Estimating blood oxygenation from photoacoustic images: can a simple linear spectroscopic inversion ever work?

Abstract

Linear spectroscopic inversions, in which photoacoustic amplitudes are assumed to be directly proportional to absorption coefficients, are widely used in photoacoustic imaging to estimate blood oxygen saturation because of their simplicity. Unfortunately, they do not account for the spatially varying wavelengthdependence of the light fluence within the tissue, which introduces "spectral coloring," a potentially significant source of error. However, accurately correcting for spectral coloring is challenging, so we investigated whether there are conditions, e.g., sets of wavelengths, where it is possible to ignore the spectral coloring and still obtain accurate oxygenation measurements using linear inversions. Accurate estimates of oxygenation can be obtained when the wavelengths are chosen to (i) minimize spectral coloring, (ii) avoid ill-conditioning, and (iii) maintain a sufficiently high signal-to-noise ratio (SNR) for the estimates to be meaningful. It is not obvious which wavelengths will satisfy these conditions, and they are very likely to vary for different imaging scenarios, making it difficult to find general rules. Through the use of numerical simulations, we isolated the effect of spectral coloring from sources of experimental error. It was shown that using wavelengths between 500 nm and 1000 nm yields inaccurate estimates of oxygenation and that careful selection of wavelengths in the 620-to 920-nm range can yield more accurate oxygenation values. However, this is only achievable with a good prior estimate of the true oxygenation. Even in this idealized case, it was shown that considerable care must be exercised over the choice of wavelengths when using linear spectroscopic inversions to obtain accurate estimates of blood oxygenation. This suggests that for a particular imaging scenario, obtaining accurate and reliable oxygenation estimates using linear spectroscopic inversions requires careful modeling or experimental studies of that scenario, taking account of the instrumentation, tissue anatomy, likely sO 2 range, and image formation process.

Oxygen Saturation Imaging Using LED-Based Photoacoustic System

“…As one can see, pulse energy of 750 nm is just J, which is half of that of 850 nm, and this is the main factor hindering the imaging depth in this work. The higher amount of background noise present in the 750 nm images (when compared to 850 nm) may very well be the reason for the slight inaccuracies (especially in deeper features) in our sO estimates [ 49 ]. Our future work will include the design and development of new LED arrays with optimal wavelengths and improved optical energy.…”

Learned spectral decoloring enables photoacoustic oximetry

“…One reason for this is that there is a domain gap between simulated and real photoacoustic data. Therefore, accurately calculating blood oxygenation values from multispectral photoacoustic signals in a clinical context remains challenging 25 .…”

Spherical-view photoacoustic tomography for monitoring in vivo placental function

“…Imaging at deeper region of tissue increases the error in estimation of sPO 2 due to depth dependency of fluence. Hochuli et al [ 44 ] explained the effect of spectral coloring in estimation of sPO 2 and showed that a fluence correction along with careful choice of wavelength selection and experimental condition can improve accuracy of sPO 2 estimation using a linear inversion. Numerical models-based fluence correction methods such as Monte Carlo simulations [ 45 , 46 ], and finite element method (FEM) of the diffusion approximation of radiative transfer equation [ 47 ], can estimate the photon distribution in the tissue provided there is prior knowledge of the optical properties of the tissue.…”