Conditional Random Fields as Recurrent Neural Networks for 3D Medical Imaging Segmentation
See where this sits in the topic map →Summary AI-generated
- TL;DR
- This paper investigates whether a technique that improves 2D natural image segmentation—Conditional Random Fields formulated as Recurrent Neural Networks—can similarly enhance 3D medical image segmentation.
- Problem
- While Conditional Random Field as a Recurrent Neural Network layers improve semantic segmentation for 2D RGB images when placed atop Fully-Convolutional Neural Networks, their effectiveness on 3D multi-modal medical images remained untested.
- Method
- The authors developed a flexible implementation of the algorithm supporting any number of spatial dimensions, input/output channels, and reference channels, and tested it on two distinct 3D medical imaging datasets.
- Results
- The evaluation concluded that the performance differences observed after applying the algorithm were not statistically significant.
- Contributions
- The work provides the first publicly available implementation of this algorithm capable of operating across arbitrary spatial dimensions, input/output image channels, and reference image channels.
- Limitations
- Not specified in the abstract.
- Takeaways
- The study discusses the underlying reasons why this technique transfers poorly from natural images to medical images.
- Applications
- 3D multi-modal medical imaging segmentation.
- Topics
- Conditional Random Fields; Recurrent Neural Networks; 3D Medical Imaging Segmentation
- For industry
- Healthcare and medical imaging.
- Why it matters
- Provides empirical insight into the limitations of transferring computer vision techniques from natural images to medical imaging domains.
Abstract
The Conditional Random Field as a Recurrent Neural Network layer is a recently proposed algorithm meant to be placed on top of an existing Fully-Convolutional Neural Network to improve the quality of semantic segmentation. In this paper, we test whether this algorithm, which was shown to improve semantic segmentation for 2D RGB images, is able to improve segmentation quality for 3D multi-modal medical images. We developed an implementation of the algorithm which works for any number of spatial dimensions, input/output image channels, and reference image channels. As far as we know this is the first publicly available implementation of this sort. We tested the algorithm with two distinct 3D medical imaging datasets, we concluded that the performance differences observed were not statistically significant. Finally, in the discussion section of the paper, we go into the reasons as to why this technique transfers poorly from natural images to medical images.