Home
This Title All WIREs
WIREs RSS Feed
How to cite this WIREs title:
WIREs Data Mining Knowl Discov
Impact Factor: 2.111

Clustering techniques for neuroimaging applications

Full article on Wiley Online Library:   HTML PDF

Can't access this content? Tell your librarian.

Clustering has been proven useful for knowledge discovery from massive data in many applications ranging from market segmentation to bioinformatics. In this study, we focus on clustering large amounts of medical image data of the human brain to identify structures of interest. Advanced Magnetic Resonance Imaging techniques enable unprecedented insights into the complex processes in the brain. However, especially for clinical studies, a huge amount of data has to be processed in order to find patterns characterizing the structure and function of the healthy brain and its alternations associated with diseases. We survey clustering methods specifically designed for neuroimaging applications such as segmentation of fiber tracks and lesions, as well as methods that can deal with multimodal imaging data. Furthermore, we will illustrate how clustering enables knowledge discovery from data by enhancing the performance of supervised techniques and discovering meaningful subgroups of subjects. The main purpose of this study is to give an introduction on how versatile clustering techniques can be applied in neuroimaging to tackle different applications where automated methods are desired. WIREs Data Mining Knowl Discov 2016, 6:22–36. doi: 10.1002/widm.1174

Illustration of common MRI sequences from left to right: T1‐weighted, T2‐weigthed, T1‐Gadulinium, and FLAIR of one slice in the brain.
[ Normal View | Magnified View ]
Illustration of SVM in order to distinguish between healthy and MCI patients.
[ Normal View | Magnified View ]
Illustration of activated brain regions in healthy controls compared to patients with somatoform pain.
[ Normal View | Magnified View ]
Illustration of an extensive stroke lesion in MRI FLAIR brain scan labeled in blue which appear hyper‐intense.
[ Normal View | Magnified View ]
Illustration of fibers, on the very left the ground truth fibers are shown and the different levels of the fiber tracking using the approach of Mai et al. The very last level has similar fiber bundles found according to the ground truth annotation.
[ Normal View | Magnified View ]
Illustration of the types of points DBSCAN determines. In this example, a core object has to have at least MinPts = 3 objects in its ε‐neighborhood.
[ Normal View | Magnified View ]
Illustration of (a) three clusters generated from different gaussian distribution, (b) Gaussian mixture model fitted with the EM algorithm on the given data points, and (c) dendrogram as output of a hierachical clustering.
[ Normal View | Magnified View ]
Illustration of several slices of DTI images (a) and corresponding diffusion tensor field (b) of the labeled area. Fibers extracted from tractography (c).
[ Normal View | Magnified View ]
A simple interaction pattern where a, b, and c depict the data objects of the multivariate time series. Signal d represents a linear combination of the other given signals over time.
[ Normal View | Magnified View ]

Browse by Topic

Technologies > Structure Discovery and Clustering
Algorithmic Development > Biological Data Mining
Application Areas > Health Care

Access to this WIREs title is by subscription only.

Recommend to Your
Librarian Now!

The latest WIREs articles in your inbox

Sign Up for Article Alerts