Logo of the Physikalisch-Technische Bundesanstalt

Compressed AFM-IR hyperspectral nanoimaging


We introduced a novel approach involving low-rank matrix reconstruction, applied to the sub-diffraction technique of atomic force microscopy-based infrared spectroscopy (AFM-IR), thereby enabling highly efficient hyperspectral infrared nanoimaging. Its practical utility was demonstrated on Leishmania parasites as a realistic target of biological importance.


In the field of materials and life sciences, infrared hyperspectral imaging has proven to be a powerful method. Nevertheless, when it comes to expanding its utility to modern sub-diffraction nanoimaging, it encounters significant inefficiencies due to its inherent reliance on sequential data acquisition processes. To address this concern, the working groups of "IR Spectrometry" and "Data Analysis and Measurement Uncertainty" joined forces as part of a DFG project in collaboration with FU Berlin, alongside the "Global Health and Tropical Medicine Institute" and Nova Universität in Lisbon.


We used a so-called AFM-IR instrument where pulsed IR radiation from a quantum cascade laser (QCL) leads to wavelengths dependent absorption, heating and expansion of the sample. This can be sensed by the probe of an atomic force microscopy (AFM), read out by a deflection laser (DF) in combination with a four-quadrant-photodiode (FQD) (see image).

For our imaging demonstration on Leishmania parasites, we focused on the mid-infrared spectral range spanning from 1300 to 1900 cm-1, utilizing a 220 nm spacing for the nanoimaging of individual parasites. We employed k-means cluster analysis to identify chemically distinct spatial locations. Following this, we reduced the initially collected data cube consisting of 134 (x) X 50 (y) X 148 (spectral) AFM-IR measurements to just 10% of its original size and then restored the complete dataset using low-rank matrix reconstruction. This demonstrates agreement within the cluster regions between the complete and reconstructed data cubes. Additionally, we illustrate that the outcomes of the low-rank reconstruction outperform alternative interpolation methods when evaluated in terms of error metrics, cluster quality, and spectral interpretation across different subsampling ratios.


In summary, we find that the utilization of low-rank matrix reconstruction can significantly reduce data acquisition time from, in this case, over 14 hours to just 1-2 hours. These findings can significantly boost the practical applicability of hyperspectral nanoimaging in both academic and industrial settings involving nano- and bio-materials.



Kästner, B., Marschall, M., Hornemann, A., Metzner, S., Patoka, P., Cortes, S., Wübbeler, G., Hoehl, A., Rühl, E., & Elster, C. (2023). Compressed AFM-IR hyperspectral nanoimaging. Measurement Science and Technology, 35(1), 015403. https://doi.org/10.1088/1361-6501/acfc27



B. Kästner, FB 7.1, Bernd.Kaestner(at)ptb.de 

C. Elster, FB 8.4, Clemens.Elster(at)ptb.de