MENU

Click here to

×

Are you sure ?

Yes, do it No, cancel

Removing Noise Bias in Photoacoustic-Based SO2 Estimates: A Simple Empirical SNR-Adaptive Thresholding Approach

D Sampaio1,2*, M Naser1 , T Mitcham1 , C Wood1 , J Pavoni2 , T Pavan2 , R Bouchard1 , (1) The University of Texas MD Anderson Cancer Center, Houston, TX, (2) University of Sao Paulo, Ribeirao Preto, SP

Presentations

(Tuesday, 7/31/2018) 4:30 PM - 6:00 PM

Room: Room 207

Purpose: Multi-wavelength photoacoustic imaging (PAI) tends to be signal-limited, primarily as a result of the attenuation of light by biological tissue over centimeter distances, which can cause pixels within an intended region of interest to give noise-biased assessments of absorber concentration (e.g., oxy/deoxyhemoglobin). Therefore, there exists a critical need to impose a quantitative approach to exclude low-/no-SNR pixels, particularly as imaging depths increase due PAI making its way into the clinic. Herein we provided an empirical approach that characterizes noise of PAI to obtain SNR estimates, presenting a methodology to select a thresholding cut-off that optimizes the trade-off between imaging depth and SNR.

Methods: No-laser-irradiation PAI acquisitions (n=1000) were performed in a water bath to collect noise-data. Normal, Log-normal, Rician, and Rayleigh distributions were fit to the pixel-wise histograms of noise data; goodness of fit was evaluated using a Kolmogorov-Smirnov test (α=0.05) to select the optimal characterization. Next, multi-wavelength (710, 734, 760, 800, and 850 nm) PAI data were acquired (with the same parameters as the noise-data acquisition) of inclusions containing oxygenated blood of 85% and 47% in an ex-vivo rat liver. Finally, an Otsu threshold of SNR (i.e., rat-data divided by noise-data standard deviation) was applied, and change of oxygen saturation (sO2) in the inclusions was estimated for validation.

Results: Based on the Kolmogorov-Smirnov test, a Log-normal (p=0.71) distribution is the optimal choice to characterize PAI noise behavior; however, it requires more than one hundred samples to be estimated accurately. Moreover, among all wavelengths, an average threshold of 6.83±0.68 dB was obtained. Such thresholding reduced the absolute error of the change of sO2 from 20% to 5% at 0.6-cm depth.

Conclusion: A simple empirically determined SNR-based threshold approach has the potential to be an initial choice to avoid highly-noise-biased assessments of absorber concentration from PAI at clinically relevant depths.

Funding Support, Disclosures, and Conflict of Interest: The authors acknowledge FAPESP for supporting Diego R. T. Sampaio (grant 2016/22720-3).

Keywords

Ultrasonics, Photoacoustics, Quantitative Imaging

Taxonomy

IM- Optical : Optoacoustics

Contact Email