Molecular Imaging: a Hands-on Tutorial based on Open Access Datasets

Arianna Sala Organizer
Université De Liège
Université De Liège
Liege, Sart-Tilman 
Igor Yakushev Co Organizer
Dept. of Nuclear Medicine, Technical University of Munich
Munich, Bavaria 
Sunday, Jun 23: 1:30 PM - 5:30 PM
Educational Course - Half Day (4 hours) 
Room: Grand Ballroom 105 
Despite an established role of molecular neuroimaging in the diagnosis of neurodegenerative, inflammatory disorders, and epilepsy, it is only recently that the broader neuroimaging community has started integrating molecular features into research on healthy brain function. The increasing interest in molecular imaging by the neuroscientific community is testified by a growing number of multimodal studies, integrating molecular features with structural and functional connectomics, and by the development of methods (conveniently translated into easy-to-use toolboxes like e.g. neuromaps, juspace) aimed at facilitating multimodal integration.
Still, most of these efforts rely on molecular features as derived from group-averaged templates and are therefore limited in the research questions they can answer, currently articulated mostly in terms of colocalization of (average) molecular and MRI-derived measures. In order to expand such multimodal efforts, i.e. moving towards a framework that allows for mechanistic and causal inference, future studies will necessarily have to rely on multimodal molecular and nonmolecular data at the individual level, rather than in the form of group-averaged templates.
While not all neuroimaging centers have the chance to acquire molecular data, a growing adherence to open science practices is making more and more open-access or easily accessible multimodal datasets (including molecular imaging data) available to the broader neuroscientific community. Currently, several thousands of multimodal data can be easily accessed by any neuroimager. Still, handling molecular data and interpreting molecular findings correctly, requires a basic understanding of principles of molecular imaging.
This Educational Course aims to provide a basic introduction to molecular imaging, covering its basic principles and main tools, illustrating similarities and critical differences in the main molecular measures commonly encountered in open access/easily accessible molecular datasets. Different hands-on, interactive sessions will allow the attendees to get first-hand experience on data access, processing and modelling of different types of molecular imaging data.


The attendees will be able to:
1) understand the basic principles of brain molecular imaging
2) discover how to access, process and model different types of brain molecular imaging data
3) critically distinguish between different types of brain molecular imaging measures 

Target Audience

The target audience for this educational course are researchers in the field of neuroimaging of all levels of expertise, wishing to expand their knowledge and know-how on brain function to include “hot” techniques, beyond the most commonly used MR and EEG. 


1. What is molecular imaging?

Molecular imaging is becoming increasingly popular in the research community and in clinical settings. In my talk I will first explain the principles of molecular imaging, the approach of tracing different biological processes in the living human brain. Afterwards, I will give an overview of established molecular targets, from perfusion to oxygen and glucose metabolism, pre- and post-synaptic neurotransmission as well as proteinopathies and neuroinflammation. Of note, I will position molecular imaging among other neurophysiological techniques, in terms of spatial, temporal resolution, validity, and reproducibility. I will also point out limitations of molecular imaging. To apply molecular imaging efficiently, knowledge on basics, strengths and limitations of this approach is essential. 


Arianna Sala, Université De Liège
Université De Liège
Liege, Sart-Tilman 

2. Relative and absolute measures in molecular imaging: an overview

PET analysis involves both semi-quantitative and fully quantitative assessment of measured radioactivity concentrations.
For data generated in clinical settings, the widely used standard uptake value (SUV) and its ratio version (SUVr) are simple semi-quantitative indices derived from the ratio of PET measurements and injected dose normalized by subject characteristics and, for SUVr, by a reference region. However, SUV's specific biological interpretation is limited due to various factors, encompassing both physiological and technical aspects.
Transitioning from static to dynamic PET data, most commonly generated in specialized research settings, essential mathematical models are required to extract additional information. Each of these models provides distinct biological significance and reliability, utilizing different units of measurement, rendering them non-comparable and non-interchangeable. For instance, when analyzing PET data related to radioligand binding to protein targets, it becomes possible to quantify various aspects such as the total concentration of receptors (Bmax, [pmol/mg]), binding potential (BPF, [ml/cm3]) as a product of Bmax and tracer affinity, and the total volume of distribution (VT, [ml/cm3]) calculated as the ratio of radioligand concentration in the tissue to that in plasma. Each of these measures holds a unique biological meaning and cannot be readily substituted.
This presentation aims to deliver a comprehensive overview of semi-quantitative and quantitative PET measures, critically evaluating their respective biological meanings and usability in a multimodal research environment.


