Clinical application of infrared fibre-optic probes for the discrimination of colorectal cancer tissues and cancer grades

• Published in: Vibrational spectroscopy 2017-07, Vol.91, p.99-110
• Main Authors: Ollesch, Julian, Zaczek, Michael, Heise, H. Michael, Theisen, Oliver, Großerüschkamp, Frederik, Schmidt, Ralf, Morgenroth, Konrad, Philippou, Stathis, Kemen, Matthias, Gerwert, Klaus
• Format: Article
• Language: English
• Subjects: ATR fiber-optics; Cancer; Colorectal cancer; Data analysis; Decision analysis; Decision trees; Diamonds; Discriminant analysis; Discrimination; Evaluation; Fiber optics; FTIR-spectroscopy; Infrared analysis; Linear discriminant analysis; Multivariate classification; Optical fibers; Random forest; Remote sensing; Sensitivity analysis; Spatial resolution; Spectra; Surgery; Visceral surgery
• ISSN: 0924-2031; 1873-3697
• DOI: 10.1016/j.vibspec.2016.07.003

Infrared spectral histopathology is a well-established method for label-free tissue classification. Flexible fibre-optic probes allow a remote sensing for in-vivo tissue annotation. The performance of infrared spectral analysis of colorectal tissue with millimetre spatial resolution for cancer lesion identification and grading within a surgical setting was assessed. Colorectal tissue was removed during routine therapeutic surgery. By consecutive spectroscopy, luminal positions within and outside the cancer lesion were analysed using fibre-coupled probes for attenuated total reflection (ATR) measurements using a diamond-prism or cone, respectively. Subsequent routine histopathology provided the gold standard for diagnosis and grading. For spectral data analysis, two feature selection algorithms were applied. Results from linear discriminant analysis and ensemble random forest classifiers based on leave-one-third-out cross-validation and test-set validation with independent data are presented. The spectral discrimination of tumour versus normal tissue under the cross-validation scheme was achieved with an accuracy of 90±5% (sensitivity of 89±7%, specificity of 90±7%), whereas respective test-set validation led to an accuracy of 80% (sensitivity of 83% and specificity of 78%). Low versus high tumour grading was assessed under cross-validation with an accuracy of 81±8% (sensitivity of 80±16%, and specificity of 81±14%). Thus, fibre-optic infrared spectroscopic tissue analysis has the potential of supporting clinical decisions by providing immediate tissue and grading information during surgery.