Machine-Learning Classifiers in Discrimination of Lesions Located in the Anterior Skull Base

Abstract

Purpose To identify the optimal machine-learning methods to preoperatively differentiate common lesions located in the anterior skull base with radiomic features and clinical features. Method A total of 235 patients diagnosed with pituitary adenoma, meningioma, craniopharyngioma, or Rathke’s cleft cyst were enrolled in current study. The discrimination on lesions was divided into three groups: pituitary adenoma＆craniopharyngioma; meningiomas＆craniopharyngioma, and pituitary adenoma＆Rathke’s cleft cyst. For each group, five selection methods were performed to select the suitable features for classifier, and nine classifiers were tested for performance in discrimination. The diagnostic performance for each combination was evaluated with receiver operating characteristic curve (ROC), with which the accuracy, sensitivity, specificity, and area under curve (AUC) were calculated. Result In each group, the classifiers represented feasible diagnostic performance with AUC more than 0.800 when combined with suitable selection method. Moreover, the combination that least absolute shrinkage and selection operator (LASSO) as selection method, and linear discriminant analysis (LDA) as classifier represented the best comprehensive discriminative ability. Conclusion The pattern recognition techniques could provide feasible diagnostic value in the discrimination of common lesions in anterior skull base; moreover, the selection method LASSO + classifier LDA had the best general performance.

Ähnliche Arbeiten