講演抄録/キーワード |
講演名 |
2023-03-06 13:41
Comparative study of the Small Intestine segmentation based on 2D and 3D U-Nets ○Qin An(Nagoya Univ.)・Hirohisa Oda(UoS)・SiRui Chen・Yuichiro Hayashi(Nagoya Univ.)・Takayuki Kitasaka(Aichi Institute Univ)・Aitaro Takimoto・Akinari Hinoki・Hiroo Uchida(Nagoya Univ.)・Kojiro Suzuki(Aichi Medical Univ.)・Masahiro Oda・Kensaku Mori(Nagoya Univ.) MI2022-82 |
抄録 |
(和) |
(まだ登録されていません) |
(英) |
In this paper, we aim to compare and analyze the small intestine segmentation methods based on U-Net and 3D U-Net. Segmenting the small intestine accurately is critical to help clinicians diagnose the intestinal diseases from CT volumes accurately. However, small intestine segmentation is still challenging due to its complex spatial structure. We trained the 3D U-Net and U-Net by four CT volumes (total number of slices 1834) from two cases, which were labeled by a specialized engineer. The input of U-Net is two-dimensional (2D) slices and the input of 3D U-Net is three-dimensional(3D) CT patches. Experimental results showed that the averaged Dice score of the 3D U-Net was 0.363±0.102 and the averaged Dice Score of the U-Net was 0.331±0.045. The Dice score of 3D U-Net was higher than that of U-Net. By analyzing the result, we got the advantages and disadvantages of different methods (U-Net and 3D U-Net) and tried to propose a new method to achieve more accurate segmentation results. |
キーワード |
(和) |
/ / / / / / / |
(英) |
Intestine segmentation / Deep Learning / U-Net / 3D U-Net / / / / |
文献情報 |
信学技報, vol. 122, no. 417, MI2022-82, pp. 46-51, 2023年3月. |
資料番号 |
MI2022-82 |
発行日 |
2023-02-27 (MI) |
ISSN |
Online edition: ISSN 2432-6380 |
著作権に ついて |
技術研究報告に掲載された論文の著作権は電子情報通信学会に帰属します.(許諾番号:10GA0019/12GB0052/13GB0056/17GB0034/18GB0034) |
PDFダウンロード |
MI2022-82 |