Modified U-Net (mU-Net) with Incorporation of Object-Dependent High Level Features for Improved Liver and Liver-Tumor Segmentation in CT Images (1911.00140v1)

Abstract: Segmentation of livers and liver tumors is one of the most important steps in radiation therapy of hepatocellular carcinoma. The segmentation task is often done manually, making it tedious, labor intensive, and subject to intra-/inter- operator variations. While various algorithms for delineating organ-at-risks (OARs) and tumor targets have been proposed, automatic segmentation of livers and liver tumors remains intractable due to their low tissue contrast with respect to the surrounding organs and their deformable shape in CT images. The U-Net has gained increasing popularity recently for image analysis tasks and has shown promising results. Conventional U-Net architectures, however, suffer from three major drawbacks. To cope with these problems, we added a residual path with deconvolution and activation operations to the skip connection of the U-Net to avoid duplication of low resolution information of features. In the case of small object inputs, features in the skip connection are not incorporated with features in the residual path. Furthermore, the proposed architecture has additional convolution layers in the skip connection in order to extract high level global features of small object inputs as well as high level features of high resolution edge information of large object inputs. Efficacy of the modified U-Net (mU-Net) was demonstrated using the public dataset of Liver tumor segmentation (LiTS) challenge 2017. The proposed mU-Net outperformed existing state-of-art networks.

Analysis of "Modified U-Net (mU-Net) with Incorporation of Object-Dependent High Level Features for Improved Liver and Liver-Tumor Segmentation in CT Images"

The paper presents an advancement in the design of neural network architectures for medical image segmentation, specifically targeting liver and liver-tumor segmentation in computed tomography (CT) images. The authors propose a modified U-Net, referred to as mU-Net, which seeks to address specific limitations of the conventional U-Net by incorporating object-dependent high-level features.

Technical Contributions

The conventional U-Net architecture has demonstrated significant utility in segmentation tasks across a range of medical imaging applications. However, it faces key challenges: the tendency to duplicate low-resolution information through skip connections, inadequate feature extraction for high-resolution edge information, and suboptimal pooling for objects of varying sizes. The proposed mU-Net attempts to overcome these limitations through several architectural innovations:

1. Residual Path Modification: The mU-Net includes an additional residual path integrated with deconvolution and activation functions, connected to the skip pathways. This approach effectively mitigates the redundant transfer of low-resolution features, a common issue in conventional U-Net designs.
2. Object-Dependent Feature Extraction: The architecture adapts the feature extraction process based on object size, allowing for the efficient handling of both small and large object features in CT scans. The skip connection feature maps are refined using additional convolutional layers, enabling the network to better capture high-level edge and global features.
3. Performance Enhancement: The paper reports on the quantitative performance improvements achieved by mU-Net over traditional architectures. On the LiTS dataset, the mU-Net achieved a Dice similarity coefficient (DSC) of 89.72% for liver-tumor segmentation and 98.51% for liver segmentation. These results represent substantial improvements, indicating the network's capability to enhance segmentation accuracy, especially in regions with ambiguous boundaries.

Implications

From a theoretical standpoint, the mU-Net validates the hypothesis that integrating object-dependent feature extraction within the network architecture can significantly enhance segmentation outcomes. The architectural modifications demonstrate that when high-resolution edge features are preserved and appropriately emphasized, they contribute to improved network decision-making in complex segmentation tasks.

Practically, this innovation holds promise for applications in radiotherapy, where precise segmentation of liver tumors and organs-at-risk is crucial. The improvement in segmentation accuracy can lead to enhanced treatment planning, potentially resulting in better patient outcomes.

Future Directions

The proposed mU-Net opens avenues for further research into adaptive network designs that can dynamically modulate their architecture based on input characteristics. Future work could explore similar concepts in other medical imaging contexts, such as MRI or ultrasound, where objects of interest often display variable resolution and contrast. Additionally, integrating more sophisticated loss functions that emphasize boundary and structural accuracy could further enhance segmentation performance.

In conclusion, this research significantly contributes to the evolving field of medical image segmentation by proposing a novel network architecture that successfully addresses the inherent limitations of conventional U-Nets. The positive results suggest the potential for broader applicability and inspire further investigation into adaptive feature extraction methodologies in deep learning frameworks.