High-frequency longitudinal white matter diffusion- and myelin-based MRI database: Reliability and variability.

Assessing the consistency of quantitative MRI measurements is critical for inclusion in longitudinal studies and clinical trials. Intraclass coefficient correlation and coefficient of variation were used to evaluate the different consistency aspects of diffusion- and myelin-based MRI measures. Multi-shell diffusion and inhomogeneous magnetization transfer data sets were collected from 20 healthy adults at a high-frequency of five MRI sessions. The consistency was evaluated across whole bundles and the track-profile along the bundles. The impact of the fiber populations on the consistency was also evaluated using the number of fiber orientations map. For whole and profile bundles, moderate to high reliability of diffusion and myelin measures were observed. We report higher reliability of measures for multiple fiber populations than single. The overall portrait of the most consistent measurements and bundles drawn from a wide range of MRI techniques presented here will be particularly useful for identifying reliable biomarkers capable of detecting, monitoring and predicting white matter changes in clinical applications and has the potential to inform patient-specific treatment strategies.

Tractostorm 2: Optimizing tractography dissection reproducibility with segmentation protocol dissemination.

The segmentation of brain structures is a key component of many neuroimaging studies. Consistent anatomical definitions are crucial to ensure consensus on the position and shape of brain structures, but segmentations are prone to variation in their interpretation and execution. White-matter (WM) pathways are global structures of the brain defined by local landmarks, which leads to anatomical definitions being difficult to convey, learn, or teach. Moreover, the complex shape of WM pathways and their representation using tractography (streamlines) make the design and evaluation of dissection protocols difficult and time-consuming. The first iteration of Tractostorm quantified the variability of a pyramidal tract dissection protocol and compared results between experts in neuroanatomy and nonexperts. Despite virtual dissection being used for decades, in-depth investigations of how learning or practicing such protocols impact dissection results are nonexistent. To begin to fill the gap, we evaluate an online educational tractography course and investigate the impact learning and practicing a dissection protocol has on interrater (groupwise) reproducibility. To generate the required data to quantify reproducibility across raters and time, 20 independent raters performed dissections of three bundles of interest on five Human Connectome Project subjects, each with four timepoints. Our investigation shows that the dissection protocol in conjunction with an online course achieves a high level of reproducibility (between 0.85 and 0.90 for the voxel-based Dice score) for the three bundles of interest and remains stable over time (repetition of the protocol). Suggesting that once raters are familiar with the software and tasks at hand, their interpretation and execution at the group level do not drastically vary. When compared to previous work that used a different method of communication for the protocol, our results show that incorporating a virtual educational session increased reproducibility. Insights from this work may be used to improve the future design of WM pathway dissection protocols and to further inform neuroanatomical definitions.

What matters in reinforcement learning for tractography.

Recently, deep reinforcement learning (RL) has been proposed to learn the tractography procedure and train agents to reconstruct the structure of the white matter without manually curated reference streamlines. While the performances reported were competitive, the proposed framework is complex, and little is still known about the role and impact of its multiple parts. In this work, we thoroughly explore the different components of the proposed framework, such as the choice of the RL algorithm, seeding strategy, the input signal and reward function, and shed light on their impact. Approximately 7,400 models were trained for this work, totalling nearly 41,000 h of GPU time. Our goal is to guide researchers eager to explore the possibilities of deep RL for tractography by exposing what works and what does not work with the category of approach. As such, we ultimately propose a series of recommendations concerning the choice of RL algorithm, the input to the agents, the reward function and more to help future work using reinforcement learning for tractography. We also release the open source codebase, trained models, and datasets for users and researchers wanting to explore reinforcement learning for tractography.

Validate your white matter tractography algorithms with a reappraised ISMRM 2015 Tractography Challenge scoring system.

