Fitting Segmentation Networks on Varying Image Resolutions using Splatting

Mikael Brudfors, Yaël Balbastre, John Ashburner, Geraint Rees, Parashkev Nachev, Sebastien Ourselin, M. Jorge Cardoso

Keywords: Image segmentation, Splatting, Resolution invariance, CNN, UNet.

TL;DR: We propose a method to fit segmentation networks to native space image data (varying resolution, field-of-view and orientation) by introducing splat layers. We show on MR images that this method improves segmentation results over standard resampling.

Abstract: Data used in image segmentation are not always defined on the same grid. This is particularly true for medical images, where the resolution, field-of-view and orientation can differ across channels and subjects. Images and labels are therefore commonly resampled onto the same grid, as a pre-processing step. However, the resampling operation introduces partial voluming and blurring, thereby changing the effective resolution and reducing the contrast between structures. In this paper we propose a splat layer, which allows networks to work in native image space, without requiring equal image grids. This layer pushes each image onto a mean space where the prediction is performed -- without performing interpolation. As the splat operator is the adjoint to the resampling operator, the mean-space prediction can be pulled back to the native label space, where the loss function is computed. This combination makes the end-to-end architecture resolution-agnostic, allowing it to be trained on data with varying resolution. We show on two publicly available datasets, with simulated and real thick-sliced magnetic resonance (MR) images, that this model improves segmentation results compared to resampling as a pre-processing step.

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.

Paper Type: methodological development

Primary Subject Area: Segmentation

Secondary Subject Area: Integration of Imaging and Clinical Data

Confidentiality And Author Instructions: I read the call for papers and author instructions. I acknowledge that exceeding the page limit and/or altering the latex template can result in desk rejection.