Arthroscopy - 2026-05-14 - Journal Article
Distinct 3-Dimensional Anatomic Patterns Including Flatter Surfaces and Greater Sagittal Inclinations of Intra-articular Structures Are Reliably Identified Through an Artificial Intelligence-Based Pipeline in Anterior Cruciate Ligament-Injured Knees.
Meyer O, Tarouco Amaro J, Cohen Kaleka C, Debieux P, Gomes Oliveira Filho N, Cohen M
Topics
Key Takeaway
An AI pipeline analyzing 19 3D anatomic variables from standard MRI discriminated ACL-injured from control knees with 80% sensitivity, 70% specificity, and AUC-ROC of 0.81, identifying flatter articular surfaces and greater sagittal tibial inclination as the dominant risk features.
Summary Depth
Choose how much analysis to show on this article page.
Summary
This retrospective case-control study asked whether an automated AI pipeline extracting 19 3D morphologic variables from T2 sagittal MRI could differentiate ACL-injured from healthy knees. ACL-injured knees showed significantly flatter femoral and tibial cartilage curvatures and greater sagittal inclination of both tibial plateaus and menisci (all P<0.05). The classification model trained on 80 cases and tested on 20 achieved 75% accuracy and AUC-ROC 0.81.
Key Limitation
The 20-patient test set is too small to yield stable classifier performance estimates, and the case-control design with no reported mechanism, activity level, or chronicity matching limits generalizability to clinical screening populations.
Original Abstract
PURPOSE
To evaluate whether an automated AI-based pipeline can identify 3-dimensional (3D) anatomic patterns associated with anterior cruciate ligament (ACL) injury from conventional magnetic resonance imaging (MRI) and accurately discriminate ACL-injured from control knees.
METHODS
Retrospective case-control study including 50 patients with ACL rupture and 50 healthy controls. T2-weighted sagittal MRI scans were processed using automated artificial intelligence models to segment cartilage, meniscal, and bone structures, reconstruct 3D models, and extract 19 anatomical variables, including curvatures and inclinations of the tibial plateaus and menisci in three planes. Group comparisons identified significant variables (P < .05), which trained an artificial intelligence classification model (80 training; 20 test set) to differentiate ACL-injured from control knees on previously unseen cases.
RESULTS
Patients with ACL rupture exhibited significantly flatter articular surfaces compared with controls: mean curvature of femoral cartilage (0.1277 ± 1.2430 vs -0.5263 ± 1.2328; P = .0061), medial tibial cartilage (0.0623 ± 0.8854 vs -0.3386 ± 0.7525; P = .0253), lateral tibial cartilage (0.1028 ± 0.8562 vs -0.3400 ± 0.7847; P = .0071), and medial meniscus (0.0894 ± 0.9519 vs -0.3583 ± 0.8845; P = .0383). Sagittal inclination was greater in the case group for the lateral plateau (82.73 ± 5.21 vs 79.87 ± 6.09; P = .0019), medial plateau (81.54 ± 5.87 vs 78.90 ± 4.16; P = .0016), lateral meniscus (83.19 ± 4.77 vs 81.83 ± 4.22; P = .0498), and medial meniscus (83.22 ± 4.87 vs 80.67 ± 4.89; P = .0008). Additional significant differences were observed in coronal meniscus-plateau angles (P = .0475 medial, P = .0445 lateral) and sagittal angle between lateral meniscus and plateau (P = .0022). Artificial intelligence model achieved 80% sensitivity, 70% specificity, 75% accuracy and AUC-ROC 0.81 in testing set.
CONCLUSIONS
Automated 3D MRI analysis revealed distinct anatomical risk patterns in patients with ACL rupture, characterized by flatter articular surfaces and increased sagittal inclination of structures. The exploratory automated 3D MRI pipeline proved feasible for extracting and analyzing combined anatomical features for differentiating ACL injured from control knees.
LEVEL OF EVIDENCE
Level III, retrospective case-control study.