The University of ManchesterKonrad Matthew DorlingDoctor of Philosophy'The Development of FTIR-Imaging for the Study of Human Prostate Cancer Biopsies'2012The potential of using FTIR imaging as analytical technique combined with pre-processing and multivariate analysis methods was investigated. FTIR spectroscopy has been used in the past to investigate aspects of prostate cancer cells and tissues, successfully showing the separation of spectral data taken from benign prostate samples and cancerous prostate samples of varying Gleason grade. This work was ground-breaking, diagnosing different grades of cancer in the same way to the original histopathologist-assigned grades of the tissue. The advent of FTIR imaging and its recent commercial availability has allowed the much more specific collection of FTIR spectra in the form of infrared images corresponding to hyperspectral cubes of data. Optimised protocols for FTIR imaging were developed for the collection of such images from prostate cancer tissue samples, so that the highest quality data could be obtained as time efficiently as possible. With the recent development of a resonant Mie scattering correction algorithm, the pre-processing of data could be done rigorously, eliminating all physical effects from spectral data for the first time. Hierarchical cluster analysis and K-means cluster analysis were employed as image clustering methods to classify the tissue based on morphology. Imaging data that had RMieS-EMSC, vector normalisation and a second derivative applied showed the best cluster assignment as advised by an experienced histopathologist.A large scale study was devised based on the author's previous work to try and classify metastatic from non-metastatic prostate cancer epithelium using FTIR images. A method for the isolation of epithelial spectra was devised by the immunohistochemical staining of the tissue sample after data collection to highlight the epithelium, and overlaying the optical image of the stained tissue with the FTIR image. Resulting epithelial spectra were extracted from the FTIR images of the two tissue classes. Principal component analysis was applied to the data, and artificial neural network were constructed using training and test sets of patient-associated spectra. Experiments were done investigating whether non-metastatic cancer epithelium classified differently to non-cancerous epithelium, and whether epithelial spectra from patients with metastatic cancer would classify differently to patients with non-metastatic cancer. The non-metastatic data did not separate well from the non-cancer data. PCA results showed the metastatic data separated from the non-metastatic data very well, and seemingly robust ANNs were also developed to classify the data.