<title>Optimum procedure for construction of spectral classification algorithms for medical diagnosis</title>

Abstract

Optical spectroscopic detection of early malignancy is becoming more widely accepted in academic circles, however much work remains to be done before full recognition by the medical community is achieved. The majority of published studies to date have demonstrated the potential of optical diagnosis techniques using small sample numbers in a selected patient population. Many are completed without a solid understanding of the shortcomings of histopathology, the 'gold standard' for cancer detection. For the development of a new technique to improve diagnosis it is vital that more rigorous protocols are employed in large-scale clinical trials. The prospect of utilizing NIR-Raman spectroscopy for the analysis of neoplastic gastrointestinal tissue has been thoroughly explored by a multi-disciplinary team including surgeons, pathologists, and spectroscopists. This study demonstrates the need for rigorous experimental protocols and histopathological analysis by a panel of expert pathologists. Measurements of tissue specimens from nine different pathological groups describing the full spectrum of disease in the oesophagus have been made. Only homogeneous samples with consensus pathology opinion were used to construct a training data set of Raman spectra. Models were constructed using multivariate analysis techniques and tested using cross-validation.© (2002) COPYRIGHT SPIE--The International Society for Optical Engineering. Downloading of the abstract is permitted for personal use only.