Neuroinformatics Strategies for Open Data Repositories and Enhanced Accessibility

Many of the existing software platforms for reproducible neuroimaging data processing are centralized (i.e. requiring data to be uploaded to a third-party server) [1,2], which is not always possible due to concerns about data privacy and ownership. Moreover, processing of prospective studies with ongoing data collection is challenging: since different software tools and versions can produce different results [3], care needs to be taken to ensure that the new data are processed in the same way as the already processed data. We introduce Nipoppy, a collaborative and open framework that can help achieve decentralized processing of ongoing studies with neuroimaging and clinical data. Nipoppy aims to facilitate every stage of data organization and processing, be flexible and extensible to handle various types of datasets and pipelines, and promote methods transparency and reusability of neuroimaging-clinical data. 

Growing availability of open-access, large-scale neuroimaging data in healthy development and disease allows for rapid discovery of radiologic, neurologic, and psychiatric insights . This is especially true in the context of machine learning (ML), which promises improved prediction of diagnoses, prognoses, disease subtypes, and more. However, harnessing ML to pursue such precision medicine efforts remains a challenge for many neuroimaging scientists – barriers in coding skills, field-specific knowledge of state-of-the-art methodology, and access to large-scale neuroimaging data all limit the rate of biomarker discovery. We introduce niCHART (NeuroImaging Computational Harmonization and ARtificial intelligence Toolbox), a mutually-compatible ecosystem of state-of-the-art methods allowing for holistic processing of multi-modal MRI images as well as calculation of statistical and ML-based imaging-derived phenotypes (IDPs). Ultimately, niCHART will allow for improved reproducibility and accessibility of neuroimaging analysis as well as allow end-users to contextualize their own data among open-access, curated neuroimaging big data. 

The image processing pipeline for UK Biobank brain imaging (1), has been used to process more than 72k UKB datasets, and >500 papers have used its outputs. The pipeline works with T1, T2 FLAIR, swMRI, dMRI, rfMRI, tfMRI (and ASL in recent subjects) and consists of a mix of scripts in bash, Python and Matlab, with the underlying tools coming primarily from FSL and FreeSurfer (Fig 1). It applies image processing and QC, and generates thousands of Imaging-Derived Phenotypes (IDPs). The pipeline was optimised for UKB data and has been adapted for other studies with some effort (2, 3, 4).

By 2023, 72k brain imaging datasets had been acquired and processed. 63k were usable data from the first imaging visit and 5k from the second. The goal is 100k first-visit and 60k repeat-visit scans. Fig 2 shows pairwise associations between 4k brain IDPs and 27k non-imaging variables. 

Advances in fetal fMRI represent, for the first time, an opportunity for neuroscience to study functional brain connectivity at the time of its emergence [1,2]. The unique challenges of in utero imaging require a community-wide effort to develop tailored methods for image preprocessing and analysis. The progress however has been hampered by the lack of openly available datasets that could be exploited by researchers across disciplines. The dHCP closes this gap by releasing the first open-access and largest-to-date fetal fMRI dataset at https://nda.nih.gov, processed using state-of the art methods. 

Historically, neuroimaging data have been stored in a variety of unique file formats and directory structures, presenting obstacles in data sharing, scientific clarity, and rigor. The introduction of the Brain Imaging Data Structure (BIDS) [1] has been pivotal in addressing these issues by standardizing file system structures and metadata for raw neuroimaging data, leading to its widespread adoption [2]. Over time, BIDS has evolved beyond its original scope of MRI data, encompassing a broader range of imaging modalities, thanks to contributions from the community [3]. However, due to this evolution being mostly centered around raw data, BIDS currently lacks detailed descriptions for advanced data derivatives, particularly in brain connectivity research.

To address this gap, the BIDS connectivity project (https://pestillilab.github.io/bids-connectivity) is expanding the scope of BIDS derivatives. This extension includes both raw and minimally processed data, as well as more sophisticated derivatives from brain connectivity experiments. The project aims to establish standard descriptions for connectivity derivatives across six key data modalities: anatomical, diffusion-weighted, and functional MRI, along with PET, M/EEG, and iEEG. This initiative will significantly bolster research capabilities in terms of data generation, sharing, and replication of studies using published data derivatives. Additionally, It will streamline neuroimaging pipelines and processing, thereby accelerating research and development. 

A deviation of 2 millimeters (mm) in deep brain stimulation (DBS) electrode positioning can result in variability of upwards of 60% in therapeutic benefit [1]. Suboptimal targeting may require reimplantation, which can pose additional risks. Localizing DBS targets is not always possible because of their small size, lack of contrast, and patient motion. Tools which involve non-linear alignment of an atlas to patient images, considered 'gold standard' for automatic localization, yield errors on the order of 2-3 mm [2] and highly depend on image quality (Figure 1A). Gadolinium enhanced T1w MRI (MRI-gad) is employed during DBS planning, as it helps with avoiding blood vessels. However, MRI-gad presents challenges during non-linear alignment [3]. Automatic localization of brain structures via machine learning (ML) offers faster and generalizable alternatives to registration approaches. However, limited studies cater ML to DBS targets while demonstrating generalizability in clinic (e.g. on MRI-gad) [4].
We validate an ML model (Figure 1) to localize surgical targets from the coordinates of salient brain landmarks [5] in patient space. Our approach is agnostic to field strength, generalizable to other brain regions, and enables more nuanced understanding of brain morphology that can be expressed in millimeters (Figure 2).