Segmentation of Living and ablated Tumor parts in CT images Using ResLU-Net

Abstract

Computed tomography (CT) is widely used as the imaging modality for the treatment of tumors in Microwave Ablation (MWA) therapy. In order to accurately perform ablation of liver tumors and prevent tumor recurrence it is necessary to segment both the living tumor and the ablated tissue on the CT images. The U-Net model has outperformed other methods in biomedical image segmentation. However, because of the low contrast between tumor and liver tissue texture, the traditional U-net network cannot perform an accurate segmentation of the CT images of liver during MWA therapy. The aim of this study is to improve the U-net model network to achieve a higher segmentation performance on the CT images of liver tumor inMWA therapy. To achieve this, residual block is added in the first steps of up-sampling to deepen the network depth and enhance the segmentation result. We compare the proposed method named as ‘ResLU-Net’ with a conventional U-Net model. The results show that the ResLU-Net method has a good performance in tumor segmentation with a structure similarity index (SSIM) value of 0.97. This new method can help physicians in the MWA therapy process.

Keywords

• ct image of liver
• computed tomography
• segmentation
• u-net
• deep learning