Machine Learning-Based Prediction of Pediatric Scoliosis Using Foot Pressure and Body Angles

Machine Learning-Based Prediction of Pediatric Scoliosis Using Foot Pressure and Body Angles | Research Square window.SnipcartSettings = { analytics: { enabled: false } }; (function() { var accessVector = localStorage.getItem('access_vector') || ''; window.dataLayer = window.dataLayer || []; if (accessVector) { window.dataLayer.push({ user: { profile: { profileInfo: { snid: accessVector } } } }); } })(); (function(w,d,s,l,i){w[l]=w[l]||[];w[l].push({'gtm.start':new Date().getTime(),event:'gtm.js'});var f=d.getElementsByTagName(s)[0],j=d.createElement(s),dl=l!='dataLayer'?'&l='+l:'';j.async=true;j.src='https://www.googletagmanager.com/gtm.js?id='+i+dl;f.parentNode.insertBefore(j,f);})(window,document,'script','dataLayer','GTM-K279D39R'); Browse Preprints In Review Journals COVID-19 Preprints AJE Video Bytes Research Tools Research Promotion AJE Professional Editing AJE Rubriq About Preprint Platform In Review Editorial Policies Our Team Advisory Board Help Center Sign In Submit a Preprint Cite Share Download PDF Research Article Machine Learning-Based Prediction of Pediatric Scoliosis Using Foot Pressure and Body Angles QiGang Chen, YanPing Song, XingYu Chen, RanTing Tao, Xin Li, Ang Ding, and 1 more This is a preprint; it has not been peer reviewed by a journal. https://doi.org/ 10.21203/rs.3.rs-6224546/v1 This work is licensed under a CC BY 4.0 License Status: Published Journal Publication published 12 Nov, 2025 Read the published version in European Spine Journal → Version 1 posted 7 You are reading this latest preprint version Abstract Purpose: This study aims to analyze the key features of scoliosis based on children's foot pressure and body angle data and to develop a predictive model to support early screening for scoliosis in ediatric populations. Methods: Data were collected from 5970 primary school students in Kunming. After excluding incomplete and anomalous samples, 3723 students were included. Machine learning algorithms, including Logistic Regression(LR), Support Vector Machine(SVM), Random Forest(RF), and LightGBM, were employed to construct classification models. Model performance was evaluated using accuracy, ROC curves, AUC, and other metrics. The optimal model was selected based on model performance comparisons, and SHAP (Shapley Additive exPlanations) interpretability analysis was conducted. Results: Statistical analysis revealed significant differences between children with scoliosis and healthy controls in toe out angle and Center of Pressure (COP) sway area. The LightGBM model outperformed others with an AUC of 0.971, significantly higher than LR (0.566), SVM (0.723), and RF (0.899). The LightGBM model also reached an accuracy of 92%. SHAP analysis identified that the toe out angle and COP sway area were the most important features for scoliosis prediction. This finding is consistent with the two sets of differential features identified in the statistical analysis. Conclusion: This study developed a predictive model for pediatric scoliosis using foot pressure and body posture data, demonstrating high accuracy and efficiency in predicting scoliosis. The study further confirmed the key role of foot pressure distribution and body posture features in scoliosis prediction, providing a new technical approach and reference for early screening of scoliosis. Scoliosis Child Artificial Intelligence Machine Learning Predictive Learning Models Figures Figure 1 Figure 2 Introduction Scoliosis is an abnormal lateral curvature of the spine, primarily occurring in the coronal plane. It is most commonly diagnosed during childhood or early adolescence[1]. Epidemiological data indicate that the incidence of scoliosis in children ranges from 1.5% to 5.2%[2]. Pediatric scoliosis not only leads to abnormal body posture but may also significantly impact growth, bodily functions, and psychological well-being[3, 4].Given the multifaceted effects of scoliosis on children, early screening and timely intervention are essential. Clinical screening follows a two-step model: "primary screening - diagnosis". initial screening is conducted through visual inspection, the Adams forward bending test, and measurement of the trunk rotation angle (ATR), where ATR ≥ 5° is considered a positive indicator. In the diagnostic phase, a standing full-spine X-ray is required, and Cobb’s angle is used for classification and risk assessment, with a diagnostic threshold of Cobb’s angle ≥10°[5]. While these methods are highly accurate, they are limited in large-scale screenings due to their complexity, high costs, and potential radiation risks. In recent years, advancements in sensor technology, wearable devices, and data analysis methods have facilitated the use of biological features, such as foot pressure distribution and body angles, as effective indicators for scoliosis screening[6–8]. In recent years, machine learning technologies have made significant advancements in the medical field, particularly in image recognition, disease prediction, and early diagnosis. By leveraging machine learning models, researchers can uncover intricate connections in large-scale data, thereby improving the accuracy of scoliosis prediction. However, while some studies have explored machine learning-based scoliosis prediction models, most have focused on imaging data analysis, with limited research specifically targeting pediatric populations[9, 10]. Therefore, developing multi-feature machine learning models tailored for predicting pediatric prediction remains a crucial research direction. This study aims to analyze the data from 5,970 primary school students in Kunming to investigate the applicability of foot pressure and body angle features in predicting pediatric scoliosis. By performing data preprocessing, statistical analysis, and machine learning modeling, this study aims to develop a predictive model suitable for early screening of pediatric scoliosis, providing a low-cost, convenient, and efficient screening method. Materials and Methods Data Source The study data were collected from 5,970 students (ages 6-12) at a primary school in Kunming. Foot pressure and body angle data were measured for all participants using dedicated sensors and an intelligent posture analysis device, respectively. Data Processing Two primarily feature categories were extracted: 1. Foot Pressure Features: These include pressure measurements from the front, back, left, and right foot regions, which help analyze differences in foot pressure distribution between children with scoliosis and healthy controls; 2. Body Angle Features: These encompass various posture-related angles, including ear level, shoulder level, pelvis level, knee level, head tilt (left and right), trunk tilt (left and right), toe-out angle, and Center of Pressure (COP) sway area.Data cleaning was performed to remove samples with missing or abnormal values (≥30%) due to equipment malfunctions, and missing values were filled using the mean. As a result, 3723 valid samples were included in the analysis. All data were anonymized to ensure the privacy and data security of the participants. To ensure data quality, a rigorous cleaning process was implemented. Samples with ≥30% missing or anomalous values due to equipment malfunctions were excluded, while remaining missing values were imputed using the mean. After preprocessing, a total of 3,723 valid samples were retained for analysis. All data were anonymized to protect participant privacy and ensure data security. Given the severe class imbalance in the dataset—where only 16.4% of samples represented scoliosis cases—the Synthetic Minority Over-sampling Technique (SMOTE) was applied to enhance model performance. SMOTE generates synthetic samples by interpolating between existing minority class instances, thereby balancing the proportion of positive and negative cases in the training set [11]. To ensure stable and unbiased model training, all numerical features were standardized using Z-score normalization, adjusting each feature to a mean of 0 and a standard deviation of 1. This normalization mitigates the risk of certain features disproportionately influencing the learning process due to scale differences[12]. Statistical Analysis Descriptive statistics were used to compute the mean and standard deviation for each feature. For normally distributed features, an independent samples t-test was performed to compare differences between children with scoliosis and healthy controls, with a significance level set at α = 0.05. Predictive Model Construction and Comparison Several machine learning algorithms were employed to develop predictive models for scoliosis. Common classification models, including Logistic Regression (LR), Support Vector Machine (SVM), Random Forest (RF), and LightGBM, were selected for comparison. To mitigate overfitting and improve generalization, k-fold cross-validation was applied during training and validation.Model performance was assessed