 Okay, hi, my name is Lucy Moore and I'm a postdoctoral scholar in the developmental cognition and neuroimaging lab in Portland, Oregon. And today I'm going to be talking about open science in infant neuroimaging and initiatives undertaken by my lab to push forward open science practices within our field. Let me just check on something. Okay, so I just want to introduce myself a little first. I received my PhD early last year using mice as a model organism to study the auditory brainstem. And this involved long hours of sitting at an electrophysiology rig and trying to tease apart the organization of local circuits by recording from individual neurons. And after experiencing how time and resource intensive biological research is, I began to think about whether the data being produced from this kind of work is being maximally utilized to drive forward our understanding. This is when I became involved in open science because this community doesn't shy away from asking these really difficult questions about the archaic structures and practices within academia. And by doing this work to challenge these structures, we can not only make our science better, but as many others from this conference have pointed out during this time, we can build our communities to be more inclusive and equitable, which is really important to me. So with these values in mind I got involved in neuroimaging because this research has more direct impact on human health and society. So I currently work with doctors Alice Graham and Damien Fair, who both work hard to integrate open science collaboration and social justice into their work. And in my current role I no longer focus on data collection, but rather analysis and building tools for image processing functional magnetic resonance imaging is a safe non invasive means of examining brain function. It measures changes in blood flow associated with neural signaling. So it has good spatial resolution for capturing activation in different regions of the brain compared to other techniques. And it's been the primary tool used in neuroscience to identify alterations in brain function underlying different mental disorders, leading to a lot of really great research. So in the decant lab we use fMRI and infants, because if we can study the neuro development of brain circuitry underlying mental health issues, then we can clinically intervene early on to prevent this circuitry from forming in the first place. Because fetal and infant brain development is so rapid, this period of extreme plasticity makes the brain vulnerable to negative environmental factors. So for example, we know that the effect of high maternal stress on fetal brain development is associated with a higher risk of the child developing a mental disorder later in life. And in our current clinical trial, we're testing the effect of a cognitive behavioral therapy based on mindfulness to reduce stress and pregnant people. Pregnant people are often hesitant to take medication to help with their stress and depression during pregnancy because of the increased risk of birth defects. And this specific therapy has the strongest evidence base for effectively reducing stress and the risk of developing postpartum depression. So we expect for this intervention to foster healthy brain development and mental health outcomes in the children via reduced stress in the mother. And fMRI will give us some indication as to how the intervention impacts brain circuitry during development so that we can be maximally informed about the success as well as the mechanism of the intervention. But unfortunately, there are pretty significant challenges to this type of research, because our field has a general lack of established standards for processing and analysis. To the extent that we basically had to build our own image processing pipeline because the available tools either didn't work or produce really variable outputs. And this general lack of standardization is concerning because like other fields of science neuroimaging has come under fire for serious issues with reproducibility. Preeminent researchers in the field are trying to tackle this issue head on. For instance, here's an article from 2017 published by Russell Polak from Stanford who is a leader in the field of cognitive neuroscience. And in this paper, he outlined several sources of non reproducibility specific to our field. In our lab we've been trying to heavily incorporate open science into our research practices so that we can feel confident that our results and conclusions from our clinical trial are robust and reproducible. So for the remainder of the talk I'll be highlighting some sources of bias and variability in infant neuroimaging regarding data processing and analytical workflows and how we've tried to address each of these issues using open science. And these practices include organizing our infant data to follow a new data standard data standard called bids. Adapting the standard adult pipeline to work with a diversity of infant data sets and releasing this on GitHub. And finally collaborating with a new standard open source fMRI pipeline called fMRI prep to integrate our infant specific processing methods. In terms of MRI data itself the biggest issue is the lack of standard file formatting and organization. So first there's a large variability and just the raw data depending on what type scanner was used to collect it. And then before the raw scan data can go through an image processing pipeline. It has to be converted to something the pipeline you're using can actually recognize. So several groups have basically just created their own file formats to work with their own pipelines. So this makes data sharing very difficult and causes mistakes in data handling by collaborators as you can imagine. And data sharing is a hugely important part of neuroimaging because data collection is very time consuming expensive technically challenging and requires a large amount of resources and not to mention access to imaging course. So for instance, even though we are one of the main labs that do infant neuroimaging in the world, about half of the data sets I work with were collected by collaborators. So to address this issue the decal