technical solution, the technical effects and advantages provided by the present invention are: The present invention realizes the complementary fusion of parameter-driven and data-driven through the dual-channel collaborative design of logistic regression clinical prediction and three-dimensional convolutional neural network image analysis, effectively improving the accuracy and reliability of the diagnosis of obstructive sleep apnea in children.

[0017] Based on the modeling of child-specific maxillofacial anatomical parameters, the system takes into account the physiological characteristics of age groups and individual differences, enhances the adaptability and generalizability to different child samples, and uses the CBCT raw image data in parallel in parameter extraction and image data cube generation, taking into account both structural parameters and raw image information, thereby improving data utilization efficiency and diagnostic value.

[0018] Relying on CBCT as the image input source, combined with deep learning technology and regression modeling analysis, it provides a non-invasive, standardized assessment method for the early screening of obstructive sleep apnea in children. The collaborative analysis module integrates the classification results of the dual-channel model through a risk factor weighting mechanism to generate a unified risk classification output, reducing the subjective errors and inconsistencies in manual interpretation. BRIEF DESCRIPTION OF THE DRAWINGS

[0019] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments. Obviously, the drawings described below are only some embodiments described in the present invention. For ordinary technicians in this field, other drawings can also be obtained based on these drawings.

[0020] Figure 1 This is a flowchart of the steps of the method for assessing the risk of sleep apnea in children according to the present invention; Figure 2 Schematic diagram of clinical feature extraction points of the method for assessing the risk of sleep apnea in children according to the present invention; Figure 3 Schematic diagram of adenoid size assessment in the method for assessing the risk of sleep apnea in children according to the present invention; Figure 4 This is a quantitative schematic diagram of the degree of upper airway obstruction in the method for assessing the risk of sleep apnea in children according to the present invention; Figure 5 This is the Nomogram of the clinical prediction model of the present invention; Figure 6 is the ROC curve diagram of the clinical prediction model of the present invention; Figure 7 Schematic diagram of the 3D Resnet architecture of the present invention; Figure 8 This is a diagram showing the distribution of the data set of the present invention; Figure 9 This is the ROC curve of the optimal 3D Resnet prediction model of the present invention. DETAILED DESCRIPTION

[0021] In order to make the purpose, technical solutions and advantages of the implementation of this application clearer, the technical solutions in the implementation of this application will be described in more detail below in conjunction with the drawings in the implementation of this application.

[0022] In the accompanying drawings, the same or similar reference numerals throughout represent the same or similar elements or elements with the same or similar functions. The described embodiments are part of the embodiments of the present application, rather than all of the embodiments. The embodiments described below with reference to the accompanying drawings are exemplary and are intended to be used to explain the present application, and should not be understood as limiting the present application. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of the present application. The embodiments of the present application are described in detail below with reference to the accompanying drawings.

[0023] Example 1 See also Figure 1 As shown, this embodiment provides a method for assessing the risk of sleep apnea in children based on CBCT dual-channel analysis, including the following steps: S101: performing head position correction and three-dimensional positioning tracing processing on the CBCT original image data to obtain a maxillofacial image parameter set, wherein the maxillofacial image parameter set includes sagittal area and volume of different upper airway segments, as well as maxillofacial multi-dimensional angle parameters and multi-dimensional distance parameters; CBCT raw image data is acquired through cone beam computed tomography (CBCT). CBCT technology uses a pulsed X-ray source and a detector array to acquire sectional images of the craniofacial region. CBCT images can clearly display the structural boundaries between craniofacial hard tissues and upper airway voids, and generate equidistant three-dimensional data without structural overlap, which facilitates subsequent accurate upper airway parameter extraction and structural analysis. CBCT overcomes the problems of traditional two-dimensional cephalometric measurements, eliminates structural overlap, and can clearly and non-magnified display craniofacial tissues. Furthermore, the spatial relationships of the structures are not affected by head position.