using multiple evaluation metrics, including accuracy, precision, recall, F1-score, AUC, and ROC curves [13–15]. The model with the highest overall performance was selected as the optimal classification model. Model Interpretation Analysis To gain deeper insights into feature contributions to scoliosis prediction, SHAP (Shapley Additive exPlanations) was employed to interpret the best-performing model (LightGBM). SHAP, a game theory-based method, quantifies and visualizes the impact of each feature on model predictions. It is derived from Shapley values, introduced by Lloyd Shapley in 1953, which fairly allocate cooperative gains and are widely applied in cooperative game theory[16]. By computing the marginal contribution of each feature under varying input conditions, SHAP provides a comprehensive understanding of the model's decision-making process[17]. Ethical Statement This study was conducted in accordance with the ethical principles of the Declaration of Helsinki. Informed consent was obtained from the parents or guardians of all participants, and all research data were anonymized. The study protocol was approved by the institutional review board (Approval Number: 202401001). Results Comparison of Features Between Two Groups Foot pressure and body angle data were collected from 5,970 primary school students in Kunming. After excluding missing and anomalous data, 3,723 valid samples were included in the analysis, with 16.4% of the children diagnosed with scoliosis and 83.6% classified as normal. A comparative analysis between the two groups revealed that t-test p-values for foot pressure across all four regions exceeded 0.05, indicating no statistically significant differences in foot pressure distribution. Similarly, body angle features exhibited minimal differences between groups, with large standard deviations, reflecting substantial individual variability. The t-test results for body angle metrics also yielded p-values greater than 0.05, suggesting no significant group differences. However, significant differences were observed in foot deviation angle and COP (Center of Pressure) sway area, with p-values less than 0.05, indicating their potential relevance in distinguishing scoliosis cases. Table 1 Comparison of Features Between Children with Scoliosis and Normal Children Meassurement Scolicosis（n=611） mean ± std Normal（3112） mean ± std t-value p-value Forefoot Pressure(%) 0.383 ± 0.161 0.378 ± 0.147 0.515 0.607 Hindfoot Pressure(%) 0.617 ± 0.161 0.620 ± 0.147 -0.515 0.607 Left Foot Pressure(%) 0.561 ± 0.101 0.557 ± 0.108 0.890 0.373 Right Foot Pressure(%) 0.438 ± 0.101 0.443 ± 0.108 -0.890 0.373 Toe Out Angle(°) 5.832 ± 2.930 6.287 ± 3.693 -2.873 0.004 Ear Level (°) -0.252 ± 3.949 -0.230 ± 6.274 -0.086 0.931 Shoulder Level (°) -0.085 ± 2.304 -0.191 ± 3.518 0.709 0.478 Pelvic Level (°) 0.807 ± 1.862 0.734 ± 2.564 0.661 0.509 Knee Level (°) -0.364 ± 4.047 -0.316 ± 5.000 -0.223 0.823 Head Lateral Tilt (°) -1.338 ± 9.169 -1.782 ± 9.929 1.021 0.307 Trunk Lateral Tilt (°) -0.042 ± 1.389 -0.093 ± 2.478 0.492 0.623 Trunk Rotaion Angle (°) -0.960 ± 10.766 -1.172 ± 10.438 -0.447 0.654 COP Sway Area(mm 2 ) 3232.290 ± 5087.129 2652.607 ± 4333.653 -2.635 0.008 Model Performance Comparison This study compared the performance of four machine learning models—Logistic Regression (LR), Support Vector Machine (SVM), Random Forest (RF), and LightGBM—for scoliosis prediction. Performance metrics, including precision, recall, F1-score, accuracy, and AUC, are presented in Table 2, and the ROC curves for each model are shown in Figure 1. LR performed poorly overall. While it achieved a precision of 0.86 for the normal category, its recall was low (0.50), indicating a high rate of false negatives. For scoliosis, LR’s precision was only 0.18, and recall was 0.57, resulting in a weak F1-score of 0.28. The model's overall accuracy was 0.51, with an AUC of 0.566, reflecting poor performance, particularly with imbalanced data. SVM showed slight improvement over LR, with a precision of 0.92 and recall of 0.57 for the normal category, yielding a balanced F1-score of 0.71. However, its performance on scoliosis cases was still lacking, with precision at 0.25 and recall at 0.74, resulting in an F1-score of 0.38. SVM's overall accuracy was 0.67, with an AUC of 0.723, still struggling with class imbalance. RF performed well for the normal category, achieving a precision of 0.93, recall of 0.88, and F1-score of 0.90. However, its performance for scoliosis prediction dropped, with precision at 0.51, recall at 0.64, and F1-score of 0.57. RF’s overall accuracy was 0.84, and its AUC was 0.899, demonstrating solid performance, but still struggled with detecting scoliosis. LightGBM outperformed all models. It achieved excellent results for the normal category with a precision of 0.93, recall of 0.98, and an F1-score of 0.95. For scoliosis, LightGBM maintained strong performance with precision of 0.89, recall of 0.81, and F1-score of 0.84. Overall, LightGBM reached an accuracy of 0.92 and an AUC of 0.971, excelling in handling imbalanced data and optimizing predictions for both categories. Table 2 Model Performance Comparison Model Precision（Normal） Recall（Nomraml） F1-score（Normal） Precision（Scoliosis） Recall（Scoliosis） F1-score（Scoliosis） Accuracy AUC LR 0.86 0.50 0.63 0.18 0.57 0.28 0.51 0.566 SVM 0.92 0.57 0.71 0.25 0.74 0.38 0.67 0.723 RF 0.93 0.88 0.90 0.51 0.64 0.57 0.84 0.899 LightGBM 0.93 0.98 0.95 0.89 0.81 0.84 0.92 0.971 Model Interpretability Analysis After evaluating model performance, LightGBM—the best-performing model—was selected for interpretability analysis. Figure 2a presents the SHAP feature importance ranking for LightGBM, which quantifies the contribution of each feature to the model's predictions. Larger SHAP values indicate a greater impact on the model's predictions. SHAP feature importance analysis revealed that the foot deviation angle and COP sway area were the most influential features in the model’s predictions. Body angle features, while contributing equally, played a critical role in the model. Among the foot pressure features, pressure on the front of the foot had the largest impact, while the pressure from other regions (left, right, and back foot) contributed less, suggesting that front foot pressure is more important in predicting scoliosis. The gender feature showed the lowest SHAP value, indicating minimal contribution to the model’s predictions. This may be due to the weak relationship between gender and scoliosis, providing limited information for the model. Figure 2b illustrates the SHAP beeswarm plot, where features are ordered by importance along the y-axis. The color of the distribution points indicates feature values, with red representing higher values and blue representing lower values. The shape of the points shows each feature's contribution across different samples. The SHAP value distribution for foot deviation angle is broad, highlighting its strong influence on the model. The majority of red points are found in the negative SHAP value region, suggesting that higher foot deviation angles negatively impact the model's predictions. In contrast, the SHAP values for COP sway area are concentrated around 0, indicating a weaker effect on the model's output. Body angles such as ear level, shoulder level, pelvis level, knee level, head tilt, trunk tilt, and trunk rotation reflect the body’s balance and deviation direction. The SHAP value distribution shows varied contributions of positive and negative angles to the model. For body angle features (pelvis level, ear level, shoulder level), SHAP values are more balanced, with both positive and negative extremes significantly influencing the model. For foot pressure, high values of front foot pressure predominantly fall in the negative SHAP value region, indicating that increased pressure on the front of the foot reduces the likelihood of scoliosis. The SHAP distribution for gender shows minimal variation, confirming its secondary role in the model. Discussion Model Performance In this study, the LightGBM model demonstrated the best performance, achieving an accuracy of 92% and an AUC of 0.95. Compared to previous studies, such as those by Yasuhito et al. [18] and Chui et al. [19] , who used X-ray image data and convolutional neural networks (CNN) to predict the Cobb angle, our approach avoids the drawbacks of radiation exposure and high costs associated with X-ray imaging. While their methods achieved accurate predictions, the use of X-rays limits large-scale screening applications. In contrast, our LightGBM model, relying on body angle and foot pressure data, provides a safer, cost-effective, and real-time method for scoliosis screening. Additionally, Kim et al. [20] used wearable inertial sensors and clinical factors to predict scoliosis, achieving 92% accuracy, similar to our findings. However, their focus was on gait analysis, while our study emphasized biomechanical features like foot pressure and body angles. Feature Importance Analysis