Chemical imaging and machine learning for sub‐classification of oesophageal tissue histology

Abstract

Abstract Fourier Transform Infrared (FTIR) based chemical imaging is a powerful, non‐destructive and label‐free biophotonic technique, which spatially acquires bio‐molecularly relevant information in histopathology. Cancer detection with objective chemical imaging techniques is relatively well established, though detection of pre‐cancer stages within a continuum from normal tissue to cancer remains challenging. Here machine learning with chemical imaging was used to provide an objective classification pipeline for oesophageal tissues pathologically classified as normal, oesophagitis, dysplasia, Barrett's disease and cancer. Spectral images were segmented using a k‐means cluster validity indices approach and clustered spectra were classified using partial least squares discriminant analysis. Classification performances approached a receiver operator characteristic area‐under‐the‐curve (ROC‐AUC) of 0.90 for binary classification tasks (eg, normal vs Barrett's). Isolated histopathological substructures were identified which delivered a ROC‐AUC in of ~0.69 in classifying into each of the five‐classes. This work may provide the means to assist pathologist diagnoses of intermediate pre‐cancer stages.

Keywords

• Barrett's Oesophagus
• FTIR spectroscopy
• oesophageal adenocarcinoma
• PLSDA