[0024] CBCT uses pulsed exposure, with a single scan dose of 30-120 μSv and a localized focus (maxillofacial area). This is different from the 200-2000 μSv dose of traditional multi-slice CT for the whole body / large area, and complies with the ALARA principle. Furthermore, the exposure time is only 5-20 seconds, making it a non-invasive procedure for almost all children. Figure 2 As shown, this provides a key basis for collecting 3D parameters for the logistic regression clinical model. This standardized basis for extracting clinical features such as angles, distances, areas, and volumes addresses the challenges of standardized identification and parameter extraction of maxillofacial and upper airway structures in CBCT images. A unified anatomical landmark positioning method ensures repeatable and clinically consistent subsequent parameter extraction.

[0025] Specifically, the steps of obtaining the maxillofacial image parameter set include: Correct the head position of the CBCT original image data based on standard anatomical landmarks; Based on the corrected CBCT original image data, the sagittal areas and volumes of the nasopharynx, velopharynx, glossopharynx, and hypopharynx were extracted. Measure the multi-dimensional angle parameters and multi-dimensional distance parameters of the mandible, hyoid bone and craniomaxillofacial skeleton in a three-dimensional coordinate system; The above area, volume, angle and distance parameters are combined to form the maxillofacial imaging parameter set.

[0026] S102: Based on the maxillofacial imaging parameter set, a logistic regression clinical prediction model for obstructive sleep apnea in children was constructed using univariate regression screening and multivariate logistic regression modeling; Specifically, if Figure 3 As shown in the figure, the adenoid size assessment diagram supports the collection of parameters such as volume and area of ​​different regions of the upper airway (especially the nasopharyngeal area) in the logistic regression model, strengthens the model's specific recognition ability of adenoid hypertrophy, and solves the problem of difficulty in quantifying adenoid volume and large subjective judgment errors in children with OSA. By visualizing and quantifying adenoid size, its performance in different obstruction levels is clarified. The steps of constructing the logistic regression clinical prediction model include: The maxillofacial imaging parameter set was randomly divided into a training set and a validation set; The univariate regression method was used to select parameters with P values ​​less than or equal to 0.1 as modeling variables; Multivariate logistic regression modeling was performed based on the screened modeling variables to construct an obstructive sleep apnea risk prediction model; A nomogram score chart is drawn based on the model regression coefficient to achieve individual risk scoring and prediction for children.

[0027] S103: Performing equal-interval sampling and pixel value normalization on the CBCT raw image data to generate an equal-dimensional three-dimensional image data cube for deep learning; Specifically, the step of generating an equal-dimensional three-dimensional image data cube for deep learning includes: The CBCT original image data is sampled at equal intervals to generate a three-dimensional data cube of preset size; According to the set window width and window position, the image pixel value is normalized to the grayscale value range of [0, 255] in the interval [-1000HU, 1000HU]; Combines normalized pixel matrices into a 3D image data cube with multi-channel input.

[0028] S104: Using a three-dimensional convolutional neural network model with the three-dimensional image data cube as input, a deep learning classification result of images of obstructive sleep apnea in children is obtained, wherein the three-dimensional convolutional neural network model is a 3D ResNet network with an identical shortcut connection structure; 3D-ResNet is a deep learning architecture that integrates a three-dimensional convolutional neural network (3D CNN) and a residual learning network (ResNet), and is specifically designed for processing temporal spatial data (such as medical image sequences).

[0029] Specifically, the steps for obtaining the image deep learning classification results include: Building a 3D ResNet network based on an identical shortcut connection structure, wherein the 3D ResNet network includes a residual module and a global average pooling layer; Input the three-dimensional image data cube and its corresponding obstructive sleep apnea label, train the 3DResNet network, and output corresponding classification features after multi-layer convolution and residual learning; Generating obstructive sleep apnea classification categories using the multi-classification probability results output by the 3D ResNet network, performing supervised training based on the classification features and preset classification labels, and optimizing weight parameters of the 3D ResNet network; The classification accuracy, precision, and recall of the 3D ResNet network were evaluated based on the training set and the validation set.