lab has adapted our infant MRI data to conform to the brain imaging data structure or bids. And this is a new data standard introduced in 2016 that will make it possible for data to be easily shared and understood between groups. And again this new standard has been introduced by some of the biggest people in the field. And figure one from this paper just shows a snapshot of how the raw MRI data is converted into open file formats with logical folder structure and human readable file names. So bids enable standardization and reproducibility in many ways. It helps with data reusability because it stores information about how data was processed, both in the file name itself as well as in the metadata. So it's easy to reference and understand how data were processed or analyzed if you want to go back to it. It makes data sharing easier because it provides standardization of documentation so that collaborators can easily understand the data organization and be able to use it correctly. And finally study helps us study reproducibility because it streamlines the process to replicate study results and write analysis pipelines that can be used across different data sets. Okay, another huge source of non reproducibility in our field is the lack of a standard robust image processing pipeline specifically designed for infants. Here is a schematic of what the fMRI prep image processing pipeline looks like. Image processing is necessary to remove non biological sources of noise, correct for image distortion introduced by the scanner and put subject data all into the same coordinate space so that we can actually make comparisons across the population. And in order to do these things, as you can see from this figure, processing pipelines are very complex and require multiple steps. The human connection project funded the creation of a standard image processing pipeline that was released in 2013, but there are a few problems with this pipeline. One is that it was written in bash, which isn't designed for complex workflows. So it takes about 24 hours to process the subject. And if the pipeline fails you have to restart it from the beginning of a stage instead of wherever the pipeline left off. And most importantly, the pipeline was written to work for adult data and doesn't accommodate infant data at all. We call babies aliens because babies, their brains are just weird and the data is really hard to work with. For instance, the data typically has really poor image contrast compared to adults. And there are typically a lot of motion artifacts since subjects are supposed to be pulled pretty much perfectly skill is still in the scanner, and babies, of course, you know, don't take direction. So the inherently messy data makes image processing very challenging and introduces more variability into study results. One part of the adult pipeline that we had to rebuild was brain segmentation, which is supposed to delineate different subcortical and cortical structures. So if you look at the image on the left, the red and black lines are supposed to delineate cortical gray and white matter. So this red line should basically be following the outline of the brain within the skull, and it totally misses the mark, particularly in the superior parietal area. So we rebuilt the part of the pipeline that handles brain segmentation and now the red line outlines the brain pretty perfectly as you can see on the right. Five minutes. Thanks. So I helped build this feature. The brain segmentation to handle infant data. And we use this Python toolkit called me pipe that is specifically built for image processing pipelines. So need pipe interfaces with many image processing softwares required in these pipelines and officially manages the complex workflow. So with our modifications, they see people pipeline cannot only process infant and adult data, but it also uses open source tools, specifically designed to handle neuro imaging pipelines. Our group released our pipeline on GitHub as a bids up, which is a portable neuro imaging pipeline that understands its data sets. And because the pipeline is containerized using Docker, the version control massively improves reproducibility of pipeline outputs. Okay, so to the last point the issue still remains that there are too many groups using different pipelines of variable quality. So as they stay in the recent paper publish on the fMRI prep pipeline, despite the wealth of accessible software and multiple attempts to outline best practices for pre processing. The large variety of data acquisition protocols has led to the use of ad hoc pipelines customized for nearly every study. fMRI prep is an open source standard for image imaging pipelines, and it was developed by some of the top cognitive neuroscientists from Stanford who I mentioned earlier. It basically has the same overall steps as the HCP pipeline but it's written in Python uses workflow managers like me pipe. And it integrates a variety of processing options tested for reproducibility, so that it can work across multiple data sets. So we're currently collaborating with them to integrate our infant specific processing methods, so that someday you'll be able to use this single interface to process and analyze any type of fMRI data. The open science frameworks adopted by our lab are essential, not only for our analysis to even work, but also to assure reproducibility in our study results. And ultimately lead this leads to effectively informing medical practice and policy to foster healthy brain development. And this is really important in relation to our clinical trial, because although the official recommendation from the US preventative services task force is to use therapeutic interventions and pregnant women considered at risk for developing depression. There currently isn't a lot of infrastructure to actually support this intervention. And with that, thank you so much to my team, our collaborators and funding sources, and especially to my mentors Alice Graham Damien Farron Eric Fesco. And finally, thank you to the conference organizers for thinking on their feet and putting in so much effort to put on a great conference. And here are my ways you can contact me. Thank you. Thank you so much.