Predictive Model for Selection of Upper Treated Vertebra Using a Machine Learning Approach.

OBJECTIVES: To train and validate an algorithm mimicking experienced surgeons' decision making regarding upper instrumented vertebra (UIV) selection in the surgical correction of thoracolumbar adult spinal deformity (ASD).

METHODS: A retrospective review was conducted of a single center database of ASD patients who underwent fusion of at least the lumbar spine (UIV>L1 to pelvis) from 2013-2018. Data collection included demographic and radiographic information. The sample was stratified into 3 groups: 70% for training, 15% for validation and 15% for performance testing. Using a deep learning algorithm, a neural network model was trained to select between upper thoracic (UT: T1-T6) and lower thoracic (LT: T7-T12) UIV. Parameters used in the deep learning algorithm included demographic, coronal, and sagittal pre-operative alignment and post-operative PI-LL.

RESULTS: 143 patients (mean age: 63.3yo±10.6, 81.8% F) with moderate to severe deformity (MaxCobb: 43°±22; TPA: 27°±14 ; PI-LL: 22°±21) were included. Patients underwent a significant change in lumbar alignment (ΔPI-LL: 21°±16 p<0.001); 35.0% had UIV in the UT, and 65.0% in the LT. At 1Y, revision rate was 11.9% and rate of radiographic proximal junctional kyphosis (PJK) was 29.4%. Neural network was composed of 8 inputs, 10 hidden neurons and 1 output (UT or LT). After training, results demonstrated an accuracy of 81.0%, precision of 87.5%, and recall of 87.5% on testing.

CONCLUSIONS: An artificial neural network successfully mimicked two lead surgeons' decision making in the selection of UIV for ASD correction. Future models integrating surgical outcomes should be developed.