Image-based classifier calls out cancer cells

Ji and colleagues used a microscopy technique called stimulated Raman scattering, or SRS, to image cancer cells in human brain tissue. SRS produces different signals for proteins and lipids, which can then be assigned a colour (blue and green, respectively), allowing the authors to differentiate brain cortex from tumour from white matter. Biopsies from adult and paediatric patients with glioblastoma revealed not only distinctive features with SRS microscopy but also the presence of infiltrating cells in tissues that appeared otherwise normal with traditional staining. Such infiltrating cells are important to catch early because leaving them behind after surgery nearly always leads to cancer recurrence. To make this SRS microscopy approach amenable to routine use in neuropathology, the authors also created an objective classifier that integrated different image characteristics, such as the protein/ lipid ratio, axonal density, and degree of cellularity, into one output, on a scale of 0 to 1, that would alert the pathologist to tumour infiltration. The classifier was built using more than 1400 images from patients with glioblastoma and epilepsy, and could distinguish between tumour-infiltrated and non-tumour regions with >99% accuracy, regardless of tumour grade or histologic subtype. This label-free imaging technology could therefore be used to complement existing neurosurgical workflows, allowing for rapid and objective characterization of brain tissues and, in turn, clinical decision-making.

Science Translation Medicine http://tinyurl.com/oxq8ojx