Spectral Shape Recovery and Analysis Via Data-driven Connections.

We introduce a novel learning-based method to recover shapes from their Laplacian spectra, based on establishing and exploring connections in a learned latent space. The core of our approach consists in a cycle-consistent module that maps between a learned latent space and sequences of eigenvalues. This module provides an efficient and effective link between the shape geometry, encoded in a latent vector, and its Laplacian spectrum. Our proposed data-driven approach replaces the need for ad-hoc regularizers required by prior methods, while providing more accurate results at a fraction of the computational cost. Moreover, these latent space connections enable novel applications for both analyzing and controlling the spectral properties of deformable shapes, especially in the context of a shape collection. Our learning model and the associated analysis apply without modifications across different dimensions (2D and 3D shapes alike), representations (meshes, contours and point clouds), nature of the latent space (generated by an auto-encoder or a parametric model), as well as across different shape classes, and admits arbitrary resolution of the input spectrum without affecting complexity. The increased flexibility allows us to address notoriously difficult tasks in 3D vision and geometry processing within a unified framework, including shape generation from spectrum, latent space exploration and analysis, mesh super-resolution, shape exploration, style transfer, spectrum estimation for point clouds, segmentation transfer and non-rigid shape matching. SUPPLEMENTARY INFORMATION: The online version supplementary material available at 10.1007/s11263-021-01492-6.

Classification of first-episode psychosis in a large cohort of patients using support vector machine and multiple kernel learning techniques.

First episode psychosis (FEP) patients are of particular interest for neuroimaging investigations because of the absence of confounding effects due to medications and chronicity. Nonetheless, imaging data are prone to heterogeneity because for example of age, gender or parameter setting differences. With this work, we wanted to take into account possible nuisance effects of age and gender differences across dataset, not correcting the data as a pre-processing step, but including the effect of nuisance covariates in the classification phase. To this aim, we developed a method which, based on multiple kernel learning (MKL), exploits the effect of these confounding variables with a subject-depending kernel weighting procedure. We applied this method to a dataset of cortical thickness obtained from structural magnetic resonance images (MRI) of 127 FEP patients and 127 healthy controls, who underwent either a 3Tesla (T) or a 1.5T MRI acquisition. We obtained good accuracies, notably better than those obtained with standard SVM or MKL methods, up to more than 80% for frontal and temporal areas. To our best knowledge, this is the largest classification study in FEP population, showing that fronto-temporal cortical thickness can be used as a potential marker to classify patients with psychosis.

Classification of first-episode psychosis: a multi-modal multi-feature approach integrating structural and diffusion imaging.

Currently, most of the classification studies of psychosis focused on chronic patients and employed single machine learning approaches. To overcome these limitations, we here compare, to our best knowledge for the first time, different classification methods of first-episode psychosis (FEP) using multi-modal imaging data exploited on several cortical and subcortical structures and white matter fiber bundles. 23 FEP patients and 23 age-, gender-, and race-matched healthy participants were included in the study. An innovative multivariate approach based on multiple kernel learning (MKL) methods was implemented on structural MRI and diffusion tensor imaging. MKL provides the best classification performances in comparison with the more widely used support vector machine, enabling the definition of a reliable automatic decisional system based on the integration of multi-modal imaging information. Our results show a discrimination accuracy greater than 90 % between healthy subjects and patients with FEP. Regions with an accuracy greater than 70 % on different imaging sources and measures were middle and superior frontal gyrus, parahippocampal gyrus, uncinate fascicles, and cingulum. This study shows that multivariate machine learning approaches integrating multi-modal and multisource imaging data can classify FEP patients with high accuracy. Interestingly, specific grey matter structures and white matter bundles reach high classification reliability when using different imaging modalities and indices, potentially outlining a prefronto-limbic network impaired in FEP with particular regard to the right hemisphere.

Infinite Feature Selection: A Graph-based Feature Filtering Approach.

We propose a filtering feature selection framework that considers subsets of features as paths in a graph, where a node is a feature and an edge indicates pairwise (customizable) relations among features, dealing with relevance and redundancy principles. By two different interpretations (exploiting properties of power series of matrices and relying on Markov chains fundamentals) we can evaluate the values of paths (i.e., feature subsets) of arbitrary lengths, eventually go to infinite, from which we dub our framework Infinite Feature Selection (Inf-FS). Going to infinite allows to constrain the computational complexity of the selection process, and to rank the features in an elegant way, that is, considering the value of any path (subset) containing a particular feature. We also propose a simple unsupervised strategy to cut the ranking, so providing the subset of features to keep. In the experiments, we analyze diverse settings with heterogeneous features, for a total of 11 benchmarks, comparing against 18 widely-known comparative approaches. The results show that Inf-FS behaves better in almost any situation, that is, when the number of features to keep are fixed a priori, or when the decision of the subset cardinality is part of the process.

