Quality control in clinical raster-scan optoacoustic mesoscopy.

Optoacoustic (photoacoustic) mesoscopy bridges the gap between optoacoustic microscopy and macroscopy and enables high-resolution visualization deeper than optical microscopy. Nevertheless, as images may be affected by motion and noise, it is critical to develop methodologies that offer standardization and quality control to ensure that high-quality datasets are reproducibly obtained from patient scans. Such development is particularly important for ensuring reliability in applying machine learning methods or for reliably measuring disease biomarkers. We propose herein a quality control scheme to assess the quality of data collected. A reference scan of a suture phantom is performed to characterize the system noise level before each raster-scan optoacoustic mesoscopy (RSOM) measurement. Using the recorded RSOM data, we develop a method that estimates the amount of motion in the raw data. These motion metrics are employed to classify the quality of raw data collected and derive a quality assessment index (QASIN) for each raw measurement. Using simulations, we propose a selection criterion of images with sufficient QASIN, leading to the compilation of RSOM datasets with consistent quality. Using 160 RSOM measurements from healthy volunteers, we show that RSOM images that were selected using QASIN were of higher quality and fidelity compared to non-selected images. We discuss how this quality control scheme can enable the standardization of RSOM images for clinical and biomedical applications.      «

Optoacoustic (photoacoustic) mesoscopy bridges the gap between optoacoustic microscopy and macroscopy and enables high-resolution visualization deeper than optical microscopy. Nevertheless, as images may be affected by motion and noise, it is critical to develop methodologies that offer standardization and quality control to ensure that high-quality datasets are reproducibly obtained from patient scans. Such development is particularly important for ensuring reliability in applying machine learn...     »