Brain Tumor Classification Using Hyperspectral Image Analysis

The capability of Hyperspectral Imaging (HSI) in rapidly acquiring abundant reflectance data in a non-invasive manner, makes it an ideal tool for obtaining diagnostic information about tissue pathology. Identifying features that provide discriminatory clues for specific pathologies will assist in understanding their underlying biochemical characteristics. In this thesis, brain cancer HSI data was analyzed to arrive at two approaches for brain tumor classification: (1) identification of simple morphological features from the pixel spectra (2) use of a computationally efficient data decomposition method to automatically identify discriminatory bands. Using a database of 26 brain cancer HS images and simple morphological features, maximum pixel classification accuracies of 87.9% for binary (tumor, normal) and 78.6% for multigroup (tumor, normal, hypervascular, and background) classification were achieved. The proposed band selection approach achieved maximum pixel classification accuracies of 99.6% for binary and 94.7% for multigroup classification with 3-times reduction in feature-set compared to the benchmark.