Automatic classification of autism spectrum disorder in children using cortical thickness and support vector machine.

OBJECTIVE: Autism spectrum disorder (ASD) is a neurodevelopmental condition with a heterogeneous phenotype. The role of biomarkers in ASD diagnosis has been highlighted; cortical thickness has proved to be involved in the etiopathogenesis of ASD core symptoms. We apply support vector machine, a supervised machine learning method, in order to identify specific cortical thickness alterations in ASD subjects. METHODS: A sample of 76 subjects (9.5 ± 3.4 years old) has been selected, 40 diagnosed with ASD and 36 typically developed subjects. All children underwent a magnetic resonance imaging (MRI) examination; T1-MPRAGE sequences were analyzed to extract features for the characterization and parcellation of regions of interests (ROI); average cortical thickness (CT) has been measured for each ROI. For the classification process, the extracted features were used as input for a classifier to identify ASD subjects through a "learning by example" procedure; the features with best performance was then selected by "greedy forward-feature selection." Finally, this model underwent a leave-one-out cross-validation approach. RESULTS: From the training set of 68 ROIs, five ROIs reached accuracies of over 70%. After this phase, we used a recursive feature selection process in order to identify the eight features with the best accuracy (84.2%). CT resulted higher in ASD compared to controls in all the ROIs identified at the end of the process. CONCLUSION: We found increased CT in various brain regions in ASD subjects, confirming their role in the pathogenesis of this condition. Considering the brain development curve during ages, these changes in CT may normalize during development. Further validation on a larger sample is required.

DCE-MRI data analysis for cancer area classification.

OBJECTIVES: The paper aims at improving the support of medical researchers in the context of in-vivo cancer imaging. Morphological and functional parameters obtained by dynamic contrast-enhanced MRI (DCE-MRI) techniques are analyzed, which aim at investigating the development of tumor microvessels. The main contribution consists in proposing a machine learning methodology to segment automatically these MRI data, by isolating tumor areas with different meaning, in a histological sense. METHODS: The proposed approach is based on a three-step procedure: i) robust feature extraction from raw time-intensity curves, ii) voxel segmentation, and iii) voxel classification based on a learning-by-example approach. In the first step, few robust features that compactly represent the response of the tissue to the DCE-MRI analysis are computed. The second step provides a segmentation based on the mean shift (MS) paradigm, which has recently shown to be robust and useful for different and heterogeneous clustering tasks. Finally, in the third step, a support vector machine (SVM) is trained to classify voxels according to the labels obtained by the clustering phase (i.e., each class corresponds to a cluster). Indeed, the SVM is able to classify new unseen subjects with the same kind of tumor. RESULTS: Experiments on different subjects affected by the same kind of tumor evidence that the extracted regions by both the MS clustering and the SVM classifier exhibit a precise medical meaning, as carefully validated by the medical researchers. Moreover, our approach is more stable and robust than methods based on quantification of DCE-MRI data by means of pharmacokinetic models. CONCLUSIONS: The proposed method allows to analyze the DCE-MRI data more precisely and faster than previous automated or manual approaches.

A framework for the automatic detection and characterization of brain malformations: Validation on the corpus callosum.

In this paper, we extend the one-class Support Vector Machine (SVM) and the regularized discriminative direction analysis to the Multiple Kernel (MK) framework, providing an effective analysis pipeline for the detection and characterization of brain malformations, in particular those affecting the corpus callosum. The detection of the brain malformations is currently performed by visual inspection of MRI images, making the diagnostic process sensible to the operator experience and subjectiveness. The method we propose addresses these problems by automatically reproducing the neuroradiologist's approach. One-class SVMs are appropriate to cope with heterogeneous brain abnormalities that are considered outliers. The MK framework allows to efficiently combine the different geometric features that can be used to describe brain structures. Moreover, the regularized discriminative direction analysis is exploited to highlight the specific malformative patterns for each patient. We performed two different experiments. Firstly, we tested the proposed method to detect the malformations of the corpus callosum on a 104 subject dataset. Results showed that the proposed pipeline can classify the subjects with an accuracy larger than 90% and that the discriminative direction analysis can highlight a wide range of malformative patterns (e.g., local, diffuse, and complex abnormalities). Secondly, we compared the diagnosis of four neuroradiologists on a dataset of 128 subjects. The diagnosis was performed both in blind condition and using the classifier and the discriminative direction outputs. Results showed that the use of the proposed pipeline as an assisted diagnosis tool improves the inter-subject variability of the diagnosis. Finally, a graphical representation of the discriminative direction analysis was proposed to enhance the interpretability of the results and provide the neuroradiologist with a tool to fully and clearly characterize the patient malformations at single-subject level.

