High throughput assessment of cells and tissues: Bayesian classification of spectral metrics from infrared vibrational spectroscopic imaging data

Abstract

Vibrational spectroscopy allows a visualization of tissue constituents based on intrinsic chemical composition and provides a potential route to obtaining diagnostic markers of diseases. Characterizations utilizing infrared vibrational spectroscopy, in particular, are conventionally low throughput in data acquisition, generally lacking in spatial resolution with the resulting data requiring intensive numerical computations to extract information. These factors impair the ability of infrared spectroscopic measurements to represent accurately the spatial heterogeneity in tissue, to incorporate robustly the diversity introduced by patient cohorts or preparative artifacts and to validate developed protocols in large population studies. In this manuscript, we demonstrate a combination of Fourier transform infrared (FTIR) spectroscopic imaging, tissue microarrays (TMAs) and fast numerical analysis as a paradigm for the rapid analysis, development and validation of high throughput spectroscopic characterization protocols. We provide an extended description of the data treatment algorithm and a discussion of various factors that may influence decision-making using this approach. Finally, a number of prostate tissue biopsies, arranged in an array modality, are employed to examine the efficacy of this approach in histologic recognition of epithelial cell polarization in patients displaying a variety of normal, malignant and hyperplastic conditions. An index of epithelial cell polarization, derived from a combined spectral and morphological analysis, is determined to be a potentially useful diagnostic marker.

Electric field standing wave effects in FT-IR transflection spectra of biological tissue sections: Simulated models of experimental variability

“…Resonant and nonresonant reflection and scattering have been identified as the source of some of these effects, and correction algorithms have been developed to account for them [1][2][3][4][5][6][7]. 50 FT-IR imaging has already proven its worth in many biological applications including measurement for both cells and tissues [8][9][10][11][12][13][14][15][16]. However, measurement in transmission mode requires the use of relatively expensive IR transparent substrates (e.g.…”

High-definition Fourier Transform Infrared (FT-IR) Spectroscopic Imaging of Human Tissue Sections towards Improving Pathology

“…In addition, in an IR image, every pixel comprises of the entire IR spectrum, with different peaks corresponding to different biomolecules which can give information about the biochemical properties of cell types or disease states (Figure 1). Here, we have shown how to compare spectral signatures between classes, however more advanced automated classification is possible using additional algorithms 3,[43][44][45][46][47][48][49][50] , such as Bayesian classification, Random Forests, Artificial Neural Networks, and Hierarchal Cluster Analysis can be performed on the data. Supervised classification approaches will allow for the construction of a classifier that can be trained to allow for automated recognition of cell types or disease states.…”

Towards better than human capability in diagnosing prostate cancer using infrared spectroscopic imaging

“…A needle is inserted into the tissue and several (6)(7)(8)(9)(10)(11)(12)(13)(14)(15)(16)(17)(18)(19)(20)(21)(22)(23) samples are obtained from different positions. Biopsy, followed by manual examination under a microscope is the primary means to definitively diagnose prostate cancer as well as most internal cancers in the human body.…”

“…The first difference is that no external dyes are needed and the contrast in images can be directly obtained from the chemical composition of the tissue. The second is that each pixel in the visible image contains RGB values but in IR imaging contains several thousand values across a bandwidth (2000 − 14000nm) that is ∼ 40 times larger than the visible spectrum (400 − 700nm) [7].…”

“…In this approach, Fourier transform infrared imaging (FTIR) is employed to provide the entire vibrational spectroscopic information from every pixel of a sample's microscopy image. While the first steps of developing novel imaging and sampling technologies is now reliable, [7] the computational challenge of providing robust classification algorithms that can rapidly provide decisions remains. Due to the above advances in imaging and sampling, data from thousands of patients is available to train and validate algorithms for different disease states.…”