Data-driven approach to optimum wavelength selection for diffuse optical imaging

The production of accurate and independent images of the changes in concentration of oxyhemoglobin and deoxyhemoglobin by diffuse optical imaging is heavily dependent on which wavelengths of near-infrared light are chosen to interrogate the target tissue. Although wavelengths can be selected by theoretical methods, in practice the accuracy of reconstructed images will be affected by wavelength-specific and system-specific factors such as laser source power and detector sensitivity. We describe the application of a data-driven approach to optimum wavelength selection for the second generation of University College London's multichannel, time-domain optical tomography system (MONSTIR II). By performing a functional activation experiment using 12 different wavelengths between 690 and 870 nm, we were able to identify the combinations of 2, 3, and 4 wavelengths which most accurately reproduced the results obtained using all 12 wavelengths via an imaging approach. Our results show that the set of 2, 3, and 4 wavelengths which produce the most accurate images of functional activation are [770, 810], [770, 790, 850], and [730, 770, 810, 850] respectively, but also that the system is relatively robust to wavelength selection within certain limits. Although these results are specific to MONSTIR II, the approach we developed can be applied to other multispectral near-infrared spectroscopy and optical imaging systems.