Improved mid-infrared imaging spots signature of cancers

29 May 2019

ICFO and DTU project upconverts mid-IR to near-IR for faster computer-assisted biopsy screening.

Hyperspectral imaging in the mid-IR wavelength region is of interest in many sectors, thanks to the number of chemical species with a spectroscopic response in that range. But efficiently detecting that response has proven to be challenging.

A common approach has been to apply Fourier transform IR (FTIR) imaging, which in turn requires the detectors to be cooled cryogenically to achieve the levels of performance needed, adding to the operation's cost and complexity. Modern laser sources and novel detectors can improve things, but such platforms still rely on direct detection of the mid-IR signal.

A project at DTU Fotonik in Denmark and IFCO in Spain has now developed a possible solution, in which the mid-IR spectral information is shifted into the near-IR while preserving all the spatial information. The work was published in Optica.

"Although mid-infrared spectroscopy is recognized as a powerful tool for chemical analysis, its applicability has been hampered by a lack of affordable light sources and sensitive detectors," said Peter Tidemand-Lichtenberg of DTU Fotonik.

"To overcome this barrier, we used an approach that translates information from the mid-infrared region, where the chemical signatures are most distinct, to the near-infrared, where today's camera technology is most mature and sensitive."

The team's system uses a mid-IR illumination source developed at ICFO delivering 20-picosecond pulses; and a lithium niobate crystal as the non-linear medium for the upconversion process, transferring mid-IR imaging signals to the near-IR range. A standard CCD camera handles image acquisition.

The nature of the non-linear medium usually puts a limit on the field of view (FoV) achievable in upconversion operations, but the project team tackled this by using a galvoscanner to rotate the crystal, an approach known to allow concentric rings in the object plane to effectively be upconverted separately.

Even small rotations can have a significant impact on the field of view, and the project improved things further by adjusting the crystal rotation time to match the camera integration time, removing the need for post-processing of the images. In practice the team found that the FoV was increased by a factor of five compared to a static design, corresponding to an increase in the number of spatially resolvable elements by a factor of 25.

"This approach is generic in nature and constitutes a major simplification in realizing video-frame-rate mid-IR monochromatic imaging," commented the project in its published paper.

Computer-assisted biopsies

One area where improved mid-IR hyperspectral imaging could be most valuable is histopathology, where absorbance by chemical bonds within tissues can provide molecular-specific contrast from unstained tissues. The new system may offer a route to simplifying current biopsy workflows and encouraging objective, rather than subjective, decision making from the spectral data, potentially allowing automated platforms to be used.

To test the system, a pilot study conducted alongside the UK's University of Exeter and Gloucestershire Hospitals NHS Foundation Trust applied the technique to esophageal tissues, evaluating cancerous and healthy samples through a computer-assisted biopsy classification.

It found that morphology and spectral classification using the system matched well with the standard stained histopathology images, and that images from only 62 wavelengths could provide enough spectral data to enable preliminary, unsupervised clustering of tissue types with a performance similar to FTIR imaging. In its paper the team suggests that the number of images needed for computer-assisted classification of biopsies may be as low as 10 to 20.

"Our upconversion imaging approach is generic and constitutes a major simplification in realizing video-frame-rate, mid-infrared monochromatic imaging," said Tidemand-Lichtenberg. "The spectral information provided by this technique could be combined with machine learning to enable faster, and possibly more objective, medical diagnostics based on chemical signatures without the need for staining."