Feature importance analysis from the LightGBM model highlighted the significant impact of foot pressure distribution and body posture features on scoliosis prediction, which aligns with previous research. Studies by Azevedo et al. [21] and Zhu et al. [22] have also underscored the critical role of foot pressure distribution and gait features in early scoliosis detection. Our study found that the foot deviation angle is an important biomechanical indicator of scoliosis, reflecting mechanical imbalances in the spine and lower limbs, which are closely linked to scoliosis development. Additionally, the COP sway area, which measures body balance and stability, was identified as a crucial predictor. Larger COP sway areas have been associated with a higher likelihood of developing scoliosis-related issues[23], confirming its importance in our model. Limitations and Future Research Directions Despite the high accuracy of our machine learning-based scoliosis prediction model, several limitations remain. First, the data were collected from a primary school in Kunming, limiting the model’s generalizability. Second, although the sample size was large, the low proportion of children with scoliosis may have introduced bias, potentially affecting the model's ability to detect scoliosis cases. Despite employing SMOTE to address class imbalance, the dataset remains skewed, which may lead to the model favoring normal children. Furthermore, scoliosis is influenced by a variety of factors, including genetics, physical activity, and lifestyle, in addition to foot pressure and body angles[24, 25].Future research could address these limitations and expand the study in several ways: 1. Cross-regional validation and dataset expansion: To improve generalizability, future studies should validate the model across different regions and age groups, enhancing its broader applicability; 2. Multimodal data fusion: This study focused solely on foot pressure and body angle data. Future research could combine these features with imaging data, biomarkers, and lifestyle factors to build multimodal predictive models, improving early scoliosis detection. Clinical Application The machine learning model developed in this study offers significant clinical potential. Unlike traditional imaging-based screening methods, the approach using foot pressure and body angle data provides advantages such as low cost, no radiation, and ease of implementation. In large-scale population screenings, this model can serve as an effective supplementary tool for the early detection of pediatric scoliosis. Furthermore, it could be integrated into smart wearable devices for continuous, dynamic assessment, enabling early prediction and timely intervention for scoliosis.This version is more concise and maintains the professional tone. Let me know if you'd like any further revisions! Conclusion This study developed a predictive model for pediatric scoliosis using foot pressure and body angle data, employing machine learning methods. The model demonstrated high efficiency and accuracy on the test set. Compared to other studies, this approach offers convenient data collection. Furthermore, the study highlights the significant role of biomechanical indicators, such as foot pressure distribution and body posture features, in predicting scoliosis. References Baker C, Morris N, Tsirikos A, et al (2024) Adolescent idiopathic scoliosis: interdisciplinary creative art practice and nature connections. Medical Humanities 50:41–51. https://doi.org/10.1136/medhum-2023-012796 Li M, Nie Q, Liu J, Jiang Z (2024) Prevalence of scoliosis in children and adolescents: a systematic review and meta-analysis. Front Pediatr 12:. https://doi.org/10.3389/fped.2024.1399049 Wang H, Li T, Yuan W, et al (2019) Mental health of patients with adolescent idiopathic scoliosis and their parents in China: a cross-sectional survey. BMC Psychiatry 19:147. https://doi.org/10.1186/s12888-019-2128-1 Amăricăi E, Suciu O, Onofrei RR, et al (2020) Respiratory function, functional capacity, and physical activity behaviours in children and adolescents with scoliosis. J Int Med Res 48:300060519895093. https://doi.org/10.1177/0300060519895093 Fainardi V, Nora M, Salghetti A, et al (2024) Prevalence of Scoliosis in Children and Adolescents with Cystic Fibrosis. Children 11:321. https://doi.org/10.3390/children11030321 Haddas R, Lawlor M, Moghadam E, et al (2023) Spine patient care with wearable medical technology: state-of-the-art, opportunities, and challenges: a systematic review. The Spine Journal 23:929–944. https://doi.org/10.1016/j.spinee.2023.02.020 Zhu F, Hong Q, Guo X, et al (2021) A comparison of foot posture and walking performance in patients with mild, moderate, and severe adolescent idiopathic scoliosis. PLoS One 16:e0251592 Zhang J, Han Y, Yin X, et al (2025) Characterization of dynamic features in the walking videos of patients with adolescent idiopathic scoliosis based on moving entropy. Chaos: An Interdisciplinary Journal of Nonlinear Science 35:013140. https://doi.org/10.1063/5.0238864 Li L, Wong M-S (2024) The application of machine learning methods for predicting the progression of adolescent idiopathic scoliosis: a systematic review. BioMed Eng OnLine 23:80. https://doi.org/10.1186/s12938-024-01272-6 Wang H, Zhang T, Cheung KM-C, Shea GK-H (2021) Application of deep learning upon spinal radiographs to predict progression in adolescent idiopathic scoliosis at first clinic visit. eClinicalMedicine 42:. https://doi.org/10.1016/j.eclinm.2021.101220 Chawla NV, Bowyer KW, Hall LO, Kegelmeyer WP (2002) SMOTE: Synthetic Minority Over-sampling Technique. Journal of Artificial Intelligence Research 16:321–357. https://doi.org/10.1613/jair.953 Singh D, Singh B (2020) Investigating the impact of data normalization on classification performance. Applied Soft Computing 97:105524. https://doi.org/10.1016/j.asoc.2019.105524 Hanley JA, McNeil BJ (1982) The meaning and use of the area under a receiver operating characteristic (ROC) curve. Radiology 143:29–36. https://doi.org/10.1148/radiology.143.1.7063747 Collins GS, Reitsma JB, Altman DG, Moons KGM (2015) Transparent Reporting of a Multivariable Prediction Model for Individual Prognosis or Diagnosis (TRIPOD): The TRIPOD Statement. Circulation 131:211–219. https://doi.org/10.1161/CIRCULATIONAHA.114.014508 Powers D, Ailab (2011) Evaluation: From precision, recall and F-measure to ROC, informedness, markedness & correlation. J Mach Learn Technol 2:2229–3981. https://doi.org/10.9735/2229-3981 Shapley LS (1953) A value for n-person games. Contribution to the Theory of Games 2: Allgaier J, Mulansky L, Draelos RL, Pryss R (2023) How does the model make predictions? A systematic literature review on the explainability power of machine learning in healthcare. Artificial Intelligence in Medicine 143:102616 Yahara Y, Tamura M, Seki S, et al (2022) A deep convolutional neural network to predict the curve progression of adolescent idiopathic scoliosis: a pilot study. BMC Musculoskelet Disord 23:610. https://doi.org/10.1186/s12891-022-05565-6 Chui C-S (Elvis), He Z, Lam T-P, et al (2024) Deep Learning-Based Prediction Model for the Cobb Angle in Adolescent Idiopathic Scoliosis Patients. Diagnostics 14:1263. https://doi.org/10.3390/diagnostics14121263 Kim Y-G, Kim S, Park JH, et al (2024) Explainable Deep-Learning-Based Gait Analysis of Hip–Knee Cyclogram for the Prediction of Adolescent Idiopathic Scoliosis Progression. Sensors 24:4504. https://doi.org/10.3390/s24144504 Azevedo N, Ribeiro JC, Machado L (2022) Balance and posture in children and adolescents: A cross-sectional study. Sensors 22:4973 Zhu F, Hong Q, Guo X, et al (2021) A comparison of foot posture and walking performance in patients with mild, moderate, and severe adolescent idiopathic scoliosis. PLoS One 16:e0251592 Siwiec A, Domagalska-Szopa M, Kwiecień-Czerwieniec I, Szopa A (2024) The Effect of the Direction of Primary Lateral Spinal Curvature on Postural Stability in Children with Scoliosis. Journal of Clinical Medicine 13:1690 Marya S, Tambe AD, Millner PA, Tsirikos AI (2022) Adolescent idiopathic scoliosis: a review of aetiological theories of a multifactorial disease. The Bone & Joint Journal 104-B:915–921. https://doi.org/10.1302/0301-620X.104B8.BJJ-2021-1638.R1 Khadour FA, Khadour YA, Albarroush D (2024) Association between postural habits and lifestyle factors of adolescent idiopathic scoliosis in Syria. Sci Rep 14:26784. https://doi.org/10.1038/s41598-024-77712-z Additional Declarations No competing interests reported. Cite Share Download PDF Status: Published Journal Publication published 12 Nov, 2025 Read the published version in European Spine Journal → Version 1 posted Editorial decision: Revision requested 08 Sep, 2025 Reviews received at journal 02 Sep, 2025 Reviewers agreed at journal 21 Aug, 2025 Reviewers invited by journal 02 Jun, 2025 Editor assigned by journal 15 Mar, 2025 Submission checks completed at journal 15 Mar, 2025 First submitted to journal 14 Mar, 2025 You are reading this