Comparison of U-net-based Convolutional Neural Networks for Liver Segmentation in CT

TL;DR

This study compares 2D and 3D U-net convolutional neural networks for liver segmentation in CT scans, finding that slice-wise 2D approaches perform remarkably well and may be more practical than 3D methods.

Contribution

It provides a comprehensive evaluation of U-net-based neural networks for liver segmentation, highlighting the effectiveness of 2D slice-wise approaches over 3D methods.

Findings

Slice-wise 2D U-net approaches achieve Dice coefficients above 0.97.

2D approaches may be preferable due to current hardware and software limitations.

The study offers a performance comparison of various neural network classifiers for liver segmentation.

Abstract

Various approaches for liver segmentation in CT have been proposed: Besides statistical shape models, which played a major role in this research area, novel approaches on the basis of convolutional neural networks have been introduced recently. Using a set of 219 liver CT datasets with reference segmentations from liver surgery planning, we evaluate the performance of several neural network classifiers based on 2D and 3D U-net architectures. An interesting observation is that slice-wise approaches perform surprisingly well, with mean and median Dice coefficients above 0.97, and may be preferable over 3D approaches given current hardware and software limitations.