Machine Learning of Medical Images

MABU SHINGO, Kido Shoji, Hashimoto Noriaki, Hirano Yasushi, Kuremoto Takashi

doi:10.11239/jsmbe.annual56.s220-2

Bibliographic Information

Other Title

医用画像の機械学習

Abstract

<p>Research on Computer-aided diagnosis of medical images using machine learning, especially deep learning, has been actively conducted. However, many machine learning techniques are based on supervised learning; thus, they need a large number of training data with correct annotations. Especially, deep learning sometimes requires tens of thousands of annotated data, and it is quite tough work for radiologists to give annotations to the images. This research aims to classify opacities of diffuse lung diseases in lung CT images, and we introduce an unsupervised classification algorithm without using annotated data, and semi-supervised classification algorithm using only a small number of annotated data. The unsupervised algorithm combines feature extraction using deep autoencoder and Bag-of-features and K-means clustering. Semi-supervised algorithm uses the same feature extraction as the above, but self-training and active learning are applied to Support Vector Machine. In the experiments, six kinds of opacities are classified and the results are analyzed.</p>

Journal

Transactions of Japanese Society for Medical and Biological Engineering

Transactions of Japanese Society for Medical and Biological Engineering Annual56 (Abstract), S220-2-S220-2, 2018

Japanese Society for Medical and Biological Engineering

Details 詳細情報について

CRID: 1390845713000194176

NII Article ID: 130007483708

DOI: 10.11239/jsmbe.annual56.s220-2

ISSN: 18814379; 1347443X

Text Lang: ja

Data Source

JaLC
CiNii Articles

Abstract License Flag: Disallowed

Export