Radiomics-based MRI models for predicting breast cancer axillary lymph node involvement in comparison with Node-RADS:...

Area of Science:

• Oncology
• Radiology
• Artificial Intelligence

Background:

• Accurate detection of axillary lymph node (LN) involvement is critical for breast cancer staging and treatment.
• Current methods for LN assessment can be invasive, necessitating exploration of non-invasive alternatives.

Purpose of the Study:

• To evaluate the feasibility of magnetic resonance imaging (MRI) radiomics-based machine learning models for predicting axillary LN involvement in breast cancer.
• To compare the performance of these radiomics models against the established Node-RADS classification system.

Main Methods:

• Retrospective analysis of multiparametric MRI data from breast cancer patients.
• Extraction of stable radiomic features from contrast-enhanced, subtracted, and T2-weighted MRI sequences.
• Development and external validation of machine learning models for binary LN involvement classification, with histopathology as the reference standard.

Main Results:

• The best-performing radiomics model achieved an 81% AUROC on the internal dataset and 94% AUROC on the external validation cohort.
• The model demonstrated high specificity (92% internal, 100% external) in predicting LN involvement.
• Performance metrics, including accuracy, sensitivity, and specificity, were comparable to Node-RADS in the small external validation cohort, with moderate agreement.

Conclusions:

• MRI radiomics-based machine learning models show potential as a non-invasive tool for assessing axillary LN involvement in breast cancer.
• The comparable performance to Node-RADS suggests these models could aid clinical decision-making, potentially enabling axillary de-escalation strategies.
• Further validation is warranted, but radiomics offers a promising avenue for personalized breast cancer treatment by reducing the need for invasive procedures.