Deep representation learning for domain adaptable classification of infrared spectral imaging data.

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 whole-slide 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. To validate the robustness of the resulting classifier, we demonstrate that a network trained on embedded tissue can be transferred to classify fresh frozen tissue. The features obtained from unsupervised pretraining thus generalize across the large spectral differences between embedded and fresh frozen tissue, where under previous approaches separate classifiers had to be trained from scratch. AVAILABILITY AND IMPLEMENTATION: Our implementation can be downloaded from https://github.com/arnrau/SCAE_IR_Spectral_Imaging. SUPPLEMENTARY INFORMATION: Supplementary data are available at Bioinformatics online.

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

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.

Microsatellite instability (MSI-H) is associated with a high immunoscore but not with PD-L1 expression or increased survival in patients (pts.) with metastatic colorectal cancer (mCRC) treated with oxaliplatin (ox) and fluoropyrimidine (FP) with and without bevacizumab (bev): a pooled analysis of the AIO KRK 0207 and RO91 trials.

INTRODUCTION: In a retrospective analysis of two randomized phase III trials in mCRC patients treated first line with oxaliplatin, fluoropyrimidine with and without Bevacizumab (the AIO KRK 0207 and R091 trials) we evaluated the association of high microsatellite instability (MSI-H), immunoscore (IS) and PD-L1 expression in relation to overall survival (OS). METHODS: In total, 550 samples were analysed. Immunohistochemical analysis of the MMR proteins and additionally fragment length analysis was performed, molecular examinations via allele-discriminating PCR in combination with DNA sequencing. Furthermore PD-L1 and IS were assessed. RESULTS: MSI-H tumors were more frequent in right sided tumors (13.66% vs. 4.14%) and were correlated with mutant BRAF (p = 0.0032), but not with KRAS nor NRAS mutations (MT). 3.1% samples were found to be PD-L1 positive, there was no correlation of PDL1 expression with MSI-H status, but in a subgroup analysis of MSI-H tumors the percentage of PD-L1 positive tumors was higher than in MSS tumors (9.75% vs. 2.55%). 8.5% of samples showed a positive IS, MSI-H was associated with a high IS. The mean IS of the pooled population was 0.57 (SD 0.97), while the IS of MSI-H tumors was significantly higher (mean of 2.4; SD 1.4; p =< 0.0001). DISCUSSION: Regarding OS in correlation with MSI-H, PD-L1 and IS status we did not find a significant difference. However, PD-L1 positive mCRC tended to exhibit a longer OS compared to PD-L1 negative cancers (28.9 vs. 22.1 months).