[0030] The steps of training the 3D ResNet network include: Two network architectures, 3D ResNet50 and 3D ResNet101, were constructed respectively, and the cross entropy loss function and Adam optimizer were used to jointly control the training process of the 3D ResNet network; The 3D ResNet50 and 3D ResNet101 are trained respectively using the training set; The classification performance of the two networks was compared on the validation set, and the network with the highest classification accuracy on the validation set was selected as the final three-dimensional convolutional neural network model.

[0031] S105: Collaboratively analyze the output results of the logistic regression clinical prediction model and the image deep learning classification results of the three-dimensional convolutional neural network model to obtain a risk classification result for obstructive sleep apnea in children.

[0032] Specifically, if Figure 4The quantitative diagram of upper airway obstruction degree shown here serves as a reference for 3D data cube generation and spatial understanding of anatomical structures, enhancing the model's ability to identify different degrees of obstruction and used to determine OSA severity. This solves the problem of traditional 2D images being unable to accurately express the three-dimensional degree of upper airway obstruction and improves the intuitiveness and accuracy of key indicators such as the minimum cross-sectional area and volume of the upper airway. The logic for generating risk classification results is as follows: The risk prediction result of the logistic regression clinical prediction model is used as the first risk factor, and the image deep learning classification result is used as the second risk factor; generating a risk classification result of the obstructive sleep apnea in children by fusing the first risk factor and the second risk factor based on a weighted collaborative decision-making mechanism; Among them: the weights of the first risk factor and the second risk factor are obtained by dynamic adjustment based on the classification accuracy of historical clinical samples.

[0033] For example, the collaborative analysis step includes: When both the logistic regression clinical prediction model and the three-dimensional convolutional neural network model output negative results, the child was determined to be in the low-risk category; When both the logistic regression clinical prediction model and the three-dimensional convolutional neural network model output positive results, the child was determined to be in the high-risk category; When the output results of the two models are inconsistent, a weighted fusion judgment is performed based on the model confidence and output weight to generate the risk classification result.

[0034] The child-centered machine learning model shows good diagnostic performance for OSA. It is optimized for children aged 6-12 years old and targets parameters specific to pediatric OSA, who benefit greatly from early diagnosis and treatment. Deep learning can quantify high-dimensional radiological phenotypes that are beyond the range of human perception. This embodiment constructs a non-invasive prediction model for OSA clinical scenarios, combines common clinical imaging data, covers the difficulties in processing and running high-dimensional data, and provides a fast, convenient, comfortable, and highly relevant AI tool for the diagnosis and prognosis of OSA, allowing existing data to provide new clinical value. The channel one clinical model is better at capturing static anatomical indicators, while the imaging model can comprehensively discover disease characteristics. This embodiment realizes the establishment of a clinical parameter model and an imaging deep learning model for pediatric OSA for the first time, retaining the anatomical parameter system familiar to doctors, integrating the high-dimensional feature extraction capabilities of deep learning, and guiding the clinical decision-making path for prediction results.

[0035] Example 2 This embodiment is based on Example 1. The target population of this embodiment is children aged 6-12 years old. After nocturnal polysomnography, OSA patients and non-OSA patients who meet the inclusion and exclusion criteria of this embodiment are included, including basic demographic information and CBCT original image data collected from the target population.

[0036] Diagnostic criteria for OSA: Children undergo overnight sleep monitoring in the hospital, which includes continuous breathing at night, arterial oxygen saturation, electroencephalogram, electrocardiogram, heart rate and other indicators. Based on the results of polysomnography, a diagnosis of sleep apnea-hypopnea syndrome and its severity is made.

[0037] Children's interpretation indicators: 1≤AHI<5 is mild OSA; 5≤AHI<10 is moderate, and ≥10 is severe.

[0038] S102: A logistic regression clinical prediction model for obstructive sleep apnea in children was constructed based on the maxillofacial imaging parameter set, which was obtained by performing univariate regression screening and multivariate logistic regression modeling on the original CBCT imaging data; Channel 1: Logistic regression clinical prediction model based on highly correlated features Unique selection of clinical parameters: Based on highly clinically relevant medical knowledge, a child-specific measurement system is established to collect image data of OSA patients, perform head position correction based on CBCT data, and extract key feature points.

