Frozen tissue analysis using supervised pattern recognition. (a) H&E stained image of positive lymph node PN2. The IR imaged area (11.2 mm × 3.1 mm) was mapped using a step size and aperture of 25 μm for a total 55,552 individual IR spectra. The positive node comprises large regions of cancerous breast tissue (2) and remnants of healthy nodal tissue (1). (b) ANN image of positive lymph node PN2. The red and blue colours represent the correctly classified cancerous (2) and healthy nodal tissues (1) respectively. Black pixels within the image describe spectra that were not identifiable by the neural net. (c) H&E stained image of negative lymph node NN1. The IR imaged area (2.8 mm × 2.5 mm) was mapped using a step size and aperture of 25 μm for a total 11,000 individual IR spectra. The negative node comprises typical anatomical features of a healthy node. These include the surrounding capsule (1), primary follicles (2) and medullary sinuses (3). (d) ANN image of negative node NN2. The blue colour is representative of healthy nodal tissue and thus correctly classifies the tissue section. The red colour describes the very small number of pixels incorrectly classified as cancerous node by the analysis. Black pixels within the image describe spectra that were not able to be classified by the neural net.
Bird et al. BMC Clinical Pathology 2008 8:8 doi:10.1186/1472-6890-8-8