An analysis of signal processing techniques for intrinsic optical signal brain imaging

Intrinsic Optical Signal (IOS) imaging uses light in the visible spectrum to visualise activity related changes within the cerebral cortex. A limiting issue with IOS is the presence of large disturbance components. Appropriate signal processing is therefore important in obtaining quality IOS imaging results. Our aim was to simplify the choice of post processing techniques, by comparing a variety of those previously performed. To reach our goal, we have designed and constructed our own IOS imaging system, including hardware and software. We carried out preliminary tests on real mice to improve our understanding of the IOS imaging modality and the signal components involved. We then created a mathematical model of the mouse brains' IOS response and analysed the performance of a variety of signal processing techniques, including those typically applied to IOS imaging. Next, we compared these techniques on real imaging data. These include: averaging over trials, spatial Gaussian filtering, temporal low pass filtering, temporal band pass filtering, global signal regression, principal component analysis, truncated differences. We were particularly focused on comparing the effectiveness of each technique’s ability to remove noise and extract the IOS signal for both spatial and temporal responses, both quantitatively and qualitatively. We performed careful analysis of the techniques and the advantages and disadvantages are presented such that the best choice can be made for a given IOS imaging setup. Our results concluded that Gaussian filtering is the most effective choice for improving the spatial IOS response with minimal complexity. For temporal data, low pass filtering and band pass filtering provided a significant improvement over averaging with reasonable complexity, however if periodic stimuli is not possible, GSR or truncated difference should be considered as these techniques also significantly improve the signal and do not require a periodic stimuli. Other PCA variations can be considered but require careful consideration of the component selection technique to be used and are highly dependent on the quality of the raw data.