Identification of Prolactinoma in Pituitary Neuroendocrine Tumors Using Radiomics Analysis Based on Multiparameter...

Area of Science:

• Oncology
• Radiology
• Artificial Intelligence
• Medical Imaging

Background:

• Accurate pre-operative histological subtyping of pituitary neuroendocrine tumors (PitNETs) is crucial for treatment planning.
• Current diagnostic methods often rely on post-operative pathological examination, delaying definitive treatment strategies.
• Multiparameter Magnetic Resonance Imaging (MRI) offers rich data for non-invasive tumor characterization.

Purpose of the Study:

• To evaluate the feasibility of using machine learning and radiomics on multiparameter MRI for pre-operative prediction of PitNET histological subtypes.
• To develop and validate models for both multi-class (six subtypes) and binary (PRL vs. non-PRL) classification of PitNETs.
• To construct a clinical-radiomics nomogram for enhanced predictive accuracy.

Main Methods:

• Retrospective analysis of 1206 PitNET patients from four medical centers (January 2016 - May 2022).
• Automated segmentation of PitNETs using a cfVB-Net deep learning model on multiparameter MRI.
• Extraction of radiomics features, calculation of radiomics scores (Radscore), and application of Gaussian Process (GP) machine learning classifiers.
• Development of a clinical-radiomics nomogram integrating clinical factors and Radscores using logistic regression.

Main Results:

• The automated segmentation model achieved a mean Dice similarity coefficient of 0.888.
• The GP model using T2WI demonstrated the highest Area Under the ROC Curve (AUC) for multi-classification (0.711 in the external test set).
• A combination of T2WI and contrast-enhanced T1WI in the GP model showed strong performance for binary classification (AUC of 0.791 in the external test set).
• The clinical-radiomics nomogram identified Radscores and Hardy grade as significant predictors for Prolactin (PRL) expression.

Conclusions:

• Machine learning and radiomics analysis of multiparameter MRI are effective for pre-operative prediction of PitNET histological subtypes.
• The developed models and nomogram show significant clinical application value in improving diagnostic accuracy before surgery.
• This non-invasive approach can aid in personalized treatment strategies for patients with PitNETs.