[0039] Model variables include: positioning tracing, angle measurement, and distance measurement on three-dimensional data; as well as area and volume calculations for the anatomical structures of different segments of the upper airway.

[0040] Characteristic points of the upper airway: total sagittal area of ​​the nasopharynx, horizontal sagittal area of ​​the hard palate, total volume of the nasopharynx, total sagittal area of ​​the velopharynx, horizontal sagittal area of ​​the soft palate tip, volume of the velopharynx, total sagittal area of ​​the glossopharynx, horizontal sagittal area of ​​the epiglottis, volume of the glossopharynx, total sagittal area of ​​the laryngopharynx, horizontal sagittal area of ​​the hyoid bone, volume of the laryngopharynx, total sagittal area of ​​the upper airway, minimum cross-sectional area of ​​the upper airway, and total volume of the upper airway.

[0041] Three-dimensional maxillofacial distance index (mm): SN - Anterior Cranial Base, S-Ba - Posterior Cranial Base, Pog-Na Perp R, Pog-Na Perp L, A-Na Perp R, A-Na Perp L, ANS-Me, S-GoR, S-Go L, Hy-MP R, Hy-MP L, Hy-PP, Hy-FH R, Hy-FH L, PNS-Sp, Tn-EpPp R, Tn-EpPp L, EpR-EpL, PpR-PpL, UR4-UL4, UR6-UL6.

[0042] Three-dimensional maxillofacial angle index (°): FMA R, FMA L, SN-GoGn R, SN-GoGn, Y-axis (SGn-FH)R, Y-axis (SGn-FH) L, NS-Co R, NS-Co L, Cranial Flexure (NS-Ba), Gonial Angle(Co-Go-Me) R, Gonial Angle (Co-Go-Me) L, Facial Angle (FH-NPog) R, Facial Angle (FH-NPog) L, AB to Facial Plane, Convexity (NA-APog).

[0043] Screening and construction of model variables All data were included in the dataset, and the dataset was randomly divided into training set and validation set according to the ratio of 7:3.

[0044] Based on R language, univariate regression was used to screen predictive variables, and predictive variables with P < 0.1 were screened for multivariate logistic regression to construct a prediction model for OSA in children. The clinical prediction model for OSA in children was fitted based on multivariate logistic regression, and the Nomogram of the prediction model was drawn, as shown in the figure below. Figure 5 As shown in the figure. Based on the contribution of each influencing factor to the outcome variable (the size of the regression coefficient), each value level of each influencing factor is assigned a score. The individual scores are then summed to obtain a total score. Finally, through the functional conversion relationship between the total score and the probability of the outcome event, the predicted value of the individual outcome event is calculated. This makes the prediction model results more readable and facilitates patient evaluation.

[0045] Model validation: draw ROC curve, such as Figure 6 As shown in the figure, the area under the ROC curve (AUC) is calculated to obtain the C-statistic. When the C-statistic is greater than 0.5, it proves that the prediction model has predictive ability, and the closer the statistic is to 1, the better the discrimination of the model.

[0046] Channel 2: A three-dimensional convolutional neural network model for diagnosing and grading OSA in children based on deep learning. The three-dimensional convolutional neural network model includes three parts: algorithm selection, data processing, and model performance evaluation: Analysis Algorithm Selection: CBCT data is a dense 3D voxel dataset. 3D Convolutional Neural Networks (CNNs) are an effective tool for processing this type of 3D data. 3D CNNs can automatically learn features from 3D medical images, eliminating the need for tedious manual feature design. They are also capable of effectively modeling complex 3D structures, making them widely used in tasks such as distinguishing normal from diseased tissue and performing regional segmentation.

[0047] In view of the above characteristics, this embodiment selects a 3D ResNet network based on 3D CNN to classify and identify the severity of sleep breathing abnormalities in children using CBCT images with OSA (obstructive sleep apnea) symptoms. Figure 7 A simplified architecture diagram of the 3D ResNet network is shown.

