Studies in neural data science. StartUp research 2017, Siena, Italy, June 25--27, 2017 (Q1626841)

The StartUp Research held 2017 in Siena, Italy, was a satellite meeting to the Statistical Conference of the Italian Statistical Society held June 2017 in Florence, Italy. A group of 28 junior researchers in statistics divided into seven groups, leaded scientifically and personally by international well-known scientists, involved themselves in the development of new statistical methods for complex and multimodal brain imaging data. The research focused on the brain imaging data provided by a pilot project of the Enhanced Nathan Kline Institute-Rockland (NKI1), (see \url{http://fcon_1000.projects.nitrc.org/indi/enhanced/}). The imaging data were pre-processed at NeuroData of John Hopkins University. The provided datasets have been analyzed from different perspectives and corresponding statistical methods have been developed. The obtained results have been presented at the StartUp Research meeting and in form of seven peer-reviewed papers have been published in the present book. The first contribution titled ``Understanding dependency patterns in structural and functional brain connectivity through fMRI and DTI data'' [ibid. 1--22 (2018; Zbl 1414.92162)] has as authors \textit{M. Crispino}, \textit{S. D'Angelo}, \textit{S. Ranciati}, and \textit{A. Mira}. The article focuses on two frameworks of multimodal brain imaging, the functional magnetic resonance imaging (fMRI) and the diffusion tensor imaging (DTI). The data consists of 24 subjects whose brain activity and structural connectivity were captured through DTI and resting state fMRI scan. The raw data have been pre-processed and the second areas of the brain have been parceled to determine a set of regions of interest, the ROIs. The aim of the work is to combine results from structural and functional observed data in order to enhance the interpretation of each separate finding and to learn possible patterns of dependencies among regions of interest of the brain. Separate statistical models for the two datasets are considered. The second section of the article is devoted to a descriptive statistics on the datasets. In the third section, the latent space model for DTI dataset and results on the DTI dataset are presented. Time-varying dynamic Bayesian networks for the fMRI dataset and results on the fMRI dataset are shown in the forth section. The paper ends with some comments. The second contribution titled ``Hierarchical graphical model for learning functional network determinants'' [ibid. 23--36 (2018; Zbl 1414.92152)] has as authors \textit{E. Aliverti}, \textit{L. Forastiere}, \textit{T. Padellini}, \textit{S. Paganin} and \textit{E. Wit}. The goal of the work is to relate functional connectivity patterns with subject specific features and brain constraints and to study the way the phenotypes affect the neurophysiological dynamics. A modular approach is considered. As a measure of brain activity, functional magnetic resonance imaging is applied. An hierarchical model is defined. The modularization procedure consists in decomposing it in three sub-models: a smoothing procedure to remove noise from the fMRI signal, a graphical model to encode functional brain connectivity and a regression model to investigate the relation between phenotypes and functional connectivity patterns. The hierarchical model is introduced in the second section and the third section is devoted to the modular estimation using connectome data. The third contribution titled ``Three testing perspectives on connectome data'' [ibid. 37--55 (2018; Zbl 1414.92158)] has as authors \textit{A. Cabassi}, \textit{A. Casa}, \textit{M. Fontana}, \textit{M. Russo} and \textit{A. Farcomeni}. This paper focuses on three different aspects of the analysis of MRI scans. One develops a bootstrap-based inferential tool to test if the functional connectivity among different brain areas corresponds to their structural connectivity and anatomical characteristics. Further, one introduces a Bayesian framework to estimate the fiber count number obtained from DTI data. In the last section, an object-oriented nonparametric test for the quality of two or more groups of functional networks derived from the functional magnetic resonance imaging data is presented. Results on different tests are shown. The forth contribution titled ``An object oriented approach to multimodal imaging data in neuroscience'' [ibid. 57--73 (2018; Zbl 1414.92160)] has as authors \textit{A. Cappozzo}, \textit{F. Ferraccioli}, \textit{M. Stefanucci} and \textit{P. Secchi}. The article focuses on several techniques to study complex multimodal imaging data in neuroscience with the aim to identify a data driven group structure in the patient samples. One provides a large set of appropriate procedures including clustering procedures, low-dimensional representation and hypothesis testing for correlation structures. Test results are shown and analyzed. The paper ends with conclusions and future research directions. The fifth contribution titled ``Curve clustering for brain functional activity and synchronization'' [ibid. 75--90 (2018; Zbl 1414.92156)] has as authors \textit{G. Bertarelli}, \textit{A. Corbella}, \textit{J. Di Iorio}, \textit{A. Gorshechnikova} and \textit{M. Scott}. This work focuses on the problem of identifying brain areas with similar behavior in the time domain by means of functional curve clustering methods. After a discussion on data selection, in the frame of the developed methodology, the \(k\)-means clustering of fMRI time series , the smoothing procedure and the functional boxplot method are described. Implementation results are largely discussed. The sixth contribution titled ``Robust methods for detecting spontaneous activations in fMRI data'' [ibid. 91--110 (2018; Zbl 1414.92168)] has as authors \textit{F. Gasperoni} and \textit{A. Luati}. The aim here is to identify spontaneous activations in resting state fMRI time series and to estimate the hemodynamic response function (HRF) at the ROI level. Two methods are considered, one is based on classical Gaussian assumption for the data generation process of fMRI data and the other is based on the assumption that the data may be generated by a heavy tailed distribution. One estimates and compares the HRF obtained from the two methods and tests the results on two patients and four ROIs. By this approach, groups of ROIs with the highest similarity can be studied and detected. The seventh contribution titled ``Hierarchical spatio-temporal modeling of resting state fMRI data.'' [ibid. 111--130 (2018; Zbl 1414.92159)] has as authors \textit{A. Caponera}, \textit{F. Denti}, \textit{T. Rigon}, \textit{A. Sottosanti} and \textit{A. Gelfand}. In this article, a spatio-temporal Bayesian factor model for the analysis of RS-fMRI data is introduced. One shows how to obtain posterior inference using an M-H algorithm. The developed model is then applied to a real RS-fMRI dataset and results are largely discussed. The eighth article titled ``Challenges in the analysis of neuroscience data'' written by \textit{M. Guindani} and \textit{M. Vannucci} [ibid. 131--156 (2018; Zbl 1414.92170)] analyses and summarizes different approaches, methods, results and proposals developed in this book and provides an outlook over trends and new research directions in the analyses of brain imaging data. An excellent book! Congratulations for the way research has been done!