"Any sample can be analysed for desired stains without material cost, time or effort, while leaving precious tissue pristine for downstream analyses," Bhargava has been quoted as saying. To study tissue samples, pathologists use dyes that stick to the particular molecule they are looking for. Staining can be a long and exacting process and the chemicals can damage cells. Pathologists also have to choose which things to test for, because it is not always possible to obtain multiple samples for multiple stains from one biopsy.
Instead of using stains, the new infrared imaging technique scans the sample with infrared light to measure the chemical composition of the cells. The computer then translates spectral information from the microscope into chemical stain patterns, obviating the need to apply dyes to the cells.
"We're relying on the chemistry to generate the ground truth and act as the 'supervisor' for a supervised learning algorithm," says David Mayerich, first author of the study. "One of the bottlenecks in automated pathology is the extensive processing that must be applied to stained images to correct for staining artefacts and inconsistencies. The ability to apply stains uniformly across multiple samples could make these initial image processing steps significantly easier and more robust".
The researchers reproduced a wide array of molecular stains by computationally isolating the spectra of specific molecules.
This allows the user to simply tune to a required stain, for as many different stains as are necessary -- all without damaging the original tissue sample, which can then be used for other tests. "This approach promises to have immediate and long-term impact in changing pathology to a multiplexed molecular science -- in both research and clinical practice," Bhargava said.