Child OSA auxiliary diagnosis model construction method and system based on craniofacial features

Abstract

The invention discloses a construction method and system of a child OSA auxiliary diagnosis model based on craniofacial features, which utilizes OSA related symptoms and clinical signs of subjects of all OSA child patients. According to the invention, the basic information, birth mode, birth weight, full-term infant, feeding mode and other information of all subjects including OSA child patients and non-OSA children, as well as craniofacial features and nasopharynx side X-film parameters are fully utilized, so that a plurality of information can be integrated while the low-radiation imaging technology is utilized; the optimal model is constructed and screened based on the artificial intelligence technology, and an auxiliary diagnosis system is further realized, so that early-stage rapid triage of OSA children is realized, and guidance is provided for orthodontic treatment of OSA patients to improve the craniofacial structure.

Description

Technical Field

[0001] This invention belongs to the field of combining artificial intelligence with obstructive sleep apnea (OSA) in children, and specifically relates to a method and system for constructing an auxiliary diagnostic model for OSA in children based on craniofacial features. Background Technology

[0002] Obstructive sleep apnea (OSA) is a common sleep disorder in children (2-18 years old), with a prevalence of approximately 1%-6%. Compared to adults, children with OSA have unique manifestations, such as hyperactivity, emotional difficulties, declining academic performance, and inattention. Furthermore, childhood OSA can cause a variety of serious complications, leading to cardiovascular disease, growth and developmental disorders, cognitive impairment, craniofacial developmental abnormalities, and secretory otitis media.

[0003] Currently, the diagnosis of childhood OSA relies on polysomnography (PSG). However, PSG testing has many limitations, such as requiring a specialized sleep laboratory, connecting multiple measurement channels, and requiring overnight parental care. In summary, PSG testing is complex, uncomfortable, time-consuming, costly, and often results in poor child cooperation, making it unsuitable for rapid or large-scale screening.

[0004] Today, artificial intelligence, represented by machine learning (ML) and deep learning (DL), is widely used in medical data analysis, but there are no reports of using such technologies to accurately assist in the diagnosis of pediatric OSA.

[0005] Therefore, there is an urgent need in this field to utilize artificial intelligence technology to achieve more accurate solutions for assisted diagnosis of OSA in children.

[0006] The information disclosed in the background section is only intended to enhance the understanding of the background of the present invention, and therefore may contain information that does not constitute prior art known to those skilled in the art. Summary of the Invention

[0007] To address the shortcomings of existing technologies, this invention proposes a method for constructing an auxiliary diagnostic model for childhood OSA based on craniofacial developmental characteristics. Childhood OSA refers to obstructive sleep apnea. The method includes the following steps: S1: To identify all subjects diagnosed with OSA and all subjects without OSA by using the obstructive apnea-hypopnea index (OAHI) after overnight polysomnography (PSG); S2: Obtain OSA-related symptoms and clinical signs from all subjects with OSA, and obtain basic information on all subjects, including those with OSA and those without OSA, such as mode of birth, birth weight, term status, and feeding method. Methods of delivery include: vaginal delivery and cesarean section; Whether an infant is full-term includes: yes or no; Feeding methods include: breastfeeding, bottle feeding, and mixed feeding; S3: Craniofacial features of all subjects were obtained using low-radiation nasopharyngeal lateral radiographs. S4: Based on all the information obtained from S2 to S3, perform univariate logistic regression to screen for variables related to childhood OSA, wherein the number of variables related to childhood OSA is the first number; S5: Based on the first number of variables related to childhood OSA, perform multivariate logistic regression to further screen for independent risk factors of childhood OSA, wherein the number of independent risk factors of childhood OSA is the second number; S6: For all subjects, including children with OSA and children without OSA, training, validation and test sets were randomly assigned. Multiple different artificial intelligence models were trained, taking into account the independent risk factors for OSA in the children, and the performance of each model was compared to find the best model.

