Conquering Data Variations in Resolution: A Slice-Aware Multi-Branch Decoder Network.

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Computer Vision

Background:

• Fully convolutional neural networks show promise for liver and liver tumor segmentation.
• Existing methods often debate 2D vs. 3D network architectures, overlooking resolution variations.

Purpose of the Study:

• To address the performance obstacle caused by variations in intra- and inter-slice resolutions in medical image segmentation.
• To propose a novel slice-aware 2.5D network for improved liver and liver tumor segmentation.

Main Methods:

• Developed a slice-aware 2.5D network focusing on both in-plane semantics and out-of-plane coherence.
• Introduced a slice-wise multi-input multi-output architecture with a Multi-Branch Decoder (MD) and Slice-centric Attention Block (SAB).
• Utilized a Densely Connected Dice (DCD) loss for coherent and continuous inter-slice predictions.

Main Results:

• Achieved state-of-the-art performance on the MICCAI 2017 Liver Tumor Segmentation (LiTS) dataset.
• Demonstrated robustness and generalizability on the ISBI 2019 Segmentation of THoracic Organs at Risk (SegTHOR) dataset.

Conclusions:

• The proposed slice-aware 2.5D network effectively tackles resolution mismatches in medical image segmentation.
• The method offers a robust and generalizable solution for liver tumor segmentation and other medical imaging tasks.