An MRI-Based Clinical-Perfusion Model Predicts Pathological Subtypes of Prevascular Mediastinal Tumors

Area of Science:

• Oncology
• Radiology
• Medical Imaging

Background:

• Prevascular mediastinal tumors (PMTs) encompass diverse subtypes like lymphoma, thymoma, and thymic carcinoma, necessitating accurate preoperative differentiation.
• Accurate pathological subtyping of PMTs is crucial for guiding appropriate clinical management and treatment strategies.

Purpose of the Study:

• To develop and evaluate machine learning (ML) models for predicting the pathological subtypes of PMTs.
• To identify key clinical variables and dynamic contrast-enhanced MRI (DCE-MRI)-derived perfusion parameters that contribute to accurate PMT subtyping.

Main Methods:

• Retrospective analysis of clinical data and preoperative DCE-MRI images from 62 PMT patients (17 lymphoma, 31 thymoma, 14 thymic carcinoma).
• Calculation of six perfusion parameters from DCE-MRI as candidate predictors.
• Development of multi-class and binary classification ML models using univariate receiver-operating-characteristic curve analysis for performance evaluation.

Main Results:

• The multi-class model achieved sensitivities of 52.9% for lymphoma, 74.2% for thymoma, and 92.8% for thymic carcinoma.
• Binary classification models showed 75% sensitivity for distinguishing Hodgkin from non-Hodgkin lymphoma and 71.4% for invasive versus noninvasive thymoma.
• Essential predictive parameters included two perfusion parameters (efflux rate constant and extravascular extracellular space volume), age, and tumor volume.

Conclusions:

• Machine learning models integrating clinical data and DCE-MRI perfusion parameters show promise for the differential diagnosis of PMT subtypes.
• The identified key parameters (perfusion metrics, age, tumor volume) can enhance the accuracy of non-invasive PMT subtyping.
• This ML-based approach offers a potential tool to improve preoperative diagnosis and guide patient management for mediastinal tumors.