[0008] Preferred, Basic information includes: age, gender, and body mass index (BMI).

[0009] Preferred, OSA-related symptoms include: snoring, mouth breathing, and breath-holding.

[0010] Preferred, Craniofacial features include: maximum adenoid width (A); width of the upper airway (N); width of the airway narrowing corresponding to the most prominent adenoid (PAS); radius measurement of tonsil tissue (T); airway width corresponding to the most prominent tonsil (P); angle of the skull base plane (NS-S.Ba); length of the anterior skull base (ASL); length of the middle skull base (MSL); angle between the anterior skull base plane and the mandibular body plane (ASL-Go-Me); mandibular angle (Go.Me-Go.Ar); correlation index between facial height and mandibular position (Go-Me-N.Pog); length of the mandibular body (Go-Me); length of the mandibular ramus (Ar-Go); anteroposterior position of the maxilla relative to the skull base (SNA); anteroposterior position of the mandible relative to the skull base (SNB); relative position difference between the maxilla and mandible (ANB); anteroposterior diameter of the maxilla (ANS-PNS); total anterior facial height (N-Me); upper facial height (N-ANS); posterior facial height (S-Go); hyoid bone to C3 Morphological parameters including vertebral body distance (HC); vertical distance from hyoid bone to mandibular plane (H-MP); soft palate tissue thickness (SPT); soft palate-posterior pharyngeal wall angle (ASP); soft palate length (SPL); soft palate thickness (SPT); width of airway narrowing corresponding to the most prominent adenoid (PAS); and soft palate-posterior pharyngeal wall angle (ASP).

[0011] Preferred, The variables related to childhood OSA include the following 14 variables: age, body mass index (BMI), mouth breathing, tonsil grading, maximum adenoid width (A), A / N ratio (the ratio of the maximum adenoid width (A) to the width of the upper respiratory tract (N), width of the airway narrowing corresponding to the most prominent adenoid (PAS), radius measurement of tonsil tissue (T), anteroposterior position of the maxilla relative to the skull base (SNA), anteroposterior position of the mandible relative to the skull base (SNB), vertical distance from the hyoid bone to the mandibular plane (H-MP), distance from the hyoid bone to the C3 vertebra (HC), soft palate tissue thickness (SPT), and soft palate-posterior pharyngeal wall angle (ASP).

[0012] Preferred, The independent risk factors for childhood OSA include the following nine factors: Age, Body Mass Index (BMI), mouth breathing, tonsil grading, SNA, SNB, H-MP, HC, ASP.

[0013] Preferably, in step S7, The effectiveness of each model is compared by comprehensively evaluating AUC, accuracy, sensitivity, specificity, and F1 score to find the optimal model.

[0014] Furthermore, the present invention also discloses a computer storage medium comprising computer instructions that, when executed on a computer, cause the computer to execute any of the foregoing methods for constructing a child OSA-assisted diagnostic model based on craniofacial features.

[0015] Furthermore, this invention also discloses a system for an auxiliary diagnostic model of pediatric OSA based on craniofacial features, wherein the system comprises: The system includes a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the program to implement any of the aforementioned methods for constructing a child OSA auxiliary diagnostic model based on craniofacial features, and to perform auxiliary diagnosis on suspected OSA patients based on the optimal model.

[0016] Compared with the prior art, the present invention has the following advantages: In addition to utilizing OSA-related symptoms and clinical signs of all OSA patients, this invention also fully leverages basic information, birth method, birth weight, term status, feeding method, craniofacial features, and nasopharyngeal lateral X-ray parameters of all subjects, including both OSA patients and non-OSA children. This allows the invention to integrate a wealth of information using low-radiation imaging technology, enabling the construction and screening of optimal models based on artificial intelligence technology. Furthermore, it realizes an auxiliary diagnostic system, which facilitates early and rapid triage of OSA children and provides guidance for orthodontic treatment of OSA patients to improve craniofacial structure. Attached Figure Description