Since 2015, research groups have sought to produce the ne plus ultra of tractography algorithms using the ISMRM 2015 Tractography Challenge as evaluation. In particular, since 2017, machine learning has made its entrance into the tractography world. The ISMRM 2015 Tractography Challenge is the most used phantom during tractography validation, although it contains limitations. Here, we offer a new scoring system for this phantom, where segmentation of the bundles is now based on manually defined regions of interest rather than on bundle recognition. Bundles are now more reliably segmented, offering more representative metrics for future users. New code is available online. Scores of the initial 96 submissions to the challenge are updated. Overall, conclusions from the 2015 challenge are confirmed with the new scoring, but individual tractogram scores have changed, and the data is much improved at the bundle- and streamline-level. This work also led to the production of a ground truth tractogram with less broken or looping streamlines and of an example of processed data, all available on the Tractometer website. This enhanced scoring system and new data should continue helping researchers develop and evaluate the next generation of tractography techniques.

Impact of follow ups, time interval and study duration in diffusion & myelin MRI clinical study in MS.

It is currently unknown how quantitative diffusion and myelin MRI designs affect the results of a longitudinal study. We used two independent datasets containing 6 monthly MRI measurements from 20 healthy controls and 20 relapsing-remitting multiple sclerosis (RR-MS) patients. Six designs were tested, including 3 MRI acquisitions, either over 6 months or over a shorter study duration, with balanced (same interval) or unbalanced (different interval) time intervals between MRI acquisitions. First, we show that in RR-MS patients, the brain changes over time obtained with 3 MRI acquisitions were similar to those observed with 5 MRI acquisitions and that designs with an unbalanced time interval showed the highest similarity, regardless of study duration. No significant brain changes were found in the healthy controls over the same periods. Second, the study duration affects the sample size in the RR-MS dataset; a longer study requires more subjects and vice versa. Third, the number of follow-up acquisitions and study duration affect the sensitivity and specificity of the associations with clinical parameters, and these depend on the white matter bundle and MRI measure considered. Together, this suggests that the optimal design depends on the assumption of the dynamics of change in the target population and the accuracy required to capture these dynamics. Thus, this work provides a better understanding of key factors to consider in a longitudinal study and provides clues for better strategies in clinical trial design.

Manifold-aware synthesis of high-resolution diffusion from structural imaging.

The physical and clinical constraints surrounding diffusion-weighted imaging (DWI) often limit the spatial resolution of the produced images to voxels up to eight times larger than those of T1w images. The detailed information contained in accessible high-resolution T1w images could help in the synthesis of diffusion images with a greater level of detail. However, the non-Euclidean nature of diffusion imaging hinders current deep generative models from synthesizing physically plausible images. In this work, we propose the first Riemannian network architecture for the direct generation of diffusion tensors (DT) and diffusion orientation distribution functions (dODFs) from high-resolution T1w images. Our integration of the log-Euclidean Metric into a learning objective guarantees, unlike standard Euclidean networks, the mathematically-valid synthesis of diffusion. Furthermore, our approach improves the fractional anisotropy mean squared error (FA MSE) between the synthesized diffusion and the ground-truth by more than 23% and the cosine similarity between principal directions by almost 5% when compared to our baselines. We validate our generated diffusion by comparing the resulting tractograms to our expected real data. We observe similar fiber bundles with streamlines having <3% difference in length, <1% difference in volume, and a visually close shape. While our method is able to generate diffusion images from structural inputs in a high-resolution space within 15 s, we acknowledge and discuss the limits of diffusion inference solely relying on T1w images. Our results nonetheless suggest a relationship between the high-level geometry of the brain and its overall white matter architecture that remains to be explored.

Track-to-Learn: A general framework for tractography with deep reinforcement learning.

Diffusion MRI tractography is currently the only non-invasive tool able to assess the white-matter structural connectivity of a brain. Since its inception, it has been widely documented that tractography is prone to producing erroneous tracks while missing true positive connections. Recently, supervised learning algorithms have been proposed to learn the tracking procedure implicitly from data, without relying on anatomical priors. However, these methods rely on curated streamlines that are very hard to obtain. To remove the need for such data but still leverage the expressiveness of neural networks, we introduce Track-To-Learn: A general framework to pose tractography as a deep reinforcement learning problem. Deep reinforcement learning is a type of machine learning that does not depend on ground-truth data but rather on the concept of "reward". We implement and train algorithms to maximize returns from a reward function based on the alignment of streamlines with principal directions extracted from diffusion data. We show competitive results on known data and little loss of performance when generalizing to new, unseen data, compared to prior machine learning-based tractography algorithms. To the best of our knowledge, this is the first successful use of deep reinforcement learning for tractography.