講演抄録/キーワード |
講演名 |
2022-02-21 15:35
Liver Tumor Segmentation by Using a Massive-Training Artificial Neural Network (MTANN) and its Analysis in Liver CT. ○Yuqiao Yang・Muneyuki Sato・Ze Jin・Kenji Suzuki(Tokyo Tech) ITS2021-33 IE2021-42 |
抄録 |
(和) |
Based on a 3D massive-training artificial neural network (MTANN) combined with a Hessian-based ellipse enhancer, a small-sample-size deep learning technique for semantic segmentation of liver tumors in contrast-enhanced CT is proposed. To show the proposed model's efficiency in a small-sample size dataset, we trained the proposed models with only 7 tumors from 7 patients, and 14 tumors from 12 patients. The proposed model achieved a Dice score of 0.703 with the training set of 12 patients. The accuracy was comparable to the CNN-based method with 131 patients in the MICCAI 2017 competition. The proposed model is essential in deep learning applications in medical imaging where a large database is not available. |
(英) |
Based on a 3D massive-training artificial neural network (MTANN) combined with a Hessian-based ellipse enhancer, a small-sample-size deep learning technique for semantic segmentation of liver tumors in contrast-enhanced CT is proposed. To show the proposed model's efficiency in a small-sample size dataset, we trained the proposed models with only 7 tumors from 7 patients, and 14 tumors from 12 patients. The proposed model achieved a Dice score of 0.703 with the training set of 12 patients. The accuracy was comparable to the CNN-based method with 131 patients in the MICCAI 2017 competition. The proposed model is essential in deep learning applications in medical imaging where a large database is not available. |
キーワード |
(和) |
deep learning / small-sample-size / medical image / semantic segmentation / / / / |
(英) |
deep learning / small-sample-size / medical image / semantic segmentation / / / / |
文献情報 |
信学技報, vol. 121, no. 374, IE2021-42, pp. 49-54, 2022年2月. |
資料番号 |
IE2021-42 |
発行日 |
2022-02-14 (ITS, IE) |
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 |
ITS2021-33 IE2021-42 |