Alessandra Bertoldo, University of Padova
Padova Neuroscience Center
Padova, Italy 

3. Open access databases in molecular imaging (or how to access more than 30.000 PET scans!)

Notwithstanding an increasing interest in integrating molecular information into brain mapping studies, the literature based on molecular imaging remains relatively limited. One reason is a relative scarcity of available molecular imaging data. Ionizing radiation and high costs limit its applicability, especially, in healthy subjects. At present, however, a growing adherence to open science practices is making more and more molecular imaging data available to the broader neuroscientific community.
In the first part of the talk, I will give an overview of openly or easily accessible molecular imaging datasets. I will explain how to find and access more than 60 datasets including brain molecular imaging data (either cross-sectional, longitudinal or multimodal) of more than 30,000 subjects. Data are available for a broad range of biological targets (metabolism, neurotransmission, pathology), in healthy subjects of all ages and patients with >20 different clinical conditions. Most datasets are multimodal with 85% of datasets also including magnetic resonance (MR) data in the same subjects. In the second part of the talk, I will give practical examples on how to access these datasets, how to practically deal with different data formats (dicom, ecat, nifti), data acquisition types (static, dynamic) and level of processing of the data (raw, pre-processed, quantified). At the end of the talk, the attendees will have obtained essential knowledge on how to identify, access, download and correctly manage molecular imaging data from the major available datasets. 


Martin Nørgaard, PhD, University of Copenhagen
Computer Science
Copenhagen, NA 

4. PET imaging of Glucose Metabolism: data access, pre-processing, and quantification using SPM and MATLAB

In this session, I will showcase the procedures for processing and quantifying Fluorodeoxyglucose Positron Emission Tomography ([18F]FDG-PET) data. To begin with, I'll elaborate on the considerations and quality controls essential when obtaining data from open-access repositories. Following that, we'll use the data from a single subject to demonstrate the steps involved in quality control, registration, spatial normalization, and aspects related to global signal normalization. These tasks will be executed using both GUI and scripts in statistical parametric mapping (SPM) within the MATLAB environment. Later on, I will explore the application of a multi-subject dataset to elucidate the computation of simple connectivity measures using both the seed-based correlation approach and the regions of interest approach.  


Xin Di, New Jersey Institute of Technology Newark, NJ 
United States

5. PET Imaging of Neurotransmitter Systems: access and pre-processing

Positron Emission Tomography (PET) is a powerful imaging technique to measure biological targets of interest in the living human brain. In particular, the interest in dynamic PET data focusing on neurotransmitter systems is rapidly increasing with the generation and release of freely available neuroreceptor maps from various receptor systems (GABA, dopamine, serotonin etc.) to the neuroimaging community in both healthy and diseased conditions, and/or during pharmacological interventions.
In this first part of the PET Imaging of Neurotransmitter Systems tutorial, I will go through how these neuroreceptor maps can be generated (focusing on the serotonin transporter using the radiotracer [11C]DASB), ranging from download of open data (from OpenNeuro) and preprocessing such as motion correction, registration, and segmentation. I will also cover the newest developments in open-source software for accessing and preprocessing of PET data (e.g. PETSurfer and PETPrep), ultimately providing the attendee with all the necessary skills to develop their own PET pipeline, generate neuroreceptor maps, and share these valuable resources back to the community.


Martin Nørgaard, PhD, University of Copenhagen
Computer Science
Copenhagen, NA 

6. PET Imaging of Neurotransmitter Systems: quantification

After preprocessing, PET data consists of time series of radioactivity concentrations in each region of the brain called time activity curves (TACs). Full quantification of PET data involves fitting pharmacokinetic models to these TACs in each target region, or voxel, in order to estimate scalar outcome measures, e.g. binding potential, which serve as proxies for the concentrations of neurotransmitter receptors and transporters. In this tutorial, I will outline some of the relevant practical considerations for PET quantification such as choosing an appropriate model and outcome measure, as well as providing a hands-on demonstration of how these models can be fitted to TAC data to derive estimates of binding using open-source tools including bloodstream and kinfitr. I will also discuss examples of how these modeled PET outcomes can be combined with data from functional magnetic resonance imaging (fMRI), electroencephalography (EEG) and diffusion weighted imaging (DWI) to facilitate a multimodal view of brain structure and function. Altogether this will provide attendees with the basics required to be able to begin to perform all stages of PET analysis from images, to estimates of receptor and transporter concentrations, to their integration with other brain imaging data. 


Granville Matheson, Columbia University / Karolinska Institutet Solna, Stockholms län