BME Frontiers / 2022 / Article / Fig 5

Research Article

Weakly- and Semisupervised Probabilistic Segmentation and Quantification of Reverberation Artifacts

Figure 5

(a) Comparison between our entire approach against other algorithms and human labels. (b) Comparison of our second network against other soft-label probabilistic networks. (c) Results comparison: the first five columns are images on phantom data, last three columns are images on chicken breast data. From top to bottom: input images (zoomed in), Human Labels, Our results, U-Net, USVS-Net, and HPU-Net. Notice that our method differentiates the reverberations better, produces fewer false positives, picks up unlabeled artifacts, and at the same time, models the exponential decay of artifact intensity.