[0017] Figure 1 This is a flowchart illustrating the method for constructing a child OSA-assisted diagnostic model based on craniofacial features, according to one embodiment of the present invention. Figures 2a to 2d This is a schematic diagram of the confusion matrix and ROC curve of two machine learning models in several embodiments of the present invention, wherein: Figure 2a Confusion matrix of LightGBM model Figure 2b Confusion matrix of XGBoost model Figure 2c ROC curve of the LightGBM model Figure 2d ROC curve of XGBoost model; Figures 3a to 3b This is a schematic diagram of the confusion matrix of the deep learning model LSTM and the ROC curve in the validation set in one embodiment of the present invention, wherein: Figure 3a The confusion matrix of the LSTM model. Figure 3b ROC curve of LSTM model; Figures 4a to 4bThese are two types of SHAP graphs for machine learning in various embodiments of the present invention, wherein: Figure 4a SHAP diagram of LightGBM; Figure 4b XGBoost SHAP diagram. Detailed Implementation

[0018] The following will refer to the appendix. Figures 1 to 4b Specific embodiments of the invention are described in detail below. While specific embodiments of the invention are shown in the accompanying drawings, it should be understood that the invention can be implemented in various forms and should not be limited to the embodiments set forth herein. Rather, these embodiments are provided to enable a more thorough understanding of the invention and to fully convey the scope of the invention to those skilled in the art.

[0019] It should be noted that certain terms are used in the specification and claims to refer to specific components. Those skilled in the art will understand that different terms may be used to refer to the same component. This specification and claims do not distinguish components based on differences in terminology, but rather on differences in function. The terms "comprising" or "including" used throughout the specification and claims are open-ended and should be interpreted as "comprising but not limited to." The following descriptions are preferred embodiments for carrying out the invention; however, these descriptions are for the purpose of understanding the general principles of the specification and are not intended to limit the scope of the invention. The scope of protection of this invention is determined by the appended claims.

[0020] To facilitate understanding of the embodiments of the present invention, further explanations and descriptions will be provided below with reference to the accompanying drawings and specific embodiments. The accompanying drawings do not constitute a limitation on the embodiments of the present invention.

[0021] In one embodiment, the present invention proposes a method for constructing a diagnostic model for childhood OSA based on craniofacial developmental features, wherein childhood OSA refers to obstructive sleep apnea, and the method includes the following steps: S1: To identify all subjects diagnosed with OSA and all subjects without OSA by using the obstructive apnea-hypopnea index (OAHI) after overnight polysomnography (PSG); S2: Obtain OSA-related symptoms and clinical signs from all subjects with OSA, and obtain basic information on all subjects, including those with OSA and those without OSA, such as mode of birth, birth weight, term status, and feeding method. Methods of delivery include: vaginal delivery and cesarean section; Whether an infant is full-term includes: yes or no; Feeding methods include: breastfeeding, bottle feeding, and mixed feeding; S3: Craniofacial features of all subjects were obtained using low-radiation nasopharyngeal lateral radiographs. S4: Based on all the information obtained from S2 to S3, perform univariate logistic regression to screen for variables related to childhood OSA, wherein the number of variables related to childhood OSA is the first number; S5: Based on the first number of variables related to childhood OSA, perform multivariate logistic regression to further screen for independent risk factors for childhood OSA, wherein the number of independent risk factors for childhood OSA is the second number; S6: For all subjects, including children with OSA and non-OSA children, training, validation and test sets were randomly assigned. Multiple different artificial intelligence models were trained, taking into account the independent risk factors for OSA in the children, and the performance of each model was compared to find the best model.

