This article demonstrates machine learning’s utility in extracting quantitative radiomics data from non-contrast CT to predict postoperative recovery for paraplegic dogs with acute thoracolumbar intervertebral disc extrusion (IVDE). This retrospective study was performed on 214 paraplegic dogs with acute IVDE from 2018-2025. 215 total features (including radiomics features from spinal canal and vertebral column segmentations) were used to train and test the eXtreme Gradient Boosting (XGBoost) model on an 80:20 split. Deep pain perception (DPP) was the clinical reference standard to assess IVDE, with deep pain negative (DPN) indicating severe spinal cord injury. The model demonstrated strong performance with an AUC of 0.91 and 86.1% accuracy in the full data set, with a positive predictive value (PPV) of 100%. This meant that in every case where a dog successfully recovered, the model was able to accurately predict this, outperforming DPP as a recovery metric. In DPN dogs, specifically, the model had high specificity but moderate sensitivity, indicating it is more reliable in predicting a dog’s chances of recovery compared to chances of not recovering. Neurological grade was the most impactful feature in prognosis, but collective radiotics and clinical assessment led to the highest prognostic confidence beyond traditional neurological assessments alone.