Quantification Of 3D Knee Morphology In Patients With Patellar Instability

Patellar instability accounts for 3% of adolescent knee injuries requiring surgery. Two crucial radiographic measures in treatment decision-making are the Caton-Deschamps Index (CDI), a measure of patellar height, and the tibial tubercle-trochlear groove distance (TT-TG), a measure of tibial tubercle lateralization relative to the deepest point in the trochlear groove. These measures rely on 2D anatomic landmark projections. The 3D drivers of these measures are not clearly understood. PURPOSE: We aimed to 1) define patellar instability shape features to quantify variations in 3D bony knee morphology and 2) determine the joint alignment and shape determinants of CDI and TT-TG. METHODS: We developed a statistical shape model from knee MRI bone segmentations of 141 adolescents with patellar instability. Joint alignment parameters (scale, translations, and rotations of the tibia and patella relative to the femur) were extracted during bone registration. Shape features, which represent the modes of greatest shape variation, were determined using principal component analysis of the registered bone mesh coordinates (Figure A). We developed linear regressions to estimate 2D radiologist-measured CDI and TT-TG from our 3D shape model. We identified joint alignment and shape drivers of these measures by comparing regression coefficient magnitudes. RESULTS: Joint alignment parameters alone estimate CDI and TT-TG better than shape features alone and nearly as well as the inputs combined (Figure B). Model feature importance suggests that 1) CDI is driven not only by patellar height but also anterior translation of the patella and 2) TT-TG is driven mainly by external tibial rotation and not by trochlear groove medialization on the femur or tibial tubercle lateralization on the tibia. CONCLUSIONS: We developed a shape model that quantifies complex 3D bony knee morphology for detecting the drivers behind 2D radiographic measures used in patellar instability treatment. Support from the Arthroscopy Association of N Am, CONMED, the Pediatric Orthopaedic Society of N Am, the University of Cincinnati, the Mobilize Center (NIH P41EB027060), and Stanford Data Science. Machine learning tools (scikit-learn, scipy, vtk) were used to develop our models.