2019 School on Brain Connectomics

An introduction to modeling in Diffusion MRI: Tensors, Propagators, and Compartments

Diffusion MRI acquisitions sample the Fourier space in a given spatial frequency and direction in order to get data on the water molecules displacement profile. In order to transform diffusion MRI data into useful information the most common approach is to fit a continuous parametric model to the discrete samples. The main idea behind this approach is that the model parameters provide a more accurate and complete description of the diffusion process and the underlying neural tissues architecture that generate it compared to the raw data. Across the years, several models have been proposed by the researchers in order to capture more information from diffusion MRI acquisition compared to the state of the arts of the time. From Diffusion Tensor Imaging to Mean Apparent Propagator models, Ball & Stick to CHARMED and NODDI. Some of these models like Spherical Deconvolution are designed to be a better starting point for tractography. Others focus on trying to estimate microstructural features such as axonal diameter and neurite density. The goal of this presentation is to present the most popular Diffusion MRI models helping the listener to understand the motivations that drove researchers to design such models, their strong points and their limits.

The elusive goal of extracting imaging biomarkers from diffusion MRI: Historical overview and current challenges

A popular definition of biomarker is: “a characteristic that is objectively measured and evaluated as an indicator of normal biological processes, pathogenic processes, or pharmacologic responses to a therapeutic intervention.(2001, Biomarkers Definition Working Group)". The diagnostic process heavily relies on the evaluation of medical signs, and biomarkers can be considered objective, quantifiable, and reproducible medical signs. Biomarkers that are validated to serve as surrogate clinical outcomes in treatment trials are very valuable and are also ideally suited for artificial intelligence-based data analysis. Magnetic resonance imaging (MRI) plays an important role in diagnostic radiology, but most of the clinically relevant information is obtained from a subjective interpretation of images, not from quantitative and biologically validated MRI metrics. Since the early days of diffusion tensor MRI, more than 20 years ago, researchers and clinicians have worked with enthusiasm in attempting to extract quantitative information form diffusion MRI signals. In this talk, we would review the efforts of developing diffusion MRI based biomarkers in the domains of stroke, white matter disorders, degenerative disorders, development and aging. We will also discuss the practical and theoretical challenges that we need to address to transition diffusion MRI from a research technique to a reliable and informative clinical tool.

Brain computer interfaces for the industrial application of cognitive neuroscience

Are the cognitive neuroscience ready to be used in advanced industrial contexts? In this talk it will be depicted a possible path for the use of advanced findings in cognitive neuroscience by using the electroencephalogram not only in the medical environment (e.g. to improve the limb's rehabilitation path for patients affected by stroke). In particular, applications of advanced EEG signal processing technique will be illustrated in the marketing context (neuromarketing) as well as in the aerospace-aeronautic environment, through the on-line monitoring of the mental workload of pilots, air traffic controllers and other category of professional drivers during their actual operations. Four main areas will be described:

1. Brain Computer Interface. In this part of the talk different applications of the brain computer interfaces (BCI) technology will be first presented. All the applications will be obtained by using the computerized analysis of the electrical activity of the human brain, gathered by a net of electrodes placed on the scalp surface (electroencephalogram, EEG). It will be described how by using the voluntary modulation of EEG activity normal subjects could control external devices such as a cursor on the screen, a mobile robot as well as a wheelchair. Successively, it will be illustrated how the BCI technology could be inserted within the rehabilitation path of the patients suffered of brain strokes. In particular, it will be showed how BCI technologies could enhance the rehabilitation exercise, by including the presentation of the attempt of the movement as early as possible to the patients, although they were not yet able to move their limbs. However, BCI applications could be extended beyond the use in the clinical context, and application in the area of the "synthetic telepathy" are already investigated by DARPA and by the present research group.
2. Neuromarketing: Application of neuroscience in the evaluation of relevant marketing stimuli will be described. The main cognitive neuroscience indicators for the appreciation of an audiovisual sensory stimuli (e.g. a TV commercials) will be also described. The use of such EEG-based indicators in practical situation will be also illustrated.
3. Online detection of mental workload: Successively, it will be showed different applications of the collection of brain activity in working contexts related to the aircraft’s pilots. It will be described as it is possible to detect the brain activity related to the insurgence of mental workload. It will be speculated that such detection could be employed in a short future to generate devices able to warn the operators about their perceived workload. Example of such detection of mental workload will be presented in three different conditions: on civil airline pilots, on military pilots and on car drivers.
4. Neuroaestethic: The issue of how we perceive the beauty will be also addressed by the talk, through the presentation of main results obtained by monitoring the EEG activity during the visit of two art galleries with the pictures of Tiziano Vecellio and Jan Veermer.

