StoopwenpueNq/jackie-fmri-sim

What it solves

Traditional brain imaging studies look at group averages and static snapshots. This project instead builds a living, queryable model of one individual dog's brain—"Jackie"—that continuously estimates her emotional state and regional brain activity from real recordings. It lets researchers and curious users type in a stimulus (like "chicken meat" or "neighbor dog") and get back a predicted emotional response with supporting brain activation data.

How it works

The system is built on six real recording sessions (J001–J006) that contain BOLD signals from key brain regions (olfactory bulbs, amygdala, nucleus accumbens), plus EEG, heart rate, breathing, and movement data. Each session was triggered by a known stimulus (food, other dogs, walking).

From these data, the model learns a mapping between any text stimulus and a six-dimensional emotional vector (joy, curiosity, calm, excitement, anxiety, affection) and a 25-region activation map. The core dynamics use an exponential decay toward baseline, BOLD injection scaled by intensity, and social-condition boosts. When a user types or imports a stimulus, the model applies these rules to produce a live "state vector" and the most likely behavioral phenotype (e.g., anxiety, affection, calm).

Three real axial fMRI slices are included as visual ground truth for three social conditions (novel street dog, best friend, neighbor after repeated exposure), showing actual activation patterns.

Who it's for

This is a research prototype aimed at neuroscientists, animal cognition researchers, and AI/ML developers interested in building digital twins or individualized brain models. It requires basic Python knowledge to install and run, and the code is designed to be extended with new sessions or regions. It could also appeal to dog enthusiasts curious about how a dog's brain might respond to different situations.

Highlights

• Built on real, multi-session fMRI, EEG, and physiological recordings from a single dog—no synthetic data for the core sessions.
• Produces a continuously updating state vector (joy, curiosity, calm, excitement, anxiety, affection) and a 25-region activation map.
• Interactive command-line queries: type a stimulus and see the predicted emotional response and arousal.
• Includes three authentic axial fMRI slices as visual ground truth for social conditions.
• Modular, extendable architecture (e.g., add new regions or stimuli) and only uses Python's standard library.