Phase Attention Mask R-CNN for Lesion Detection and Segmentation
Liver cancer is the sixth most commonly diagnosed cancer and the fourth cause of cancer-related deaths worldwide. Multi-phase computed tomography (CT), which is also known as contrast-enhanced dynamic CT, is usually used for early liver cancer diagnosis. Lesion detection and segmentation are essential pretreatment steps in computer-aided diagnosis of liver cancers. In this study, we proposed a phase attention mask R-CNN based method (Fig.1) for simultaneous detection and segmentation of liver lesions in multi-phase CT images. Each feature of the triple phase image is selectively extracted by the attention network for each scale. Typical segmentation results are shown in Fig.2. The segmentation accuracy (Dice value) is about 0.60 ~ 0.66 for single-phase CT images, and the accuracy can be improved to about 0.77 by the proposed method using attention network with multi-phase CT images.
This work is supported in part by the Grant-in Aid for Scientific Research from the Japanese Ministry for Education, Science, Culture and Sports (MEXT) under the Grant No. 20KK0234, 20K21821, 18H03267.
Fig.1. Overview of the proposed phase attention mask R-CNN network for lesion detection and segmentation
Fig.2. Typical segmentation results
Related Publications：
1. Ryo Hasegawa, Yutaro IWAMOTO, Xianhua HAN, *Lanfen LIN, *Hongjie HU, Xiujun CAI, and *Yen-Wei CHEN, “Automatic Detection and Segmentation of Liver Tumors in Multi-Phase CT Images by Phase Attention Mask R-CNN,” Proc. of 39th IEEE International Conference on Consumer Electronics (IEEE ICCE2021), pp.1-4, online, Jan. 10-12, 2021.
2. Ryo Hasegawa, Yutaro IWAMOTO, Lanfen LIN, Hongjie HU, and Yen-Wei CHEN, “Automatic Segmentation of Liver Tumor in Multiphase CT Images by Mask R-CNN,” 2020 IEEE 2nd Global Conference on Life Sciences and Technologies (LifeTech), Kyoto, Japan, March, 2020.(Link)