latest preprint version Research Square lets you share your work early, gain feedback from the community, and start making changes to your manuscript prior to peer review in a journal. As a division of Research Square Company, we’re committed to making research communication faster, fairer, and more useful. We do this by developing innovative software and high quality services for the global research community. Our growing team is made up of researchers and industry professionals working together to solve the most critical problems facing scientific publishing. Also discoverable on Platform About Our Team In Review Editorial Policies Advisory Board Help Center Resources Author Services Accessibility API Access RSS feed Manage Cookie Preferences © Research Square 2026 | ISSN 2693-5015 (online) Privacy Policy Terms of Service Do Not Sell My Personal Information {"props":{"pageProps":{"initialData":{"identity":"rs-6224546","acceptedTermsAndConditions":true,"allowDirectSubmit":false,"archivedVersions":[],"articleType":"Research Article","associatedPublications":[],"authors":[{"id":465646680,"identity":"b15adaf9-6891-4e97-8c83-58718d39ed4b","order_by":0,"name":"QiGang Chen","email":"data:image/png;base64,iVBORw0KGgoAAAANSUhEUgAAAZAAAAAyAQMAAABI0h/eAAAABlBMVEX///8AAABVwtN+AAAACXBIWXMAAA7EAAAOxAGVKw4bAAAA5UlEQVRIie3PoWrDUBTG8RMO3JpTYm8hrK9wSyGqULfnuDFRG2zuioiElEQs8y17icnJlgOtufOR2Rtkrq6Nb0kyN3H/+vzgOwAu139s4rd8VivyJ1nWaJOMIAgLBhMHs4pZNfY4iiwZLK9U/RjPfjY4LOY5aH4tkKCm0ESpAL98073Ey2HPu0KQ91GFdfQVgLTfn/270Ev304IIA9sRK0DJ534iEKEjkoR8Cl+iAocJoQAmq4hkHMMoIrvHD1ujSRKz1PZIg7/MywrbVl3W61OW/Z5N8uCX7/3kdunfzl0ul8t1tyvCPUegGWRWggAAAABJRU5ErkJggg==","orcid":"","institution":"Kunming Municipal Hospital of Traditional Chinese Medicine","correspondingAuthor":true,"prefix":"","firstName":"QiGang","middleName":"","lastName":"Chen","suffix":""},{"id":465646683,"identity":"f9af68f9-c80b-4af6-8322-1e07dcf04aa7","order_by":1,"name":"YanPing Song","email":"","orcid":"","institution":"Kunming Municipal Hospital of Traditional Chinese Medicine","correspondingAuthor":false,"prefix":"","firstName":"YanPing","middleName":"","lastName":"Song","suffix":""},{"id":465646686,"identity":"82bfc2f3-3698-46f5-b303-eb45e1705a49","order_by":2,"name":"XingYu Chen","email":"","orcid":"","institution":"Kunming Municipal Hospital of Traditional Chinese Medicine","correspondingAuthor":false,"prefix":"","firstName":"XingYu","middleName":"","lastName":"Chen","suffix":""},{"id":465646687,"identity":"8d32fc19-a198-43a7-baee-9132f4495a64","order_by":3,"name":"RanTing Tao","email":"","orcid":"","institution":"Kunming Municipal Hospital of Traditional Chinese Medicine","correspondingAuthor":false,"prefix":"","firstName":"RanTing","middleName":"","lastName":"Tao","suffix":""},{"id":465646688,"identity":"ebde5315-0f65-4c0d-b30a-3c22c8740fb9","order_by":4,"name":"Xin Li","email":"","orcid":"","institution":"Kunming Municipal Hospital of Traditional Chinese Medicine","correspondingAuthor":false,"prefix":"","firstName":"Xin","middleName":"","lastName":"Li","suffix":""},{"id":465646690,"identity":"a22ad839-4ea1-45b3-960b-7f20b7f6e807","order_by":5,"name":"Ang Ding","email":"","orcid":"","institution":"Kunming Municipal Hospital of Traditional Chinese Medicine","correspondingAuthor":false,"prefix":"","firstName":"Ang","middleName":"","lastName":"Ding","suffix":""},{"id":465646692,"identity":"09e6a14c-b100-40e3-b6ef-bfb580a5a4b6","order_by":6,"name":"Sen Wang","email":"","orcid":"","institution":"Kunming Municipal Hospital of Traditional Chinese Medicine","correspondingAuthor":false,"prefix":"","firstName":"Sen","middleName":"","lastName":"Wang","suffix":""}],"badges":[],"createdAt":"2025-03-14 08:23:25","currentVersionCode":1,"declarations":"","doi":"10.21203/rs.3.rs-6224546/v1","doiUrl":"https://doi.org/10.21203/rs.3.rs-6224546/v1","draftVersion":[],"editorialEvents":[{"content":"https://doi.org/10.1007/s00586-025-09548-8","type":"published","date":"2025-11-12T15:57:48+00:00"}],"editorialNote":"","failedWorkflow":false,"files":[{"id":84185753,"identity":"41b029ac-dc55-4c04-8702-4e29a80142bf","added_by":"auto","created_at":"2025-06-09 05:29:05","extension":"png","order_by":1,"title":"Figure 1","display":"","copyAsset":false,"role":"figure","size":47181,"visible":true,"origin":"","legend":"\u003cp\u003eROC Curves of Models on the Test Set\u003c/p\u003e","description":"","filename":"1.png","url":"https://assets-eu.researchsquare.com/files/rs-6224546/v1/5697bff33ad37d4dad2e0822.png"},{"id":84185752,"identity":"ef80ae02-7358-4b9a-9b36-232e1a65de46","added_by":"auto","created_at":"2025-06-09 05:29:05","extension":"png","order_by":2,"title":"Figure 2","display":"","copyAsset":false,"role":"figure","size":90296,"visible":true,"origin":"","legend":"\u003cp\u003eVisual Explanation of LightGBM Model for Prostate Cancer Prediction\u003cbr\u003e\n(A) Bar chart of average feature importance based on SHAP values;\u003cbr\u003e\n(B) Beeswarm plot showing the contribution and distribution of each feature to the model’s predictions.\u003c/p\u003e","description":"","filename":"2.png","url":"https://assets-eu.researchsquare.com/files/rs-6224546/v1/7e2b3d43d8b3228c2da19627.png"},{"id":96105027,"identity":"db642d0e-770c-457a-b992-064280d30a7b","added_by":"auto","created_at":"2025-11-17 16:07:29","extension":"pdf","order_by":0,"title":"","display":"","copyAsset":false,"role":"manuscript-pdf","size":671844,"visible":true,"origin":"","legend":"","description":"","filename":"manuscript.pdf","url":"https://assets-eu.researchsquare.com/files/rs-6224546/v1/1950db85-bcaa-45b3-8a03-320b0ce796ea.pdf"}],"financialInterests":"No competing interests reported.","formattedTitle":"Machine Learning-Based Prediction of Pediatric Scoliosis Using Foot Pressure and Body Angles","fulltext":[{"header":"Introduction","content":"\u003cp\u003eScoliosis is an abnormal lateral curvature of the spine, primarily occurring in the coronal plane. It is most commonly diagnosed during childhood or early adolescence[1]. Epidemiological data indicate that the incidence of scoliosis in children ranges from 1.5% to 5.2%[2]. Pediatric scoliosis not only leads to abnormal body posture but may also significantly impact growth, bodily functions, and psychological well-being[3, 4].Given the multifaceted effects of scoliosis on children, early screening and timely intervention are essential. Clinical screening follows a two-step model: \"primary screening - diagnosis\". initial screening is conducted through visual inspection, the Adams forward bending test, and measurement of the trunk rotation angle (ATR), where ATR ≥ 5° is considered a positive indicator. In the diagnostic phase, a standing full-spine X-ray is required, and Cobb’s angle is used for classification and risk assessment, with a diagnostic threshold of Cobb’s angle ≥10°[5]. While these methods are highly accurate, they are limited in large-scale screenings due to their complexity, high costs, and potential radiation risks.\u003c/p\u003e\n\u003cp\u003eIn recent years, advancements in sensor technology, wearable devices, and data analysis methods have facilitated the use of biological features, such as foot pressure distribution and body angles, as effective indicators for scoliosis screening[6–8].\u0026nbsp;\u003c/p\u003e\n\u003cp\u003eIn recent years, machine learning technologies have made significant advancements in the medical field, particularly in image recognition, disease prediction, and early diagnosis. By leveraging machine learning models, researchers can uncover intricate connections in large-scale data, thereby improving the accuracy of scoliosis prediction. However, while some studies have explored machine learning-based scoliosis prediction models, most have focused on imaging data analysis, with limited research specifically targeting pediatric populations[9, 10]. Therefore, developing multi-feature machine learning models tailored for predicting pediatric prediction remains a crucial research direction.\u003c/p\u003e\n\u003cp\u003eThis study aims to analyze the data from 5,970 primary school students in Kunming to investigate the applicability of foot pressure and body angle features in predicting pediatric scoliosis. By performing data preprocessing, statistical analysis, and machine learning modeling, this study aims to develop a predictive model suitable for early screening of pediatric scoliosis, providing a low-cost, convenient, and efficient screening method.\u003c/p\u003e"},{"header":"Materials and Methods","content":"\u003cp\u003e\u003cstrong\u003eData Source\u003cbr\u003e\u003c/strong\u003e The study data were collected from 5,970 students (ages 6-12) at a primary school in Kunming. Foot pressure and body angle data were measured for all participants using dedicated sensors and an intelligent posture analysis device, respectively.