Raman Spectroscopy and Artificial Intelligence to Predict the Bayesian Probability of Breast Cancer

Abstract

effects to account for the accumulation of axial conformational strain and the anisotropic lateral coupling between individual dimers. Analysis of the four different MT kinetic phases (growing, catastrophe, shrinking, rescue) shows that catastrophe and rescue correlate with a small change in the number of GTP-dimers and the ruggedness of the MT tip. Contrary to what is hypothesized, our results reveal that exposure of GDP-dimers at the tip does not correlate with an increase in the probability of catastrophe when the ruggedness of the MT tip is below a certain threshold, thereby providing new insights into the microscopic mechanism of catastrophe and rescue.

Review of Laser Raman Spectroscopy for Surgical Breast Cancer Detection: Stochastic Backpropagation Neural Networks

“…If a nonlinear separation is expected, one or more layers are inserted between the input and output layers to extract higher order terms. Figure 3 illustrates a simple data flow we have used in previous work beginning with the initial analysis of variance and choice of spectral regions of interest used as qualitative input exploration with PCA, HCA, and k-means, through to a stochastic backpropagation NN using a single intermediary layer between the input and output layers (known as the hidden layer) [52].…”

Discriminating cell line specific features of antibiotic‐resistant strains of Escherichia coli from Raman spectra via machine learning analysis

“…Classical machine learning algorithms such as kmeans clustering [13,14], discriminant analysis [15,16] and support vector machines (SVMs) [15,17] are commonly used to classify Raman spectra. Recently, neural network (NN) approaches have also been increasingly utilized for this purpose [11,18].…”

“…Recently, neural network (NN) approaches have also been increasingly utilized for this purpose [11,18]. NN algorithms originated as a pattern recognition tool designed to work like the human brain [13] and have been studied extensively in the past several decades [19][20][21][22]. The ability of these models has grown from simple digit recognition tasks to complex biomedical classification tasks such as the prediction of postoperative mortality and tumor identification from medical images [23,24].…”

“…In general, datasets containing hundreds to thousands of samples per class [25,26] are needed to adequately construct the more flexible NN models. Raman spectral datasets are not commonly as large and typically include tens to hundreds of samples per class [13,18,27], although new approaches to RS [28][29][30][31][32][33][34] will increasingly make larger datasets feasible.…”

“…Goodacre et al and Ho et al classify bacterial Raman data by artificial neural network (ANN) and convolutional neural network (CNN) analysis, respectively, but do not extract feature weights from the NN for evaluation [11,18]. Other studies use NNs to analyze Raman spectral data for mineral identification, breast cancer diagnosis and component identification of mixtures, but again do not perform feature extraction [13,35,36]. Thrift et al examine bacterial species by NN analysis to determine antibiotic susceptibility from surface enhanced Raman spectroscopy (SERS) data but use a generative model, a variational autoencoder to extract the relative importance of spectral features [37].…”