Real-time cancer diagnosis using random projection ensemble classification

Hyperspectral imaging (HSI) is a new technique that enables the spatial and spectral dimensions of light to be recorded simultaneously. HSI offers great potential for use in medical imaging through the ability to characterise the biochemical properties of normal tissue and identify changes that occur early in the development of disease [1]. The implementation of HSI in medical imaging has previously been limited fundamentally by the inability to acquire the data in real time during an imaging examination. Recently, this limitation has been overcome with the ability to perform ‘snapshot’ HSI [2]. However, in order to provide diagnostically relevant information to a clinician, the HSI data must be classified into different disease states, for example, different stages of cancer development.  This project will investigate the use in this context of a new statistical methodology [3] for high-dimensional classification based on the aggregation of the results of applying a low-dimensional base classifier on many random projections of the original data.

This project could appeal to a candidate with either a physics or a statistical background.

[1] Lu, G. and Fei, B. (2014) Medical hyperspectral imaging: a review.  J. Biomedical Optics, 19, 010901.

[2] Hagen, N. and Kudenov, M. W. (2013) Review of snapshot spectral imaging technologies. Optical Engineering, 52, 090901.

[3] Cannings, T. I. and Samworth, R. J. (2015) Random projection ensemble classification.  Available athttp://arxiv.org/abs/1504.04595.

Papers, Publications & Software