The use of dynamic susceptibility contrast (DSC) MRI to automatically classify patients with first episode psychosis.

Hemodynamic changes in the brain have been reported in major psychosis in respect to healthy controls, and could unveil the basis of structural brain modifications happening in patients. The study of first episode psychosis is of particular interest because the confounding role of chronicity and medication can be excluded. The aim of this work is to automatically discriminate first episode psychosis patients and normal controls on the basis of brain perfusion employing a support vector machine (SVM) classifier. 35 normal controls and 35 first episode psychosis underwent dynamic susceptibility contrast magnetic resonance imaging, and cerebral blood flow and volume, along with mean transit time were obtained. We investigated their behavior in the whole brain and in selected regions of interest, in particular the left and right frontal, parietal, temporal and occipital lobes, insula, caudate and cerebellum. The distribution of values of perfusion indexes were used as features in a support vector machine classifier. Mean values of blood flow and volume were slightly lower in patients, and the difference reached statistical significance in the right caudate, left and right frontal lobes, and in left cerebellum. Linear SVM reached an accuracy of 83% in the classification of patients and normal controls, with the highest accuracy associated with the right frontal lobe and left parietal lobe. In conclusion, we found evidence that brain perfusion could be used as a potential marker to classify patients with psychosis, who show reduced blood flow and volume in respect to normal controls.

Detection of corpus callosum malformations in pediatric population using the discriminative direction in multiple kernel learning.

In this paper we propose a Multiple Kernel Learning (MKL) classifier to detect malformations of the Corpus Callosum (CC) and apply it in a pediatric population. Furthermore, we extend the concept of discriminative direction to the linear MKL methods, implementing it in a single subject analysis framework. The CC is characterized using different measures derived from Magnetic Resonance Imaging (MRI) data and the MKL approach is used to efficiently combine them. The discriminative direction analysis highlights those features that lead the classification for each given subject. In the case of a CC with malformation this means highlighting the abnormal characteristics of the CC that guide the diagnosis. Experiments show that the method correctly identifies the malformative aspects of the CC. Moreover, it is able to identify dishomogeneus, localized or widespread abnormalities among the different features. The proposed method is therefore suitable for supporting neuroradiologists in the decision-making process, providing them not only with a suggested diagnosis, but also with a description of the pathology.

Mapping brains on grids of features for schizophrenia analysis.

This paper exploits the embedding provided by the counting grid model and proposes a framework for the classification and the analysis of brain MRI images. Each brain, encoded by a count of local features, is mapped into a window on a grid of feature distributions. Similar sample are mapped in close proximity on the grid and their commonalities in their feature distributions are reflected in the overlap of windows on the grid. Here we exploited these properties to design a novel kernel and a visualization strategy which we applied to the analysis of schizophrenic patients. Experiments report a clear improvement in classification accuracy as compared with similar methods. Moreover, our visualizations are able to highlight brain clusters and to obtain a visual interpretation of the features related to the disease.

Machine learning approaches: from theory to application in schizophrenia.

In recent years, machine learning approaches have been successfully applied for analysis of neuroimaging data, to help in the context of disease diagnosis. We provide, in this paper, an overview of recent support vector machine-based methods developed and applied in psychiatric neuroimaging for the investigation of schizophrenia. In particular, we focus on the algorithms implemented by our group, which have been applied to classify subjects affected by schizophrenia and healthy controls, comparing them in terms of accuracy results with other recently published studies. First we give a description of the basic terminology used in pattern recognition and machine learning. Then we separately summarize and explain each study, highlighting the main features that characterize each method. Finally, as an outcome of the comparison of the results obtained applying the described different techniques, conclusions are drawn in order to understand how much automatic classification approaches can be considered a useful tool in understanding the biological underpinnings of schizophrenia. We then conclude by discussing the main implications achievable by the application of these methods into clinical practice.

Free Energy Score Spaces: Using Generative Information in Discriminative Classifiers.

