Infrared hyperspectral imaging for point-of-care wound assessment

Mellors, Ben O L (2022). Infrared hyperspectral imaging for point-of-care wound assessment. University of Birmingham. Ph.D.

Preview

Mellors2022PhD.pdf
Text - Accepted Version
Available under License All rights reserved.
Download (5MB) | Preview

Abstract

Wound healing assessment and management are both important in ensuring a correct healing sequence. Most of these assessment techniques involve simple observation with the naked eye, which causes two main issues: the parameters assessed are highly subjective, and they rely upon the knowledge and experience of a trained medical professional. Any failure or incorrect management can result in further complications and even fatality, therefore quantitative wound assessment techniques are the next step towards a more accessible and reliable wound management strategy.
Current research in this field is focused on utilising non-invasive imaging techniques, mainly within the visible and infrared (IR) range, to identify the biological and chemical changes during the wound healing process. Any abnormalities can then be identified earlier to aid in the correct diagnosis and treatment of the wound. Technologies that utilise concepts of non-contact imaging, such as optical imaging and spectroscopy can be used to obtain spatial and spectral maps of biomarkers, which provide valuable information on the wound (e.g., precursors to improper healing or delineate viable and necrotic tissue). This work extends this research further by investigating two different imaging modalities, Negative Contrast Imaging (NCI), along with Spatial Frequency Domain Imaging (SFDI) for the applications of point of care wound assessment.
Intelligent data analysis algorithms, in the form of k-means clustering and principal component analysis were applied to spectral data, collected from wound biopsies as part of a previous study, highlighting the ability to diagnose wound healing status from the contrast of spectral information, which is not reliant upon a subjective clinical diagnosis. These methods provided the motivation for a larger cell culture trauma study, in which the NCI was utilised to obtain spectral reflectance maps across a 2.5- 3.5 μm wavelength region of both healthy and traumatised human epidermal fibroblasts, induced via chemical assays. Using the same intelligent analysis tools, along with pre-processing methods including spectral derivatives, the resulting clusters can be utilised as a diagnostic tool for the assessment of cellular health and were quantifiable metrics were defined to compare the different analysis methods
Near infrared (NIR) methodologies were also investigated, with two areas of SFDI identified for further advancements. Current SFDI acquisition and optical property parameter recovery is performed via a pixel-wise process, generating large amounts of data and a high computational burden for parameter recovery. Data reduction, through the application of Compressive Sensing (CS) at both the image acquisition and data analysis stages provided up to a 90% reduction in data, whilst maintaining <10% error in recovered absorption and reduced scattering optical maps.
This pixel-wise methodology also affects the forward modelling and inverse problem (imaging), based upon the diffusion approximation or Monte-Carlo methods due to their pixel-independent nature. NIRFAST, an existing FEM based NIR modelling tool, was adapted to produce pixel-dependent forward modelling for heterogenic samples, providing a mechanism towards a pixel dependent SFDI image modelling and parameter recovery system.

Type of Work:

Thesis (Doctorates > Ph.D.)

Award Type:

Doctorates > Ph.D.

Supervisor(s):