A representation learning approach for recovering scatter‐corrected spectra from Fourier‐transform infrared spectra of tissue samples

Abstract

Infrared spectra obtained from cell or tissue specimen have commonly been observed to involve a significant degree of scattering effects, often Mie scattering, which probably overshadows biochemically relevant spectral information by a nonlinear, nonadditive spectral component in Fourier transform infrared (FTIR) spectroscopic measurements. Correspondingly, many successful machine learning approaches for FTIR spectra have relied on preprocessing procedures that computationally remove the scattering components from an infrared spectrum. We propose an approach to approximate this complex preprocessing function using deep neural networks. As we demonstrate, the resulting model is not just several orders of magnitudes faster, which is important for real‐time clinical applications, but also generalizes strongly across different tissue types. Using Bayesian machine learning approaches, our approach unveils model uncertainty that coincides with a band shift in the amide I region that occurs when scattering is removed computationally based on an established physical model. Furthermore, our proposed method overcomes the trade‐off between computation time and the corrected spectrum being biased towards an artificial reference spectrum.

Attenuated Total Reflection (ATR) Micro-Fourier Transform Infrared (Micro-FT-IR) Spectroscopy to Enhance Repeatability and Reproducibility of Spectra Derived from Single Specimen Organic-Walled Dinoflagellate Cysts

“…37,51–59 In the past few years, deep learning approaches have made their advance in FT-IR spectroscopy. 60–63 Worth noticing is that recently developed deep neural network architectures can successfully approximate complex, often computationally heavy, model-based correction algorithms. 64 The currently available correction algorithms are often optimized for transmission spectra, as the distortions in transflection spectra are in general more severe.…”

Encoder–decoder neural networks for predicting future FTIR spectra – application to enzymatic protein hydrolysis

“…To obtain more robust models than the composite ones discussed here possible alternative strategies include (1) tuning the encoder-decoder networks with added AMW, substrate and enzyme information (as mentioned above), (2) retraining the AMW prediction on the output from the networks and (3) combining the FTIR networks with AMW prediction modelling for a joint optimization. We remark that for the combined approach (3), exploring the pre-training and fine-tuning strategy [18,19] based on FTIR autoencoders appears fruitful.…”

“…For process monitoring, Jo et al [14] and Jinadasa et al [15] apply AE for feature/ signal extraction in near-infrared and Raman spectroscopy. In several recent papers [16,17] AEs and/or encoder-decoder networks [18,19] are used for denoising and scattering removal in FTIR spectra or to obtain classifiers for spectral histopathology, for example, Raulf et al [19]. In particular, the networks presented in Raulf et al's papers [18,19] are obtained by fine-tuning the encoder-decoder models, where the encoders (architecture and initial parameters) originate from an autoencoder obtained in a pre-training step.…”

“…In several recent papers [16,17] AEs and/or encoder-decoder networks [18,19] are used for denoising and scattering removal in FTIR spectra or to obtain classifiers for spectral histopathology, for example, Raulf et al [19]. In particular, the networks presented in Raulf et al's papers [18,19] are obtained by fine-tuning the encoder-decoder models, where the encoders (architecture and initial parameters) originate from an autoencoder obtained in a pre-training step. That is, the pre-training provides an initial guess for embedding of the raw spectra suitable for a given problem (specified by the decoder), for example, regression [18] or classification [19].…”

“…In particular, the networks presented in Raulf et al's papers [18,19] are obtained by fine-tuning the encoder-decoder models, where the encoders (architecture and initial parameters) originate from an autoencoder obtained in a pre-training step. That is, the pre-training provides an initial guess for embedding of the raw spectra suitable for a given problem (specified by the decoder), for example, regression [18] or classification [19]. Using pure data-driven approaches Mete et al [20] construct surrogate models of bioreactors.…”