Deep representation learning for domain adaptatable classification of infrared spectral imaging data

Abstract

Motivation:Applying infrared microscopy in the context of tissue diagnostics heavily relies on computationally preprocessing the infrared pixel spectra that constitute an infrared microscopic image. Existing approaches involve physical models, which are non-linear in nature and lead to classifiers that do not generalize well, e.g. across different types of tissue preparation. Furthermore, existing preprocessing approaches involve iterative procedures that are computationally demanding, so that computation time required for preprocessing does not keep pace with recent progress in infrared microscopes which can capture wholeslide images within minutes. Results:We investigate the application of stacked contractive autoencoders as an unsupervised approach to preprocess infrared microscopic pixel spectra, followed by supervised fine-tuning to obtain neural networks that can reliably resolve tissue structure.

Single‐layer multiple‐kernel‐based convolutional neural network for biological Raman spectral analysis

“…Representation learning will allow SLMK‐CNN to achieve better generalization by learning representations of the input Raman data that make it easier to extract useful Raman spectral information. The generalization performance of representation learning has been demonstrated in many previous studies in which the following learning algorithms were used: unsupervised learning algorithms such as deep belief net or stacked denoising autoencoder and transfer learning algorithm . In addition, generalization performance is usually improved by providing a larger quantity of representative data .…”

Label-free, automated classification of microsatellite status in colorectal cancer by infrared imaging

“…Owing to the current study design, we selected retrospective formalin-fixed, paraffine-embedded (FFPE) samples from sporadic UICC stage II/III patients, but this approach could also be applied to fresh frozen tissue by re-training RF classifiers or switching to deep learning algorithms. Recent progress in applying deep learning algorithms to hyperspectral data allows transfer learning from FFPE samples to fresh frozen ones for the same entity 54 . Following a standard dewaxing procedure for FFPE samples, no further processing of the thin tissue sections is needed.…”