BME Frontiers / 2022 / Article / Fig 4

Research Article

Weakly- and Semisupervised Probabilistic Segmentation and Quantification of Reverberation Artifacts

Figure 4

(a) Intermediate results from the transform function. From left to right: input images , hand labels , outputs of the first network , results after artifact removal , results after creating exponential decay , results for alternate measurements to compensate for pixels not following exponential decay , results after applying exponential decay and compensation , results after differentiating between reverberations . We can observe that each step of the transform function achieves the desired outcome. (b) Visual results of using different hyperparameters for the transform function. Left to right: original ultrasound images, hand labels, the output of the first network, the output of the transform function using the hyperparameters in the paper, the output of the transform function using the hyperparameters in the paper, the output of the transform function using the hyperparameters in the paper. It can be seen that the outputs from different hyperparameters are really similar visually. There is only small difference in the details.