Abstract

Vibrational spectroscopic imaging methods are novel tools to visualise chemical component in tissue without staining. Fourier transform infrared (FTIR) imaging is more frequently applied than Raman imaging so far. FTIR images recorded with a FPA detector have been demonstrated to identify the primary tumours of brain metastases. However, the strong absorption of water makes it difficult to transfer the results to non-dried tissues. Raman spectroscopy with near infrared excitation can be used instead and allows collecting the chemical fingerprint of native specimens. Therefore, Raman spectroscopy is a promising tool for tumour diagnosis in neurosurgery. Scope of the study is to compare FTIR and Raman images to visualize the tumour border and identify spectral features for classification. Brain metastases were obtained from patients undergoing surgery at the university hospital. Brain tissue sections were shock frozen, cryosectioned, dried and the same areas were imaged with both spectroscopic method. To visualise the chemical components, multivariate statistical algorithms were applied for data analysis. Furthermore classification models were trained using supervised algorithms to predict the primary tumor of brain metastases. Principal component regression (PCR) was used for prediction based on FTIR images. Support vector machines (SVM) were used for prediction based on Raman images. The principles are shown for two specimens. In the future, the study will be extended to larger data sets.

© 2011 OSA/SPIE