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eData Processing\u003cbr\u003e\u003c/strong\u003e Two primarily feature categories were extracted: 1. Foot Pressure Features: These include pressure measurements from the front, back, left, and right foot regions, which help analyze differences in foot pressure distribution between children with scoliosis and healthy controls; 2. Body Angle Features: These encompass various posture-related angles, including ear level, shoulder level, pelvis level, knee level, head tilt (left and right), trunk tilt (left and right), toe-out angle, and Center of Pressure (COP) sway area.Data cleaning was performed to remove samples with missing or abnormal values (≥30%) due to equipment malfunctions, and missing values were filled using the mean. As a result, 3723 valid samples were included in the analysis. All data were anonymized to ensure the privacy and data security of the participants. To ensure data quality, a rigorous cleaning process was implemented. Samples with ≥30% missing or anomalous values due to equipment malfunctions were excluded, while remaining missing values were imputed using the mean. After preprocessing, a total of 3,723 valid samples were retained for analysis. All data were anonymized to protect participant privacy and ensure data security.\u003c/p\u003e\n\u003cp\u003eGiven the severe class imbalance in the dataset—where only 16.4% of samples represented scoliosis cases—the Synthetic Minority Over-sampling Technique (SMOTE) was applied to enhance model performance. SMOTE generates synthetic samples by interpolating between existing minority class instances, thereby balancing the proportion of positive and negative cases in the training set [11]. To ensure stable and unbiased model training, all numerical features were standardized using Z-score normalization, adjusting each feature to a mean of 0 and a standard deviation of 1. This normalization mitigates the risk of certain features disproportionately influencing the learning process due to scale differences[12].\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eStatistical Analysis\u003cbr\u003e\u003c/strong\u003e Descriptive statistics were used to compute the mean and standard deviation for each feature. For normally distributed features, an independent samples t-test was performed to compare differences between children with scoliosis and healthy controls, with a significance level set at α = 0.05.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003ePredictive Model Construction and Comparison\u003cbr\u003e\u003c/strong\u003e Several machine learning algorithms were employed to develop predictive models for scoliosis. Common classification models, including Logistic Regression (LR), Support Vector Machine (SVM), Random Forest (RF), and LightGBM, were selected for comparison. To mitigate overfitting and improve generalization, k-fold cross-validation was applied during training and validation.Model performance was assessed using multiple evaluation metrics, including accuracy, precision, recall, F1-score, AUC, and ROC curves [13–15]. The model with the highest overall performance was selected as the optimal classification model.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eModel Interpretation Analysis\u003c/strong\u003e\u003cbr\u003eTo gain deeper insights into feature contributions to scoliosis prediction, SHAP (Shapley Additive exPlanations) was employed to interpret the best-performing model (LightGBM). SHAP, a game theory-based method, quantifies and visualizes the impact of each feature on model predictions. It is derived from Shapley values, introduced by Lloyd Shapley in 1953, which fairly allocate cooperative gains and are widely applied in cooperative game theory[16]. By computing the marginal contribution of each feature under varying input conditions, SHAP provides a comprehensive understanding of the model's decision-making process[17].\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eEthical Statement\u003cbr\u003e\u003c/strong\u003e This study was conducted in accordance with the ethical principles of the Declaration of Helsinki. Informed consent was obtained from the parents or guardians of all participants, and all research data were anonymized. The study protocol was approved by the institutional review board (Approval Number: 202401001).\u003c/p\u003e"},{"header":"Results","content":"\u003cp\u003e\u003cstrong\u003eComparison of Features Between Two Groups\u003cbr\u003e\u003c/strong\u003e Foot pressure and body angle data were collected from 5,970 primary school students in Kunming. After excluding missing and anomalous data, 3,723 valid samples were included in the analysis, with 16.4% of the children diagnosed with scoliosis and 83.6% classified as normal. A comparative analysis between the two groups revealed that t-test p-values for foot pressure across all four regions exceeded 0.05, indicating no statistically significant differences in foot pressure distribution. Similarly, body angle features exhibited minimal differences between groups, with large standard deviations, reflecting substantial individual variability. The t-test results for body angle metrics also yielded p-values greater than 0.05, suggesting no significant group differences. However, significant differences were observed in foot deviation angle and COP (Center of Pressure) sway area, with p-values less than 0.05, indicating their potential relevance in distinguishing scoliosis cases.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eTable 1\u003c/strong\u003e Comparison of Features Between Children with Scoliosis and Normal Children\u003c/p\u003e\n\u003cdiv align=\"center\"\u003e\n \u003ctable border=\"1\" cellspacing=\"0\" cellpadding=\"0\" width=\"100%\"\u003e\n \u003ctbody\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003eMeassurement\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003eScolicosis（n=611）\u003cbr\u003e\u0026nbsp;mean \u0026plusmn; std\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003eNormal（3112）\u003cbr\u003e\u0026nbsp;mean \u0026plusmn; std\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003et-value\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003ep-value\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003eForefoot Pressure(%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e0.383 \u0026plusmn; 0.161\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e0.378 \u0026plusmn; 0.147\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e0.515\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e0.607\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003eHindfoot Pressure(%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e0.617 \u0026plusmn; 0.161\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e0.620 \u0026plusmn; 0.147\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e-0.515\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e0.607\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003eLeft Foot Pressure(%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e0.561 \u0026plusmn; 0.101\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e0.557 \u0026plusmn; 0.108\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e0.890\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e0.373\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003eRight Foot Pressure(%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e0.438 \u0026plusmn; 0.101\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e0.443 \u0026plusmn; 0.108\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e-0.890\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e0.373\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003eToe Out Angle(\u0026deg;)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e5.832 \u0026plusmn; 2.930\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e6.287 \u0026plusmn; 3.693\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e-2.873\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e0.004\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003eEar Level\u0026nbsp;(\u0026deg;)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e-0.252 \u0026plusmn; 3.949\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e-0.230 \u0026plusmn; 6.274\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e-0.086\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e0.931\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003eShoulder Level\u0026nbsp;(\u0026deg;)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e-0.085 \u0026plusmn; 2.304\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e-0.191 \u0026plusmn; 3.518\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e0.709\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e0.478\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003ePelvic Level\u0026nbsp;(\u0026deg;)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e0.807 \u0026plusmn; 1.862\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e0.734 \u0026plusmn; 2.564\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e0.661\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e0.509\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003eKnee Level\u0026nbsp;(\u0026deg;)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e-0.364 \u0026plusmn; 4.047\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e-0.316 \u0026plusmn; 5.000\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e-0.223\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e0.823\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003eHead Lateral Tilt\u0026nbsp;(\u0026deg;)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e-1.338 \u0026plusmn; 9.169\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e-1.782 \u0026plusmn; 9.929\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e1.021\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e0.307\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003eTrunk Lateral Tilt\u0026nbsp;(\u0026deg;)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e-0.042 \u0026plusmn; 1.389\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e-0.093 \u0026plusmn; 2.478\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e0.492\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e0.623\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003eTrunk Rotaion Angle\u0026nbsp;(\u0026deg;)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e-0.960 \u0026plusmn; 10.766\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e-1.172 \u0026plusmn; 10.438\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e-0.447\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e0.654\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003eCOP Sway Area(mm\u003csup\u003e2\u003c/sup\u003e)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e3232.290 \u0026plusmn; 5087.129\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e2652.607 \u0026plusmn; 4333.653\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e-2.635\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e0.008\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003c/tbody\u003e\n \u003c/table\u003e\n\u003c/div\u003e\n\u003cp\u003e\u003cstrong\u003eModel Performance Comparison\u003c/strong\u003e\u003cbr\u003eThis study compared the performance of four machine learning models\u0026mdash;Logistic Regression (LR), Support Vector Machine (SVM), Random Forest (RF), and LightGBM\u0026mdash;for scoliosis prediction. Performance metrics, including precision, recall, F1-score, accuracy, and AUC, are presented in Table 2, and the ROC curves for each model are shown in Figure 1.\u003c/p\u003e\n\u003cp\u003e LR performed poorly overall. While it achieved a precision of 0.86 for the normal category, its recall was low (0.50), indicating a high rate of false negatives. For scoliosis, LR\u0026rsquo;s precision was only 0.18, and recall was 0.57, resulting in a weak F1-score of 0.28. The model\u0026apos;s overall accuracy was 0.51, with an AUC of 0.566, reflecting poor performance, particularly with imbalanced data. \u0026nbsp; \u0026nbsp; \u0026nbsp; \u0026nbsp; \u0026nbsp; \u0026nbsp;\u0026nbsp;\u003c/p\u003e\n\u003cp\u003e SVM showed slight improvement over LR, with a precision of 0.92 and recall of 0.57 for the normal category, yielding a balanced F1-score of 0.71. However, its performance on scoliosis cases was still lacking, with precision at 0.25 and recall at 0.74, resulting in an F1-score of 0.38. SVM\u0026apos;s overall accuracy was 0.67, with an AUC of 0.723, still struggling with class imbalance.\u003c/p\u003e\n\u003cp\u003e RF performed well for the normal category, achieving a precision of 0.93, recall of 0.88, and F1-score of 0.90. However, its performance for scoliosis prediction dropped, with precision at 0.51, recall at 0.64, and F1-score of 0.57. RF\u0026rsquo;s overall accuracy was 0.84, and its AUC was 0.899, demonstrating solid performance, but still struggled with detecting scoliosis.\u003c/p\u003e\n\u003cp\u003e LightGBM outperformed all models. It achieved excellent results for the normal category with a precision of 0.93, recall of 0.98, and an F1-score of 0.95. For scoliosis, LightGBM maintained strong performance with precision of 0.89, recall of 0.81, and F1-score of 0.84. Overall, LightGBM reached an accuracy of 0.92 and an AUC of 0.971, excelling in handling imbalanced data and optimizing predictions for both categories.\u003c/p\u003e\n\u003cp\u003eTable 2 Model Performance Comparison\u003c/p\u003e\n\u003ctable border=\"1\" cellspacing=\"0\" cellpadding=\"0\" width=\"111%\"\u003e\n \u003ctbody\u003e\n \u003ctr\u003e\n \u003ctd\u003e\n \u003cp\u003eModel\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003ePrecision（Normal）\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003eRecall（Nomraml）\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003eF1-score（Normal）\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003ePrecision（Scoliosis）\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003eRecall（Scoliosis）\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003eF1-score（Scoliosis）\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003eAccuracy\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003eAUC\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd\u003e\n \u003cp\u003eLR\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e0.86\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e0.50\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e0.63\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e0.18\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e0.57\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e0.28\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e0.51\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e0.566\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd\u003e\n \u003cp\u003eSVM\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e0.92\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e0.57\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e0.71\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e0.25\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e0.74\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e0.38\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e0.67\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e0.723\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd\u003e\n \u003cp\u003eRF\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e0.93\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e0.88\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e0.90\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e0.51\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e0.64\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e0.57\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e0.84\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e0.899\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd\u003e\n \u003cp\u003eLightGBM\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e0.93\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e0.98\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e0.95\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e0.89\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e0.81\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e0.84\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e0.92\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e0.971\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003c/tbody\u003e\n\u003c/table\u003e\n\u003cp\u003e\u003cbr\u003e\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eModel Interpretability Analysis\u003cbr\u003e\u003c/strong\u003e After evaluating model performance, LightGBM\u0026mdash;the best-performing model\u0026mdash;was selected for interpretability analysis. Figure 2a presents the SHAP feature importance ranking for LightGBM, which quantifies the contribution of each feature to the model\u0026apos;s predictions. Larger SHAP values indicate a greater impact on the model\u0026apos;s predictions.\u003c/p\u003e\n\u003cp\u003e SHAP feature importance analysis revealed that the foot deviation angle and COP sway area were the most influential features in the model\u0026rsquo;s predictions. Body angle features, while contributing equally, played a critical role in the model. Among the foot pressure features, pressure on the front of the foot had the largest impact, while the pressure from other regions (left, right, and back foot) contributed less, suggesting that front foot pressure is more important in predicting scoliosis. The gender feature showed the lowest SHAP value, indicating minimal contribution to the model\u0026rsquo;s predictions. This may be due to the weak relationship between gender and scoliosis, providing limited information for the model.\u003c/p\u003e\n\u003cp\u003e Figure 2b illustrates the SHAP beeswarm plot, where features are ordered by importance along the y-axis. The color of the distribution points indicates feature values, with red representing higher values and blue representing lower values. The shape of the points shows each feature\u0026apos;s contribution across different samples. The SHAP value distribution for foot deviation angle is broad, highlighting its strong influence on the model. The majority of red points are found in the negative SHAP value region, suggesting that higher foot deviation angles negatively impact the model\u0026apos;s predictions. In contrast, the SHAP values for COP sway area are concentrated around 0, indicating a weaker effect on the model\u0026apos;s output.