[0022] The above embodiments fully illustrate that, in addition to utilizing the OSA-related symptoms and clinical signs of all OSA patients, this invention also fully utilizes the basic information, birth method, birth weight, whether the infant is full-term, feeding method, and other information of all subjects, including OSA patients and non-OSA children, as well as craniofacial features. This allows the invention to integrate a wealth of information while utilizing low-radiation nasopharyngeal lateral radiographs, so as to construct and screen the best model based on artificial intelligence technology. This facilitates the realization of an auxiliary diagnostic system, which not only helps in the early and rapid triage of OSA children but also provides guidance for orthodontic treatment of OSA patients to improve craniofacial structure.

[0023] In another embodiment, in step S1, the subjects are classified according to the following criteria: For children without OSA, OAHI < 1 time / hour; for children with OSA, OAHI ≥ 1 time / hour.

[0024] In another embodiment, Basic information includes: age, gender, and body mass index (BMI).

[0025] In another embodiment, OSA-related symptoms include: snoring, mouth breathing, and breath-holding.

[0026] In another embodiment, Craniofacial features, including: maximum adenoid width (A); width of the upper airway (N); width of the airway narrowing corresponding to the most prominent adenoid (PAS); radius measurement of tonsil tissue (T); airway width corresponding to the most prominent tonsil (P); angle of the skull base plane (NS-S.Ba); length of the anterior skull base (ASL); length of the middle skull base (MSL); angle between the anterior skull base plane and the mandibular body plane (ASL-Go-Me); mandibular angle (Go.Me-Go.Ar); correlation index between facial height and mandibular position (Go-Me-N.Pog); length of the mandibular body (Go-Me); length of the mandibular ramus (Ar-Go); anteroposterior position of the maxilla relative to the skull base (SNA); anteroposterior position of the mandible relative to the skull base (SNB); relative position difference between the maxilla and mandible (ANB); anteroposterior diameter of the maxilla (ANS-PNS); total anterior facial height (N-Me); upper facial height (N-ANS); posterior facial height (S-Go); hyoid bone to C3 Vertebral distance (HC); vertical distance from the hyoid bone to the mandibular plane (H-MP); soft palate tissue thickness (SPT); soft palate-posterior pharyngeal wall angle (ASP); soft palate length (SPL).

[0027] In another embodiment, The variables related to childhood OSA include the following 14 variables: age, body mass index (BMI), mouth breathing, tonsil grading, maximum adenoid width (A), A / N ratio (the ratio of the maximum adenoid width (A) to the width of the upper respiratory tract (N), width of the airway narrowing corresponding to the most prominent adenoid (PAS), radius measurement of tonsil tissue (T), anteroposterior position of the maxilla relative to the skull base (SNA), anteroposterior position of the mandible relative to the skull base (SNB), vertical distance from the hyoid bone to the mandibular plane (H-MP), distance from the hyoid bone to the C3 vertebra (HC), soft palate tissue thickness (SPT), and soft palate-posterior pharyngeal wall angle (ASP).

[0028] In another embodiment, The independent risk factors for childhood OSA include the following nine factors: Age, Body Mass Index (BMI), mouth breathing, tonsil grading, SNA, SNB, H-MP, HC, ASP.

[0029] In another embodiment, in step S7, The effectiveness of each model is compared by comprehensively evaluating AUC, accuracy, sensitivity, specificity, and F1 score to find the optimal model.

