Radiomics-Based Detection of Radionecrosis Using Harmonized Multiparametric MRI

Area of Science:

• Radiology
• Medical Imaging
• Oncology

Background:

• Differentiating between radiation necrosis and tumor progression after treatment is crucial for patient management.
• Multiparametric MRI offers rich data but is susceptible to variability from different scanners and protocols.
• Radiomics aims to extract quantitative features from medical images for improved diagnostic accuracy.

Purpose of the Study:

• To assess the impact of data harmonization techniques on radiomic feature variability.
• To evaluate the performance of radiomic models in distinguishing radiation necrosis from tumor progression using multiparametric MRI.
• To identify optimal MRI sequences and radiomic approaches for this classification task.

Main Methods:

• Radiomic feature analysis was performed on multiparametric MRI data from a multicentric clinical trial.
• Sensitivity of features to protocol variations was assessed.
• Image normalization and ComBat harmonization were employed to reduce scanner-related variability.
• Various radiomic models (LASSO, Random Forest) were trained and evaluated for classification performance.

Main Results:

• Harmonized radiomic models demonstrated improved predictive performance compared to models using raw data (balanced accuracy increased from 0.61 to 0.72 with ComBat harmonization).
• A model combining MR perfusion features with clinical observation achieved the highest classification accuracy (0.76).
• T1-weighted MRI before contrast administration also showed promising performance, nearing that of perfusion-based models.

Conclusions:

• Data harmonization is essential for robust radiomic analysis in multicentric MRI studies.
• Radiomics, particularly using perfusion and T1-weighted imaging, can effectively differentiate radiation necrosis from tumor progression.
• Optimized radiomic models enhance diagnostic capabilities in neuro-oncology patient follow-up.