Improved Padding in CNNs for Quantitative Susceptibility Mapping

JUAN LIU

Abstract: Deep learning methods have been proposed for quantitative susceptibility mapping (QSM) - background field removal, single-step QSM, and field-to-source inversion. However, the conventional padding mechanism used in CNNs can cause spatial artifacts, especially at the boundaries of regions of interest. To address this issue, we propose an improved padding technique which utilizes the neighboring voxels to estimate the invalid pixels at volume boundaries. Studies using simulated data show that the proposed method greatly improves estimation accuracy and reduces artifacts in the results.

Paper Type: methodological development

Primary Subject Area: Image Acquisition and Reconstruction

Secondary Subject Area: Image Synthesis

Paper Status: original work, not submitted yet

Source Code Url: The code will be open if accepted.

Data Set Url: The data will be open if accepted.

Registration: I acknowledge that publication of this at MIDL and in the proceedings requires at least one of the authors to register and present the work during the conference.

Authorship: I confirm that I am the author of this work and that it has not been submitted to another publication before.