A Novel Fusion of Radiomics and Semantic Features: MRI-Based Machine Learning in Distinguishing Pituitary Cystic...

Area of Science:

• Radiology and Medical Imaging
• Artificial Intelligence in Medicine
• Oncology

Background:

• Cystic pituitary adenomas (CPA) and Rathke's cleft cysts (RCCs) are common sellar region lesions that can be challenging to differentiate.
• Accurate distinction is crucial for appropriate clinical management and patient outcomes.

Purpose of the Study:

• To evaluate the diagnostic performance of machine learning models utilizing semantic and radiomic features from magnetic resonance imaging (MRI) for differentiating CPA from RCCs.
• To compare the efficacy of models trained on semantic features alone versus a combination of semantic and radiomic features.

Main Methods:

• A cohort of 65 patients with CPA or RCCs was analyzed using MRI.
• Semantic features were assessed by multiple observers, and radiomic features were extracted from T2-weighted, T1-weighted, and T1-contrast-enhanced sequences.
• Machine learning models (SVM, LR, LGB) were trained and validated using different feature combinations.

Main Results:

• Models integrating semantic and radiomic features outperformed those using semantic features alone.
• The highest test accuracies were achieved with combined semantic and T2-weighted radiomic features (LR: 93.8%, SVM: 92.3%, LGB: 90.8%).
• The SVM model with combined features showed statistically significant improvement over semantic features alone (p = 0.019).

Conclusions:

• Machine learning, particularly when combining semantic and radiomic MRI features, shows significant potential for accurate differentiation of CPA and RCCs.
• This approach may enhance diagnostic precision in neuroimaging for sellar region lesions.