Description: Scientists often create data derivatives (i.e., transforming, combining, and/or altering raw variables in some way) when cleaning data and preparing for analyses. Unfortunately, these derivatives are often generated within larger scripts, are the result of judgement calls that aren't written down (e.g., deriving a new variable for a subset of the entire sample), can be duplicated when team members aren't aware that the work has already been done by someone else, and can be lost when team members move on to other projects, graduate, and/or leave the group for their next positions. The goal of this project is to develop a workflow for how teams can efficiently and transparently create, document, and share data derivatives with their future selves, their team members, and their collaborators through the creation and maintenance of a derived data registry.
Project url: https://github.com/no-scientist-is-an-island/no-scientist-is-an-island.github.io
Contributors: Theresa Cheng & Megan Huibregtse
Description: NeuroNest is the brainchild of students, researchers, and enthusiasts who attended Neurohackademy 2024 at the University of Washington in Seattle. Our platform is dedicated to exploring the intersection of data science and brain science, providing a comprehensive resource for anyone interested in these fields. We believe in the power of knowledge sharing and community learning, aiming to demystify complex concepts in neuroimaging and data science, making them accessible to all. Our mission is to create an inclusive space where beginners and experts alike can find valuable resources. We offer a diverse array of content, including tutorials on MRI, EEG, and machine learning, coding snippets, project ideas, and insightful articles on the latest technologies and techniques. At NeuroNest, we understand that learning is a continuous journey, and we're here to support you every step of the way. The NeuroNest community is built on curiosity, innovation, and collaboration. We are passionate about fostering an environment where learning is both fun and rewarding. Whether you're just starting or looking to deepen your expertise, NeuroNest is the place to unlock new insights and explore the wonders of the brain and data. Join us as we delve into this exciting world and become a part of a vibrant community dedicated to advancing knowledge and innovation. Welcome to NeuroNest!
Project url: https://neurohackademy2024.github.io/NeuroNest/
Contributors: Cynthia Sopko, Annachiara Crocetta, Arianna Mordy, Sara Wong, Morgan Fitzgerald, Anne-Marie Leiby, Alan Patrick Davalos-Guzman, Elena von Perponcher, Mohamed Elsherif
Description: BIDSInspec is a python library that will help researchers summarize, visualize, and validate their BIDS datasets. With the increasing use of automated tools (e.g., BIDS apps) it is important to always examine your data prior to pre-processing. A problem that arises when curating neuroimaging data is the heterogeneity that exisits in MRI session aquisitions within and across institutions. BIDSInspec is designed to solve these problems by: 1. creating a summary report of the BIDS directory by specific features (e.g., scanner type, magnetic field stength, acquisition time, resolution, etc.) and 2. enabling researchers to validate a BIDS dataset against an acquisition protocol using features within JSON sidecar files and NIFTI image properties (e.g., image shape, resolution, etc.)
Project url(s): https://github.com/NeuroHackademy2024/BIDSInspec
Contributors: Ashley Ptinis, Nancy Ortega
Description: Our project has three parts: 1) Making BrainEffeX easier to use and understand by improving the user interface. 2) Creating a tool to visualize script pipelines and the flow of inputs and outputs throughout functions. The tool is also automated as a Github workflow to update in the README.md file every time the repository changes. 3) Nipoppy-effeX addresses the challenge of underpowered functional magnetic resonance imaging (fMRI) studies, which often rely on limited sample sizes and custom preprocessing pipelines. Our project proposes a portable, user-friendly BIDS application that facilitates the reprocessing of both public and non-public fMRI datasets. Currently, Nipoppy-effeX enables researchers to execute semi-automated, reproducible data processing pipelines solely for resting-state functional connectivity analyses. By leveraging the lightweight Nipoppy framework and robust tools like fMRIPrep, users can perform within-subject and between-group analyses efficiently. By democratizing access to comprehensive fMRI analysis, Nipoppy-effeX can pave the way for more powerful neuroimaging studies while complementing existing databases like BrainEffeX.
Project url: https://github.com/andreifoldes/nipoppy-effex, https://github.com/halleeshearer/scripts2viz
Contributors: Ashely Humphries, Alex Fischbach, Tamas Foldes, Hallee Shearer
Description: Federated learning (FL) has become a promising machine learning method for conducting large-scale analyses across multiple institutions without the need to share data. With FL, data privacy and security are maintained, as the data never leave the institution; only encrypted model parameters are exchanged and aggregated. The aim of our project was to explore the use of FL using Flower (open source federated learning framework) and scikit-learn to predict brain age using measures of grey and white matter volume. We first focus on collating FreeSurfer statistical data across multiple subjects, focusing on both cortical and subcortical brain regions from the Healthy Brain Network (HBN). We included 223 participants with demographic information and freesurfer deriviates available. Our script (client_newmodel.py) sets up a client for federated learning using the Flower (see https://flower.ai/docs/framework/tutorial-series-what-is-federated-learning.html for more details). It enables training and evaluation of machine learning models (Logistic Regression, Linear Regression, LassoCV) on different partitions of the HBN sample. The client communicates with a federated learning server to participate in model training and evaluation of tasks over multiple rounds of federated learning, and saves the aggregated model weights after each round. We use 3 to mimic different dataset and compare accuracy in models with different sample distributions.