Conclusions: The possibility to detect in a reliable way the cerebral activity during "real-life" conditions and the possibility to detect brain activity with dry electrodes will be also discussed. Quoting the scientist Martha Farah, the issue is not "if" but "when" the neuroscience will shape our future. We are thinking that such time is arrived.

Hands-on: A Practical Introduction to diffusion MRI model fitting

In this hands-on, the students of the school will start exploring the world of diffusion signal modeling. In this practical session, we will focus on two of the most popular diffusion MRI models Diffusion Tensor Imaging (DTI) and Spherical Deconvolution (SD). In the first part, the students will be able to write their own Python code for DTI fitting, the calculation of its indices (Fractional Anisotropy, Mean Diffusivity) and their visualization. In the second part, the students will calculate the fiber Orientation Distribution Function (fODF) on real data using the Diffusion Imaging with Python (DIPY) library.

Brain Morphometry using diffusion MRI data

Tensor-based morphometry (TBM) performed using T1-weighted images (T1WIs) is a well-established method for analyzing local morphological changes occurring in the brain due to normal aging and disease. However, in white matter regions that appear homogeneous on T1WIs, T1W-TBM may be inadequate for detecting changes that affect specific pathways. In these regions, diffusion tensor MRI (DTI) can identify white matter pathways on the basis of their different anisotropy and orientation. In the last couple of years, our lab has investigated the possibility of performing TBM using deformation fields constructed using all scalar and directional information provided by the diffusion tensor (DTBM) with the goal of increasing sensitivity in detecting morphological abnormalities of specific white matter pathways. In this talk I will present results that indicate that DTBM could be a powerful tool for detecting morphological changes of specific white matter pathways in normal brain development and neurological disorders.

Diffusion MRI data harmonization: experience from the international MICCAI Challenges and future directions. Can we learn anything new?

Diffusion MRI provides indices sensitive to biological properties of the imaged tissues, designed to be quantitative and scanner-independent. However, in practical terms these metrics often show large between-scanner variations due to differences in MRI pulse sequences, reconstruction algorithms, noise characteristics, and other factors that vary across scanners. MRI harmonisation techniques aim to mitigate these differences and make MRI scans/metrics obtained with different machines as comparable as possible. However, until recently benchmark data sets and systematic comparisons of harmonisation algorithms in diffusion MRI have been lacking. The 2017 and 2018 Challenges on Diffusion MRI data Harmonisation were organised as part of the MICCAI workshop "Computational Diffusion MRI", with the aim of providing the community with a unique benchmark data base of "travelling heads", i.e. a cohort of volunteers scanned with different protocols and scanners, supporting future research in diffusion MRI harmonisation. This talk will describe the experience of organising the two challenges, reporting on the strategy followed to design the experiments, the issues encountered during the organisation, the main results of the two Challenges and potential future directions in diffusion MRI harmonisation.

Overview of data acquisition and pre-processing factors that affect brain diffusion MRI quality

Diffusion-weighted magnetic resonance imaging (dMRI) measures water mobility properties that allow the characterization of brain tissue microstructure and orientations in a wide variety of neuroscience and clinical studies (human/animal, in-vivo/ex-vivo, health/disease). A generic dMRI study, after its planning, typically includes four distinct crucial steps in the following order: (i) data acquisition, (ii) data pre-processing to remove unwanted sources of noise/artifacts, (iii) data modelling for specific diffusion metrics derivation, and (iv) statistical analysis on the derived metrics. The quality of each step depends on the quality of the previous one, making a pyramid-like structure for data quality considerations. Obtaining good-quality dMRI data is challenging due to the large number of factors involved. The goal of this lecture is to focus on the first two steps, acquisition and pre-processing, providing an overview of the various factors that can affect dMRI data quality.