\u003c/p\u003e\n\u003cp\u003e Body angles such as ear level, shoulder level, pelvis level, knee level, head tilt, trunk tilt, and trunk rotation reflect the body\u0026rsquo;s balance and deviation direction. The SHAP value distribution shows varied contributions of positive and negative angles to the model. For body angle features (pelvis level, ear level, shoulder level), SHAP values are more balanced, with both positive and negative extremes significantly influencing the model. For foot pressure, high values of front foot pressure predominantly fall in the negative SHAP value region, indicating that increased pressure on the front of the foot reduces the likelihood of scoliosis. The SHAP distribution for gender shows minimal variation, confirming its secondary role in the model.\u003c/p\u003e"},{"header":"Discussion","content":"\u003cp\u003e\u003cstrong\u003eModel Performance\u003c/strong\u003e\u003cbr\u003eIn this study, the LightGBM model demonstrated the best performance, achieving an accuracy of 92% and an AUC of 0.95. Compared to previous studies, such as those by Yasuhito et al. [18] and Chui et al. [19] , who used X-ray image data and convolutional neural networks (CNN) to predict the Cobb angle, our approach avoids the drawbacks of radiation exposure and high costs associated with X-ray imaging. While their methods achieved accurate predictions, the use of X-rays limits large-scale screening applications. In contrast, our LightGBM model, relying on body angle and foot pressure data, provides a safer, cost-effective, and real-time method for scoliosis screening. Additionally, Kim et al. [20] used wearable inertial sensors and clinical factors to predict scoliosis, achieving 92% accuracy, similar to our findings. However, their focus was on gait analysis, while our study emphasized biomechanical features like foot pressure and body angles.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eFeature Importance Analysis\u003cbr\u003e\u003c/strong\u003e Feature importance analysis from the LightGBM model highlighted the significant impact of foot pressure distribution and body posture features on scoliosis prediction, which aligns with previous research. Studies by Azevedo et al. [21] and Zhu et al. [22] have also underscored the critical role of foot pressure distribution and gait features in early scoliosis detection. Our study found that the foot deviation angle is an important biomechanical indicator of scoliosis, reflecting mechanical imbalances in the spine and lower limbs, which are closely linked to scoliosis development. Additionally, the COP sway area, which measures body balance and stability, was identified as a crucial predictor. Larger COP sway areas have been associated with a higher likelihood of developing scoliosis-related issues[23], confirming its importance in our model.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eLimitations and Future Research Directions\u003cbr\u003e\u003c/strong\u003e Despite the high accuracy of our machine learning-based scoliosis prediction model, several limitations remain. First, the data were collected from a primary school in Kunming, limiting the model’s generalizability. Second, although the sample size was large, the low proportion of children with scoliosis may have introduced bias, potentially affecting the model's ability to detect scoliosis cases. Despite employing SMOTE to address class imbalance, the dataset remains skewed, which may lead to the model favoring normal children. Furthermore, scoliosis is influenced by a variety of factors, including genetics, physical activity, and lifestyle, in addition to foot pressure and body angles[24, 25].Future research could address these limitations and expand the study in several ways: 1. Cross-regional validation and dataset expansion: To improve generalizability, future studies should validate the model across different regions and age groups, enhancing its broader applicability; 2. Multimodal data fusion: This study focused solely on foot pressure and body angle data. Future research could combine these features with imaging data, biomarkers, and lifestyle factors to build multimodal predictive models, improving early scoliosis detection.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eClinical Application\u003cbr\u003e\u003c/strong\u003e The machine learning model developed in this study offers significant clinical potential. Unlike traditional imaging-based screening methods, the approach using foot pressure and body angle data provides advantages such as low cost, no radiation, and ease of implementation. In large-scale population screenings, this model can serve as an effective supplementary tool for the early detection of pediatric scoliosis. Furthermore, it could be integrated into smart wearable devices for continuous, dynamic assessment, enabling early prediction and timely intervention for scoliosis.This version is more concise and maintains the professional tone. Let me know if you'd like any further revisions!\u003c/p\u003e"},{"header":"Conclusion","content":"\u003cp\u003eThis study developed a predictive model for pediatric scoliosis using foot pressure and body angle data, employing machine learning methods. The model demonstrated high efficiency and accuracy on the test set. Compared to other studies, this approach offers convenient data collection. Furthermore, the study highlights the significant role of biomechanical indicators, such as foot pressure distribution and body posture features, in predicting scoliosis.\u003c/p\u003e\n\u003cp\u003e\u003cbr\u003e\u003c/p\u003e"},{"header":"References","content":"\u003col\u003e\n\u003cli\u003eBaker C, Morris N, Tsirikos A, et al (2024) Adolescent idiopathic scoliosis: interdisciplinary creative art practice and nature connections. Medical Humanities 50:41\u0026ndash;51. https://doi.org/10.1136/medhum-2023-012796\u003c/li\u003e\n\u003cli\u003eLi M, Nie Q, Liu J, Jiang Z (2024) Prevalence of scoliosis in children and adolescents: a systematic review and meta-analysis. Front Pediatr 12:. https://doi.org/10.3389/fped.2024.1399049\u003c/li\u003e\n\u003cli\u003eWang H, Li T, Yuan W, et al (2019) Mental health of patients with adolescent idiopathic scoliosis and their parents in China: a cross-sectional survey. BMC Psychiatry 19:147. https://doi.org/10.1186/s12888-019-2128-1\u003c/li\u003e\n\u003cli\u003eAmăricăi E, Suciu O, Onofrei RR, et al (2020) Respiratory function, functional capacity, and physical activity behaviours in children and adolescents with scoliosis. J Int Med Res 48:300060519895093. https://doi.org/10.1177/0300060519895093\u003c/li\u003e\n\u003cli\u003eFainardi V, Nora M, Salghetti A, et al (2024) Prevalence of Scoliosis in Children and Adolescents with Cystic Fibrosis. Children 11:321. https://doi.org/10.3390/children11030321\u003c/li\u003e\n\u003cli\u003eHaddas R, Lawlor M, Moghadam E, et al (2023) Spine patient care with wearable medical technology: state-of-the-art, opportunities, and challenges: a systematic review. The Spine Journal 23:929\u0026ndash;944. https://doi.org/10.1016/j.spinee.2023.02.020\u003c/li\u003e\n\u003cli\u003eZhu F, Hong Q, Guo X, et al (2021) A comparison of foot posture and walking performance in patients with mild, moderate, and severe adolescent idiopathic scoliosis. PLoS One 16:e0251592\u003c/li\u003e\n\u003cli\u003eZhang J, Han Y, Yin X, et al (2025) Characterization of dynamic features in the walking videos of patients with adolescent idiopathic scoliosis based on moving entropy. Chaos: An Interdisciplinary Journal of Nonlinear Science 35:013140. https://doi.org/10.1063/5.0238864\u003c/li\u003e\n\u003cli\u003eLi L, Wong M-S (2024) The application of machine learning methods for predicting the progression of adolescent idiopathic scoliosis: a systematic review. BioMed Eng OnLine 23:80. https://doi.org/10.1186/s12938-024-01272-6\u003c/li\u003e\n\u003cli\u003eWang H, Zhang T, Cheung KM-C, Shea GK-H (2021) Application of deep learning upon spinal radiographs to predict progression in adolescent idiopathic scoliosis at first clinic visit. eClinicalMedicine 42:. https://doi.org/10.1016/j.eclinm.2021.101220\u003c/li\u003e\n\u003cli\u003eChawla NV, Bowyer KW, Hall LO, Kegelmeyer WP (2002) SMOTE: Synthetic Minority Over-sampling Technique. Journal of Artificial Intelligence Research 16:321\u0026ndash;357. https://doi.org/10.1613/jair.953\u003c/li\u003e\n\u003cli\u003eSingh D, Singh B (2020) Investigating the impact of data normalization on classification performance. Applied Soft Computing 97:105524. https://doi.org/10.1016/j.asoc.2019.105524\u003c/li\u003e\n\u003cli\u003eHanley JA, McNeil BJ (1982) The meaning and use of the area under a receiver operating characteristic (ROC) curve. Radiology 