Abstract 4517: Comprehensive intraoperative imaging of surgical margins with AI-driven open-top light-sheet microscopy

Abstract

Surgeons rely on subjective visual and tactile feedback to differentiate tumors from adjacent normal tissue, which lacks accuracy and may make complete resection challenging. Over 20% of breast and 40% of head & neck cancer surgeries yield positive margins—cancerous cells at the edge of the resected specimen—requiring costly additional treatments with inferior outcomes. When pathologists assess margins, they quantify the tumor's proximity to the margin surface, where a margin of several millimeters is often desired. This is done by viewing formalin-fixed, paraffin-embedded (FFPE) tissue sections oriented perpendicularly to the specimen surface. FFPE sections are typically ∼5 µm thick and are obtained at 3 - 5 mm intervals, meaning that < 0.1% of the specimen surface is evaluated. We propose that for solid continuous tumors, comprehensive surface microscopy — though unable to measure the tumor’s proximity to the specimen surface — could more accurately identify positive margins and could be performed rapidly in the operating room to maintain specimen orientation. We are developing an intraoperative open-top light-sheet (OTLS) microscopy system that comprehensively assesses fresh specimen surfaces through an artificial intelligence (AI)-driven, time-efficient, multi-resolution workflow. The system resembles a flatbed scanner and images the bottom face of a specimen placed on a glass plate. Fresh specimens are rapidly stained (< 1 min) with a fluorescent agent for nuclear and stromal contrast. A profilometer quickly (< 1 min) obtains a height map of the specimen’s bottom surface to guide the OTLS microscope as it captures a thin volume (50 – 100 µm in depth) encompassing the specimen’s surface. The OTLS microscope operates at two resolution modes: low (∼5 × 5 × 5 µm xyz) and high (∼1 × 1 × 3 µm xyz). First, a comprehensive low-resolution scan is acquired rapidly (< 3 min per 10 x 10 cm area). An AI model then identifies high-risk regions to re-scan at high resolution for diagnosis. Our 3D AI-driven imaging strategy mimics the time-efficient workflow of pathologists, who first examine FFPE slides at low magnification before zooming into localized regions at high magnification. The ability to acquire a shallow amount of volumetric data (up to ∼ 100 microns deep) can improve positive margin detection over a single 2D surface due to the greater amount of cellular information and the ability to avoid artifacts from surgical damage (e.g. cautery). To develop our 3D AI algorithm, we are building a dataset from freshly excised breast and head & neck specimens, with ground-truth diagnostic labels provided by pathologists. The model will be weakly supervised, using pre-trained models for 3D feature extraction.In summary, AI-guided multi-resolution OTLS microscopy has the potential to enable rapid, comprehensive intraoperative margin assessment, reducing positive margin rates and improving patient outcomes. Citation Format: David Brenes, Qinghua Han, Rui Wang, Rauf Kareem, Michael C. Topf, Eben L. Rosenthal, Emily Marchiano, Emily Palmquist, Faisal Mahmood, Sara H. Javid, Suzanne M. Dintzis, Jonathan T. Liu. Comprehensive intraoperative imaging of surgical margins with AI-driven open-top light-sheet microscopy [abstract]. In: Proceedings of the American Association for Cancer Research Annual Meeting 2025; Part 1 (Regular s); 2025 Apr 25-30; Chicago, IL. Philadelphia (PA): AACR; Cancer Res 2025;85(8_Suppl_1): nr 4517.