2.4. Simulations to determine thresholds of Qmotion

RSOM data was recorded with consistent system noise level characterized by the SNR of the suture reference scan. The RSOM data and corresponding image quality were further assessed based on whether the amount of motion contaminating the data was likely to be less or greater than what our motion correction algorithm could handle. In this section, we investigated how motion correction algorithms performed for different amplitudes of motion and applied simulation studies to determine T_std and T_max. A base motion graph [blue line in Fig. 2g] was extracted from RSOM data acquired over a skin region of 4 × 2 mm in the lower arm of a healthy volunteer, where pulse, breathing and random motions are mixed, and was then treated as a complex vertical motion pattern. Ten artificial motion graphs were generated by multiplying weighting values (0.1 to 3 with step size of 0.3) of the base motion graph, achieving standard deviations from 0.5 µm to 15 µm indicated by the blue stars in Fig. 2j and corresponding maximum motions from 4 to 120 µm. Then, the artificial motion graphs were added to a motion-corrected RSOM image shown in Fig. 2d, obtaining a sequence of motion-corrupted RSOM datasets. The motion correction algorithm developed by our previous study [9] was then applied to correct the motion-corrupted RSOM data. The differences between the added motions and the retrieved motions from the motion-corrupted data are characterized by cross-correlation C(n):

Where n represents the number of the added motion graphs. Man is the nth added motion graph and Mrn corresponds to the nth retrieved motion graph. * is the cross-correlation operator.

Simulations to determine threshold values of Q_motion for classifying raster-scan optoacoustic mesoscopy (RSOM) images as low- or high-quality. The corrupted RSOM datasets are formed by adding artificial motion graphs with different variations to motionless RSOM raw data. (a-c) Three reconstructed maximum intensity projection (MIP) images after adding motion graphs (corresponding to labels 1–3 in j and k) to the RSOM raw data without motion correction. (d-f) Corresponding reconstructed MIP images after adding motion graphs to the RSOM raw data with motion correction. The images are color-coded to represent the two reconstructed frequency bands (red: larger structures in the bandwidth of 10–40 MHz; green: smaller structures in the bandwidth of 40–120 MHz). The skin epidermis (EP) and dermis (DR) layers are indicated in (d). (g-i) Comparisons between the added motion graphs (blue) and the corresponding retrieved motion graphs (red). (j) Relationship between changes of the contrast-to-noise ratio (CNR) and M_std of the added motion graphs. (k) Cross-correlation values between the added and retrieved motion graphs. Labels 1–3 in (j) and (k) indicate the M_std values of images (a)-(c).The red dashed lines indicate the determined value of T_std. Scale bar 500 µm.

The contrast to noise ratio (CNR) of the motion corrected images was calculated to quantify the performance of the motion correction algorithm on the ten motion-corrupted RSOM datasets. We defined CNR as:

where Ip represents the peak intensity of RSOM features inside the reconstructed image. Sb refers to the standard deviation of a background region in the reconstructed image. The values of T_std and T_max were determined based on the decrement of the CNR values, where it dropped by 1 dB. According to Eq. 2, the threshold of Q_motion (TQ_motion) equals: T_std + β * T_max.