[0030] In another embodiment, Specifically, the following craniofacial features were obtained through a lateral nasopharyngeal X-ray: A: Maximum width of the adenoids; N: Width of the upper respiratory tract; PAS: Width of the airway narrowing corresponding to the most prominent part of the adenoids; T: Radius measurement of the tonsil tissue (usually calculated from cross-sectional images); P: Airway width corresponding to the most prominent part of the tonsils; NS-S.Ba: Skull base plane angle (e.g., the angle formed by connecting the nasal root point, the center point of the sella turcica, and the skull base point); ASL: Anterior skull base length (usually the distance from N to S); MSL: Middle skull base length (distance from the center point of the sella turcica to the clivus); ASL-Go-Me: Angle between the anterior skull base plane and the mandibular body plane; Go.Me-Go.Ar: Mandibular angle (the angle between the mandibular body and the mandibular ramus); Go-Me-N.Pog: Correlation index between facial height and mandibular position (e.g., assessing the anteroposterior position of the mandible through the most prominent point of the anterior border of the chin); Go-Me: Mandibular body length (distance from the angle of the mandible to the submental point); Ar-Go: length of the mandibular ramus (distance from the articular process to the angle of the mandible); SNA: anteroposterior position of the maxilla relative to the skull base; SNB: anteroposterior position of the mandible relative to the skull base; ANB: relative position difference between the maxilla and mandible (reflecting the type of skeletal malocclusion); ANS-PNS: anteroposterior diameter of the maxilla (reflecting the longitudinal development of the maxilla); N-Me: total anterior facial height (vertical distance from the root of the nose to the submental point); N-ANS: upper facial height (vertical distance from the root of the nose to the anterior nasal spine); S-Go: posterior facial height (vertical distance from the sella turcica to the angle of the mandible); HC: distance from the hyoid bone to the C3 vertebral body; H-MP: vertical distance from the hyoid bone to the mandibular plane; SPT: soft palate tissue thickness; ASP: soft palate-posterior pharyngeal wall angle; SPL: soft palate length.

[0031] In another embodiment, The lateral nasopharyngeal radiograph examination is as follows: The Osofos X-ray OC-100 (Imaging GmbH, Finland) acquires lateral cephalometric radiographs in a standardized manner, with the distance between the measuring instrument and the head fixed at 150 cm. The subject stands naturally with their line of sight parallel to the ground, and the images are taken while the subject is not speaking or swallowing.

[0032] In another embodiment, The craniofacial feature measurements were performed by two experienced radiologists, following the procedure outlined below: Researcher A manually traced the image outline using 0.03-inch (approximately 0.76 mm) thick acetate tracing paper, and Researcher B independently reviewed it; To assess intra-observer error, 10 images were randomly selected two weeks later and measured a second time by the same researcher. If the two measurements differed significantly (linear parameter deviation > 0.5 mm or angular parameter deviation > 0.5 degrees), a third measurement was required, and the average of the two results closest to the final result was taken as the valid data.

[0033] It should be noted that due to the geometric characteristics of X-ray projection, there is a magnification effect of approximately 7-8% on linear parameters in the image. To correct for this deviation, the actual length is converted using the image's built-in scale during measurement. The calculation formula is: True length = Image measurement value × (Actual scale length / Image scale display length).

[0034] In another embodiment, All participants (n = 815) visited the Department of Otolaryngology-Head and Neck Surgery at the Second Affiliated Hospital of Xi'an Jiaotong University between June 2021 and June 2024. Written informed consent forms, demographic information, clinical symptoms, and lateral nasopharyngeal X-rays were obtained from medical records signed by the children's legal guardians. Inclusion criteria include: Children aged 2-12 years with suspected OSA; Underwent PSG testing; I have undergone lateral head radiographs. Exclusion criteria include: Baseline data is missing; Children with craniofacial diseases; Children who have received OSA treatment (including drug treatment, surgical treatment, and CPAP treatment); Those with acute or chronic inflammatory diseases, liver disease, or kidney disease; Comorbidities such as Down syndrome, Kruzon syndrome, and Pierre Robin sequence.

[0035] Ultimately, all participants included 456 children with OSA (55.60%) and 359 children without OSA (44.05%), totaling 815 participants. Among them, 50.06% (408) were boys and 49.94% (407) were girls, with no significant difference between the two groups in terms of gender.

[0036] In another embodiment, the training set, validation set, and test set are randomly divided, for example: All subjects were randomly divided into three groups in an 8:1:1 ratio: a training set for initial model training, a validation set for model tuning and hyperparameter selection, and a test set for final evaluation of the model's generalization ability.

