2022/01/10 by Daniel L. Wulff, Wulff, Daniel, Jannis Hagenah +3
Physics and Astronomy · Computer Science · #Advanced Radiotherapy Techniques #AI in cancer detection #Medical Image Segmentation Techniques
paper · pdf · doi:10.48550/arxiv.2201.03319
3D ultrasound (3DUS) becomes more interesting for target tracking in\nradiation therapy due to its capability to provide volumetric images in\nreal-time without using ionizing radiation. It is potentially usable for\ntracking without using fiducials. For this, a method for learning meaningful\nrepresentations would be useful to recognize anatomical structures in different\ntime frames in representation space (r-space). In this study, 3DUS patches are\nreduced into a 128-dimensional r-space using conventional autoencoder,\nvariational autoencoder and sliced-wasserstein autoencoder. In the r-space, the\ncapability of separating different ultrasound patches as well as recognizing\nsimilar patches is investigated and compared based on a dataset of liver\nimages. Two metrics to evaluate the tracking capability in the r-space are\nproposed. It is shown that ultrasound patches with different anatomical\nstructures can be distinguished and sets of similar patches can be clustered in\nr-space. The results indicate that the investigated autoencoders have different\nlevels of usability for target tracking in 3DUS.\n