Cascaded Deep Convolutional Neural Networks as Improved Methods of Preprocessing Raman Spectroscopy Data

Abstract

Machine learning has had a significant impact on the value of spectroscopic characterization tools, particularly in biomedical applications, due to its ability to detect latent patterns within complex spectral data. However, it often requires extensive data preprocessing, including baseline correction and denoising, which can lead to an unintentional bias during classification. To address this, we developed two deep learning methods capable of fully preprocessing raw Raman spectroscopy data without any human input. First, cascaded deep convolutional neural networks (CNN) based on either ResNet or U-Net architectures were trained on randomly generated spectra with augmented defects. Then, they were tested using simulated Raman spectra, surface-enhanced Raman spectroscopy (SERS) imaging of chemical species, low resolution Raman spectra of human bladder cancer tissue, and finally, classification of SERS spectra from human placental extracellular vesicles (EVs). Both approaches resulted in faster training and complete spectral preprocessing in a single step, with more speed, defect tolerance, and classification accuracy compared to conventional methods. These findings indicate that cascaded CNN preprocessing is ideal for biomedical Raman spectroscopy applications in which large numbers of heterogeneous spectra with diverse defects need to be automatically, rapidly, and reproducibly preprocessed.

Artificial Intelligence for Surface‐Enhanced Raman Spectroscopy

“…XGBoost [97] IR and Raman spectra Predict surface-adsorbate interaction properties ET, [98] SISSO [98] Synthesis Selected synthesis descriptors Classify selected features of spectra ET [99] The citrate to gold (III) ratio, scanning velocity and radiation intensity Predict nanoparticle size ANN [ 100] Spectroscopic characteristics based on UV-vis/DLS Optimize experimental/reaction conditions GA, [ 101] BO+DNN [ 102] Reaction conditions Predict selected spectroscopic characteristics of nanoparticles SVR [103] Target molecules Propose a sequence of chemically viable reaction steps ANN [ 104] Instrumentations and spectral preprocessing Complex 2D images Optimize illumination light source parameters CNN [ 105] Patterns of weakly scattering perturbations Design transmission matrices VAE [ 106] Scattering spectra of nanostructures Encode up to 9 bits of information for high-density optical information storage CNN [ 107] Noisy Raman spectra Remove the baseline, cosmic rays, and noise simultaneously or separately CNN, [108][109][110] ResNet, [ 111] U-net, [ 111] ANN+U-net, [ 112] PCA, [113] GAN [ 114] Spectral analysis SERS spectrum Identify the existence of the molecular fingerprints…”

Manifold Learning Enables Interpretable Analysis of Raman Spectra from Extracellular Vesicle and Other Mixtures

“…Raman spectroscopy, particularly surface-enhanced Raman spectroscopy (SERS), may meet this need as it has enabled increasingly advanced applications in the chemical [17][18][19][20][21] and biological sciences [22][23][24][25][26][27][28][29][30] due to ongoing improvements in both instrumentation and analysis. The unique ability of SERS to non-destructively fingerprint chemical and molecular species within diverse samples makes it possible to carefully monitor changes in biological systems or other dynamic processes, provided that sufficiently advanced analytical tools can be applied to the acquired spectra [28][29][30][31].…”

“…Autoencoders are a subtype of deep learning designed to extract the latent features from unlabeled datasets [25,[45][46][47][48][49][50]. They accomplish this by first using an encoder to compress the dimensionality of the data into a lower dimensional space known as latent or code space, and then restoring it back to original dimensions by reconstructing the input at the output.…”

“…They accomplish this by first using an encoder to compress the dimensionality of the data into a lower dimensional space known as latent or code space, and then restoring it back to original dimensions by reconstructing the input at the output. Many autoencoder variations have been introduced for different purposes such as dimensionality reduction [45], denoising of images, preprocessing of spectroscopy data [25,[46][47][48], generative models [49], and data storage compression [50]. However, despite their potential in allowing quantitative analysis of highly heterogeneous and dynamic Raman spectral data sets, few studies have applied autoencoders to the study of EV SERS to date.…”

Semi-Supervised Deep Learning-Based Multi-component Spectral Calibration Modeling for UV–vis and Near-Infrared Spectroscopy without Information Loss

“…Deep learning is a multi-layered stack of nonlinear modules that learns patterns from raw input and gradually translates them into higher-level representations. Recently, deep neural networks have become a powerful and flexible tool for spectral analysis to extract high-level features containing high-resolution information. − Among well-developed deep neural networks, convolution neural networks (CNN) and deep autoencoders have been applied into spectral analysis. − For example, Acquarelli et al proposed an end-to-end CNN-based prediction method based on spectra data, avoiding the steps of tedious feature engineering . Malek et al successively applied a one-dimensional CNN (1D-CNN) to extract deep features, which is a milestone of extracting deeper-level features using the high-dimensional original spectral data .…”