143:29\u0026ndash;36. https://doi.org/10.1148/radiology.143.1.7063747\u003c/li\u003e\n\u003cli\u003eCollins GS, Reitsma JB, Altman DG, Moons KGM (2015) Transparent Reporting of a Multivariable Prediction Model for Individual Prognosis or Diagnosis (TRIPOD): The TRIPOD Statement. Circulation 131:211\u0026ndash;219. https://doi.org/10.1161/CIRCULATIONAHA.114.014508\u003c/li\u003e\n\u003cli\u003ePowers D, Ailab (2011) Evaluation: From precision, recall and F-measure to ROC, informedness, markedness \u0026amp; correlation. J Mach Learn Technol 2:2229\u0026ndash;3981. https://doi.org/10.9735/2229-3981\u003c/li\u003e\n\u003cli\u003eShapley LS (1953) A value for n-person games. Contribution to the Theory of Games 2:\u003c/li\u003e\n\u003cli\u003eAllgaier J, Mulansky L, Draelos RL, Pryss R (2023) How does the model make predictions? A systematic literature review on the explainability power of machine learning in healthcare. Artificial Intelligence in Medicine 143:102616\u003c/li\u003e\n\u003cli\u003eYahara Y, Tamura M, Seki S, et al (2022) A deep convolutional neural network to predict the curve progression of adolescent idiopathic scoliosis: a pilot study. BMC Musculoskelet Disord 23:610. https://doi.org/10.1186/s12891-022-05565-6\u003c/li\u003e\n\u003cli\u003eChui C-S (Elvis), He Z, Lam T-P, et al (2024) Deep Learning-Based Prediction Model for the Cobb Angle in Adolescent Idiopathic Scoliosis Patients. Diagnostics 14:1263. https://doi.org/10.3390/diagnostics14121263\u003c/li\u003e\n\u003cli\u003eKim Y-G, Kim S, Park JH, et al (2024) Explainable Deep-Learning-Based Gait Analysis of Hip\u0026ndash;Knee Cyclogram for the Prediction of Adolescent Idiopathic Scoliosis Progression. Sensors 24:4504. https://doi.org/10.3390/s24144504\u003c/li\u003e\n\u003cli\u003eAzevedo N, Ribeiro JC, Machado L (2022) Balance and posture in children and adolescents: A cross-sectional study. Sensors 22:4973\u003c/li\u003e\n\u003cli\u003eZhu F, Hong Q, Guo X, et al (2021) A comparison of foot posture and walking performance in patients with mild, moderate, and severe adolescent idiopathic scoliosis. PLoS One 16:e0251592\u003c/li\u003e\n\u003cli\u003eSiwiec A, Domagalska-Szopa M, Kwiecień-Czerwieniec I, Szopa A (2024) The Effect of the Direction of Primary Lateral Spinal Curvature on Postural Stability in Children with Scoliosis. Journal of Clinical Medicine 13:1690\u003c/li\u003e\n\u003cli\u003eMarya S, Tambe AD, Millner PA, Tsirikos AI (2022) Adolescent idiopathic scoliosis: a review of aetiological theories of a multifactorial disease. The Bone \u0026amp; Joint Journal 104-B:915\u0026ndash;921. https://doi.org/10.1302/0301-620X.104B8.BJJ-2021-1638.R1\u003c/li\u003e\n\u003cli\u003eKhadour FA, Khadour YA, Albarroush D (2024) Association between postural habits and lifestyle factors of adolescent idiopathic scoliosis in Syria. Sci Rep 14:26784. https://doi.org/10.1038/s41598-024-77712-z\u003c/li\u003e\n\u003c/ol\u003e\n\u003cp\u003e\u0026nbsp;\u003c/p\u003e"}],"fulltextSource":"","fullText":"","funders":[],"hasAdminPriorityOnWorkflow":false,"hasManuscriptDocX":true,"hasOptedInToPreprint":true,"hasPassedJournalQc":"","hasAnyPriority":false,"hideJournal":false,"highlight":"","institution":"","isAcceptedByJournal":true,"isAuthorSuppliedPdf":false,"isDeskRejected":"","isHiddenFromSearch":false,"isInQc":false,"isInWorkflow":false,"isPdf":false,"isPdfUpToDate":true,"isWithdrawnOrRetracted":false,"journal":{"display":true,"email":"[email protected]","identity":"european-spine-journal","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":false,"externalIdentity":"esjo","sideBox":"Learn more about [European Spine Journal](http://link.springer.com/journal/586)","snPcode":"586","submissionUrl":"https://submission.springernature.com/new-submission/586/3","title":"European Spine Journal","twitterHandle":"","acdcEnabled":true,"dfaEnabled":true,"editorialSystem":"stoa","reportingPortfolio":"Springer Hybrid","inReviewEnabled":true,"inReviewRevisionsEnabled":false},"keywords":"Scoliosis, Child, Artificial Intelligence, Machine Learning, Predictive Learning Models","lastPublishedDoi":"10.21203/rs.3.rs-6224546/v1","lastPublishedDoiUrl":"https://doi.org/10.21203/rs.3.rs-6224546/v1","license":{"name":"CC BY 4.0","url":"https://creativecommons.org/licenses/by/4.0/"},"manuscriptAbstract":"\u003cp\u003e\u003cstrong\u003ePurpose:\u003c/strong\u003e This study aims to analyze the key features of scoliosis based on children's foot pressure and body angle data and to develop a predictive model to support early screening for scoliosis in ediatric populations.\u003cbr\u003e\n \u003cstrong\u003eMethods: \u003c/strong\u003eData were collected from 5970 primary school students in Kunming. After excluding incomplete and anomalous samples, 3723 students were included. Machine learning algorithms, including Logistic Regression(LR), Support Vector Machine(SVM), Random Forest(RF), and LightGBM, were employed to construct classification models. Model performance was evaluated using accuracy, ROC curves, AUC, and other metrics. The optimal model was selected based on model performance comparisons, and SHAP (Shapley Additive exPlanations) interpretability analysis was conducted.\u003cbr\u003e\n \u003cstrong\u003eResults:\u003c/strong\u003e Statistical analysis revealed significant differences between children with scoliosis and healthy controls in toe out angle and Center of Pressure (COP) sway area. The LightGBM model outperformed others with an AUC of 0.971, significantly higher than LR (0.566), SVM (0.723), and RF (0.899). The LightGBM model also reached an accuracy of 92%. SHAP analysis identified that the toe out angle and COP sway area were the most important features for scoliosis prediction. This finding is consistent with the two sets of differential features identified in the statistical analysis.\u003cbr\u003e\n \u003cstrong\u003eConclusion:\u003c/strong\u003e This study developed a predictive model for pediatric scoliosis using foot pressure and body posture data, demonstrating high accuracy and efficiency in predicting scoliosis. The study further confirmed the key role of foot pressure distribution and body posture features in scoliosis prediction, providing a new technical approach and reference for early screening of scoliosis.\u003c/p\u003e","manuscriptTitle":"Machine Learning-Based Prediction of Pediatric Scoliosis Using Foot Pressure and Body Angles","msid":"","msnumber":"","nonDraftVersions":[{"code":1,"date":"2025-06-09 05:29:01","doi":"10.21203/rs.3.rs-6224546/v1","editorialEvents":[{"type":"communityComments","content":0},{"type":"decision","content":"Revision requested","date":"2025-09-08T09:14:32+00:00","index":"","fulltext":""},{"type":"editorInvitedReview","content":"","date":"2025-09-02T13:12:41+00:00","index":"hide","fulltext":""},{"type":"reviewerAgreed","content":"23970308535466143501457698746428094782","date":"2025-08-22T03:17:31+00:00","index":"hide","fulltext":""},{"type":"reviewersInvited","content":"","date":"2025-06-02T05:10:26+00:00","index":"","fulltext":""},{"type":"editorAssigned","content":"","date":"2025-03-15T06:24:08+00:00","index":"","fulltext":""},{"type":"checksComplete","content":"","date":"2025-03-15T06:21:11+00:00","index":"","fulltext":""},{"type":"submitted","content":"European Spine Journal","date":"2025-03-14T08:16:53+00:00","index":"","fulltext":""}],"status":"published","journal":{"display":true,"email":"[email protected]","identity":"european-spine-journal","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":false,"externalIdentity":"esjo","sideBox":"Learn more about [European Spine Journal](http://link.springer.com/journal/586)","snPcode":"586","submissionUrl":"https://submission.springernature.com/new-submission/586/3","title":"European Spine Journal","twitterHandle":"","acdcEnabled":true,"dfaEnabled":true,"editorialSystem":"stoa","reportingPortfolio":"Springer Hybrid","inReviewEnabled":true,"inReviewRevisionsEnabled":false}}],"origin":"","ownerIdentity":"50c13d4a-5493-4be4-97c6-2311f8c694c3","owner":[],"postedDate":"June 9th, 2025","published":true,"recentEditorialEvents":[],"rejectedJournal":[],"revision":"","amendment":"","status":"published-in-journal","subjectAreas":[],"tags":[],"updatedAt":"2025-11-17T16:01:23+00:00","versionOfRecord":{"articleIdentity":"rs-6224546","link":"https://doi.org/10.1007/s00586-025-09548-8","journal":{"identity":"european-spine-journal","isVorOnly":false,"title":"European Spine Journal"},"publishedOn":"2025-11-12 15:57:48","publishedOnDateReadable":"November 12th, 2025"},"versionCreatedAt":"2025-06-09 05:29:01","video":"","vorDoi":"10.1007/s00586-025-09548-8","vorDoiUrl":"https://doi.org/10.1007/s00586-025-09548-8","workflowStages":[]},"version":"v1","identity":"rs-6224546","journalConfig":"researchsquare"},"__N_SSP":true},"page":"/article/[identity]/[[...version]]","query":{"redirect":"/article/rs-6224546","identity":"rs-6224546","version":["v1"]},"buildId":"8U1c8b4HqxoKbykW_rLl7","isFallback":false,"isExperimentalCompile":false,"dynamicIds":[84888],"gssp":true,"scriptLoader":[]}