Figure 1

Work flow diagram for training and test phases of diagnostic algorithm. The current protocol for cancer diagnosis and grading of biopsy material involves the sectioning of samples, H&E staining, and an assessment of tissue and cellular morphology by a pathologist (left, blue shaded boxes). To implement an automated protocol of analysis using infrared micro-spectral imaging, a training phase is required to develop a robust diagnostic algorithm (right, red shaded boxes). The final paradigm for automated analysis involves infrared micro-spectral imaging of unstained tissue, followed by computer analysis using the diagnostic algorithm (right, green arrow). A two step approach for training a neural net was employed in this investigation. Spectral data sets recorded from tissues are initially scrutinised via hierarchical cluster analysis (HCA), a completely unsupervised method of analysis, to produce groups of infrared spectra that are specific to tissue type or class. This is achieved by directly correlating spectral images constructed from HCA analysis, to morphological interpretations that were made by a pathologist using the stained tissue. These tissue specific groups of spectra were then pooled into two separate data blocks, classed as being either healthy or malignant in nature. These newly compiled data blocks were then used to train a diagnostic ANN.