Functional near-infrared spectroscopy (fNIRS) is a non-invasive optical neuroimaging method, where light is used to measure changes in blood oxygenation. It is well-suited to studying dementia in its various stages of neurodegeneration, with several advantages in comparison to traditional imaging approaches used in dementia research, such as functional magnetic resonance imaging (fMRI). Similarly, high-density diffuse optical tomography (HD-DOT) is an imaging technique that collects NIRS data using a high-density array of overlapping channels, combined with a light transport model developed using subject-specific anatomical scans, to produce high-resolution, three-dimensional reconstructions of metabolic activity in the brain.
There is a distinct lack of HD-DOT studies in dementia, and as a result, its potential as a valuable, alternative neuroimaging method has not been uncovered. The aim of my PhD is to utilise HD-DOT data, in both its light intensity signal and reconstructed image formats, to perform fine-scale machine learning classification of dementia in its early, as well as advanced, stages. Interpretable deep learning models can be applied to raw HD-DOT data to achieve greater classification accuracies, while providing a means to evaluate model prediction pathways.
This would be of significant benefit to clinicians, as fast and early dementia diagnoses could be provided in the absence of subject-specific structural priors from an MRI. Additionally, the explainability of computational models used for diagnosis is crucial for all clinical applications. The long-term goal of this work is to identify early-stage biomarkers for dementia diagnosis at a point where medical intervention has the highest chance of success in preventing cognitive decline.