2020年春の年会

講演情報

一般セッション

VII. 保健物理と環境科学 » VII. 保健物理と環境科学

[1F01-05] 医学・生物学応用1

2020年3月16日(月) 10:30 〜 11:55 F会場 (共通講義棟 M棟1F M-2)

座長:真辺 健太郎(JAEA)

10:30 〜 10:45

[1F01] 3D internal points position prediction using a recurrent neural network for tumor tracking during lung cancer radiotherapy

*Michel Pohl1, Mitsuru Uesaka1, Kazuyuki Demachi1, Ritu Bhusal Chhatkuli2 (1. Univ. of Tokyo, 2. NIRS)

キーワード:Optical flow, Lung cancer, Radiotherapy, Real-time recurrent learning, Recurrent neural network

In lung cancer treatment with radiotherapy, one can choose to track implanted metal markers or diaphragm points of interest, to reduce the radiation delivered to healthy tissues. But the typical treatment machine response time of approximately 500ms leads to inaccuracies in the beam delivery, which results in turn to unwanted damage to the tumor surrounding tissues. To overcome this latency, predicting the position of the tracked surrogates is necessary.

In this study we use chest 4DCBCT images of a patient breathing, with lung cancer. The image sequence is extended artificially to provide enough training and testing data. Internal points are selected and their motion during the breathing process are computed from the Lucas Kanade pyramidal iterative optical flow registration algorithm. We evaluate the performance of a recurrent neural network for the prediction of their position.