Claudio Marrocco

Maximizing the area under the ROC curve by pairwise feature combination

The majority of the available classification systems focus on the minimization of the classification error rate. This is not always a suitable metric specially when dealing with two-class problems with skewed classes and cost distributions. In this case, an effective criterion to measure the quality of a decision rule is the area under the Receiver Operating Characteristic curve (AUC) that is also useful to measure the ranking quality of a classifier as required in many real applications. In this paper we propose a nonparametric linear classifier based on the maximization of AUC. The approach lies on the analysis of the Wilcoxon-Mann-Whitney statistic of each single feature and on an iterative pairwise coupling of the features for the optimization of the ranking of the combined feature. By the pairwise feature evaluation the proposed procedure is essentially different from other classifiers using AUC as a criterion. Experiments performed on synthetic and real data sets and comparisons with previous approaches confirm the effectiveness of the proposed method.

A computer-aided detection system for clustered microcalcifications

OBJECTIVE The aim of this paper is to describe a novel system for computer-aided detection of clusters of microcalcifications on digital mammograms. METHODS AND MATERIAL Mammograms are first segmented by means of a tree-structured Markov random field algorithm that extracts the elementary homogeneous regions of interest. An analysis of such regions is then performed by means of a two-stage, coarse-to-fine classification based on both heuristic rules and classifier combination. In this phase, we avoid taking a decision on the single microcalcifications and forward it to the successive phase of clustering realized through a sequential approach. RESULTS The system has been tested on a publicly available database of mammograms and compared with previous approaches. The obtained results show that the system is very effective, especially in terms of sensitivity. CONCLUSIONS The proposed approach exhibits some remarkable advantages both in segmentation and classification phases. The segmentation phase employs an image model that reduces the computational burden, preserving the small details in the image through an adaptive local estimation of all model parameters. The classification stage combines the results of the classifiers focused on the single microcalcification and the cluster as a whole. Such an approach makes a detection system particularly effective and robust with respect to the large variations exhibited by the clusters of microcalcifications.

Detection of microcalcifications clusters in mammograms through TS-MRF segmentation and SVM-based classification

At present, mammography is the only not invasive diagnostic technique allowing the diagnosis of a breast cancer at a very early stage. A visual clue of such disease particularly significant is the presence of clusters of microcalcifications. Reliable methods for an automatic detection of such clusters are very difficult to accomplish because of the small size of the microcalcifications and of the poor quality of the digital mammograms. A method designed for this task is described. The mammograms are firstly segmented by means of the tree structured Markov random field algorithm which extracts the elementary homogeneous regions of interest on the image. Such regions are then submitted to a further analysis (based both on heuristic rules and support vector classification) in order to reduce the false positives. The approach has been successfully tested on a standard database of 40 mammographic images, publicly available.

Design of reject rules for ECOC classification systems

ECOC is a widely used and successful technique, which implements a multi-class classification system by decomposing the original problem into several two-class problems. In this paper, we study the possibility to provide ECOC systems with a tailored reject option carried out through different schemes that can be grouped under two different categories: an external and an internal approach. The first one is based on the reliability of the entire system output and does not require any change in its structure. The second scheme, instead, estimates the reliability of the internal dichotomizers and implies a slight modification in the decoding stage. Experimental results on popular benchmark data sets are reported to show the behavior of the different schemes.

Exploiting AUC for optimal linear combinations of dichotomizers

The combination of classifiers is an established technique to improve the classification performance. The possible combination rules proposed up to now generally try to decrease the classification error rate, which is a performance measure not suitable in many real situations and particularly when dealing with two-class problems. In this case, a good alternative is given by the area under the receiver operating characteristic curve (AUC), whose effectiveness in measuring the classification quality has been proved in many recent papers. In this paper, we propose a method to achieve the optimal linear combination of two dichotomizers based on the maximization of the AUC of the resulting classification system. The effectiveness of the approach has been confirmed by the tests performed on standard datasets.