[0037] In another embodiment, Two existing machine learning models were used as alternative models.

[0038] For example, two machine learning models include Lightweight Gradient Boosting Machine (LightGBM) and Extreme Gradient Boosting Tree (XGBoost).

[0039] In another embodiment, One existing deep learning model is used as an alternative model.

[0040] For example, deep learning models include Long Short-Term Memory (LSTM) networks.

[0041] It is understood that this invention is not limited to existing artificial intelligence models or artificial intelligence models derived from existing technologies through modification and recombination. This is because the technical problem this invention aims to solve is to propose a method for constructing a child OSA-assisted diagnostic model based on craniofacial developmental characteristics, and to propose technical means for selecting the optimal model within this construction method.

[0042] In another embodiment, exemplaryly, based on all information of all subjects and the selection of 3 alternative models: Table 1. Comparison of demographic characteristics between children with and without OSA. The craniofacial features were compared between the OSA group and the non-OSA group. Table 2 shows the comparison of craniofacial features between children with and without OSA. Compared with children with OSA, children with OSA had larger values ​​for A (14.89 vs. 16.38), A / N ratio (0.71 vs. 0.77), T (10.33 vs. 11.21), H-MP (11.28 vs. 12.88), and SPT (8.26 vs. 8.52), and smaller values ​​for PAS (5.98 vs. 4.79), SNA (89.10 vs. 87.05), SNB (82.40 vs. 81.10), HC (31.43 vs. 30.34), and ASP (144.34 vs. 142.08). p <0.05). There were no statistically significant differences between the two groups in N, P, T / P, NS-S.Ba, ASL, MSL, ASL-Go-Me, Go.ME-Go.Ar, Go.Me-N.Pog, Go-Me, Ar-Go, ANB, ANS-PNS, N-Me, N-ANS, S-Go, and SPL.

[0043] Table 2 Comparison of craniofacial features between children with OSA and those without OSA. First, univariate logistic regression was used to screen for 14 variables associated with childhood OSA (Table 3), including: age, body mass index (BMI), mouth breathing, tonsil grading, A, A / N, PAS, T, SNA, SNB, H-MP, HC, SPT, and ASP (where p < 0.01 for age, T, and SPT, and p < 0.001 for the others). After multivariate logistic regression analysis, age, BMI, mouth breathing, tonsil grading, SNA, SNB, H-MP, HC, and ASP were identified as independent risk factors for childhood OSA (Table 4).

[0044] Table 3. Univariate Logistic Regression Screening of Risk Factors for Childhood OSA Table 4. Multivariate Logistic Regression Screening for Risk Factors of Childhood OSA After selecting three existing technologies—LightGBM, XGBoost, and LSTM—we used logistic regression analysis to include variables with p < 0.05 in LightGBM, XGBoost, and LSTM, and found that all of these models showed good performance. Figures 2a to 2d The confusion matrices and ROC curves of two machine learning models (LightGBM and XGBoost) are shown. The AUC values ​​of LightGBM and XGBoost are 0.941 and 0.938, respectively. Figures 3a to 3b The confusion matrix and ROC curve of the deep learning model LSTM are shown, with an AUC value of 0.809.

[0045] Among the three models, LightGBM had the highest AUC, with Acc, Sen, Spec, and F1 scores of 0.878, 0.867, 0.892, and 0.877, respectively. XGBoost and LSTM had the next highest AUCs (Table 5). The Acc, Sen, Spec, and F1 scores for Extreme Gradient Boosting Tree were 0.902, 0.889, 0.919, and 0.902, respectively, while those for Long Short-Term Memory Network were 0.724, 0.773, 0.783, and 0.700 (Table 5).

