神經架構搜尋用於大腦腫瘤分割

A Neural Architecture Search Method for Brain Tumor Segmentation

吳炘珉(Hsin-Min Wu)

指導教授 ： 周承復

國立臺灣大學/電機資訊學院/資訊網路與多媒體研究所/碩士(2020年)

https://doi.org/10.6342/NTU202002637

摘要

目前，神經網路架構搜尋已經在圖片分類問題上取得了不錯的效果。然而神經網路架構搜尋比較少被應用於其他領域的結果。本論文的目的是用神經網路架構搜尋來找到一個可以用於大腦腫瘤分割問題的神經網路架構。本論文使用基於強化學習的網路架構搜尋方法，並且使用一個遞迴神經網路作為控制器，搜尋一個固定架構之下的細胞結構。為了使搜尋過程能夠加速，在搜尋過程中使用了參數共享使所有的子模型都能共用參數。因為是圖像分割問題，搜尋的目標是在要找出一個類似於U-net的網路中的細胞結構。實驗結果顯示，在Brats的資料集中搜尋得到的架構可以達到和人所設計的架構有同樣優秀的結果。

關鍵字

神經架構搜索 ； U-net ； 神經網路 ； 強化學習

並列摘要

Recently, neural architecture search has achieved state-of-the art performance in the problem of image classification. However, neural architecture search is rarely used in the domain other than image segmentation. The purpose of this paper is to use neural network architecture search to find a neural network architecture which can be used for brain tumor segmentation. In this paper, we use a network architecture search method based on reinforcement learning, and use a recurrent neural network as a controller to search for cell structures in fixed architecture. In order to speed up the search process, parameter sharing is used in the search process so that all child models can share parameters. Because it is an image segmentation problem, we use neural architecture search to find each cell architecture in a U-net like network. The experimental results show that, the searched architecture can achieved state-of-the art performance as human-designed models in the Brats dataset.

並列關鍵字

Neural Architecture Search ； U-net ； Neural Network ； Reinforcement learning

參考文獻

[1] S. Bakas, H. Akbari, A. Sotiras, M. Bilello, M. Rozycki, J. Kirby, J. Freymann, K. Farahani, and C. Davatzikos. Segmentation labels and radiomic features for the pre-operative scans of the tcga-gbm collection. the cancer imaging archive. Nat Sci Data, 4:170117, 2017.

[2] S. Bakas, H. Akbari, A. Sotiras, M. Bilello, M. Rozycki, J. Kirby, J. Freymann, K. Farahani, and C. Davatzikos. Segmentation labels and radiomic features for the pre-operative scans of the tcga-lgg collection. The cancer imaging archive, 286, 2017.

[3] S. Bakas, H. Akbari, A. Sotiras, M. Bilello, M. Rozycki, J. S. Kirby, J. B. Freymann, K. Farahani, and C. Davatzikos. Advancing the cancer genome atlas glioma mri collections with expert segmentation labels and radiomic features. Scientific data, 4:170117, 2017.

[4] S. Bakas, M. Reyes, A. Jakab, S. Bauer, M. Rempfler, A. Crimi, R. T. Shinohara, C. Berger, S. M. Ha, M. Rozycki, et al. Identifying the best machine learning algorithms for brain tumor segmentation, progression assessment, and overall survival prediction in the brats challenge. arXiv preprint arXiv:1811.02629, 2018.

[5] G. Bender, P.-J. Kindermans, B. Zoph, V. Vasudevan, and Q. Le. Understanding and simplifying one-shot architecture search. In International Conference on Machine Learning, pages 550–559, 2018.