3D probe localization from 2D ultrasound images using an RFF-enhanced deep neural network.
Journal:
Annual International Conference of the IEEE Engineering in Medicine and Biology Society. IEEE Engineering in Medicine and Biology Society. Annual International Conference
PMID:
40039234
Abstract
Peripheral nerve blocking (PNB) via ultrasound (US) imaging offers the advantages of non-invasiveness, nonionizing radiation, and real-time visualization. However, the high cost of 3D US makes the clinicians to imagine the anatomical volume from 2D sections. Consequently, there is a need to develop a tool capable of predicting the trajectory of a US probe and reconstructing a 3D volume. This paper presents a kernel-based deep learning enhancement for estimating the freehand trajectory of an US probe from 2D US images. Specifically, we employ a random Fourier features (RFF)-based approach to enhance the generalization capability of existing models for 2D US probe localization. Training of the architecture with the proposed layer considers a public dataset of two anatomical phantoms. A cross-validation scheme validates the robustness using predictions from various training data splits. The results demonstrate that the RFF layer outperforms the baseline models in probe localization.