Abstract 2530: Multispectral imaging system for intraoperative molecular imaging with two near-infrared fluorescent probes

Abstract Image-guided surgery using near-infrared (NIR) fluorescent probes offers surgeons real-time guidance to locate tumors and sentinel lymph nodes. The past decade has seen a proliferation in NIR fluorophore development, with options for targeting different tumor markers and tissue types. Using multiple fluorophores could improve tumor detection sensitivity and specificity, allow simultaneous tumor imaging and lymph node mapping, or distinguish tumors from vital anatomical structures. Current imaging systems, however, are limited to detecting a single NIR probe. This has motivated the development of a novel multispectral camera, capable of simultaneously imaging multiple NIR fluorophores with similar emission spectra that are excited with a single excitation source. The camera features a unique spectral filter array, with filters to capture both visible light and nine overlapping spectral bands in the NIR, monolithically integrated with three vertically stacked photodiodes. The result is a compact instrument that can produce overlaid visible and NIR images without co-registration error, while sampling the NIR spectrum in a nuanced manner to distinguish similar NIR fluorophores. The pixelated filters are arranged in a 2x2 “super-pixel” pattern, comprising a visible filter and three NIR filters with several passbands. The NIR filters produce nine spectral features, which are used to classify fluorescent spectra using machine learning. Initial experiments were performed to demonstrate the camera’s ability to differentiate two NIR fluorophores in vials. Combinations of indocyanine green (ICG), pafolacianine, and JAS239 in 10% FBS solutions were mixed with a range of volumetric ratios. The fluorophore concentrations were chosen to have similar overall fluorescence intensity, indistinguishable by a traditional NIR camera. Gradient boosting classifiers were trained to predict the dye or combination of dyes that produced the fluorescence at each pixel in an image. Pure ICG and pafolacianine were classified with 97% accuracy and their mixtures with 77% accuracy. ICG and JAS239 were discriminated with 99% accuracy, and mixtures of these two dyes with 81% accuracy. The results demonstrate that our single-chip multispectral imager captures sufficient information to distinguish two NIR fluorescent probes, a task unachievable by current instruments. Future steps will test the discrimination capabilities in more complex samples, moving closer to clinical applications. Citation Format: Brianna Hajek, Zhongmin Zhu, Viktor Gruev. Multispectral imaging system for intraoperative molecular imaging with two near-infrared fluorescent probes [abstract]. In: Proceedings of the American Association for Cancer Research Annual Meeting 2025; Part 1 (Regular Abstracts); 2025 Apr 25-30; Chicago, IL. Philadelphia (PA): AACR; Cancer Res 2025;85(8_Suppl_1):Abstract nr 2530.