- Published on
Research - ML Integration of ICG Fluorescence and Raman for Precise Glioblastoma Margin Determination
- Author
-
-
- Name
- Posts
- Posts
-
We've published work in Research (a Science Partner Journal) on a dual-modal "search-and-confirm" approach to glioblastoma surgery that pairs indocyanine green (ICG) fluorescence with label-free Raman spectroscopy. ICG is widely used to highlight tumor tissue intraoperatively, but it accumulates wherever vascular boundaries are disrupted, which leads to signal outside the tumor and a systematic overestimate of the tumor boundary, risking resection of healthy brain tissue. Raman spectroscopy reads molecular composition directly and matches the true margin, but it's too slow to survey an open surgical field point by point.
The two methods are hard to combine because the 785-nm excitation commonly used in Raman spectroscopy drives strong ICG fluorescence that swamps the weak Raman signal. We resolved this through optimized ICG dosing plus E-PLS baseline correction and interpolation to suppress the fluorescence background and enable simultaneous acquisition, without dual-wavelength lasers or extra optics. With this setup, ICG does the fast wide-field "search," and Raman confirms each flagged region at the molecular level in about 3 seconds, reaching 85 to 90% point classification accuracy in vivo.
Across ex vivo and in vivo experiments, Raman-defined boundaries tracked histopathology while ICG boundaries consistently extended past the real tumor. In one telling case, a region that lit up under ICG (and would therefore have been flagged for resection) was histologically normal, and Raman correctly called more than 90% of the point checks in that region noncancerous. By turning Raman into a real-time corrective filter on fluorescence guidance, the combined approach offers a path to more precise, tissue-sparing glioblastoma resection.