Prediction of pituitary adenoma surgical consistency: radiomic data mining and machine learning on T2-weighted MRI

• DOI: 10.1007/s00234-020-02502-z
• Publication Date: 2020-07-23T09:03:53Z
• Authors: Renato Cuocolo; Lorenzo Ugga; Domenico Solari; Sergio Corvino; Alessandra D’Amico; Daniela Russo; Paolo Cappabianca; Luigi Maria Cavallo; Andrea Elefante
• Keywords: Hyperparameter; Feature (linguistics); Neuroradiology

Abstract Purpose Pituitary macroadenoma consistency can influence the ease of lesion removal during surgery, especially when using a transsphenoidal approach. Unfortunately, it is not assessable on standard qualitative MRI. Radiomic texture analysis could help in extracting mineable quantitative tissue characteristics. We aimed to assess accuracy combined with machine learning preoperative evaluation pituitary patients undergoing endoscopic endonasal surgery. Methods Data 89 (68 soft and 21 fibrous macroadenomas) who underwent MRI surgery at our institution were retrospectively reviewed. After manual segmentation, radiomic features extracted from original filtered MR images. Feature stability multistep feature selection performed. oversampling balance classes, 80% data was used for hyperparameter tuning via stratified 5-fold cross-validation, while 20% hold-out set employed its final testing, an Extra Trees ensemble meta-algorithm. The reference based surgical findings. Results A total 1118 extracted, which 741 stable. low variance ( n = 4) highly intercorrelated 625) parameters, recursive elimination identified subset 14 features. tuning, classifier obtained 93%, sensitivity 100%, specificity 87%. area under receiver operating characteristic precision-recall curves 0.99. Conclusion Preoperative T2-weighted predict consistency.