Artificial Intelligence in Biomedical Imaging Lab (AIBIL)

Research    •    People    •    Funding    •    Software    •    Jobs


  • MUSE: generates a large ensemble of candidate labels in the target image space using multiple atlases, registration algorithms and smoothness values for these algorithms. The ensemble is then fused into a final segmentation.
  • Image Harmonization: Inter-scanner variability presents challenges in pooling data from multiple centers, but it also presents an opportunity for image harmonization, the process of removing the systematic differences that exist in multi-center datasets via statistical methods.
  • Brain AGE: MRI derived Brain Age has been widely adopted by the neuroscience community as an informative biomarker of brain health at the individual level. Individuals displaying pathologic or atypical brain development and aging patterns can be identified through positive or negative deviations from typical Brain Age trajectories.
  • COMPARE: Classification Of Morphological Patterns using Adaptive Regional Elements
    COMPARE is a method for classification of structural brain magnetic resonance (MR) images, which is a combination of deformation-based morphometry and machine learning methods. Before running classification, a morphological representation of the anatomy of interest is obtained from structural MR brain images using a high-dimensional mass-preserving template warping method [1, 2]. Regions that display strong correlations between tissue volumes and classification (clinical) variables learned from training samples are extracted using a watershed segmentation algorithm. To achieve robustness to outliers, the regional smoothness of the correlation map is estimated by a cross-validation strategy. A volume increment algorithm is then applied to these regions to extract regional volumetric features. To improve efficiency and generalization ability of the classification, a feature selection technique using Support Vector Machine-based criteria is used to select the most discriminative features, according to their effect on the upper bound of the leave-one-out generalization error. Finally, SVM-based classification is applied using the best set of features, and it is tested using a leave-one-out cross-validation strategy. Although the algorithm is designed for structural brain image classification, it is readily applicable for functional brain image classification with proper feature images.
    (For more details visit COMPARE's webpage)

  • GONDOLA: Generative-Discriminative Basis Learning
    This software implements Generative-Discriminative Basis Learning (GONDOLA), GONDOLA provides a generative method to reduce the dimensionality of medical images while using class labels. It produces basis vectors that are useful for classification and also clinically interpretable. When provided with two sets of labeled images as input, the software outputs features saved in the Weka Attribute-Relation File Format (ARFF) and a MATLAB data file. The program can also save basis vectors as NIfTI-1 images. Scripts are provided to find and build an optimal classifier using Weka. The software can also be used for semi-supervised cases in which a number of subjects do not have class labels.
    (For more details visit GONDOLA's webpage)
  • SmileGAN: Semi-supervised clustering via GAN
    SmileGAN is a semi-supervised clustering method which is designed to identify disease-related heterogeneity within a patient group. In particular, many diseases and disorders are highly heterogeneous, with no single imaging signature associated with them. Identifying subtypes has been of increasing interest in neuroscience. The SmileGAN model by construction seeks multiple imaging signatures, while avoiding the common pitfall of standard clustering methods, i.e. finding variations that are unrelated to the disease and only relate to other confounding factors.  This is achieved through a GAN-based formulation that simultaneously seeks a number of (regularized) transformations from healthy controls to patients, rather than clustering patients directly, in addition to an inverse mapping that ensures that the subtype/cluster of an individual can be correctly estimated from her/his scans.
    (For more details visit