2025 ISAKOS Biennial Congress ePoster
3D-Pass (3D Patellar Instability Anatomical Severity Score): A Novel Metric Using Machine Learning To Predict Treatment Outcome In Patella Instability
Marissa Lee Sinopoli, PhD, Claremont, CA UNITED STATES
Anthony Gatti, PhD, Stanford, CA UNITED STATES
Christian E Wright, BS, St. Louis, Missouri UNITED STATES
Anna Bartsch, MD, Basel SWITZERLAND
Matthew William Veerkamp, BA, Cincinnati, OH UNITED STATES
Akshay Chaudhari, PhD, Palo Alto, California UNITED STATES
Beth Ellen Shubin Stein, MD, New York, NY UNITED STATES
Shital N. Parikh, MD, Cincinnati, OH UNITED STATES
Kevin G. Shea, MD, Palo Alto, California UNITED STATES
Scott Delp, PhD, Redwood City, CA UNITED STATES
Seth L. Sherman, MD, Redwood City, California UNITED STATES
The JUPITER Group, Cincinnati, OH UNITED STATES
Stanford University, Palo Alto, California, UNITED STATES
FDA Status Not Applicable
Summary
3D-PASS, developed from 272 MRIs of patients with patellar instability, is associated with instability history; correlates with patient-reported outcomes one year following non-operative treatment; and demonstrates that 3D bone positions are more reflective of patellar instability severity and informative in treatment outcome prediction than traditional imaging measures or 3D bone shape.
ePosters will be available shortly before Congress
Abstract
Introduction
Patellar instability treatment outcomes vary. A third of non-operative patients re-dislocate, and 58% experience activity limitations after 6 months1,2. Accurate prediction of outcomes could enhance personalized interventions. Treatment decisions rely on clinical factors and 2D measures of patellar tilt, patella alta, trochlear dysplasia, and tibial tubercle–trochlear groove (TT-TG) distance. While informative, these measures do not fully leverage 3D MRI bone data. This study aims to develop a 3D Patellar instability Anatomical Severity Score (3D-PASS) associated with instability history and post-treatment outcomes.
Methods
We retrospectively analyzed 272 patients from the JUPITER (Justifying Patellar Instability Treatment by Results) study and 26 age-matched ACL-injured controls. Patellar instability was classified as first-time or recurrent. Patients and their clinicians determined treatment (non-operative or operative).
Patellar tilt, Caton-Deschamps Index (CDI), sulcus angle, and TT-TG distance were measured from baseline imaging. Patients completed the Kujala Anterior Knee Pain Scale (Kujala) and Banff Patellofemoral Instability Instrument (BPII) at baseline and one year after.
From 3D MRIs, we developed a statistical knee model3 to quantify 3D relative bone positions and bone shape.
We compared preliminary scores of instability severity based on four feature sets:
1. 2D imaging measures: patellar tilt, CDI, sulcus angle, and TT-TG distance
2. 3D relative bone positions: rotations and translations between the knee bones (patella, femur, and tibia)
3. 3D bone shape: knee bone curvature
4. 3D relative bone positions & bone shape: Feature Sets 2 and 3 combined
Severity scores reflected distances between the control and recurrent instability cohort means. Zero corresponded to the control mean. Higher scores indicated greater instability severity.
We compared scores using t-tests across instability history and correlations with patient-reported outcomes. A sub-analysis of the best-performing preliminary score led to the 3D Patellar instability Anatomical Severity Score (3D-PASS), using the feature subset with the highest correlation to patient-reported outcomes.
Results
All four preliminary scores increased from the control to the primary instability to the recurrent instability cohorts. The preliminary 3D relative bone positions score best distinguished between the first-time and recurrent instability cohorts (p < 0.001). Of the preliminary scores, it was the only one correlated with one-year non-operative patient-reported outcomes (rKujala = -0.42, rBPII = -0.49). None of the preliminary scores correlated with one-year operative outcomes.
3D-PASS was based on the subset of 3D relative bone position features that maximized correlations with outcomes. Mean 3D-PASS differed significantly across the control (mean: 0.0), first-time instability (mean: 1.4), and recurrent instability (mean: 1.8) cohorts and achieved strong correlations (rKujala = -0.68, rBPII = -0.70) with non-operative outcomes.
Conclusion
The 3D Patellar instability Anatomical Severity Score (3D-PASS) is associated with instability history and patient-reported outcomes one year following non-operative treatment. Based on bone positions, 3D-PASS showed stronger associations with recurrence and outcomes than traditional imaging measures and 3D bone shape. Clinically, this novel metric could help identify patients at risk of poor outcomes and guide earlier surgical intervention to improve prognosis.
References:
Dixit 2017. 10.1097/JSA.0000000000000149
Atkin 2000. 10.1177/03635465000280040601
Cootes 1999. 10.1093/oso/9780199637010.003.0007