Semi-supervised cluster analysis of imaging data

Abstract

In this paper, we present a semi-supervised clustering-based framework for discovering coherent subpopulations in heterogeneous image sets. Our approach involves limited supervision in the form of labeled instances from two distributions that reflect a rough guess about subspace of features that are relevant for cluster analysis. By assuming that images are defined in a common space via registration to a common template, we propose a segmentation-based method for detecting locations that signify local regional differences in the two labeled sets. A PCA model of local image appearance is then estimated at each location of interest, and ranked with respect to its relevance for clustering. We develop an incremental k-means-like algorithm that discovers novel meaningful categories in a test image set. The application of our approach in this paper is in analysis of populations of healthy older adults. We validate our approach on a synthetic dataset, as well as on a dataset of brain images of older adults. We assess our method's performance on the problem of discovering clusters of MR images of human brain, and present a cluster-based measure of pathology that reflects the deviation of a subject's MR image from normal (i.e. cognitively stable) state. We analyze the clusters' structure, and show that clustering results obtained using our approach correlate well with clinical data.

Multimodal manifold-regularized transfer learning for MCI conversion prediction

“…Duchesne et al used the morphological factor method based on MRI data and presented an accuracy of 72.3 % on the dataset of 20 MCI-C and 29 MCI-NC subjects (Duchesne and Mouiha 2011). Unlike the approaches of reducing feature dimensionality for addressing the small-sample-size problem, several groups have applied a semi-supervised learning (SSL) method by increasing the number of training samples with unlabeled samples, which are often much easier to obtain (Cheng et al 2013a; Filipovych et al 2011a, b; Zhang and Shen 2011). …”

Semi-supervised fuzzy clustering with metric learning and entropy regularization

“…Semi-supervised clustering learning from a combination of side information and unlabeled data can be usually seen as such a class of appealing clustering approaches. Specifically, semi-supervised clustering, the use of class labels, pairwise constraints, or prior membership degrees on some instances to aid unsupervised clustering, has attracted great interest both in theory and in practice because of requiring less human effort and giving higher accuracy [1][2][3][4][5][6][7][8][9][10][11][12][19][20][21].…”

“…There have been considerable researches on SHCM over the past few years. This family of algorithms include constrained K-means [1], PCK-means [2], MPCK-means [3,4], enhancing semisupervised clustering [5], constraint-based clustering [6], kernel semi-supervised clustering [7,8], metric-based clustering [9], and other extension works [10][11][12]. Constrained K-means adjusts the cluster memberships to be consistent with pairwise constraints [1].…”

“…Kang et al [10] developed a software program for semi-supervised cluster analysis using semantic distance between gene pairs in gene ontology. Filipovych et al [11] presented a semi-supervised clustering-based framework for discovering coherent subpopulations in heterogeneous image sets. Liu et al [12] presented a novel semi-supervised sparse metric learning algorithm through employing abundant unlabeled data and pursuing sparsity of metrics.…”

Domain Transfer Learning for MCI Conversion Prediction

“…The same problem also exists in the classification between AD and NC, where the number of AD and NC subjects are also limited. Recently, to enhance the classification between AD and NC, several studies have used semi-supervised learning (SSL) methods [8] in AD diagnosis, where MCI subjects are treated as unlabeled data to aid AD classification [9][10][11]. Semi-supervised learning methods can efficiently utilize unlabeled samples to improve classification and regression performance, but it requires unlabeled samples and labeled samples coming from the same data distribution [8], which is usually not satisfied in practice.…”