Random 2.5D U-net for Fully 3D Segmentation

TL;DR

This paper introduces a novel 2.5D U-net architecture that uses projections and 2D convolutions to efficiently perform 3D segmentation without heavy 3D convolutions, enabling processing of large volumetric data.

Contribution

The authors propose a 2.5D U-net that avoids 3D convolutions by using projections and a trainable reconstruction, allowing end-to-end segmentation of large 3D data.

Findings

Outperforms standard approaches in sparse binary segmentation tasks

More resistant to artefacts compared to existing methods

Enables processing of large volumetric data without cropping or sliding windows

Abstract

Convolutional neural networks are state-of-the-art for various segmentation tasks. While for 2D images these networks are also computationally efficient, 3D convolutions have huge storage requirements and therefore, end-to-end training is limited by GPU memory and data size. To overcome this issue, we introduce a network structure for volumetric data without 3D convolution layers. The main idea is to include projections from different directions to transform the volumetric data to a sequence of images, where each image contains information of the full data. We then apply 2D convolutions to these projection images and lift them again to volumetric data using a trainable reconstruction algorithm. The proposed architecture can be applied end-to-end to very large data volumes without cropping or sliding-window techniques. For a tested sparse binary segmentation task, it outperforms already known standard approaches and is more resistant to generation of artefacts.