A radiomics pipeline dedicated to Breast MRI: validation on a multi-scanner phantom study

Area of Science:

• Medical Imaging
• Radiomics
• Biomedical Engineering

Background:

• Quantitative MRI analysis faces challenges from intensity variations and bias fields.
• These issues impact the reliability of radiomic features in breast MRI studies.

Purpose of the Study:

• To develop and assess a dedicated pipeline for standardizing breast MRI data.
• The goal is to enhance the reproducibility of radiomic studies across different scanners and acquisitions.

Main Methods:

• T1, T2, and T1-Dynamic Contrast-Enhanced (DCE) MRI data from breast phantoms were acquired.
• Images underwent bias field correction, normalization (Z-score or histogram matching), and ComBat harmonization.
• Radiomic features were extracted and statistically compared.

Main Results:

• The standardization pipeline substantially reduced intra- and inter-acquisition variability.
• ComBat harmonization decreased significantly different radiomic features between coils from 87% to 3%.
• Classification performance for lesions in phantoms was preserved or improved.

Conclusions:

• A novel standardization pipeline effectively reduces variability in breast MRI.
• This approach supports robust multi-scanner radiomic studies.
• Further validation on patient data is warranted.