Machine Learning in Patient Selection for Stent Placement and Classification of Malignant SVC Syndrome

Introduction: To use machine learning to analyze features of a multicenter cohort to train a decision tree classification model that accurately predicts patient selection for percutaneous stent placement in the treatment of superior vena cava syndrome. Head-to-head comparison between proposed SVC grading schemes.

Method(s): Double-arm retrospective multi-center study of patients who underwent either stenting or medical management for symptomatic malignant SVC syndrome. Using significance testing of multiple clinical and radiologic features via machine learning classification models, a decision tree model was created to predict patients that would be selected for stent placement. SVC syndrome grading schemes were tested to assess the accuracy in terms of their predictive association with need for stent placement.

Result(s): Preliminary accuracy of the decision tree algorithm was determined to be 80% using the key feature of emergent indication as the primary stratifying factor and subsequently the Stanford classification system. Further training of the algorithm with the entire dataset can improve the accuracy.

Conclusion(s): Machine learning analysis demonstrates the ability to derive simple yet effective algorithms to assist with clinical and procedural decision-making in the field of interventional radiology.

Publication History

Article published online:
09 February 2023