Principal Component Analysis for Nonlinear Optical Microscopic Chemical Imaging of Nitrogen Gas

We have implemented principal component analysis for microscopic wide-field chemical imaging via coherent Raman spectroscopy. Microscopic imaging of nitrogen gas has been challenging due to extremely weak signals stemming from low order Raman interaction. Wide-field coherent Raman micro-spectroscopy has demonstrated the ability to chemically distinguish nitrogen gas although it has been difficult to quantify spatial-density information due to significant levels of background noise. By subtracting the Gaussian beam shape and removing contributions from uninformative noise simultaneously from the set of images, we can reconstruct the normalized intensity fluctuations. Our analysis demonstrates that nitrogen gas within microvolume can be rapidly monitored under ambient conditions in less than 0.2 seconds. We believe that our work has the potential to improve visualization of microscopic flows due to molecular dynamics of gases and/or liquids otherwise invisible to infrared optical techniques.

💡 Research Summary

This paper presents a method for rapid, quantitative chemical imaging of nitrogen gas using wide‑field coherent anti‑Stokes Raman spectroscopy (CARS) combined with principal component analysis (PCA) for background removal. Nitrogen, which constitutes 78 % of the atmosphere, is Raman‑active at 2330 cm⁻¹ but produces extremely weak signals in conventional Raman or infrared‑based sensors because it is non‑polar and infrared‑inactive. The authors previously demonstrated that wide‑field CARS can detect the nitrogen vibrational mode, yet the Gaussian intensity profile of the excitation beams and detector artefacts (dark current, read noise, shot noise) obscure spatial density information.

To overcome these limitations, the authors recorded CARS images with three ultrafast laser beams (pump, Stokes, probe) on an EMCCD camera. Six scan sets were acquired: scans 1‑5 with all three beams present (exposure times of 0.2 s for scans 1‑3 and 0.1 s for scans 4‑5) and scan 6 with only the probe beam (0.1 s). For each scan, 99 frames were captured, then cropped to two regions of interest—30 µm × 30 µm and 100 µm × 100 µm—at a pixel size of 0.75 µm. Each cropped frame was reshaped into a row vector, forming a data matrix whose rows correspond to individual frames and columns to pixel locations.

Prior to PCA, the matrix was normalized using MATLAB’s z‑score function (subtracting the mean and dividing by the standard deviation of each frame). This step equalizes the variance across frames with different exposure times, ensuring that the subsequent PCA operates on a common scale.

PCA decomposition yielded eigen‑vectors (principal components) and corresponding scores. The first principal component (PC1) captured the dominant Gaussian illumination profile and consistent detector artefacts; the second component (PC2) represented a systematic skew of the images toward the right. By discarding PC1 and PC2 and reconstructing the data from the remaining higher‑order components, the authors effectively removed the background illumination and geometric distortion while preserving the variance associated with the nitrogen CARS signal.

Reconstructed images displayed a flattened intensity distribution, free of the Gaussian beam shape, and the nitrogen signal became clearly visible even at the shorter 0.1 s exposure. Including scan 6 (probe‑only) in the data matrix improved the separation between background and signal, because PCA could learn the variance structure of pure background frames. Surface plots and axis projections of both the original and reconstructed images confirmed that the method works for both crop sizes, demonstrating robustness against changes in field of view.

The authors conclude that PCA is a powerful, computationally inexpensive tool for background subtraction in wide‑field CARS microscopy of gases. By eliminating the beam profile and detector artefacts, they achieve quantitative imaging of nitrogen gas within a micro‑volume in less than 0.2 seconds. This capability opens new avenues for visualizing microscopic gas flows, studying gas‑liquid interfacial dynamics, and monitoring non‑infrared‑active species that are invisible to conventional optical techniques.

Future work may extend the approach to multi‑species mixtures, temperature‑ or pressure‑dependent studies, three‑dimensional volumetric imaging, and integration with more advanced machine‑learning methods (e.g., autoencoders or convolutional neural networks) for even finer noise suppression and feature extraction. The demonstrated workflow—data acquisition, z‑score normalization, PCA decomposition, selective component removal, and reconstruction—provides a reproducible pipeline that can be adapted to other nonlinear optical imaging modalities where weak signals are masked by strong, structured backgrounds.