Raman spectroscopy is a powerful technique used across the life sciences to measure the molecular composition of a sample. There has been growing interest to miniaturize Raman imaging devices for endoscopic applications, however typically these probes are based on fiber bundles which increase the overall footprint of the probe. Recent works have shown that by applying a wavefront shaping technique, a single fiber may be transformed into a sub-cellular resolution Raman endoscope. However, a single probe both exciting and collecting the signal leads to an unavoidable large background signal from the fiber itself, masking large portions of the Raman signal from the sample. Here, we adopt a data-driven approach to de-convolve the background signal from the sample. In particular, we demonstrate that by applying PCA and machine learning techniques, sub-cellular resolution Raman images of pharmaceutical clusters can be made with supervision-free analysis.
|