3131

Automatic quality assessment of high-resolution T2-weighted images used in hippocampus volumetry and functional studies – a tool developed as part of DZNE-DELCODE study.
Arturo Cardenas-Blanco1,2, Yi Chen3, Jose Pedro Valdes-Herrera4, Laura Dobisch1, Renat Yakupov1, Klaus Fliessbach5,6, Michael Wagner5,6, Annika Spottke6, Stefan Teipel7,8, Katharina Buerger9,10, Anja Schneider5,6, Oliver Peters11,12, Peter Nestor1, Josef Priller11,12, Jens Wiltfang13,14, Christoph Laske15,16, Frank Jessen6,17, and Emrah Duezel1,3,18

1German Center for Neurodegenerative Diseases (DZNE), Magdeburg, Germany, 2IKND, Magdeburg, Germany, 3Institute of cognitive neurology and dementia research, Magdeburg, Germany, 4Aging & Cognition Research Group, German Center for Neurodegenerative Diseases (DZNE), Magdeburg, Germany, 5Department of Psychiatry, University Hospital Bonn, Bonn, Germany, 6German Center for Neurodegenerative Diseases (DZNE), Bonn, Germany, 7German Center for Neurodegenerative Diseases (DZNE), Rostock, Germany, 8Department of Psychosomatic Medicine, University Medicine Rostock, Rostock, Germany, 9German Center for Neurodegenerative Diseases (DZNE), Munich, Germany, 10Institute for Stroke and Dementia Research, Ludwig-Maximillian-Universitaets, Munich, Germany, 11Department of Psychiatry, Charité-Universitätsmedizin Berlin, Berlin, Germany, 12German Center for Neurodegenerative Diseases (DZNE), Berlin, Germany, 13Department of Psychiatry and Psychotherapy, University Medical Center Göttingen, Goettingen, Germany, 14German Center for Neurodegenerative Diseases (DZNE), Goettingen, Germany, 15Department of Psychiatry and Psychotherapy, Eberhard Karls University, Tuebingen, Germany, 16Aging & Cognition Research Group, German Center for Neurodegenerative Diseases (DZNE), Tuebingen, Germany, 17Department of Psychiatry, University of Cologne, Cologne, Germany, 18Institute of Cognitive Neuroscience, University College London, London, United Kingdom

Synopsis

This abstract presents a processing pipeline developed to automatically assess the quality of specific structural T2-weighted images typically acquired in the study of the hippocampus. By combining existing neuroimaging tools, the presented pipeline generates descriptive information about the signal properties in different tissue classes of the T2-weighted image. This information could subsequently be used to detect sub-optimal volumes due to noise or motion artifacts. Similarly as it measures the angulation of the T2-weighted slices with respect to the HC, it could also be used to automatically determine whether the field of view angulation follows protocol.

Introduction

During the last years there has been a growing interest in understanding the impact of neurodegenerative diseases in the hippocampus (HC) – by following volume changes [1] or understanding its role in different cognitive processes [2]. Due to its particular shape, this analysis is typically carried out by acquiring a high-resolution T2-weighted (T2-w) sequence oriented perpendicularly to the HC’s longitudinal axis. In small, single-site studies, assessing the quality of structural images represents a challenge itself, in which motion artifacts and angulation are the main sources of problem. Obviously, the challenge increases with the number of subjects and becomes of critical importance in large – multisite studies. Recently, different initiatives have developed several strategies to automatically assess the quality of different MR images. [3, 4]. In this study, we present an automatic workflow developed to assess the quality of high resolution T2-w images acquired for HC segmentation as part of the DZNE-DELCODE study [5].

Methodology

Data corresponding to 338 subjects, part of the initial release of DELCODE, has been analysed using this pipeline - Figure 1 summarises the cohort's demographic details. Developed in python, making use of different neuroimaging packages ( fsl, ants and spm)[6, 7, 8] and integrated using Nipype [9], the proposed pipeline aims to automatically assess the quality of high resolution T2-w images. Taking structural a structural whole brain T1 weighted (T1-w) images and high in-plane resolution partial brain T2-w images as an input, the data goes through an analysis pipeline which interconnect so called “workflows” in which different data analysis processes take place (Figure 2). The processing pipeline is divided into three different workflows. The first one, “SkullStripCoreg” segments the T1-w image using spm and coregisters it to the T2-w images. The second workflow – “Workflow_quality”, brings the GM, WM, CSF and background tissue probability maps obtained in the first workflow into T2 space. Descriptive metrics for each of these signals are then extracted and stored into a csv file using pandas. The third workflow “r2First_WF” segments the HC from the T1-w images using fsl FIRST and estimate the angulation of the T2-w sequence with respect to the main longitudinal axis of the HC. The presence of motion and artifacts is assessed by analysing the signal distribution in the gray matter (GM), white matter (WM) and backgroung (BG) regions. The angulation of the T2-w images to the HC is assessed by measuring the angle between a vector normal to the T2-w slices (information obtained from the dicom header) and the HC's first eigenvector obtained after running principal components analysis of the voxels that belong to each of the HC masks. The signal to noise ratio (SNR) and the contrast to noise ratio (CNR) are estimated in the T2-w space as follows.

