Hyperspectral imaging for cancer detection and classification

The design and implementation of a classification system for hyperspectral images of cancer cell cultures is discussed. The ability to distinguish between different types of cancers is of particular importance in this study. This possibility allows for metastasised tumours to be identified, in the near infrared regions of 920 nm–2514 nm and thus the origin of a tumour. Using Principal Component Analysis (PCA) to find the features for Artificial Neural Networks (ANNs) and Support Vector Machines (SVMs), different cancer types could be distinguished with an overall accuracy of 87.4 % using an ANN solution whereas the SVM accuracy ranged from 73 %–88.9 % due to the One-Vs-One (OVO) multiclass technique implemented.