[0046] Table 5 Comparison of the performance of the three models We visualized the risk factors in two machine learning models using Shapley and Interpretation (SHAP), and the relative importance of risk factors was largely consistent in both models. We found that BMI played the greatest role in diagnosing childhood OSA, followed by SNA, SNB, HC, SAP, H-MP, and mouth breathing. Figures 4a to 4b ).

[0047] In another embodiment, the present invention also discloses a computer storage medium comprising computer instructions that, when executed on a computer, cause the computer to perform any of the methods described above.

[0048] In another embodiment, the present invention also discloses a system for a pediatric OSA-assisted diagnostic model based on craniofacial features, wherein the system comprises: The system includes a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the program to implement any of the aforementioned methods for constructing a child OSA auxiliary diagnostic model based on craniofacial features, and to perform auxiliary diagnosis on suspected OSA patients based on the optimal model.

[0049] The applicant has provided a detailed description of the embodiments of the present invention in conjunction with the accompanying drawings. However, those skilled in the art should understand that the above embodiments are merely preferred examples of the present invention and are not limited to the specific embodiments described above. The detailed description is intended to help readers better understand the spirit of the present invention and is not intended to limit the scope of protection of the present invention. On the contrary, any improvements or modifications made based on the inventive spirit of the present invention should be included within the scope of protection of the present invention.

Claims

1. A method for constructing an auxiliary diagnostic model for childhood OSA based on craniofacial developmental features, wherein childhood OSA refers to obstructive sleep apnea. Includes the following steps: S1: To identify all subjects diagnosed with OSA and all subjects without OSA by using the obstructive apnea-hypopnea index (OAHI) after overnight polysomnography (PSG); S2: Obtain OSA-related symptoms and clinical signs of all subjects with OSA, as well as basic information of all subjects, including those with OSA and those without OSA, such as mode of birth, birth weight, whether they are full-term, and feeding method. S3: Craniofacial features of all subjects were obtained using low-radiation nasopharyngeal lateral radiographs. S4: Based on all the information obtained in steps S2 to S3, perform univariate logistic regression to screen for variables related to childhood OSA, wherein the number of variables related to childhood OSA is a first number; S5: Based on the first number of variables related to childhood OSA, perform multivariate logistic regression to further screen for independent risk factors for childhood OSA, wherein the number of independent risk factors for childhood OSA is the second number; S6: For all subjects, including children with OSA and non-OSA children, training, validation and test sets were randomly assigned. Multiple different artificial intelligence models were trained, taking into account the independent risk factors for OSA in the children, and the performance of each model was compared to find the best model.

2. The construction method according to claim 1, wherein, Preferred, Basic information includes: age, gender, and body mass index (BMI).

3. The construction method according to claim 1, wherein, OSA-related symptoms include: snoring, mouth breathing, and breath-holding.

4. The construction method according to claim 1, wherein, The craniofacial features include: maximum width of the adenoids (A); width of the upper respiratory tract (N).

5. The construction method according to claim 1, wherein, The variables related to childhood OSA include the following: age, body mass index (BMI), mouth breathing, and tonsil grading.

6. The construction method according to claim 1, wherein, The independent risk factors for childhood OSA include the following factors: SNA, SNB, H-MP, HC, ASP.

7. The construction method according to claim 1, wherein, In step S7, The effectiveness of each model is compared by comprehensively evaluating AUC, accuracy, sensitivity, specificity, and F1 score to find the optimal model.

8. A computer storage medium, wherein, The storage medium includes computer instructions that, when executed on a computer, cause the computer to perform the method for constructing a child OSA-assisted diagnostic model based on craniofacial features as described in any one of claims 1 to 7.

9. A system for an auxiliary diagnostic model of pediatric OSA based on craniofacial features, wherein, The system includes: A memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor, when executing the program, implements the method for constructing a child OSA auxiliary diagnostic model based on craniofacial features as described in any one of claims 1 to 7, and performs auxiliary diagnosis on suspected OSA patients based on the optimal model.

Images