$$SNR=(mean(GM)/std(BG))$$

$$CNR=(mean(GM)-mean(WM))/std(BG)$$


Results

Figure 3 shows the SNR and CNR distribution values, for all subjects, across sites. Even though we expect to have signal variation across sites, the information shown in the figure can be used to identify outliers in each of the sites and flag them for meticulous visual inspection. Figure 4 shows the orientation information of the T2-w sequence with respect to the left and right hippocampi. Ideally, we would expect to find our values distributed around 0 degrees. Nevertheless, the main longitudinal axis of the hippocampus is not contained in a true sagittal plane, its tail is closer to the brain’s midline that its head. We believe this is causing a bias in the orientation measurement which requires further investigation. An ANOVA statistical test was used to assess whether the different in angulation of the T2-weighted slices is different across diagnostic groups. Results were negative for both, left and right HC with p=0.75, f=0.53 and p=0.70, f =0.54 respectively.

Discussion

This abstract presents a processing pipeline developed to automatically assess the quality of specific structural T2-w images acquired in the study of the HC as part of the DZNE-DELCODE study. By combining existing neuroimaging tools, the presented pipeline generates descriptive information about the signal properties in different tissue classess of the T2-w image which ultimately can be used to detect sub-optimal volumes due to noise or motion artifacts. Similarly it measures the angulation of the T2-w slices with respect to the structure of interest, the HC. The fact that no significant differences in orientation of the T2-w slices across the groups were found could suggest that this method to determine the HC's longitudinal axes is robust versus atrophy, given that no differences were found between AD and the other diagnostic groups.

Acknowledgements

We gratefully thank all the participants in this research study together with our radiographers for their dedicated efforts in scanning standardisation.

References

1. Schuff N., et al. MRI of hippocampal volume loss in early Alzheimer's disease in relation to ApoE genotype and biomarkers. Brain - 2009

2. Maas A., Laminar activity in the hippocampus and entorhinal cortex related to novelty and episodic encoding. Nature Communications - 2014

3. Shehzad Z., et al.The Preprocessed Connectomes Project Quality Assessment Protocol - a resource for measuring the quality of MRI data. Frontiers in Neuroscience - 2015.

4. Alfaro-Almagro F., et al. Image processing and Quality Control for the first 10,000 brain imaging datasets from UK Biobank. Neuroimage - 2017

5. Jessen, F., et al. Design and first baseline data of the DZNE multicenter observational study on predementia Alzheimer‘s Disease (DELCODE). Alzheimer‘s Research & Therapy (in press)

6. Patenaude B., et al. A Bayesian Model of Shape and Appearance for Subcortical Brain. NeuroImage - 2011.

7. Avants B., et al. A reproducible evaluation of ANTs similarity metric performance in brain image registration. Neuroimage - 2011

8. Ashburner, J., et al. Unified segmentation. NeuroImage - 2005.

9. Gorgolewski K., et al. Nipype: a flexible, lightweight and extensible neuroimaging data processing framework in Python. Frontiers in Neuroinformatics - 2011

Figures

Top left- Sex and age distribution of the subjects analysed in this study. Top right- Number of subjects per diagnostic group. AD alzheimers, HC: Controls, MCI: Mild cognitive impairment, Relative: AD subject’s relative, SCD: Subjective cognitive decline. Bottom- Subject distribution according to site and diagnosis

Overview of the processing pipeline presented in this abstract. The data analysis process is divided in three main “workflows”. 1- ”SkullStripCoreg”: combining tools from spm and ants, the algorithm segments the T1weighted image into different tissue classes, e.g. WM, GM, CSF, background and coregisters the T1 and the T2 weighted images. 2 – “r2First_WF”: using fsl first it segments the HC and estimates the angles between the T2 weighted images and the HC’s longitudinal axes. 3 – “workflow_quality”: using ants tools extracts the descriptive metrics from the different tissue classes in T2 weighted space. All relevant output is exported into text for its posterior analysis.

Violin plots showing the distribution of SNR and CNR values across sites. This figure as well as the table used to generate it can be used to detect outliers which would in turn be further scrutinised.

Violin plots showing the orientation of the T2-weighted images with respect to the HC for the different diagnostic groups. Ideally, we would expect to find our values distributed around 0 degrees. Nevertheless, the main longitudinal axis of the hippocampus is not contained in a true sagittal plane, its tail is closer to the brain’s midline that its head. We believe this is causing this bias in the orientation measurement and requires further investigation.

Proc. Intl. Soc. Mag. Reson. Med. 26 (2018)
3131