Identification of at-risk prostate cancer patients using Fourier transform infrared spectroscopy and machine learning

Fourier Transform Infrared Spectroscopy (FTIR) has been shown to be a useful tool to complement the histopathological assessment of biomedical tissue samples, allowing for diagnostic and prognostic applications based solely on chemical imaging of the tissues. This technique can be used to assist in determining the prognosis of prostate cancer patients, aiding the treatment decision protocols employed by clinicians. We report a stratification protocol to identify at-risk prostate cancer patients with poor outcomes from a large patient study (<i>n</i>=183) through the usage of label-free chemical imaging (without chemical de-waxing or staining) of numerous prostate cancer biopsy cores (<i>n</i>=1440) paired with machine learning techniques, without consideration of additional clinical variates beyond patient age and PSA levels. Distinctly different patient outcome groups are identified using infrared hyperspectral data, closely matching patient groups separated by tumour stage.