3d morhphological or anatomical landmark detection method and device using deep reinforcement learning

[특허 요약] Provided are 3D morphological or anatomical landmark detection method and device using deep reinforcement learning, in which the method includes performing volume rendering on 3D image data to generate 3D volume rendered model data, and applying a single-stage deep reinforcement learning (DRL), including obtaining a 2D coordinate value corresponding to a landmark point on a 2D view image, which is constructed by the 3D volume rendered model data, by using a deep reinforcement learning agent, and obtaining a remaining 1D coordinate value corresponding to the landmark point based on an 1D profile of the 3D image data which is obtained based on the 2D coordinate value. The method may further include applying a multi-stage DRL, including detecting the landmark point by repeating the operations of obtaining the 2D coordinate value corresponding to the landmark point on the 2D view image, which is constructed by the 3D volume rendered model data, by using the deep reinforcement learning agent, and, from a 2D view image which is determined based on the obtained 2D coordinate value, obtaining a 2D coordinate value corresponding to the landmark point by using a deep reinforcement learning agent.

특허 등록번호

4016470A

특허 출원번호

21214467

출원일

2021.12.14

등록일

2022.06.22

국가

유럽_특허청

출원인

연세대학교 산학협력단, 기초과학연구원

발명자

Sang Hwy Lee, Sung Ho Kang, Ki Wan Jeon

중요 키워드

coordinate value, landmark, reinforcement learning, deep reinforcement, drl