A score function induced by a generative model of the data can provide a feature vector of a fixed dimension for each data sample. Data samples themselves may be of differing lengths (e.g., speech segments or other sequential data), but as a score function is based on the properties of the data generation process, it produces a fixed-length vector in a highly informative space, typically referred to as "score space." Discriminative classifiers have been shown to achieve higher performances in appropriately chosen score spaces with respect to what is achievable by either the corresponding generative likelihood-based classifiers or the discriminative classifiers using standard feature extractors. In this paper, we present a novel score space that exploits the free energy associated with a generative model. The resulting free energy score space (FESS) takes into account the latent structure of the data at various levels and can be shown to lead to classification performance that at least matches the performance of the free energy classifier based on the same generative model and the same factorization of the posterior. We also show that in several typical computer vision and computational biology applications the classifiers optimized in FESS outperform the corresponding pure generative approaches, as well as a number of previous approaches combining discriminating and generative models.

Statistical 3D Shape Analysis by Local Generative Descriptors.

In this paper, we propose a new approach for surface representation. Generative models are exploited for encoding the variations of local geometric properties of 3D shapes. Surfaces are locally modeled as a stochastic process which spans a neighborhood area through a set of circular geodesic pathways, captured by a modified version of a Hidden Markov Model (HMM) named multicircular HMM (MC-HMM). The approach proposed consists of two main phases: 1) local geometric feature collection and 2) MC-HMM parameter estimation. The effectiveness of our proposal is demonstrated by several applicative scenarios, all using well-known benchmark data sets, such as multiple view registration, matching of deformable shapes, and object recognition on cluttered scenes. The results achieved are very promising and open up the use of generative models as geometric descriptors in an extensive range of applications.

A new shape diffusion descriptor for brain classification.

In this paper, we exploit spectral shape analysis techniques to detect brain morphological abnormalities. We propose a new shape descriptor able to encode morphometric properties of a brain image or region using diffusion geometry techniques based on the local Heat Kernel. Using this approach, it is possible to design a versatile signature, employed in this case to classify between normal subjects and patients affected by schizophrenia. Several diffusion strategies are assessed to verify the robustness of the proposed descriptor under different deformation variations. A dataset consisting of MRI scans from 30 patients and 30 control subjects is utilized to test the proposed approach, which achieves promising classification accuracies, up to 83.33%. This constitutes a drastic improvement in comparison with other shape description techniques.

Visual MRI: merging information visualization and non-parametric clustering techniques for MRI dataset analysis.

OBJECTIVE: This paper presents Visual MRI, an innovative tool for the magnetic resonance imaging (MRI) analysis of tumoral tissues. The main goal of the analysis is to separate each magnetic resonance image in meaningful clusters, highlighting zones which are more probably related with the cancer evolution. Such non-invasive analysis serves to address novel cancer treatments, resulting in a less destabilizing and more effective type of therapy than the chemotherapy-based ones. The advancements brought by Visual MRI are two: first, it is an integration of effective information visualization (IV) techniques into a clustering framework, which separates each MRI image in a set of informative clusters; the second improvement relies in the clustering framework itself, which is derived from a recently re-discovered non-parametric grouping strategy, i.e., the mean shift. METHODOLOGY: The proposed methodology merges visualization methods and data mining techniques, providing a computational framework that allows the physician to move effectively from the MRI image to the images displaying the derived parameter space. An unsupervised non-parametric clustering algorithm, derived from the mean shift paradigm, and called MRI-mean shift, is the novel data mining technique proposed here. The main underlying idea of such approach is that the parameter space is regarded as an empirical probability density function to estimate: the possible separate modes and their attraction basins represent separated clusters. The mean shift algorithm needs sensibility threshold values to be set, which could lead to highly different segmentation results. Usually, these values are set by hands. Here, with the MRI-mean shift algorithm, we propose a strategy based on a structured optimality criterion which faces effectively this issue, resulting in a completely unsupervised clustering framework. A linked brushing visualization technique is then used for representing clusters on the parameter space and on the MRI image, where physicians can observe further anatomical details. In order to allow the physician to easily use all the analysis and visualization tools, a visual interface has been designed and implemented, resulting in a computational framework susceptible of evaluation and testing by physicians. RESULTS: Visual MRI has been adopted by physicians in a real clinical research setting. To describe the main features of the system, some examples of usage on real cases are shown, following step by step all the actions scientists can do on an MRI image. To assess the contribution of Visual MRI given to the research setting, a validation of the clustering results in a medical sense has been carried out. CONCLUSIONS: From a general point of view, the two main objectives reached in this paper are: (1) merging information visualization and data mining approaches to support clinical research and (2) proposing an effective and fully automated clustering technique. More particularly, a new application for MRI data analysis, named Visual MRI, is proposed, aiming at improving the support of medical researchers in the context of cancer therapy; moreover, a non-parametric technique for cluster analysis, named MRI-mean shift, has been drawn. The results show the effectiveness and the efficacy of the proposed application.