Project url:https://github.com/mollyolzinski/Federated-Learning.git
Contributors: Michelle Wang, Emma Corley, Eren Kafadar, Molly Olzinski, Aoife Warren, Audrey Weber, Maya Lakshman
Description: The NeuroNav dashboard is an interactive tool for visualizing a subset of data from the Human Connectome Project Young Adult dataset. Users of this tool can display and interact with different types of plots in each panel of the tool. One panel shows the age and sex distribution of the sample. Another panel allows users to plot performance on behavioral tasks against one another or against various structural brain measures such as white or gray matter volume. Users have the options to display a line of fit over this data and to display data for a chosen age or sex bucket. In the final panel, users can display plots representing results from tractometry analysis. This project uses Jupyter Widgets to render the elements of each plot, including the interactive components.
Project url: https://github.com/NeuroHackademy2024/neuro-nav
Contributors: Eyerusalem Abebaw, Rohini Kumar, Sophia Mehdizadeh, Jonathan Wehnert
Description: Psychopathology arising from enhanced negative emotion or from the loss of positive emotional experience affects over 400 million people globally. Such states of disordered emotion cut across multiple diagnostic categories and are compounded by accompanying disruptions in cognitive function. Not surprisingly, therefore, these forms of psychopathology are a leading cause of disability. The Human Connectome Project (HCP) contains multi-modal neuroimaging data of over 1000 people aged 18-35 years who are experiencing varying degrees of acute threat, loss of reward valuation/responsiveness, and difficulties in working memory. This project aims to develop a suite of open-source machine learning predictive and multimodal frameworks for structural and functional neuroimaging data from the HCP to evaluate aspects of emotional self-perception and recognition.
Project url: https://github.com/dbraun31/nh2024-hcp1200-ml
Contributors: Angelina, Da Yea, Dave, Diana, Dickson, Haruka, Jacqueline, Jiayue, Jony, Kristen, Laura, Melissa, Mingcong, Omair, Ruize, Shangcheng, Shuhei, Theresa, Yuxiang. :)
Description: Our project focused on classifying eye states (open or closed) using EEG data collected from 109 volunteers through the BCI2000 system. We analyze 1500 one- and two-minute 64-channel EEG recordings to develop and compare the performance of both traditional machine learning (ML) and deep learning (DL) models. Using methods like Random Forest (RF) and Gradient Boosting Machine (GBM), we manually extracted features from the EEG signals for ML classification. Simultaneously, we employed Convolutional Neural Networks (CNNs) and Recurrent Neural Networks (RNNs) in our DL approach to automatically learn and classify the data. This dual approach allowed us to explore the effectiveness of different techniques in accurately detecting eye states from EEG signals.
Project url(s): https://github.com/vs-turner/eeg-hackathon
Contributors: Arwa Adib, Bobby Tromm, Victoria Sayo Turner, Simona Vaitekunaite\
Description: Reading an article takes time, money, and effort. Before you know it you are taking an afternoon nap while reading through a 20-page lit review on the inner workings of magnetic resonance spectroscopy. Wouldn’t it be nice to have a “too long, didn’t read” version of the articles and data you need that pulls all the necessary information? This project leverages natural language processing and machine learning techniques to create a search engine that extracts and summarizes key information from a large repository of scientific articles based on a user question. When a user inputs a query, the engine identifies and retrieves relevant articles, presenting the extracted data in a tabular format. Additionally, the tool offers a detailed panel that provides supplementary demographic and methodological information for each selected article, enabling a more comprehensive understanding without the need to navigate the entire text.
Project url: https://github.com/cmcurran410/tldr2024/
Contributors: Cristian Curran, Magdalena Martinez-Garcia, Keerthi Stanley, Loreta Sutkus & Aydin Tasevac
Description: The aims of this project were to test QSIprep for preprocessing, reconstruct dMRI data, and compare different reconstruction models. Here, we provide step-by-step instructions for true beginners to diffusion imaging preprocessing and reconstruction. For starters, we compiled background resources on diffusion imaging into an introductory slide deck. We have very detailed steps for downloading the data from an existing database (in this case: Healthy Brain Network, 2 subjects), instructions for creating a singularity image for the latest version of QSIPrep, and additional details on BIDS compliance for QSIPrep. For example, the fMRI field maps need to be removed, and a ‘dataset_description.json’ file is necessary. We have line-by-line instructions on how to update the shell scripts for running QSIPrep preprocessing and reconstruction. We provide an explanation of QSIPrep outputs. We finally take the preprocessed data to further fit the appropriate diffusion models to estimate fiber bundles. We provide a python tutorials showing how to perform reconstruction model comparison for differend preprocessed inputs, as well as how to compare the fit of diffusion tensor imaging and constrained spherical deconvolution approaches.
Project url: https://github.com/NeuroHackademy2024/diffusion-mri
Contributors: Allesandra Iadipaolo, Lya K. Paas Oliveros, Elle Murata, Gaby Ojeda Valencia, Lupita Yañez-Ramos, Claudia Tato, Qingqing Yang, Luis Álvarez :)
Description: ThinkShareCare is an interactive web app designed for mental health clinicians to help assess their patients' odds of specific outcomes, such as potential diagnoses, subtypes, and treatment responses. The app uses models created by the community, including contributions from mental health clinics, hospitals, and researchers. Contributors can upload model weights based on their patient data, which then get integrated into similar models using a meta-regression technique. This process allows the models to theoretically become more accurate over time as they are informed by increasing amounts of patient data. Importantly, no Protected Health Information (PHI) is shared with this technique, as only the model weights, generated on the user's browser, are sent to the server. Critically, while this tool may offer valuable insights and help clinicians integrate quantitative data, it is intended to be an additional resource for clinical decision-making, not a standalone diagnostic tool.
Project URL: https://github.com/hughesdy/ThinkShareCare
Contributors: Dylan Hughes & Sarah Zapetis