Improved quantification of photoacoustic biomarkers through residual-based noise masking.

Spectroscopic photoacoustic (PA) imaging enables non-invasive, label-free estimation of in vivo biomarkers such as tissue oxygen saturation (SO2). Quantitative accuracy is limited by low signal-to-noise ratio (SNR) in deeper regions from fluence attenuation and heterogeneous absorber concentration. In these regions, electronic noise can dominate PA spectra, yielding physiologically-plausible but spurious SO2 estimates. Since biomarkers are reported as values averaged over a region-of-interest (ROI), including noise-dominated pixels introduces systematic bias and reduces sensitivity to physiological changes. We present a pixel-wise noise masking framework: at each pixel, a linear least-squares residual compares the measured PA spectrum and a reference noise spectrum, and residual statistics within a signal-free ROI define a masking threshold. Unlike denoising, this approach excludes unreliable pixels prior to ROI-based estimation without modifying measured signals. Phantom and in vivo experiments demonstrate improved sensitivity to SO2 dynamics following noise masking, highlighting the need for robust noise exclusion in quantitative PA imaging.