An automated approach for fringe frequency estimation and removal in infrared spectroscopy and hyperspectral imaging of biological samples

Analytical and experimental solutions for Fourier transform infrared microspectroscopy measurements of microparticles: A case study on Quercus pollen

BACKGROUND

FTIR microspectroscopy is a popular non-destructive technique for chemical analysis and identification of microparticles, such as microplastics, pollen, spores, microplankton organisms, sediments and microfossils. Unfortunately, measured spectra of microparticles are usually distorted by Mie-type scattering interferents thus hindering the analysis of spectral data. To retrieve chemical absorbance spectra, two different approaches are regularly employed: analytical (application of scatter-correction preprocessing methods), and experimental (measurement in an embedding matrix). The comparative studies of preprocessing spectral strategies are needed to determine pros and cons of these approaches, and when they are most suitable for use.

RESULTS

We conducted the first-ever comparative study on 12 different analytical and experimental approaches for FTIR measurements of microparticles, as demonstrated on classification and chemical characterisation of pollen of four Quercus species. Individual pollen grains were measured on 1) microscope slides and 2) embedded in a paraffin-polyethylene (PEP) matrix. For analytical approaches, we have applied simple model-based algorithm (EMSC: extended multiplicative signal correction), Mie-theory model-based algorithm (ME-EMSC: Mie-extinction EMSC) and deep learning-based algorithm (DCNN: deep convolutional neural network). Moreover, we applied algorithms for the correction of the embedded spectra: fringe-correction EMSC and two different paraffin-correction EMSC algorithms. The best classification accuracy is obtained for simple preprocessing, where scattering information is not completely removed, as well as for complex algorithms where scattering information is parameterized and retained. In chemical characterisation studies, strong scattering signals hinder valuable chemical information, and it is imperative to suppress them either by embedding or by an analytical approach.

SIGNIFICANCE

The results show that scattering spectral interferents are not necessarily detrimental for classification studies of biological microparticles. In fact, they have considerable diagnostic value even in closely related microorganisms due to species-specific physical properties. The results clearly show that analytical and experimental solutions for FTIR measurements of microparticles should be carefully selected, taking into account the origin of the microparticles (i.e., biological or artificial) and purpose of the study (classification or chemical characterisation).

Full fingerprint hyperspectral imaging of prostate cancer tissue microarrays within clinical timeframes using quantum cascade laser microscopy

One of the major limitations for clinical applications of infrared spectroscopic imaging modalities is the acquisition time required to obtain reasonable images of tissues with high spatial resolution and good signal-to-noise ratio (SNR). The time to acquire a reasonable signal to noise spectroscopic scan of a standard microscope slide region of tissue can take many hours. As a trade-off, systems can allow for discrete wavenumber acquisitions, sacrificing potentially vital chemical bands in order to reach specific acquisition targets. Recent instrumentation developments now allow for the full fingerprint imaging of entire microscope slides in under 30 minutes, enabling rapid, high quality spectroscopic imaging of tissues within clinical timeframes without sacrificing frequency bands. Here we compare the data from a novel QCL microscope to an FTIR microscope covering multiple aspects of spectroscopic imaging of a large, clinically relevant, prostate cancer tissue cohort (N = 1281). Comparisons of hyperspectral data acquisition quality in both achieved signal to noise and image contrast alongside the capacity for unsupervised and supervised modelling of tissue constituents are reported. We conclude that it is now possible to collect full fingerprint spectra and derive clinically relevant data in a timeframe suitable for translation into the pathology laboratory without the need to resort to discrete frequency imaging with subsequent loss of information.

The Use of Constituent Spectra and Weighting in Extended Multiplicative Signal Correction in Infrared Spectroscopy

Extended multiplicative signal correction (EMSC) is a widely used preprocessing technique in infrared spectroscopy. EMSC is a model-based method favored for its flexibility and versatility. The model can be extended by adding constituent spectra to explicitly model-known analytes or interferents. This paper addresses the use of constituent spectra and demonstrates common pitfalls. It clarifies the difference between analyte and interferent spectra, and the importance of orthogonality between model spectra. Different normalization approaches are discussed, and the importance of weighting in the EMSC is demonstrated. The paper illustrates how constituent analyte spectra can be estimated, and how they can be used to extract additional information from spectral features. It is shown that the EMSC parameters can be used in both regression tasks and segmentation tasks.

Preprocessing Strategies for Sparse Infrared Spectroscopy: A Case Study on Cartilage Diagnostics

The aim of the study was to optimize preprocessing of sparse infrared spectral data. The sparse data were obtained by reducing broadband Fourier transform infrared attenuated total reflectance spectra of bovine and human cartilage, as well as of simulated spectral data, comprising several thousand spectral variables into datasets comprising only seven spectral variables. Different preprocessing approaches were compared, including simple baseline correction and normalization procedures, and model-based preprocessing, such as multiplicative signal correction (MSC). The optimal preprocessing was selected based on the quality of classification models established by partial least squares discriminant analysis for discriminating healthy and damaged cartilage samples. The best results for the sparse data were obtained by preprocessing using a baseline offset correction at 1800 cm-1, followed by peak normalization at 850 cm-1 and preprocessing by MSC.