Breast Lesion Classification with Multiparametric Breast MRI Using Radiomics and Machine Learning: A Comparison with...

Area of Science:

• Radiology
• Medical Imaging
• Artificial Intelligence in Medicine

Background:

• Accurate breast tumor classification is crucial for patient management.
• Multiparametric MRI offers comprehensive data for lesion assessment.
• Radiomics and machine learning (ML) present novel tools for image analysis.

Purpose of the Study:

• To compare the diagnostic performance of radiomics-ML models with experienced radiologists in classifying breast tumors.
• To evaluate the effectiveness of radiomics features combined with clinical data (BI-RADS, DWI score, ADC) for breast lesion discrimination.

Main Methods:

• Retrospective analysis of 104 breast lesions from 93 women using multiparametric MRI.
• Radiologists assigned BI-RADS categories and provided DWI scores and ADC values.
• Ten radiomics-ML models were developed using features extracted from MRI data.

Main Results:

• Radiomics models incorporating DWI score or ADC values with BI-RADS achieved high accuracy (88.5%) and AUC (0.93-0.96).
• These models demonstrated significant improvement over basic radiomics models.
• Performance was comparable to that of experienced radiologists (accuracy 85.6%, AUC 0.03).

Conclusions:

• Radiomics analysis coupled with ML can significantly aid in breast lesion discrimination.
• This AI-driven approach shows potential to support radiologists, especially those with less experience.
• Multiparametric MRI data, when analyzed with radiomics and ML, enhances diagnostic capabilities for breast tumors.