BREAST AI-PLAN: Prompt-Driven AI Assistance for Breast Surgery Planning - A Retrospective Single-center Study

Area of Science:

• Oncoplastic and reconstructive breast surgery
• Artificial intelligence in medicine
• Clinical decision support systems

Background:

• Preoperative planning in breast surgery involves complex patient and tumor factors, alongside aesthetic goals.
• Variability in documentation and expertise can affect the reproducibility of surgical planning across institutions.
• Large language models (LLMs) offer potential for standardizing outputs from clinical data.

Purpose of the Study:

• To assess the capability of a prompt-engineered AI assistant (BREAST AI-PLAN) to replicate multidisciplinary team (MDT) decisions.
• To evaluate the AI's ability to reproduce aesthetic scores using routinely collected clinical data.
• To determine the agreement between AI-generated plans and established clinical decisions in breast cancer surgery.

Main Methods:

• A retrospective single-center study involving 30 women undergoing breast cancer surgery.
• Standardized data collection included demographics, comorbidities, breast anatomy, tumor biology, and staging.
• Structured prompts were generated for the AI assistant, which provided outputs on therapy suitability, surgical plans, reconstruction, and aesthetic scores; MDT decisions served as the reference standard.

Main Results:

• The AI demonstrated very good agreement for neoadjuvant therapy suitability (κ=0.91) and good/moderate agreement for planned surgery (κ=0.58), surgical approach (κ=0.43), reconstruction (κ=0.36), and adjuvant therapy (κ=0.29).
• Aesthetic scoring showed high observed concordance (0.778-0.896) but slight weighted agreement (κ=0.14-0.16).
• Patient demographics included a mean age of 59.2 years, median BMI of 24.2, with varying comorbidities and tumor subtypes.

Conclusions:

• The BREAST AI-PLAN system successfully replicated key elements of MDT surgical planning and aesthetic assessments with significant agreement.
• AI-driven assistance has the potential to enhance standardization and auditability in surgical planning, benefiting both current practice and training.
• Further prospective validation, including integration with imaging data, is recommended to fully establish the clinical utility of this AI tool.