Non-contrast MRI based machine learning and radiomics signature can predict the severity of primary lower limb lymphedema

Motivation: At present, there is no application of radiomics in the diagnosis of the severity of primary lower limb lymphedema(PLE). Goal(s): Develop machine learning models to predict severe PLE. Approach: 119 patients were divided into non severe group (mild, moderate) and severe group. Used five commonly used machine learning models: Logistic Regression, Support Vector Machine, RandomForest, ExtraTrees, and Light Gradient Boosting Machine. Results: The ExtraTree model performed the best in the test set, with an AUC of 0.938, sensitivity of 75%, and specificity of 100%. The net benefit of the ExtraTree model was greater than that of the two extreme curves. Impact: All five models performed well in distinguishing between the nonsevere group and the severe group. NCMRI based machine learning radiomics signature can predict the severity of primary lower limb lymphedema.