Back to 2026 Abstracts
Standardizing Breast Symmetry Assessment: Development of a Five-Domain Metric with Machine Learning-Based Automation
Lisa Duan, Erin Abbott, Roshan Parikh, Sheuli Chowdhury
*, Peter W. Henderson
Icahn School of Medicine, New York City, NY
Background: Breast symmetry is a critical determinant of aesthetic outcomes in reconstructive and cosmetic breast surgery, particularly in unilateral breast reconstruction. However, symmetry assessment remains subjective, with poor inter-rater reliability and no standardized composite metric. Recent advances in machine learning have enabled automated landmark detection, but integration with a comprehensive symmetry metric remains limited. This study aims to develop a five-domain breast symmetry metric, evaluate inter-rater reproducibility, and establish a machine learning framework for automated assessment.
Methods: A symmetry metric was constructed across five domains: breast width, inframammary fold (IMF) curvature shape and position, and nipple-areolar complex (NAC) shape and position. Each domain incorporated multiple linear, angle, or area measurements. Two independent raters applied the metric to an initial cohort of deidentified postoperative photographs, with ongoing expansion toward a larger cohort (n = 200). Inter-rater agreement was assessed using intraclass correlation coefficients (ICC) and absolute agreement metrics, including mean absolute difference and percent difference. A convolutional neural network based on a pretrained YOLO architecture is being fine-tuned to detect anatomical landmarks and automate symmetry scoring.
Results: In an initial cohort, inter-rater agreement was high across all domains, with mean percent differences ranging from 1.2 to 3.6%. Composite scores demonstrated minimal inter-rater variability, with a mean absolute difference of 0.85 points and a mean difference of 0.9% (range: 0.9-1.9%). A convolutional neural network is under development to automate landmark detection and symmetry scoring, with training and validation in progress.
Conclusion: This study introduces a reproducible, multi-domain breast symmetry metric and establishes a framework for automated assessment. Ongoing work includes validation against manual measurements, assessment of time efficiency, and correlation with subjective patient-reported outcomes.
Models adjusted for BMI, tissue expander size, and axillary irradiation. OR > 1 indicates higher odds with early exchange.
Back to 2026 Abstracts
Models adjusted for BMI, tissue expander size, and axillary irradiation. OR > 1 indicates higher odds with early exchange.