An effective learning strategy for cascaded object detection

Object detection is frequently a complex, severely unbalanced classification problem.A cascade of node classifiers allows us to efficiently handle the complexity.In our proposal, each node classifier is trained with a ranking-based algorithm.Ranking effectively faces the imbalance between object and non-object patches.Our method is effective if compared to other learning strategies for skewed classes. To distinguish objects from non-objects in images under computational constraints, a suitable solution is to employ a cascade detector that consists of a sequence of node classifiers with increasing discriminative power. However, among the millions of image patches generated from an input image, only very few contain the searched object. When trained on these highly unbalanced data sets, the node classifiers tend to have poor performance on the minority class. Thus, we propose a learning strategy aimed at maximizing the node classifiers ranking capability rather than their accuracy. We also provide an efficient implementation yielding the same time complexity of the original Viola-Jones cascade training. Experimental results on highly unbalanced real problems show that our approach is both efficient and effective when compared to other node training strategies for skewed classes.

MCS-based balancing techniques for skewed classes: An empirical comparison

The class imbalance is a critical problem in classification tasks related to many real world applications. A large number of solutions were proposed in literature, both at the algorithmic and data levels. In this paper we analyze the second kind of approach and, in particular, we focus our attention on the use of Multiple Classification Systems where each classifier is trained on a dataset containing the minority class and a subset of the majority class samples. The aim of this approach is to avoid the drawbacks of other methods, commonly used in this context, which force a balanced distribution by oversampling the minority class. We compare the results obtained applying different realizations of the method on the UCI Repository datasets.

Automatic segmentation of the pectoral muscle in mediolateral oblique mammograms

When mammograms are analyzed through a Computer Aided Diagnosis (CAD) system the presence of the pectoral muscle can affect the results of the automatic detection of breast lesions. This problem is particularly evident in mediolateral oblique (MLO) view where the pectoral muscle appears as a high intensity region across the margin of the mammogram. An automatic identification of the pectoral muscle is an essential step because of its similar characteristics with the abnormal tissue that can interfere with the detection of suspicious regions or bias the estimation of breast tissue density. This paper presents a new approach for the detection of pectoral muscle in MLO view of the mammo-graphic images. It is based on a preprocessing step useful to normalize the image and highlight the boundary between the muscle and the mammary tissue. A subsequent step including edge detection and regression via RANSAC provides the final contour of the muscle area. The experiments performed on a standard data set show very encouraging results.

Algorithms for detecting clusters of microcalcifications in mammograms

Mammography is a not invasive diagnostic technique widely used for early cancer detection in women breast. A particularly significant clue of such disease is the presence of clusters of microcalcifications. The automatic detection of such clusters is a very difficult task because of the small size of the microcalcifications and of the poor quality of the digital mammograms. In literature, all the proposed method for the automatic detection focus on the single microcalcification. In this paper, an approach that moves the final decision on the regions identified by the segmentation in the phase of clustering is proposed. To this aim, the output of a classifier on the single microcalcifications is used as input data in different clustering algorithms which produce the final decision. The approach has been successfully tested on a standard database of 40 mammographic images, publicly available.

Detection of Clusters of Microcalcifications in Mammograms: A Multi Classifier Approach

Mammography is a not invasive diagnostic technique widely used for early cancer detection in women breast. A particularly significant clue of such disease is the presence of clusters of microcalcifications. The automatic detection and classification of such clusters is a very difficult task because of the small size of the microcalcifications and of the poor quality of the digital mammograms. In literature, all the proposed methods for the automatic detection focus on the single microcalcification. In this paper, an approach that moves the final decision on the regions identified by the segmentation in the phase of clustering is proposed. To this aim, the output of a classifier on the single microcalcifications is used as input data in a clustering algorithms which produce the detected clusters. As final output the system highlights the suspicious clusters, leaving to the specialist the diagnosis responsibility. The approach has been successfully tested on a standard database of 40 mammographic images, publicly available.