CHEMOMETRIC ANALYSIS WITH CROSS-VALIDATION FOR SPECTROSCOPIC SIGNALS OBTAINED FROM MEDICAL SPECIMENS

Vilmantas Gėgžna1, 2, Ignas Čiplys1, 2, Darius Varanius1, 2, Rūta Kurtinaitienė3, Gunaras Terbetas4, Aurelija Vaitkuvienė1, Juozas Vidmantis Vaitkus1, 5

1 Institute of Applied Research, Vilnius University, Vilnius, Lithuania

2 Faculty of Natural Sciences, Vilnius University, Vilnius, Lithuania

3 Vilnius University Hospital Santariskiu Klinikos, Vilnius, Lithuania

4 Clinics of Neurology and Neurosurgery, Faculty of Medicine, Vilnius University, Vilnius, Lithuania

5 Faculty of Physics, Vilnius University, Vilnius, Lithuania

Spectroscopic techniques may be a convenient tool for medical diagnostics as they are non-invasive and can be implemented in modern spectroscopic devices that are small and portable. However, before it is possible to use these techniques in medical diagnostics, there is a challenge to develop an appropriate computational model that allows extracting relevant biomedical information about a certain type of medical conditions from spectroscopic data. Discipline called chemometrics [1] provides a set of mathematical and statistical tools dedicated tho deal with described issue. Thus the aim of this study was to apply chemometrical methods to analyze spectroscopic data obtained from human intervertebral disc (HID) and uterine cervix (UC) specimens and to evaluate if there is an agreement between specroscopic and more conventional biomedical evaluation.

Methods. This investigation was approved by Lithuanian Bioethical Comitee. Samples of HID were obtained during operations of intervertebral disc hernia in Clinics of Neurology and Neurosurgery (Vilnius, Lithuania) and samples of UC were collected in Vilnius University Hospital Santariskiu Klinikos (Vilnius, Lithuania) during patient's examination in doctor's office. All specimens were taken as a part of conventional medical procedure and used in the investigation under written consent of each patient. So called biomedical information (information about samples localization inside a boddy of patient and/or results of biochemical, e. g., quantity of collagen, proteoglycans, as well as medical, e. g., histological and cytological, tests) was collected.

In the experimental part of analysis auto-fluorescence of the samples was excited in-vitro with 355 nm radiation (laser "STA-01-TH", Standa, ltd.) and registered with a spectrometer “OceanOptics USB2000" (Ocean Optics, Inc.) through a fiber based optical system.

Chemometric analysis was carried out with software "R" (https://www.R-project.org/). Its packages "alsace"[2], "nmf" [3], "hyperSpec”[4], and "caret" [5] had a key role in spectra manipulation, extraction of spectroscopic features and development of computational spectroscopy-based classification model. An agreement between biomedical and spectroscopic information was evaluated by Cohen's kappa (${\kappa}$) [6] statistic. The results were verified by k-fold cross-validation [7].

Results. Positive values of kappa (${\kappa}>0$) were achieved. This fact indicates that there is at least some degree of agreement between spectroscopic and biomedical data.


[1] K. Varmuza and P. Filzmoser. Introduction to Multivariate Statistical Analysis in Chemometrics. (CRC Press, 2009).

[2] R. Wehrens, et al. Metabolite profiling in LC-DAD using multivariate curve resolution: the alsace package for R. Metabolomics 11(1):143-54 (2014).

[3] R. Gaujoux, C. Seoighe. A flexible R package for nonnegative matrix factorization. BMC Bioinformatics 11:367 (2010).

[4] C. Beleites, C. Krafft, J. Popp and V. Sergo. hyperSpec: working with spectroscopic data. (The R User Conference useR! 2011).

[5] M. Kuhn. Building Predictive Models in R Using the caret Package. Journal of Statistical Software 28(5)1-26 (2008).

[6] J.R Landis, G.G. Koch. "The measurement of observer agreement for categorical data". Biometrics 33 (1):159–74 (1977).

[7] J.D. Rodríguez et al. Sensitivity Analysis of k-Fold Cross Validation in Prediction Error Estimation. IEEE Trans Pattern Anal Mach Intell 32(3):569-575 (2010)