Rapid AI-assisted intraoperative diagnosis for biliary tract adenocarcinoma using femtosecond laser label-free microscopy.

Abstract

e16019 Background: Intraoperative histopathological assessment is crucial in hepatobiliary surgery, particularly for determining resection margins in biliary tract adenocarcinoma (BTA). Conventional frozen-section analysis is limited by time-consuming processing, structural artifacts, and subjective interpretation, potentially impacting surgical decisions. Methods: To address these limitations, we developed an integrated diagnostic platform combining femtosecond laser label-free microscopy (FLI) with artificial intelligence (AI). This system performs rapid, nondestructive imaging of fresh, unprocessed tissue specimens in under three minutes. The AI model was trained and validated to analyze the FLI-derived images. Results: The AI-enhanced FLI system demonstrated high diagnostic accuracy. It effectively differentiated benign from malignant tissue at biliary margins, achieving an area under the curve (AUC) of 0.915-0.940. Beyond binary classification, the platform predicted key prognostic features: perineural invasion (AUC: 0.817), vascular invasion (AUC: 0.900), and Ki-67 proliferation index (AUC: 0.94), while also providing a multiplexed immuno-profile. The entire workflow, from specimen acquisition to a comprehensive diagnostic report, was completed within five minutes. Conclusions: Our AI-FLI platform represents a transformative tool for intraoperative diagnosis. It delivers rapid, accurate, and multifaceted pathological assessment directly from fresh tissue, overcoming major constraints of frozen-section analysis. This technology provides surgeons with critical, real-time guidance to optimize the extent of resection during BTA surgery, potentially improving oncologic outcomes.