Description: Using RSA and brain encoding model to study the differences in brain responses during perception and imagery.
Project url: https://github.com/mynanshan/NHprojectNSDimagery
Contributors: Muye Nanshan
Description: Deep learning models for the construction of retinotopic maps from structural data have achieved promising results, but they still struggle at capturing fine-grained individual differences and generalising across datasets. We hypothesise that these struggles can partially be attributed to the limited number of samples available to train these models in fully supervised models. Self-supervised learning, where the model is first trained to learn a general representation of the phenomenon of interest on unlabelled data that can later be fine-tuned for the task of interest, can alleviate this problem, as large, heterogeneous collections of unlabelled structural data are now available This project therefore aims at exploring the feasibility of this approach for retinotopy maps modelling by
- Defining a transformer model that can be trained in a self-supervised manner on the HCP structural data.
- Fine-tune the resulting model to predict retinotopy maps from structural data. We aim for a fully functioning pipeline that can be further extended in future works.\
Project url: https://github.com/Junebeomstics/surface-vision-transformers
Contributors: Junbeom Kwon & Federico Giacardi
Description: Precision fMRI uses resting-state data to define individualized functional network. One of the benefits of individualized network definition is that it elegantly handles regions whose functional organizations have huge across-subject variability, region such as the Default Mode Network (DMN). The current project attempts to run multivariate analysis during task fMRI using template/group/individual network definition, and quantify the benefit of precision fMRI for the purpose of multivariate analysis. The current project uses PAN Dataset.
Project url: https://github.com/henrfo/panmvpa
Contributors: Tong-Le Cai & Kareem Al-Khalil & DT Nguyen & Sophia Martin & Henrik Formoe & Ziyuan Chen
Description: An open-source educational platform dedicated to improving data quality in fMRI and EEG research. Through accessible explanations, annotated examples of common artifacts, practical quality-assurance workflows, and carefully curated learning resources, the project helps students and researchers build the skills needed to recognize, evaluate, and address data-quality issues. By combining education with practical guidance, the website aims to make neuroimaging quality assurance more transparent, consistent, and approachable for the broader research community!
Project URL: NeuroQA
Contributors: Tien Yang & Morgan Barnes & Chenye Bao & Henrik Formoe & Tamaya Levy & Merel van der Thiel
Description: Predict delay discounting (as measured by questionnaire) from gambling task functional connectivity OR brain structure (DWI)
Project url: https://github.com/pganon32/nh26_td_gamb_diff_FC_best
Contributors: Jonas Granzow, Caroline Raymond, Daniel Porta-Casteras, Emily Wertheimer, Youngeun Park, Hamid Hemmatr, Jessica Tai, Pouneh Baniasad, Patrick Gagnon
Description: Use psilocybin dataset to estimate brain state dynamics using HMMs, as well as attractor landscapes and key features to distinguish altered from normal states
Project url: FunctionalAttractors
Contributors: Monica Vogel, Daniel Bazan, Yicheng Zheng, Donisha Smith & Udbhav Singhal
Description: Brain function emerges from the interaction between local neurochemical signaling and large-scale network dynamics. While Magnetic Resonance Spectroscopy (MRS) quantifies the biochemical substrates of neuronal and glial function within specific brain regions, functional MRI reveals how these regions interact as distributed networks to support cognition and behavior. Combining MRS with functional connectivity therefore provides a powerful multiscale framework for understanding how regional excitatory-inhibitory balance, neuronal health, and metabolic processes shape the architecture and efficiency of functional brain networks. This integrated approach moves beyond describing brain activity to elucidating the neurochemical mechanisms underlying network organization and dysfunction.
This project therefore aims to explore the feasibility of this approach by:\
- Utilizing the HCP-aging dataset to build baseline regression models for MRS prediction.
- Using linear mixed models (LMM) and Generalized Estimating Equations (GEE) to test if longitudinal scans do better than baseline data.
We aim for a fully functioning pipeline that can be further extended in future works.
Project url: github.com/prank24/neurohackademy_project_mrs_func
Contributors: Siddharth Nayak & Prankur Saxena & Maria Perica & Poorvi Balaji & [anca.e.p]
Description: Analyze fMRI data from 21 participants, beginning with a first-level general linear model, followed by whole-brain and ROI-specific representational similarity analysis approaches, to compare neural representations of different stimuli types. Project url: https://github.com/LuciZR/Hyperface-Dynamic-Task-Localizer Contributors: Lucia Z-Rivera, Jillian O'Malley, Bailey Harris, Natalia Pallis-Hassani, Heather Laurel Jensen, & Emily Fitzgerald
Description: Most insights from white matter connectivity needed for face perception come from brain lesioned patients. Yet, there is an opportunity to reproduce the same insights by using big data coming from healthy adults. The Human Connectome Project (HCP) data allows us to do just that, by correlating face memory performance with statistics from reconstructed tracts. This project aimed at correlating face memory performance of HCP participants with their structural and functional connectomes. Structural connectivity was estimated from the stream line counts between ROIs, and also from the track profiles. Functional connectivity was estimated from two types of tasks: resting state (rs), and blocks where faces were shown to participants (task).
Project url: Structural and Functional Connectivity of Face Recognition Pathways
Contributors: Manuel Mejia, Ryan Dean, Celina Alba, Emilia Zaldivar, Chenyuan Li, Nick Wellman & Shivaram Karandikar
Description: Film
Project url: https://youtu.be/82dTs6mZ7XM
Contributors: Henrik Formoe, Daniel Porta-Casteras, Lara Ressin, Kelly Chang & John Pyles
Description: Develops pipeline for using pre-trained LLMs to predict clinical phenotypes across several datasets and evaluates performance.
Project url: https://github.com/stvsever/multi_agent_decision_support_system/tree/main
Contributors: Ziyuan Chen, Stijn Van Severen, Gavin Schneider
Description: This project explores the StudyForrest dataset, in which participants experience the Forrest Gump movie during fMRI. Our team develops preliminary brain encoding and decoding analyses that relate time-varying movie information—such as emotion, audiovisual content, and behavioural annotations—to brain responses. We use the dataset's shared naturalistic stimulus, anatomical derivatives, and multi-subject recordings to study both within-participant and cross-participant representations.
Project url: ForrestGump-NeuroHackademy
Contributors: tamarj8592, juz031, joosenli, Konstantinos14, samtorrisi, rcastolayo, and snpushpi.
Description: A python package for anatomical labeling, functional decoding, and interactive exploration of brain parcellations. Additional features include summarization of statistical brain maps using a parcellation.
Project url: https://github.com/amishavyas/neurolabel
Contributors: Amisha Vyas, Mia Casburn, Hyesun Choi, Anna Isberg, Alex Litovchenko, Kaitlyn Mundy, Rajikha Raja, Lara Ressin, & America Romero
Description Using features from resting state EEG (eyes open and closed) to predict contraceptive use (vs non use)
Project URL https://github.com/ekaplan7/menstrual-aperiodic
Contributors: Liz Kaplan, Nicole Liddle, Aleyna Gross
Description: This notebook searches multiple scholarly sources, ranks records using keyword and semantic methods, discovers topics with BERTopic, maps topic relationships, identifies potential gaps within the retrieved corpus, and exports a complete data dictionary.
Project url: https://github.com/peterald/ai_interactive_evidence_mapper_Peter_A/tree/ai_interactive_evidence_mapper_peteraldana
Contributors: Peter Aldana