Uncertainty-Aware Machine Learning for Predicting Axillary Lymph Node Metastasis Using Breast MRI

Area of Science:

• Oncology
• Radiology
• Artificial Intelligence

Background:

• Axillary lymph node metastasis (ALNM) is a critical prognostic factor in breast cancer, influencing staging, treatment, and survival outcomes.
• Accurate preoperative prediction of ALNM is essential for personalized treatment strategies and improved patient management.

Purpose of the Study:

• To develop and validate a machine learning model for predicting preoperative axillary lymph node metastasis (ALNM) using breast MRI and clinicopathological data.
• To achieve probability calibration and uncertainty-aware predictions for enhanced reliability in risk assessment.

Main Methods:

• A retrospective study of 204 breast cancer patients who underwent contrast-enhanced MRI.
• Development and evaluation of machine learning models, including probability calibration and conformal prediction for uncertainty quantification.
• Dataset split into training (60%), calibration (20%), and testing (20%) sets for robust model assessment.

Main Results:

• The Conformal-Calibrated Interpretable Risk Model (CCIRM) demonstrated superior performance with a test accuracy of 0.9268 and AUC of 0.937.
• CCIRM achieved high precision (0.9545) and recall (0.9130) for the ALNM-positive class.
• The model provides calibrated risk estimates and uncertainty quantification, aiding interpretation in borderline cases.

Conclusions:

• The proposed CCIRM framework offers a reliable and trustworthy approach for preoperative ALNM risk estimation.
• The model's ability to provide uncertainty-aware predictions supports clinical decision-making.
• Prospective, multicentre external validation is recommended before clinical implementation.