A U-Net based multi-scale feature extraction for liver tumour segmentation in CT images

Gong, Ming and Soraghan, John and Di Caterina, Gaetano and Grose, Derek (2021) A U-Net based multi-scale feature extraction for liver tumour segmentation in CT images. In: 10th International Conference on Communications, Signal Processing, and Systems, 2021-08-21 - 2021-08-22.

[thumbnail of Gong-etal-CSPS2021-A-U-Net-based-multi-scale-feature-extraction-for-liver-tumour]
Preview
 Text. Filename: Gong_etal_CSPS2021_A_U_Net_based_multi_scale_feature_extraction_for_liver_tumour.pdf
Accepted Author Manuscript
Download (846kB)| Preview

Abstract

A new method for automatic liver tumour segmentation from computed tomography (CT) scans based on deep neural network is presented. Two cascaded deep convolu-tional neural networks are used to segment the CT image of the abdominal cavity. The first U-net is used for coarse segmentation to obtain the approximate position of the liver and tumour. Using this as a prediction the original image is cropped to reduce its size in order to increase the segmentation accuracy. The second modified U-net is employed for accurate segmentation of the actual liver tumours. Residual modules and dense connections are added to U-net to help the network train faster while pro-ducing more accurate results. In addition, multi-dimensional information fusion is introduced to make the network more comprehensive in restoring information. The Liver Tumour Segmentation (LiTs) dataset is used to evaluate the relative segmenta-tion performance obtaining an average dice score of 0.665 based our method.

ORCID iDs

Gong, Ming, Soraghan, John ORCID logoORCID: https://orcid.org/0000-0003-4418-7391, Di Caterina, Gaetano ORCID logoORCID: https://orcid.org/0000-0002-7256-0897 and Grose, Derek;
CORE (COnnecting REpositories)