Prediction of Pivot Shift Grade Using In-Vivo Ultrasound Bone Tracking During Sit-Stand-Sit: A Machine Learning Feasibility Study

BackgroundThe pivot shift (PS) test is the most specific clinical examination for anterolateral rotational instability in ACL-deficient knees, yet grading remains subjective, as evidenced by poor inter-observer reliability, particularly for Grade 2. Since low-grade (Grade 1) versus high-grade (Grades 2/3) PS is the threshold for recommending lateral extra-articular augmentation, performing the test in awake clinic patients limits grading reproducibility and introduces variability in surgical decision-making. Existing methods to quantify the pivot shift usually require examiner-performed testing under general anaesthesia. No prior approach has ascertained PS grading from a separate patient-performed functional movement. PurposeTo evaluate the feasibility of a machine learning (ML) classifier, trained on kinematic ultrasound bone-tracking signals acquired during a patients sit-stand-sit (SSS) knee movement, to predict their PS grade, and to clinically validate its ability to differentiate low versus high-grade PS. MethodsUltrasound bone-tracking kinematic data were collected during SSS manoeuvres in 23 ACL-injured patients using the GATOR device, and ground truth PS grades (0-3) were assigned under general anaesthesia by fellowship-trained orthopaedic sports surgeons. From the data collected, Leave-one-out cross-validation (LOOCV) was used to train the ML classifier. Clinical SSS data from 6 ACL-deficient patients was used for independent held-out validation of their low-grade (Grade 1) versus high-grade (Grade 2/3) PS. Multiple deep learning architectures (XceptionTime, InceptionTime, FCN, ResNet, ResCNN) and training strategies (including mixup augmentation and supervised contrastive learning) were tested. Performance was measured by one-versus-rest (OVR) AUC under LOOCV and by AUC (low vs high grade PS) from the held-out patient sessions. ResultsThe ML classifier achieved a maximum OVR AUC of 0.928 {+/-} 0.084 under LOOCV. Classifier performance increased with pivot-shift severity: Grade 3 was identified most reliably (AUC ~0.81; sensitivity 0.70-0.80), whereas Grade 2 remained the most challenging boundary (sensitivity 0.20-0.75 across configurations). For the clinically relevant binary classification of low-versus high-grade pivot shift, the classifier generalised well to a completely unseen patient cohort (AUC 0.889; accuracy 0.860; sensitivity 0.850; minimum-class sensitivity 0.767). ConclusionThe study demonstrates that kinematic ultrasound bone-tracking during sit-stand-sit contains transferable information about rotational instability severity in ACL-deficient patients, and represents the first reported approach to predict pivot shift grade from a patient-performed functional movement. The strong cross-validation performance confirms that the signals contain meaningful PS grade-discriminative information, but larger datasets targeting 50-100 sessions per grade will be required to achieve patient-level generalisation and advance this novel rotational instability assessment tool toward full clinical adoption. Level of EvidenceLevel IV, diagnostic feasibility study.