Computational Methods for Neuroanatomical Bundle Segmentation

Tractography of dMRI data provides a virtual representation of the paths of hundreds of thousands of fibers, known as streamlines, that compose the white matter and connect different areas of the cortical gray matter. White matter bundle segmentation aims to virtually group together streamlines that have an analogous shape, share a similar pathway and connect the same anatomical brain regions into anatomically meaningful structures, called bundles. Segmenting anatomical structures in the white matter of the human brain is useful in many different applications, such as surgical planning, population studies, and diagnosis or monitoring of brain diseases. The presentation provides an overview of the main computational methods for neuroanatomical bundles segmentation.

Diffusion MRI: Microstructure modeling and multidimensional encoding

Diffusion MRI offers a unique way to study the microstructure of living tissue in vivo in a non-invasive manner. Microstructure features such as cell sizes, shapes and orientations all leave their mark on the water diffusion. By encoding information about the diffusion process, these microstructure properties can be inferred from the MRI data. However, diffusion MRI performed with pulsed gradients puts a limit on what type of information that can be retrieved from the imaging data. Recently result have demonstrated that extended information can be retrieved by using more advanced encoding such as b-tensor encoding. In this talk, I will cover the basics of how diffusion in a simple geometry is related to the observed signal, and use this as the foundation for more advanced so-called compartment models. I will describe requirements on the encoding for accurate estimation of compartment information from diffusion MRI, and illustrate where there are still experimental limitations.

What can we learn from the diffusion MRI signal?

The diffusion MRI signal is sensitive to the incoherent motion of water molecules at a scale similar to cellular structures, giving it the potential to report on tissue microstructure and its orientation. Of note, this dependence on the orientation of the tissue forms the basis of tractography: if the orientation of brain white matter can be estimated at any location, this can be used to delineate the trajectory of long-range white matter pathways in the brain. Over the past 2 decades, many models have been proposed to characterise the diffusion signal and extract meaningful information from it. These models fall into two main camps: those that attempt to estimate characteristics of the diffusion process (e.g. diffusion tensor imaging, diffusion spectrum imaging, Q-Ball imaging, ...), and those that attempt to characterise some aspect of microstructure directly (ball and sticks, CHARMED, axCaliber, spherical deconvolution, ...). Furthermore, some of these methods aim to recover features of the microstructure of the tissue, irrespective of its orientation, while others focus of the estimation of fibre orientations. Finally, each method comes with its own data acquisition requirements (b-values, number of diffusion-weighted directions, etc), with multi-shell high angular resolution sequences becoming increasingly popular. All of these different considerations have implications for the nature and quality of the information that can be extracted from diffusion MRI data, with obvious knock-on effects for tractography and connectomics.

MRtrix3: A fast, flexible and open software framework for medical image processing and visualisation

MRtrix3 provides a set of tools to perform various types of diffusion MRI analyses, from various forms of tractography through to next-generation group-level analyses. It is designed with consistency, performance, and stability in mind, and is freely available under an open-source license. It is developed and maintained by a team of experts in the field, fostering an active community of users from diverse backgrounds. In this hands-on session, we will explore the basic functionality offered within MRtrix3 using the data provided in the BATMAN tutorial.

Seeding connectivity - cortical parcellations

In this talk I will be presenting cortical parcellation schemes and their use to infer cortical brain connectivity. The aim is to provide a overview of existing parcellation schemes that can be used in the context of connectivity inference, and to understand the fundamental and practical aspects of building a cortical parcellation. I will first introduce the motivations and the general concept of cortical parcellation. I will then discuss what can be expected from a cortical parcellation in terms of properties (e.g. resolution or geometry) and general relations with function, connectivity, of microstructure. We will then move on to the different types of information that can be used to parcellate the cortex and the different classes of atlases they lead to. Existing parcellation techniques and associated softwares will then be reviewed.

Decoding brain activity and connectivity from functional neuroimaging

In the last decades, the study of brain functional connectivity (FC) has become an increasingly active field of research providing novel and crucial insights into brain function, in resting-state as well as during tasks. FC can be derived from different non-invasive magnetic resonance imaging (MRI) modalities, such as fMRI based on the BOLD contrast or Arterial Spin Labeling (ASL). Classical connectivity analyses allow detecting regions with similar behaviours or coherent signals, indicating their membership to the same functional network. In particular, FC can be computed in both time and frequency domains exploring different approaches such as seed-based correlation, independent component analysis, spectral coherence and many others. More recently, new approaches have been proposed for extracting significant aspects of the network and quantitatively characterising its global organization. In this lecture, I will present an overview of the main non-invasive functional MRI modalities and their applications for task and resting-state paradigms. Moreover, I will illustrate the associated methods of analysis for both brain localization and connectivity estimation, giving a glimpse of their main advantages/disadvantages.