[0048] Compared to other CNN architectures, the core idea of ​​the 3D ResNet network lies in the introduction of the identity shortcut connection structure. This structure enables the network to skip one or more layers, successfully solving the problem of increasing training error rates caused by increasing network depth.

[0049] In order to meet the requirements of data privacy and the algorithm itself, the dataset was processed in three steps: Data cleaning and preprocessing: Convert the original DICOM data into a 3D array without patient, hospital, and equipment information.

[0050] Equally spaced sampling: Due to the high resolution of CBCT, in order to adapt to the input resolution of the 3D ResNet network and reduce the GPU memory usage of training, the original DICOM resolution is re-sampled into 96x96x96 cubes through uniform scaling and padding.

[0051] It should be noted that based on the differences in clinical parameters, the focus of the disease can be concentrated on the distance, angle, proportion between bony structures, as well as the size and proportion of soft tissue and air space.

[0052] The CT value of human tissue ranges from -1000 to +1000, with a total of 2000 scales. The lowest CT value of air is approximately -1000 HU. The window width is the range of CT values ​​used when displaying an image. Tissue structures within this range are divided into multiple levels (grayscale) from white to black according to their density. The window level is the anchor point of the window technique, representing the center of the window width and the center of the grayscale display.

[0053] In order to improve the detail contrast of the head and face area, the window width and window position of [-1000, 1000] were set to transform the CBCT pixel values ​​into the range of [0, 255]. The pixels below -1000 were set to 0, and the pixels above 1000 were set to 255.

[0054] Soft tissue window: window width 200~350, window position 30~50; bone window: window width 1000~1500, window position 250~500.

[0055] Dataset division: Currently, 113 valid data sets are used for display, and the AHI values ​​are converted into three categories: asymptomatic, mild, and moderate to severe according to the international standards for the interpretation of PSA in children. Figure 8 The dataset distribution shows that 30.1% of cases are asymptomatic, 56.6% are mild, and 13.3% are moderate or severe. A holdout method (4:1) was used to allocate the training and test sets.

[0056] Model Performance Selection and Evaluation: The basic residual block of a 3D ResNet network consists of two 3x3 convolutional layers, with identity mapping connecting the input directly to the output. This design principle allows the network to retain input information even at deeper layers, preventing the vanishing gradient problem.

[0057] ResNets of different depths: ResNets of different depths can be constructed according to task requirements and computing resource limitations. Taking into account the high complexity of 3D data, this embodiment compares the training effects of Resnet 50 and the deeper Resnet 101 model, and finds that the accuracy of the trained 3D Resnet50 network architecture is always between 80% and 85%, while the training effect of the 3DResnet101 network architecture is better, with an accuracy of more than 90%. The best training result achieved an accuracy of 91.3%, a precision of 93.5%, a recall rate of 93.5%, and an F1 value of 93.5%. The AUCs for the three categories of asymptomatic, mild, and moderate to severe were calculated to be 0.942, 0.944, and 1.0, respectively. Figure 9 Shows the ROC curve of the optimal 3D Resnet prediction model.

[0058] Decision-making system: Based on a dual-channel clinical and imaging verification system for pediatric OSA, disease diagnosis and prediction results are obtained. Based on the results of this model, the need for PSG examinations can be reduced or postponed to a certain extent. This allows for large-scale disease screening and pre- and post-treatment comparative studies in a rapid and widely practical manner.

[0059] When both dual-channel model predictions are negative, it is recommended to consider postponing PSG in combination with clinical signs while ensuring the accuracy and quality of the imaging data. When both dual-channel model predictions are positive, the patient is considered high-risk and, if interventional measures are needed, a portable sleep monitoring device can be used for further evaluation or PSG can be used for confirmation. When the clinical prediction model gives a negative result and the imaging prediction model gives a positive result, it is recommended to verify the imaging data quality and measurement positioning index; When the result of the clinical medicine prediction model is yes and the imaging prediction model is no, and the imaging data is accurate and of qualified quality, the patient is a low-risk population and it is recommended to select further screening methods as needed based on clinical signs.

