ULTRA maps glioma margins in 3D from human surgical tissue in 30 minutes

A brain-tumor margin can look clean on one glass slide and infiltrated on the next. That is the problem ULTRA was built to confront: the platform produced 3D virtual histology of glioma tissue in about 30 minutes, using human surgical specimens rather than conventional staining and sectioning.

The system was developed by researchers from Fudan University, Huashan Hospital of Fudan University, Zhongshan Hospital of Fudan University, Beijing Neurosurgical Institute, Capital Medical University and collaborators. It rapidly clears fresh or fixed brain tissue, images chemical signals with stimulated Raman scattering microscopy, then uses artificial intelligence to restore the image, predict a protein channel from a lipid channel and create H&E-like images that pathologists can read more easily.

That combination addresses a specific weakness of frozen-section pathology. Intraoperative analysis is fast, but it usually relies on a limited number of two-dimensional slices. Gliomas can spread through surrounding brain tissue in irregular patterns, so a slice that appears nearly normal may sit beside one containing tumor cells. ULTRA keeps the tissue volume intact and follows the architecture across depth.

The researchers evaluated the platform on samples from 17 patients undergoing brain-tumor resection, including several glioma subtypes. In one glioblastoma margin specimen, the team processed approximately 1 mm³ of tissue in about 30 minutes and used a 3D convolutional neural network to calculate an “ULTRAscore,” estimating whether each tissue block contained tumor. The resulting map divided the sample into dense tumor, sparse infiltrative tumor and non-tumor brain regions.

So what changes in practice? If the approach can eventually be adapted and validated during surgery, it could give neurosurgeons more tissue-level information about where glioma infiltration continues beyond the obvious mass. In the study, pathologists reviewing only individual 2D sections missed tumor in some depth ranges, with reported miss rates of 20% and 23%, even though adjacent planes contained tumor. For now, ULTRA remains a laboratory result from surgical specimens, not a clinical system shown to guide operations in real time.