LEAP


In LEAP, manifold learning is used to boost multi-atlas segmentation by Learning Embeddings for Atlas Propagation. The method automatically learns a set of structural labels for a large and diverse image population by propagating manually defined label maps from a subset of the population to the entire image set. The automatically generated atlas database can be used to accurately label unseen images by selecting the most appropriate atlases from the entire pool. The resulting structural segmentation allows the accurate extraction of the volumes of different brain structures. 4D graph-cuts, an extension to LEAP, furthermore generates an accurate segmentation of longitudinal image sequences, allowing the reliable measurement of atrophy. The LEAP technology is licensed to IXICO Ltd. and currently tested for the application of measuring hippocampal volume in clinical trials of several major pharmaceutical companies.

Related Publications:

  1. • R. Wolz, P. Aljabar, J.V. Hajnal, A. Hammers, D. Rueckert

   LEAP: Learning Embeddings for Atlas Propagation

   NeuroImage, 49(2):1316-1325, 2010

  1. • R. Wolz, R.A. Heckemann, P. Aljabar, J.V. Hajnal, A. Hammers, J. Lotjonen and D. Rueckert

   Measurement of hippocampal atrophy using 4D graph-cut segmentation: Application to ADNI

   NeuroImage, 52(1):109-118, 2010

  1. • L. Clerx, L. van der Pol, D. Rueckert, R. de Jong, R. v Schijndel et al.

   Comparison of measurements of medial temporal lobe atrophy in the prediction of Alzheimer's Disease in subjects with MCI

   Alzheimer's and Dementia, Volume 7, Issue 4, Supplement 1, July 2011, Page S133

  1. • R. Wolz, D. Rueckert

   Segmentation of the ADNI database into 83 anatomical regions using LEAP

   MICCAI 2011 Workshop on Multi-Atlas Labeling and Statistical Fusion, Toronto, Canada, September 2011


Manifold Classification


Learning a low-dimensional manifold space of a set of images can help to characterise the population of interest. If the clinical status of some subjects is available, it can be used to infer on the status of unlabelled subjects. In this work, the low-dimensional manifold space learned from subjects with Alzheimer’s disease and healthy controls is used to identify subjects at risk of developing the disease. Ways are proposed to incorporate subjects metadata and longitudinal MRI into the manifold learning process, increasing the accuracy of the learned space and leading to a better prediction accuracy.

Related Publications:

  1. • R. Wolz, P. Aljabar, J.V. Hajnal and D. Rueckert

   Manifold learning for biomarker discovery in MR imaging

   MICCAI 2010 Workshop on Machine Learning in Medical Imaging, Beijing, China, September 2010

  1. • R. Wolz, P. Aljabar, J. V. Hajnal, J. Lotjonen, D. Rueckert

   Manifold Learning Combining Imaging with Non-Imaging Information

   ISBI 2011, Chicago, USA, April 2011 (Finalist Best Student Paper Award)

  1. • R. Wolz, P. Aljabar, J. V. Hajnal, J. Lotjonen, D. Rueckert

   Manifold-based classification incorporating subject metadata

   MIUA 2011, London, UK, July 2011