2021年度 人工知能学会全国大会(第35回)

講演情報

IEEE CYBCONF

IEEE CYBCONF » IEEE CYBCONF

[1M4-CC] Late Breaking Research Session - A

2021年6月8日(火) 17:20 〜 19:00 M会場 (CybConf会場)

Emi Yuda

18:20 〜 18:40

[1M4-CC-04] Development of system to design the patient specific injection template for uterine cancer treatment

Mutsumi Yamada1, Kento Morita1, Ryuichi Yada2, Shuhei Sekii3, Tetsushi Wakabayashi1 (1. Mie University, 2. Anjo Kosei Hospital, 3. Kita-Harima Medical Center)

Brachytherapy is a kind of treatment for uterine cancer, in which tiny radiation sources are injected into tumor. The injection is difficult because the 3D position of tumor and radiation sources are not visible in X-ray fluoroscopy. The purpose of this research is the development of system to automatically design the patient specific injection template by building an injection position and direction optimization method. This study employs MR image of uterine cancer subject, and tumor area mask image is created by doctor. The tumor area is mapped onto spherical surface whose radius is equal to the farthest point from its origin. We mapped set of voxels that lie on lines connecting origin and spherical surface. Injection target points must be evenly distributed within tumor, we quantized the mapping surface by 1 degree and split into clusters by k-means clustering. Finally, we determined center of each cluster as injection target point. Radiation sources are injected through the bottom plane of the cylinder shape injection template, and injection start points must be evenly distributed within template not to interfere with each other. In the 2D plane, start points are distributed to having regular polygon shape. For start points optimization, the gradient descent method minimizes the objective function that evaluates the variance of distance between start points and distance from each point to outer circle. For optimization including path to target points, we will add inverse of the nearest neighbor distance between paths to the objective function.

講演PDFパスワード認証
論文PDFの閲覧にはログインが必要です。参加登録者の方は「参加者用ログイン」画面からログインしてください。あるいは論文PDF閲覧用のパスワードを以下にご入力ください。

パスワード