Showing papers by "Balaji Krishnapuram published in 2006"

Multiple Instance Learning for Computer Aided Diagnosis

Abstract: Many computer aided diagnosis (CAD) problems can be best modelled as a multiple-instance learning (MIL) problem with unbalanced data: i.e., the training data typically consists of a few positive bags, and a very large number of negative instances. Existing MIL algorithms are much too computationally expensive for these datasets. We describe CH, a framework for learning a Convex Hull representation of multiple instances that is significantly faster than existing MIL algorithms. Our CH framework applies to any standard hyperplane-based learning algorithm, and for some algorithms, is guaranteed to find the global optimal solution. Experimental studies on two different CAD applications further demonstrate that the proposed algorithm significantly improves diagnostic accuracy when compared to both MIL and traditional classifiers. Although not designed for standard MIL problems (which have both positive and negative bags and relatively balanced datasets), comparisons against other MIL methods on benchmark problems also indicate that the proposed method is competitive with the state-of-the-art.

Variational Bayes for continuous hidden Markov models and its application to active learning

Abstract: In this paper, we present a variational Bayes (VB) framework for learning continuous hidden Markov models (CHMMs), and we examine the VB framework within active learning. Unlike a maximum likelihood or maximum a posteriori training procedure, which yield a point estimate of the CHMM parameters, VB-based training yields an estimate of the full posterior of the model parameters. This is particularly important for small training sets since it gives a measure of confidence in the accuracy of the learned model. This is utilized within the context of active learning, for which we acquire labels for those feature vectors for which knowledge of the associated label would be most informative for reducing model-parameter uncertainty. Three active learning algorithms are considered in this paper: 1) query by committee (QBC), with the goal of selecting data for labeling that minimize the classification variance, 2) a maximum expected information gain method that seeks to label data with the goal of reducing the entropy of the model parameters, and 3) an error-reduction-based procedure that attempts to minimize classification error over the test data. The experimental results are presented for synthetic and measured data. We demonstrate that all of these active learning methods can significantly reduce the amount of required labeling, compared to random selection of samples for labeling.

Method, apparatus and computer program for displaying marks in an image data set

Abstract: A method and an apparatus display marks in an image data set, wherein an image data set comprising marks is provided and wherein during a review phase not all marks within the image data set are displayed at the same time. A list of the marks can be generated by sorting the marks depending on a predetermined sorting criterion and wherein the marks are displayed temporally one after another within the image data set in accordance with the generated list. The image data set is for example a medical image data set, wherein the marks are CAD marks and wherein the sorting criterion is the probability of marking illness, in particular the suspiciousness.

Batch classification with applications in computer aided diagnosis

Abstract: Most classification methods assume that the samples are drawn independently and identically from an unknown data generating distribution, yet this assumption is violated in several real life problems. In order to relax this assumption, we consider the case where batches or groups of samples may have internal correlations, whereas the samples from different batches may be considered to be uncorrelated. Two algorithms are developed to classify all the samples in a batch jointly, one based on a probabilistic analysis and another based on a mathematical programming approach. Experiments on three real-life computer aided diagnosis (CAD) problems demonstrate that the proposed algorithms are significantly more accurate than a naive SVM which ignores the correlations among the samples.

System and method for learning rankings via convex hull separation

Abstract: A method for finding a ranking function /that classifies feature points in an n- dimensional space includes providing (61) a plurality of feature points Xk derived from tissue sample regions in a digital medical image, providing (62) training data A comprising training samples Aj where formula (I), providing (63) an ordering E = {(P,Q)| AP< AQ} of at least some training data sets where all training samples xI ∈Ap are ranked higher than any sample xj ∈ AQ , and solving (64) a mathematical optimization program to determine the ranking function f that classifies said feature points x into sets A. For any two sets Ai, Aj, Ai -< Aj , and the ranking function f satisfies inequality constraints f (xi ) ≤ f (xj ) for all xI ∈ conv(Ai) and xj ∈ conv(Aj) , where conv(A) represents the convex hull of the elements of set A.

Batch classification with applications in computer aided diagnosis

Abstract: Most classification methods assume that the samples are drawn independently and identically from an unknown data generating distribution, yet this assumption is violated in several real life problems. In order to relax this assumption, we consider the case where batches or groups of samples may have internal correlations, whereas the samples from different batches may be considered to be uncorrelated. Two algorithms are developed to classify all the samples in a batch jointly, one based on a probabilistic analysis and another based on a mathematical programming approach. Experiments on three real-life computer aided diagnosis (CAD) problems demonstrate that the proposed algorithms are significantly more accurate than a naive SVM which ignores the correlations among the samples.

Addressing image variability while learning classifiers for detecting clusters of micro-calcifications

Abstract: Computer aided detection systems for mammography typically use standard classification algorithms from machine learning for detecting lesions. However, these general purpose learning algorithms make implicit assumptions that are commonly violated in CAD problems. We propose a new ensemble algorithm that explicitly accounts for the small fraction of outlier images which tend to produce a large number of false positives. A bootstrapping procedure is used to ensure that the candidates from these outlier images do not skew the statistical properties of the training samples. Experimental studies on the detection of clusters of micro-calcifications indicate that the proposed method significantly outperforms a state-of-the-art general purpose method for designing classifiers (SVM), in terms of FROC curves on a hold out test set.