Developing a radiomics model to predict tumor consistency of pituitary adenomas using multicenter MRI data

Motivation: Knowing the consistency of pituitary adenomas pre-surgery is crucial, yet current radiomic studies lack emphasis on multi-center patient data and more precise prediction. Goal(s): This study developed radiomic models to classify adenomas from multi-center patients into three consistency levels. Approach: Following image preprocessing, a novel feature engineering method called Feature Gradient was applied to select optimal feature subsets for the SVM classification models. Results: Across seven SVM classifiers, the average AUC and accuracy scores on the testing set were 0.68 and 0.66, respectively, and three-sequences radiomic model achieved AUC and accuracy scores of 0.79 and 0.8. Impact: This research underscores the importance of applying radiomics to diverse, generalized patient data, advancing its potential for real-world clinical use and demonstrating its adaptability to varied patient and imaging conditions in practical medical scenarios.