A machine learning-based predictive model for complication risks in vacuum-assisted breast biopsy

Area of Science:

• Medical imaging and interventional radiology
• Machine learning in healthcare
• Breast oncology and diagnostics

Background:

• Ultrasound-guided vacuum-assisted breast biopsy (VABB) is standard for benign breast lesions.
• Postoperative complications like bruising and residual tumors remain clinical challenges.
• Current risk assessment for VABB complications lacks precision.

Purpose of the Study:

• Develop and validate machine learning models to predict VABB complications.
• Improve precision in risk assessment for postoperative adverse events.
• Enhance preoperative planning and patient counseling for VABB procedures.

Main Methods:

• Multicenter retrospective study of 1,064 VABB procedures (2017-2025).
• Developed six machine learning models using 12 preoperative variables.
• Random forest algorithm showed superior performance in cross-validation and external validation.

Main Results:

• Machine learning model achieved high accuracy in predicting bruising (AUC 0.971, 96.7%) and operative duration.
• Key predictors identified: tumor size, blood flow grade, and distance to pectoralis muscle.
• Model demonstrated generalizability with strong external validation (AUC 0.945).

Conclusions:

• A clinically validated machine learning tool accurately predicts common VABB complications, especially bruising.
• Incorporating tumor and anatomical data aids in mitigating adverse outcomes.
• The model can enhance surgical decision-making and patient recovery expectations.