Search

Purpose: We evaluate an SGRT device (Voxelan HEV‐600 M/RMS) installed with Radixact, with the view angle of the Voxelan's camera at 74 degrees. The accuracy of Voxelan with this steep angle was evaluated with phantom experiments and inter‐fractional setup errors of patients. Methods: In the phantom experiments, the difference between the measured values of Voxelan from the truth was evaluated for translations and rotations. The inter‐fractional setup error between the setup using skin markers with laser localizer (laser setup: LS) and the setup using Voxelan (surface setup: SS) was compared for head and neck (N = 19), chest (N = 7) and pelvis (N = 9) cases. The inter‐fractional setup error was calculated by subtracting from bone matching by megavoltage computed tomography (MVCT) as ground truth. Results: From the phantom experiments, the average difference between the measured values of Voxelan from the truth was within 1 mm and 1 degree. In all cases, inter‐fractional setup error based on MVCT was not significantly different between LS and SS by Welch's t‐test (P > 0.05). The vector offset of the LS for head and neck, chest, and pelvis were 6.5, 9.6, and 9.6 mm, respectively, and that of the SS were 5.8, 8.6, and 12.6 mm, respectively. Slight improvement was observed for the head and neck, and chest cases, however, pelvis cases were not improved because the umbilical region could not be clearly visualized as a reference. Conclusion: The results show that SS in Voxelan with an installation angle of 74 degrees is equal to or better than LS. [ABSTRACT FROM AUTHOR]

Map building and path planning are essential for long-term visual navigation. Methods, such as semi-parametric topological memory (SPTM), use deep learning to build a topological map consisting of nodes sampled from an agent's past observations and edges generated by a neural network (NN) and perform path planning on the map. Such methods require neither accurate sensor nor advanced expertise for map building needed in classical metric-map-based techniques such as visual simultaneous localization and mapping. However, they often plan improper paths including teleportations and detours, even if we have collected enough training data. In this paper, we propose a topological-map-based planning method that includes two modifications to SPTM to address this problem. The first modification is that we observe omnidirectional images and construct an NN on the basis of spherical convolutional NNs, which guarantee rotation invariance on omnidirectional images. The second modification is that we train the NN on a metric learning framework. We conducted experiments to show the effectiveness and applicability to real world of our method for path planning in visual navigation. The results indicate that the first modification prevents detouring and the second one prevents teleportations in path planning. [ABSTRACT FROM AUTHOR]