Rapid infrared mapping for highly accurate automated histology in Barrett's oesophagus

pixel size and 64 scans per pixel. Principal component fed linear discriminant analysis was used to build classification models based on spectral differences, which were then tested using leave-one-sample-out cross validation. Key biochemical differences were identified by their spectral signatures: high glycogen content was seen in normal squamous (NSQ) tissue, high glycoprotein content was observed in glandular BE tissue, and high DNA content in dysplasia/adenocarcinoma samples. Classification of normal squamous samples versus 'abnormal' samples (any stage of Barrett's) was performed with 100% sensitivity and specificity. Neoplastic Barrett's (dysplasia or adenocarcinoma) was identified with 95.6% sensitivity and 86.4% specificity. Highly accurate pathology classification can be achieved with FTIR measurement of frozen tissue sections in a clinically applicable timeframe.