[0060] Example 3 The parts not described in detail in this embodiment are as shown in Example 1. This embodiment provides a children's sleep breathing abnormality risk assessment system based on CBCT dual-channel analysis, including a head position correction and image parameter extraction module, a logistic regression prediction modeling module, an image data cube generation module, an image deep learning analysis module and a collaborative analysis module. The modules are connected by wired and / or wireless connections.

[0061] The head position correction and image parameter extraction module performs head position correction and three-dimensional positioning tracing on the original CBCT image data to obtain a maxillofacial image parameter set, which includes sagittal area and volume for different sections of the upper airway, as well as maxillofacial multi-dimensional angle parameters and multi-dimensional distance parameters; A logistic regression prediction modeling module is used to construct a logistic regression clinical prediction model for obstructive sleep apnea in children based on the maxillofacial imaging parameter set. The logistic regression clinical prediction model is generated based on univariate regression screening and multivariate logistic regression modeling, and is used to output a risk prediction result for obstructive sleep apnea in children. An image data cube generation module is used to perform equal-interval sampling and pixel value normalization processing on the CBCT raw image data to generate an equal-dimensional multi-channel three-dimensional image data cube; an image deep learning analysis module, which uses a three-dimensional convolutional neural network model with the three-dimensional image data cube as input to obtain a deep learning classification result of images of obstructive sleep apnea in children, wherein the three-dimensional convolutional neural network model is a 3D ResNet network with an identical shortcut connection structure; The collaborative analysis module collaboratively analyzes the output results of the logistic regression clinical prediction model with the image deep learning classification results of the three-dimensional convolutional neural network model to obtain the risk classification results of obstructive sleep apnea in children.

[0062] The step of obtaining the maxillofacial image parameter set comprises: Correct the head position of the CBCT original image data based on standard anatomical landmarks; Based on the corrected CBCT original image data, the sagittal areas and volumes of the nasopharynx, velopharynx, glossopharynx, and hypopharynx were extracted. Measure the multi-dimensional angle parameters and multi-dimensional distance parameters of the mandible, hyoid bone and craniomaxillofacial skeleton in a three-dimensional coordinate system; The above area, volume, angle and distance parameters are combined to form the maxillofacial imaging parameter set.

[0063] The steps of constructing the logistic regression clinical prediction model include: The maxillofacial imaging parameter set was randomly divided into a training set and a validation set; The univariate regression method was used to select parameters with P values ​​less than or equal to 0.1 as modeling variables; Multivariate logistic regression modeling was performed based on the screened modeling variables to construct an obstructive sleep apnea risk prediction model; A nomogram score chart is drawn based on the model regression coefficient to achieve individual risk scoring and prediction for children.

[0064] The step of generating an equal-dimensional three-dimensional image data cube for deep learning comprises: The CBCT original image data is sampled at equal intervals to generate a three-dimensional data cube of preset size; According to the set window width and window position, the image pixel value is normalized to the preset grayscale value range in the preset pixel interval; The normalized pixel matrix is ​​combined into a three-dimensional image data cube in the form of a multi-channel input, where each channel of the three-dimensional image data cube corresponds to normalized two-dimensional slice data in axial, sagittal and coronal dimensions.

[0065] The steps of obtaining the image deep learning classification results include: Building a 3D ResNet network based on an identical shortcut connection structure, wherein the 3D ResNet network includes a residual module and a global average pooling layer; Input the three-dimensional image data cube and its corresponding obstructive sleep apnea label, train the 3DResNet network, and output corresponding classification features after multi-layer convolution and residual learning; Generating obstructive sleep apnea classification categories using the multi-classification probability results output by the 3D ResNet network, performing supervised training based on the classification features and preset classification labels, and optimizing weight parameters of the 3D ResNet network; The classification accuracy, precision, and recall of the 3D ResNet network were evaluated based on the training set and the validation set.

