Cancer detection from stained biopsies using high-speed spectral imaging.

The escalating demand for diagnosing pathological biopsies requires the procedures to be expedited and automated. The existing imaging systems for measuring biopsies only measure color, and even though a lot of effort is invested in deep learning analysis, there are still serious challenges regarding the performance and validity of the data for the intended medical setting. We developed a system that rapidly acquires spectral images from biopsies, followed by spectral classification algorithms. The spectral information is remarkably more informative than the color information, and leads to very high accuracy in identifying cancer cells, as tested on tens of cancer cases. This was improved even more by using artificial intelligence algorithms that required a rather small training set, indicating the high level of information that exists in the spectral images. The most important spectral differences are observed in the nucleus and they are related to aneuploidy in tumor cells. Rapid spectral imaging measurement therefore can bridge the gap in the machine-aided diagnostics of whole biopsies, thus improving patient care, and expediting the treatment procedure.