[19p-C32-1~12] 3.3 生体・医用光学

15:15 〜 15:30英語発表

[19p-C32-9] Neural network-based amplitude-spectral dynamic optical coherence tomography

〇(D)Yusong Liu¹, Ibrahim Abd El-Sadek^1,2, Atsuko Furukawa³, Satoshi Matsusaka³, Yoshiaki Yasuno¹ (1.COG, Univ. of Tsukuba, 2.Faculty of Science, Damietta Univ., 3.Faculty of Medicine, Univ. of Tsukuba)

キーワード：

optical coherence tomography、dynamic optical coherence tomography、deep learning

Dynamic optical coherence tomography (DOCT) is a combination of sequential OCT acquisition and subsequent analysis of the OCT signal fluctuations among multiple frames captured at each location in the sample. It has been widely used for investigating the functional activities of in-vitro samples. Despite its success, DOCT requires a long acquisition time due to the requirement of multiple (tens to hundreds of) frames per location. To make DOCT imaging faster, we trained a neural network (NN) which utilizes long short-term memory and 3-dimensional convolutional layer to generate amplitude spectrum DOCT (AS-DOCT) images, a frequency-sensitive DOCT, from a few OCT frames. NN-generated AS-DOCT images (computed from 4 frames) well resemble the ground truth AS-DOCT images (computed from 32 frames). The proposed NN method may significantly reduce the volumetric acquisition time of AS-DOCT.