A study of FT-IR spectroscopy for the identification and classifcation of haematological malignancies
Date published
Free to read from
Authors
Supervisor/s
Journal Title
Journal ISSN
Volume Title
Publisher
Department
Type
ISSN
Format
Citation
Abstract
The aim of the work presented in this thesis was to explore the use of FT-IR spectroscopy, as a complementary clinical tool for haematological laboratory analysis. FT-IR spectra were measured from air-dried and frozen cell lines derived from lymphoma, lymphoid, myeloid leukaemia and normal and chronic lymphocytic leukaemia blood samples. Multivariate statistical analysis was used to extract important spectral information with the greatest discriminative power. Principal component fed linear discriminant spectral models have been tested with leave one out cross validation procedures. A preliminary unfiltered classification model using 50 frozen and air-dried samples correctly classified 54% of 18556 spectra. The performance improved with the three cell line group datasets, with 71% of 19903 spectra correctly classified. Furthermore, the use of the frozen spectra improved the performance of the three cell line group classification model considerably. Findings showed that 73.3% of 9920 spectra were correctly classified in the frozen datasets, whereas in the air-dried only 41.5% of 9983 spectra are correctly classified. Optimisation of the spectral models by selection of principal components, application of Savitsky-Golay filters and selecting spectra using standard deviation and absorption filter tool was investigated. Using the first 25 significant PCs, a 0 th derivative Savitsky-Golay filter and the absorbance filter tool on the frozen five cell line spectral dataset were shown to be the optimal parameters for constructing a classification model. When tested with leave one batch out cross validation 90% of the spectra were correctly classified for the five cell line model. Blood component classification models tested with leave one batch out cross validation performed well. The whole blood model correctly classified 70% of 1736 spectra, measured on 22 samples. The plasma model correctly classified 80.6% of 331 spectra and the buffy coat model correctly classified 99.5% of 1438 spectra. This demonstrated that the buffy coat (containing white blood cells) holds the key biochemical information for discrimination between the pathology of the blood samples. Partial least squares analysis has been demonstrated as a method to support whole blood count tests for real time prediction of cellular constituents. These findings demonstrate the potential of FT- IR spectroscopy as a clinical tool although more work is needed if it is to be applied in clinical practice.