Gabriele Lombardi

Automatic liver segmentation from abdominal CT scans

Abdominal CT images have been widely studied in the recent years as they are becoming an invaluable mean for abdominal organ investigation. In the field of medical image processing, some of the current interests are the automatic diagnosis of liver pathologies and its 3D volume rendering. The first and fundamental step in all these studies is the automatic liver segmentation, that is still an open problem. In this paper we describe two automatic methods to segment the liver from abdominal CT data. They have been evaluated on the data of 40 patients, by comparing the automatically detected liver volumes to the liver boundaries manually traced by three experts. The best performing method achieves a symmetric volume difference of 95%, which is comparable to the inter and intra-personal variability of the manual segmentation produced by experts.

IDEA: intrinsic dimension estimation algorithm

The high dimensionality of some real life signals makes the usage of the most common signal processing and pattern recognition methods unfeasible. For this reason, in literature a great deal of research work has been devoted to the development of algorithms performing dimensionality reduction. To this aim, an useful help could be provided by the estimation of the intrinsic dimensionality of a given dataset, that is the minimum number of parameters needed to capture, and describe, all the information carried by the data. Although many techniques have been proposed, most of them fail in case of noisy data or when the intrinsic dimensionality is too high. In this paper we propose a local intrinsic dimension estimator exploiting the statistical properties of data neighborhoods. The algorithm evaluation on both synthetic and real datasets, and the comparison with state of the art algorithms, proves that the proposed technique is promising.

DANCo: An intrinsic dimensionality estimator exploiting angle and norm concentration

Abstract In the past decade the development of automatic intrinsic dimensionality estimators has gained considerable attention due to its relevance in several application fields. However, most of the proposed solutions prove to be not robust on noisy datasets, and provide unreliable results when the intrinsic dimensionality of the input dataset is high and the manifold where the points are assumed to lie is nonlinearly embedded in a higher dimensional space. In this paper we propose a novel intrinsic dimensionality estimator ( DANCo ) and its faster variant ( FastDANCo ), which exploit the information conveyed both by the normalized nearest neighbor distances and by the angles computed on couples of neighboring points. The effectiveness and robustness of the proposed algorithms are assessed by experiments on synthetic and real datasets, by the comparative evaluation with state-of-the-art methodologies, and by significance tests.

Novel IPCA-Based Classifiers and Their Application to Spam Filtering

This paper proposes a novel two-class classifier, called IPCAC, based on the Isotropic Principal Component Analysis technique; it allows to deal with training data drawn from Mixture of Gaussian distributions, by projecting the data on the Fisher subspace that separates the two classes. The obtained results demonstrate that IPCAC is a promising technique; furthermore, to cope with training datasets being dynamically supplied, and to work with non-linearly separable classes, two improvements of this classifier are defined: a model merging algorithm, and a kernel version of IPCAC. The effectiveness of the proposed methods is shown by their application to the spam classification problem, and by the comparison of the achieved results with those obtained by Support Vector Machines SVM, and K-Nearest Neighbors KNN.

PIPCAC: A Novel Binary Classifier Assuming Mixtures of Gaussian Functions

Probabilistic classifiers are among the most popular classification methods adopted by the machine learning community. They are often based on a-priori knowledge about the probability distribution underlying the data; nevertheless this information is rarely provided, so that a family of probability distribution functions is assumed to be an approximation model. In this paper we present an efficient binary classification algorithm, called Perceptron-IPCAC (PIPCAC), assuming that each class is distributed accordingly to a Mixture of Gaussian functions. PIPCAC is defined as a multilayer perceptron trained by combining different linear classifiers. The algorithm has been tested on both synthetic and real datasets, and the obtained results demonstrate the effectiveness and efficiency of the proposed method. Furthermore, the promising performances have been confirmed by the comparison of its results with those achieved by Support Vector Machines.

Curvature Estimation and Curve Inference with Tensor Voting: A New Approach

Recently the tensor voting framework ( TVF ), proposed by Medioni at al., has proved its effectiveness in perceptual grouping of arbitrary dimensional data. In the computer vision field, this algorithm has been applied to solve various problems as stereo-matching, boundary inference, and image inpainting. In the last decade the TVF was augmented with new saliency features, like curvature and first order tensors. In this paper a new curvature estimation technique is described and its effectiveness, when used with the saliency functions proposed in [1], is demonstrated. Results are shown for synthetic datasets in spaces of different dimensionalities.

The Neighbors Voting Algorithm and Its Applications

In the last ten years the tensor voting framework (TVF), proposed by Medioni at al., has proved its effectiveness in perceptual grouping of arbitrary dimensional data. In the computer vision and image processing fields, this algorithm has been applied to solve various problems like stereo-matching, 3D reconstruction, and image inpainting . In this paper we propose a new technique, inspired to the TVF, that allows to estimate the dimensionality and normal orientation of the manifolds underlying a given point set. These features are encoded in tensors that can be considered as weak classifiers, whose combination is then used as a strong classifier to solve different classification problems. To prove the effectiveness of the described algorithm, three problems are considered: clustering by dimensionality estimation, image classification by manifold learning, and image inpainting by texture learning.

3D α expansion and graph cut algorithms for automatic liver segmentation from CT images

Abdominal CT images have been widely studied in the recent years as they are becoming an invaluable mean for abdominal organ investigation. In the field of medical image processing, some of the current interests are the automatic diagnosis of liver pathologies and its 3D volume rendering. The first and fundamental step in all these studies is the automatic liver segmentation, that is still an open problem. In this paper we describe an automatic method to segment the liver from abdominal CT data, by combining an α-expansion and a graph cut algorithm. When evaluated on the data of 40 patients, by comparing the automatically detected liver volumes to the liver boundaries manually traced by three experts, the method achieves a symmetric volume difference of 94%.

Tensor Voting Fields: Direct Votes Computation and New Saliency Functions

The tensor voting framework (TVF), proposed by Medioni at at, has proved its effectiveness in perceptual grouping of arbitrary dimensional data. In the computer vision and image processing fields, this algorithm has been applied to solve various problems like stereo-matching, 3D reconstruction, and image in painting. The TVF technique can detect and remove a big percentage of outliers, but unfortunately it does not generate satisfactory results when the data are corrupted by additive noise. In this paper a new direct votes computation algorithm for high dimensional spaces is described, and a parametric class of decay functions is proposed to deal with noisy data. Preliminary comparative results between the original TVF and our algorithm are shown on synthetic data.

3D Volume Reconstruction and Biometric Analysis of Fetal Brain from MR Images

Magnetic resonance imaging (MRI) is becoming increasingly popular as a second-level technique, performed after ultrasonography (US) scanning, for detecting morphologic brain abnormalities. For this reason, several medical researchers in the past few years have investigated the field of fetal brain diagnosis from MR images, both to create models of the normal fetal brain development and to define diagnostic rules, based on biometric analysis; all these studies require the segmentation of cerebral structures from MRI slices, where their sections are clearly visible. A problem of this approach is due to the fact that fetuses often move during the sequence acquisition, so that it is difficult to obtain a slice where the structures of interest are properly represented. Moreover, in the clinical routine segmentation is performed manually, introducing a high inter and intra-observer variability that greatly decreases the accuracy and significance of the result. To solve these problems in this paper we propose an algorithm that builds a 3D representation of the fetal brain; from this representation the desired section of the cerebral structures can be extracted. Next, we describe our preliminary studies to automatically segment ventricles and internal liquors (from slices where they are entirely visible), and to extract biometric measures describing their shape. In spite of the poor resolution of fetal brain MR images, encouraging preliminary results have been obtained.