Detección de tumores cerebrales usando imágenes hiper-espectrales mediante algoritmos semi-supervisados combinando spectral unmixing y clasificación supervisada

Abstract

The detection of human brain cancer tissues by the naked eye during neurosurgical operations is one of the current challenges for neurosurgeons in a tumour resection surgery. Hyperspectral imaging provides a large amount of information about the characteristics of the materials captured due to its high spectral resolution. This paper proposes a strategy based on this type of data for brain cancer detection using semi-supervised classification in order to improve the classification results offered by supervised approach. The main goal is to find the best alternative to detect brain tumour samples taken into account the accuracy obtained. For that end, the semi-supervised algorithm proposed combines spectral unmixing techniques with supervised classification. Quantitative and qualitative experimental results have been conducted to analyse the classification results in semi-supervised fashion.