Forward and inverse problems in EEG and MEG

This course will cover the forward and inverse problems of EEG and MEG. After explaining what is a "gain matrix" which relates the source activity to the sensor measurements, we will study the ill-posedness of the inverse source reconstruction problem, and we will discuss the regularization strategies which are used to solve this problem. EEG and MEG will be compared, and basic functional connectivity measures will also be discussed.

Machine Learning for Brain Computer Interfaces

This course will introduce the main data processing methods allowing to decode brain activity in real time and convert it to a control signal for Brain Computer Interfaces. Spatial filtering, and feature classification are the main components of this processing chain. We will also discuss several important BCI paradigms (based on mental imagery, or on evoked-potentials) and their potential applications.

Hands on: Resting-state functional MRI: from pre-processing to functional brain connectivity

There are several toolboxes currently available for analysing resting-state fMRI data, in particular for performing data pre-processing and deriving the most common functional connectivity measures. In this hands-on session, we will explore the main functionalities offered by FSL, a comprehensive library of analysis tools for different neuroimaging data (e.g., dMRI, fMRI, and MRI). In particular, we will focus on resting-state fMRI aiming at:

1. understanding and visualising the 4D-BOLD signals;
2. performing some basic and advanced pipelines based on ICA for data pre-processing;
3. deriving connectivity measures at both single-subject and group level;
4. understanding how these measures can be compared across groups.

What functional connectivity is good for? A theory of the function of spontaneous activity

Noise in the brain is not random but correlated. Its function is unknown. fMRI studies of ongoing resting activity have shown that brain noise is correlated in large scale distributed spatiotemporal patterns called resting state networks (RSNs). The function of RSNs has been widely debated, but their functional significance is unknown. I will present a series of experiments that indicate that spontaneous activity as measured with fMRI RSN has a representational function, namely to code for information states of high behavioral relevance. We think that these patterns are low dimensional as they represent principal components of behavior in space and time.

Brain functional connectivity inference: models, definitions, perspectives and pitfalls

The aim of this lecture is to introduce the multifaceted concept of brain connectivity by providing an overview of different definitions (synchronicity, causality, influence) that are at the basis of the approaches currently available to estimate brain connectivity from electrophysiological data. Different approaches will be reviewed, ranging from model­free to model­based data-driven, to biologically inspired models, describing the basic concepts and the advantages and limitations of different methods when applied to neuroelectrical data. Time, frequency and time-resolved analysis at different time scales will be discussed, together with the important issue of the statistical assessment of connectivity patterns. Learning objectives will include: (i) understanding the basics and principles of the estimation of brain connectivity, with particular focus on the importance of different definitions and concepts; (ii) understanding the main advantages and pitfalls resulting from different approaches; (iii) understanding which approaches are currently available for the estimation of time-varying connectivity to capture the dynamics of brain processes and for the statistical assessment of connectivity patterns.

Graph theoretical applications to structural and functional connectivity

This lecture will introduce the fundamentals of the application of graph theory to functional and structural brain networks. We will discuss several graph metrics like density, degree, strength, clustering, path length, modularity and metrics of centrality, and discuss how together these metrics define several network architectures

Hands on: Brain Connectivity and Graphs

Part 1. Getting started

• MATLAB basics, set the MATLAB path, initiate and save a new MATLAB script.

Part 2. Gain confidence with brain connectivity data

• load and visualize connectivity matrices and their weight distribution
• understand the structure of a connectivity matrix (symmetries, diagonals, etc.)
• binarize connectivity matrices and create a group-representative network

Part 3. Network measures of integration and segregation: is the brain small-world?

• compute basic integration and segregation measures: shortest path, efficiency, clustering coefficient and small-world index

Part 4. Brain hubs and network visualization

• plot the degree sequence and the strength sequence and identify brain network hubs
• visualize a brain connectivity network

Part 5. Brain network alterations in schizophrenia

• discuss the topic of weights normalization and group comparison
• compute global and nodal measures for the two groups of subjects (healthy controls and schizophrenia patients)
• statistical testing and group-comparison: detect group-differences and discuss statistical issues and multiple comparison correction