Projects

A selection of my recent and past projects.


🧠 NeuroConText: Contrastive Learning for Neuroscience Meta-Analysis with Rich Text Representation

NeuroConText_framework_training NeuroConText_framework_inference

retrieval_leftOutArticles DiceScore_articles_test_ brain_maps_reconstruction_dice_leftOutArticles_best_median_worst

Objective:
Meta‑analysis aggregates thousands of neuroimaging studies to extract reproducible activation patterns associated with concepts like attention, language, or emotion. However, existing tools rely on manually curated keywords or sparse coordinate tables, missing the rich information in full texts. As the literature grows, scalable methods that link full text to brain data are essential.

What we proposed:
We introduce NeuroConText, a contrastive learning framework that aligns full-text articles with activation maps derived from coordinate‑based meta‑analysis (CBMA):

Advantages over prior models:

Papers:

  1. NeuroConText - Journal version - bioRxiv
  2. NeuroConText - Conference version - MICCAI’24

🧠 Peaks2Image: Reconstructing fMRI Statistical Maps from Reported Peak Coordinates

Objective:
Neuroscience articles often report peak activation coordinates instead of full statistical maps, limiting spatial modeling. Recovering full maps from peak sets allows leveraging legacy data for modern meta-analytic pipelines.

What we proposed:
We developed Peaks2Image, a neural model that:

Advantages over prior work:

Paper:
Peaks2Image