Low et al: Machine-learning-based prediction of functional recovery in deep-pain-negative dogs after decompressive thoracolumbar hemilaminectomy for acute intervertebral disc extrusion
Veterinary Surgery 4, 2025

🔍 Key Findings

The study included 162 deep-pain-negative dogs undergoing decompressive surgery (hemilaminectomy) for acute thoracolumbar intervertebral disc extrusion (IVDE).

Ambulatory recovery occurred in 53.1% of dogs (86/162).

The best performing machine-learning model was XGBoost, with an AUC of 0.9502 and accuracy of 89.1%, outperforming Ridge, AdaBoost, and Naive Bayes models.

Preoperative-only XGBoost models were less accurate, with AUC dropping to 0.8271 and accuracy to 71.9%.

Top predictive features (by SHAP analysis) included:

1. T2-weighted to L2 spinal cord signal ratio (lower values predicted better outcome)
2. Use of fenestration (presence associated with better recovery)
3. Hospitalization duration
4. Imaging modality used
5. Duration of nonambulatory status

Machine learning provided better insight into prognostic factors than traditional statistical methods.