|
KMID : 1147120140200010001
|
|
Journal of the Korean Society of Imaging Informatics in Medicine
2014 Volume.20 No. 1 p.1 ~ p.12
|
|
|
Deep Learning-based Feature Extraction for Medical Image Analysis
|
|
Lee Ye-Ha
Kim Hyun-Jun
Kim Guk-Bae
Kim Nam-Kug
|
|
|
Abstract
|
|
|
|
Extracting features that represents well the characteristics of medical image is critical for successful medical image analysis. However, most of the conventional medical image analysis methods rely on hand-crafted or case-specific feature extraction algorithms widely used in computer vision research which often lead to unstable or degraded results. This is because the assumptions or rules used for the design of hand-crafted features are not general enough to reflect the underlying image statistics or complex structure of the input data. Therefore, developing learning-based method which extracts most significant features automatically from the medical images is key to the success of robust and accurate medical image analysis. Recently, deep learning algorithms have attracted attentions from various fields for their remarkable performance gain on feature-based recognition problems. In both supervised and unsupervised manner, deep learning methods are capable of learning hierarchical representation of the patterns from low-level to high-level features. To this end, in this paper, we introduce deep learning-based feature extraction method which adaptively learns the most significant features for the given task using deep structure. We also introduce some recent literatures on medical image analysis such as image segmentation and image registration using deep learning algorithms.
|
|
|
KEYWORD
|
|
|
Deep learning, Representation learning, Medical image analysis
|
|
|
FullTexts / Linksout information
|
|
|
|
|
|
Listed journal information
|
|
|
|