Multi and hyperspectral imaging (MSI/HSI) techniques provide spatial and spectral information and can be applied to detect and understand changes in biological tissue that occur with disease. This thesis has evaluated the application of MSI/HSI to two disease applications: analysing nailfold capillaries to aid in the evaluation of systemic scleroderma, and enhancing cancer detection in the gastrointestinal tract during endoscopies.

The nailfold capillaries are the smallest blood vessels in the body, and deformations in these capillaries are indicators of systemic scleroderma, a rheumatic disease. A hypothesis was formed correlating these deformations with changes in oxygen levels. Two multispectral systems, capable of imaging the nailfold capillaries were designed: one based on LED illumination, and the other a snapshot detector-based imaging system. These systems were tested with microfluidic blood phantoms, which simulated varying oxygenation levels. Developing microfluidic blood phantoms proved critical for evaluating system performance. Phantoms with microfluidic depths used horse blood that was chemically oxygenated and deoxygenated to variable blood oxygenation levels. The LED-based system was subsequently used for imaging healthy nailfold capillaries in a proof-of-concept demonstration. Concurrently, the multispectral snapshot camera system was incorporated into a clinical trial at the University of Manchester. Data from the nailbed capillaries of healthy controls as well as those with systemic sclerosis were evaluated and classification was achieved.

Early detection of cancer in the gastrointestinal tract can lead to curative intervention, but contrast for early lesions is low. To evaluate the potential for MSI methods to address this challenge, design optimisation was performed using pre-existing hyperspectral data from endoscopies conducted in the gastrointestinal tract (oesophagus and colon). An open-source Python-based toolbox for spectral band optimisation was developed to analyse datasets in order to design the optimal imaging bands and finalise the spatial layout of multispectral filter arrays that could be deployed in clinical settings. Disease detection accuracy was optimised by selecting subsets of spectral bands and integrating machine learning methods, such as k-nearest neighbour (kNN) classification and support vector machines (SVM). The maximum classification accuracies occurred using a kNN classifier and were 0.848 and 0.999 for the oesophagus and colon, respectively. SVM performed reasonably well with accuracies of 0.811 and 0.997 for the oesophagus and colon; while spectral angle mapping classification was a good classifier of colon tissue (accuracy of 0.995), it performed poorly on oesophageal tissue (accuracy of 0.245). The toolbox was also deployed to design filters capable of imaging blood oxygenation in tissue, to improve the detection and understanding of cancers where hypoxia plays a role. This research demonstrates the promising diagnostic capabilities of spectral imaging in measuring blood oxygenation.