Prediction of Response to Spinal Cord Stimulation Using Machine Learning Based on Radiomics and Patient-Reported Outcomes

BACKGROUND AND OBJECTIVES: Chronic pain affects more patients than cancer, diabetes, and heart disease combined, resulting in high morbidity and significant healthcare costs. Spinal cord stimulation (SCS), which is an Food and Drug Administration-approved treatment for conditions such as complex regional pain syndrome and refractory back pain, has increased by 20% over the past 5 years, partially because of the opioid epidemic. Despite its growth, SCS has substantial failure rates due to inadequate patient selection criteria. To improve outcomes and reduce healthcare costs, machine learning (ML) models incorporating radiomics are developed to identify patients likely to benefit from SCS. METHODS: In this study, we developed ML models that integrate spinal imaging radiomics and clinical data to predict patient responses to SCS. We used ML models on the largest US SCS database, integrating spinal imaging with clinical data to predict patient responses accurately. RESULTS: Integrating radiomic measures with clinical variables enhanced the model's predictive capability, achieving an accuracy of 90.00%, an area under the curve of 91.40%, a sensitivity of 84.62%, and a specificity of 94.12% for the "50% Responder" target. For the "70% Responder" target, the model demonstrated consistently strong predictive performance, with an accuracy of 90.00%, an area under the curve of 86.11%, a sensitivity of 83.33%, and a specificity of 91.67%. CONCLUSION: Our study demonstrates the value of ML models combined with systematic feature selection in predicting clinical outcomes, emphasizing the importance of integrating radiomics and clinical variables for improved model interpretability and robustness.