Multi-scale anatomical awareness improves the accuracy of the real-time electric field estimation

Induced electric fields (E-field) of a coil within target areas are substantial for precise trans cranial therapy. The fast and precise estimation of a stimulation is essential for a navigation system. However, high accuracy and low time consumption are rarely satisfied at the same time in previous models. In this paper, we present an anatomical-awareness model to integrate binary, explicit anatomical structures as mediate variables. Multi-scale attention blocks are also introduced to capture the anatomical variations. The presented model mitigates the anatomy-related errors. The presented architecture not only reduces the mean relative errors of E-field to about 7%, but also has the characteristics of low time consumption, which makes it suitable for a real-time navigation system.