35 Maria Tirindelli Computer Vision News The second work focuses on the identification of augmentation techniques that better reflect the physics of the acquisition of ultrasound data, compared to standard techniques. More specifically, we identify three possible augmentation techniques. The first technique introduces augmentations by introducing synthetic deformation, where rigid structures are left undeformed while soft tissues deform, consistent with what happens in real ultrasound acquisitions. The second technique introduces an augmentation that simulates variations in contrast and signal-to-noise ratio of the input signal. Finally, the third technique introduces an augmentation that simulates the occurrence of multiple reflection artifacts in ultrasound data. We further compare our results with the training without augmentation and classical augmentation for example segmentation and classification problems. The final work focuses on the definition of optimal trajectories for robotically actuated ultrasound acquisition. To this end, we first define a method that extracts the optimal trajectory to ensure the maximal coverage of the volume to be acquired. Secondly, we utilize confidence maps to extract information on the scanned spatial points coverage. Based on this information, we define an optimization criterion that allows the extraction of an optimal trajectory in a way that the occluded points are reached from different orientations and the volume is properly scanned. We tested the proposed methods on three phantoms in a simulated environment, as well as on a phantom in a real-case scenario.
RkJQdWJsaXNoZXIy NTc3NzU=