[0066] The steps of training the 3D ResNet network include: Two network architectures, 3D ResNet50 and 3D ResNet101, were constructed respectively, and the cross entropy loss function and Adam optimizer were used to jointly control the training process of the 3D ResNet network; The 3D ResNet50 and 3D ResNet101 are trained respectively using the training set; The classification performance of the two networks was compared on the validation set, and the network with the highest classification accuracy on the validation set was selected as the final three-dimensional convolutional neural network model.

[0067] The logic for generating risk classification results: The risk prediction result of the logistic regression clinical prediction model is used as the first risk factor, and the image deep learning classification result is used as the second risk factor; generating a risk classification result of the obstructive sleep apnea in children by fusing the first risk factor and the second risk factor based on a weighted collaborative decision-making mechanism; Among them: the weights of the first risk factor and the second risk factor are obtained by dynamic adjustment based on the classification accuracy of historical clinical samples.

[0068] The above description is merely a specific embodiment of the present application, but the scope of protection of the present application is not limited thereto. Any changes or substitutions that can be easily conceived by a person skilled in the art within the technical scope disclosed in this application should be included in the scope of protection of this application. Therefore, the scope of protection of this application should be based on the scope of protection of the claims.

Claims

1. A risk assessment method for sleep apnea in children based on CBCT dual-channel analysis, characterized by: The following steps are involved: S101: performing head position correction and three-dimensional positioning tracing processing on the CBCT original image data to obtain a maxillofacial image parameter set, wherein the maxillofacial image parameter set includes sagittal area and volume of different upper airway segments, as well as maxillofacial multi-dimensional angle parameters and multi-dimensional distance parameters; S102: Based on the maxillofacial imaging parameter set, a logistic regression clinical prediction model for obstructive sleep apnea in children was constructed using univariate regression screening and multivariate logistic regression modeling; S103: Performing equal-interval sampling and pixel value normalization on the CBCT raw image data to generate an equal-dimensional three-dimensional image data cube for deep learning; S104: Obtaining a deep learning classification result of images of obstructive sleep apnea in children using a three-dimensional convolutional neural network model that uses the three-dimensional image data cube as input, wherein the three-dimensional convolutional neural network model is a 3D ResNet network with an identical shortcut connection structure; S105: Collaboratively analyze the output results of the logistic regression clinical prediction model and the image deep learning classification results of the three-dimensional convolutional neural network model to obtain a risk classification result for obstructive sleep apnea in children.

2. The method for risk assessment of sleep apnea in children based on CBCT dual-channel analysis according to claim 1, characterized in that: The steps of obtaining the maxillofacial image parameter set include: Correct the head position of the CBCT original image data based on standard anatomical landmarks; The sagittal area and volume of different upper airway segments were extracted based on the corrected CBCT original image data; the upper airway segments included the nasopharyngeal area, velopharyngeal area, glossopharyngeal area, and hypopharyngeal area. Measure the multi-dimensional angle parameters and multi-dimensional distance parameters of the maxillofacial region in a three-dimensional coordinate system; the maxillofacial region includes the mandible, hyoid bone and craniomaxillofacial skeleton; The sagittal area, volume, and maxillofacial multidimensional angle parameters and multidimensional distance parameters of different upper airway segments are combined to form the maxillofacial imaging parameter set.

3. The method for risk assessment of sleep apnea in children based on CBCT dual-channel analysis according to claim 1, characterized in that: The steps of constructing the logistic regression clinical prediction model include: The maxillofacial imaging parameter set was randomly divided into a training set and a validation set; The univariate regression method was used to select parameters with P values ​​less than or equal to 0.1 as modeling variables; Multivariate logistic regression modeling was performed based on the screened modeling variables to construct an obstructive sleep apnea risk prediction model; A nomogram score chart is drawn based on the model regression coefficient to achieve individual risk scoring and prediction for children.

