This work tackles with the problem of classifying materials from broadband coherent anti-Stokes Raman scattering (BCARS) microscopy at a pixel level. We developed a supervised classifier from a tandem of Artificial Neural Network (ANN) models that identify relevant features in Raman spectra and achieve high classification accuracy. Figure 1 summarizes our approach.
This work addresses the problem of assigning a meaningful label, from a chemical point of view, to each pixel based on their vibrational spectra. This process is commonly referred to as spectral labeling, digital staining or pseudo-coloring. The majority of chemical identification is performed manually by skilled users, capable of differentiating features consisting in unique patterns of multiple, often crowded, peaks within the vibrational spectrum.
We developed a method to derive mathematical if-then decision rules for pixel classification from Artificial Neural Networks (ANN). The proposed method combines the accuracy of ANN and the interpretability of decision rules in order bring more insight into identifying and differentiating relevant spectral patterns for various chemical components.
We used two ANNs in tandem for rule generation. This strategy improves not only the accuracy of the rules but also reduces the complexity of these rules thus making them more interpretable by human experts.
For more information, please, read the paper: Petru Manescu, Young Jong Lee, Charles Camp, Marcus Cicerone, Mary Brady, Peter Bajcsy, "Accurate and Interpretable Classification of Microspectroscopy Pixels Using Artificial Neural Networks" Medical Image Analysis, Volume 37, April 2017, Pages 37–45. (online version)
Petru Manescu, Young Jong Lee, Charles Camp, Marcus Cicerone, Mary Brady, Peter Bajcsy,
"Accurate and Interpretable Classification of Microspectroscopy Pixels Using Artificial Neural Networks"
Medical Image Analysis, Volume 37, April 2017, Pages 37–45