To elucidate whether Raman spectroscopy aided by extensive spectral database and neural network analysis can be a fast and confident biomarking tool for the diagnosis of various types of cancer.
Study included 27 patients with 11 different malignant tumors. Using Raman microscopy (RM) a total of 540 Raman spectra were recorded from histology specimens of both tumors and surrounding healthy tissues. Spectra were analyzed using the principal component analysis (PCA) and results, along with histopathology data, were used to train the neural network (NN) learning algorithm. Independent sets of spectra were used to test the accuracy of PCA/NN tissue classification.
The confident tumor identification for the purpose of medical diagnosis has to be performed by taking into account the whole spectral shape, and not only particular spectral bands. The use of PCA/NN analysis showed overall sensitivity of 96% with 4% false negative tumor classification. The specificity of distinguishing tumor types was 80%. These results are comparable to previously published data where tumors of only one tissue type were examined and can be regarded satisfactorily for a relatively small database of Raman spectra used here.
In vitro RM combined with PCA/NN is an almost fully automated method for histopathology at the level of macromolecules. Supported by an extensive tumor spectra database, it could become a customary histological analysis tool for fast and reliable diagnosis of different types of cancer in clinical settings.
The article is published online in the journal Medical Research in Biophysics and is free to access.