4. The method for risk assessment of sleep apnea in children based on CBCT dual-channel analysis according to claim 1, characterized in that: The step of generating an equal-dimensional three-dimensional image data cube for deep learning comprises: The CBCT original image data is sampled at equal intervals to generate a three-dimensional data cube of preset size; According to the set window width and window position, the image pixel value is normalized to the preset grayscale value range in the preset pixel interval; The normalized pixel matrix is ​​combined into a three-dimensional image data cube in the form of a multi-channel input, where each channel of the three-dimensional image data cube corresponds to normalized two-dimensional slice data in axial, sagittal and coronal dimensions.

5. The method for risk assessment of sleep apnea in children based on CBCT dual-channel analysis according to claim 1, characterized in that: The steps of obtaining the image deep learning classification results include: Building a 3D ResNet network based on an identical shortcut connection structure, wherein the 3D ResNet network includes a residual module and a global average pooling layer; Input the three-dimensional image data cube and its corresponding obstructive sleep apnea label, train the 3DResNet network, and output corresponding classification features after multi-layer convolution and residual learning; Generating obstructive sleep apnea classification categories using the multi-classification probability results output by the 3D ResNet network, performing supervised training based on the classification features and preset classification labels, and optimizing weight parameters of the 3D ResNet network; The classification accuracy, precision, and recall of the 3D ResNet network were evaluated based on the training set and the validation set.

6. The method for risk assessment of sleep apnea in children based on CBCT dual-channel analysis according to claim 5, characterized in that: The steps of training the 3D ResNet network include: Two network architectures, 3D ResNet50 and 3D ResNet101, were constructed respectively, and the cross entropy loss function and Adam optimizer were used to jointly control the training process of the 3D ResNet network; The 3D ResNet50 and 3D ResNet101 are trained respectively using the training set; The classification performance of the two networks was compared on the validation set, and the network with the highest classification accuracy on the validation set was selected as the final three-dimensional convolutional neural network model.

7. The method for risk assessment of sleep apnea in children based on CBCT dual-channel analysis according to claim 1, characterized in that: The logic for generating risk classification results: The risk prediction result of the logistic regression clinical prediction model is used as the first risk factor, and the image deep learning classification result is used as the second risk factor; generating a risk classification result of the obstructive sleep apnea in children by fusing the first risk factor and the second risk factor based on a weighted collaborative decision-making mechanism; Among them: the weights of the first risk factor and the second risk factor are obtained by dynamic adjustment based on the classification accuracy of historical clinical samples.

8. A system for assessing the risk of dysrespiration in children based on CBCT dual-channel analysis, based on the implementation of the method for assessing the risk of dysrespiration in children based on CBCT dual-channel analysis according to any one of claims 1 to 7, characterized in that: It includes a head position correction and image parameter extraction module, a logistic regression prediction modeling module, an image data cube generation module, an image deep learning analysis module, and a collaborative analysis module, and each module is connected via wired and / or wireless connections; The head position correction and image parameter extraction module performs head position correction and three-dimensional positioning tracing on the original CBCT image data to obtain a maxillofacial image parameter set, which includes sagittal area and volume for different sections of the upper airway, as well as maxillofacial multi-dimensional angle parameters and multi-dimensional distance parameters; A logistic regression prediction modeling module is used to construct a logistic regression clinical prediction model for obstructive sleep apnea in children based on the maxillofacial imaging parameter set. The logistic regression clinical prediction model is generated based on univariate regression screening and multivariate logistic regression modeling, and is used to output a risk prediction result for obstructive sleep apnea in children. An image data cube generation module is used to perform equal-interval sampling and pixel value normalization processing on the CBCT raw image data to generate an equal-dimensional multi-channel three-dimensional image data cube; an image deep learning analysis module, which uses a three-dimensional convolutional neural network model with the three-dimensional image data cube as input to obtain a deep learning classification result of images of obstructive sleep apnea in children, wherein the three-dimensional convolutional neural network model is a 3D ResNet network with an identical shortcut connection structure; The collaborative analysis module collaboratively analyzes the output results of the logistic regression clinical prediction model with the image deep learning classification results of the three-dimensional convolutional neural network model to obtain the risk classification results of obstructive sleep apnea in children.

Images