| 講演抄録/キーワード |
| 講演名 |
2024-03-04 10:22
Robust segmentation approach over various training-to-test ratios for gross tumor volumes of lung cancer based on fused outputs ○Yunhao Cui・Hidetaka Arimura(Kyushu Univ.)・Yuko Shirakawa(National Hospital Organization Kyushu Cancer)・Tadamasa Yoshitake(Kyushu Univ.)・Yoshiyuki Shioyama(Saga HIMAT)・Hidetake Yabuuchi(Kyushu Univ.) MI2023-68 |
| 抄録 |
(和) |
This study investigates robust deep learning (DL) methods for segmenting lung cancer from stereotactic body radiotherapy (SBRT) using the datasets with low training-to-test number ratio (TTR, the ratio of number of cases in training dataset to that in test dataset). Using 192 SBRT patients, 3D U-Net, V-Net, and dense V-Net were trained, producing nine fused outputs. Voting-fused model outperformed others, achieving Dice’s similarity coefficients of 0.798 to 0.829 and Hausdorff distances of 5.40±3.00 to 6.07±3.26 mm over 5 TTRs of 0.116 to 1.00. The findings recommend voting-fused model as a robust approach for low TTR datasets in lung cancer segmentation. |
| (英) |
This study investigates robust deep learning (DL) methods for segmenting lung cancer from stereotactic body radiotherapy (SBRT) using the datasets with low training-to-test number ratio (TTR, the ratio of number of cases in training dataset to that in test dataset). Using 192 SBRT patients, 3D U-Net, V-Net, and dense V-Net were trained, producing nine fused outputs. Voting-fused model outperformed others, achieving Dice’s similarity coefficients of 0.798 to 0.829 and Hausdorff distances of 5.40±3.00 to 6.07±3.26 mm over 5 TTRs of 0.116 to 1.00. The findings recommend voting-fused model as a robust approach for low TTR datasets in lung cancer segmentation. |
| キーワード |
(和) |
/ / / / / / / |
| (英) |
deep learning / low training-to-test number ratio / lung cancer segmentation / / / / / |
| 文献情報 |
信学技報, vol. 123, no. 411, MI2023-68, pp. 117-118, 2024年3月. |
| 資料番号 |
MI2023-68 |
| 発行日 |
2024-02-25 (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 |
MI2023-68 |
|