Application of Unsupervised Machine Learning in Analyzing Clinical Features of Patients with Anterior Cruciate Ligament Tears

Zhao Junwei; Chen Desheng; Guo Xinjing; Lei Lei; Zhu Jiawang; Ji Gang

Abstract：Objective To analyze the clinical characteristics of patients with anterior cruciate ligament(ACL)tears based on statistical description and unsupervised machine learning,aiming to explore the patterns of clinical features and concomitant injuries.Methods A retrospective analysis was conducted on 224 patients who underwent ACL reconstruction at Tianjin Hospital of Tianjin University from March 2020 to June 2024.The cohort included 161 males and 63 females,with an age range of 14 to 60 years(mean 30.74±10.39).A total of 126 data indicators were collected for each patient,including 20 general clinical characteristics,17 intraoperative indicators,and 89 laboratory and imaging indicators.Unsupervised machine learning was applied for data dimensionality reduction and clustering analysis based on statistical description,followed by intergroup comparisons.Results The patients were divided into four clusters using unsupervised machine learning.Significant differences(P＜0.05)were observed among the clusters in terms of gender,affected side,concomitant lateral meniscus posterior root tears,concomitant lateral meniscus body tears,concomitant lateral meniscus posterior horn tears,age,hospitalization duration,ligament diameter,hemoglobin levels,left ventricular end-diastolic diameter,and left ventricular end-systolic diameter.Group 1 had the lowest ligament diameter,lowest male proportion,lowest right knee proportion,and lowest proportion of posterior root injury and posterior horn injury of the lateral meniscus.Group 2 had the highest age,longest hospitalization duration,highest right knee proportion,and highest proportion of posterior root injury of the lateral meniscus.Group 3 had the lowest age,highest ligament diameter,highest male proportion,and the highest proportion of body and posterior horn injuries of the lateral meniscus.Group 4 had the shortest hospitalization duration and the lowest proportion of body injury of the lateral meniscus.Conclusion The application of unsupervised machine learning in the analysis of clinical features of ACL tear patients effectively identifies heterogeneous features and potential subgroups.It reveals the occurrence patterns of clinical features and concomitant injuries,providing a reference for clinicians to develop personalized treatment plans.