Rapid 3D passive needle localization for automatic slice alignment in MR-guided interventions

In Magnetic Resonance Imaging (MRI)-guided needle interventions, typically 2D real-time imaging is used for the visual tracking of the needle. For imaging planes insufficiently aligned with the needle, realignment can become necessary for better needle display. In this work, techniques for the visualization of metallic needles with MRI and rapid 3D needle localization for automatic imaging plane alignment were developed and investigated. First, the image artifact of a needle was characterized quantitatively for conventional gradient-echo (GRE) imaging with the fast low-angle shot (FLASH) technique, as well as for a positive, susceptibility-based imaging technique (dephased GRE). A model-based method for needle localization, using a baseline-subtracted radial FLASH k-space acquisition, was then developed and evaluated in phantom measurements, as well as demonstrated in an animal experiment. Additionally, a localization method based on undersampled dephased FLASH imaging and Convolutional Neural Network-based postprocessing was developed, investigated, and compared with the model-based method. The needle artifacts were well described by the developed quantitative models (measured deviation ≤ 2 pixels). Both developed localization methods allowed for fast 3D needle localization (under 1 s). Feasibility of automatic slice alignment of 2D imaging planes with the detected needles was demonstrated retrospectively. The developed methods for rapid, passive 3D needle localization can accelerate the workflow of MRI-guided interventions and facilitate clinical applicability.