AI-based Prediction of Neck Metastasis in Early-stage Lip Squamous Cell Carcinoma Using Preoperative...

Area of Science:

• Oncology
• Artificial Intelligence
• Medical Imaging

Background:

• Early-stage lip squamous cell carcinoma (SCC) requires accurate assessment for lymph node metastasis.
• Preoperative risk stratification is crucial for guiding treatment decisions in lip SCC.
• Current methods for assessing metastasis risk may be limited.

Purpose of the Study:

• To evaluate the feasibility of AI-based large language models (LLMs) for predicting cervical lymph node metastasis (pN+) in early-stage lip SCC.
• To assess the utility of preoperative clinical photographs and basic clinical parameters for AI-driven risk prediction.
• To compare the performance of different LLMs (ChatGPT and Gemini) in this predictive task.

Main Methods:

• Retrospective, single-center study of 14 patients with histopathologically confirmed lip SCC.
• Analysis of preoperative standardized JPEG photographs using ChatGPT and Gemini via a zero-shot approach.
• Comparison of LLM predictions with final pathologic findings from selective neck dissection.

Main Results:

• Both ChatGPT and Gemini overestimated the metastasis risk compared to actual findings (0%).
• ChatGPT demonstrated better calibration (Brier score 0.023) and weaker correlation with depth of invasion (r=0.16) than Gemini (Brier 0.064, r=0.56).
• Both models identified lower-lip and commissural lesions as higher risk.

Conclusions:

• AI-based LLMs can extract relevant risk cues from preoperative photographs for lip SCC.
• AI-assisted preoperative risk assessment is a novel and accessible approach.
• Oversensitivity and reduced specificity remain limitations for current AI models in this context.