Correction for R-Mie Effects and Data Preprocessing

The aforementioned computational procedures have been demonstrated to reduce the spectral distortions to acceptable levels. This was established by comparison of uncontaminated spectra, collected from samples with low morphological heterogeneity, with spectra that were originally strongly contaminated and processed via the correction algorithms [9]. 

All data preprocessing was carried out by software written in-house in 64-bit MATLAB (The Mathworks, Natick, MA, USA) for the automatic analysis of an entire data set at one time. The sequence of data processing steps for each of the 409,600 spectra data set is summarized later. A more detailed discussion of all steps involved can be found in the literature [16]. 

1) Conversion of transmittance spectra to absorbance units. This step was required since the PE400 stores spectra in the mapping file as raw transmission spectral vectors. The spectral vector corresponding to a given 6.25 pm X 6.25 pm pixel will be henceforth referred to as pixel spectrum. 

2) Construction of a noise data set from areas not occupied by cells. This step was accomplished using a simple integrated amide I intensity criterion to define areas of the substrate not occupied by cells. 

3) Construction of a signal data set from areas occupied by cells. These areas were selected by several criteria, including integrated intensity at the amide I band, integration over the entire spectrum, and defining minimum and maximum ranges for the amide I band position. 

---