PhD candidate, NeuroscienceUniversity of Illinois Urbana-Champaign
I build deep learning and software for brain imaging, and use it to map how the young pig brain wires itself.
I am a PhD candidate in the Neuroscience Program at the University of Illinois Urbana-Champaign, working with Ryan N. Dilger in the Piglet Nutrition and Cognition Laboratory and with the Pig Imaging Group at the Beckman Institute. I study how structural brain networks develop in the young pig, using diffusion MRI, tractography and deep learning, and how those networks relate to cognition.
The domestic pig has a gyrencephalic brain, a neurodevelopmental timeline closer to ours than that of rodents, and early nutrient needs much like a human infant. That makes it a practical window into questions about infant brain development that are difficult to answer in people. Much of my work is building the imaging tools that the model has been missing.
Before Illinois, I studied cognitive decline with functional MRI at Georgetown University and memory at UC San Diego, where I also trained in machine learning and neural computation. Alongside the science I build and ship software end to end: deep learning tools for imaging, the systems my lab runs on (from a 48-cage automated feeding system to study scheduling with an LLM assistant), and a few products of my own.
How does the developing brain build its wiring, and does that wiring predict how well an animal learns and remembers? I approach this in the young pig: first by building imaging tools that work on pig anatomy, then by using them to compare species, relate structure to behavior, and separate the effects of age from those of body growth.
Published 2025, pipeline in development
Brain extraction tools designed for humans fail on pig anatomy, so manual masking used to take hours per scan. PigBET, a 2.5D deep learning model with ImageNet pre-trained encoders, reaches a Dice coefficient of 0.98 across ages and MRI modalities and does the job in minutes. I am now building an open pipeline that takes raw diffusion MRI to structural connectivity matrices and white matter tract reconstructions.
Presented at SfN 2026
Young pigs tested on novel object recognition give a behavioral readout that can be lined up against their MRI. Recognition memory was associated with hippocampal volume and with microstructure of the fornix, the main output tract of the hippocampus, which points to specific circuits rather than whole-brain differences.
Ongoing
If the pig is a better model of human brain development than the mouse, that should show up in network organization. I compare graph properties of pig structural connectomes at several ages with human and mouse connectomes, under matched network density.
Ongoing
In a growing animal, chronological age and body weight rise together, which makes their effects on the brain hard to tell apart. Separating them matters for preterm and small-for-gestational-age infants, whose brains develop under a mismatch between the two.
A deep learning framework for MRI brain tissue segmentation across species, ages and modalities. Neuroscience 2024 (Society for Neuroscience), Chicago.
Recognition memory in young pigs is associated with hippocampal volume and fornix microstructure
A deep learning framework for MRI brain tissue segmentation across species, ages, and modalities
Establishing the pig as a translational animal model for neurodevelopment
Establishing multiscale measures of myelin in the mouse auditory cortex
Establishing multiscale measures of myelin in the rat sensorimotor cortex
A deep learning-based automated pig brain tissue segmentation tool for MRI data
A functional MRI study of impaired spatial inhibition of return in older adults with mild cognitive impairment or mild Alzheimer’s disease
Full and current list on Google Scholar.
Most of what I build started as a bottleneck in the lab: a step that took hours by hand, a spreadsheet passed around by email, a tool that assumed a human brain. These are the ones that grew into real software, from deep learning models to hardware control and LLM-powered apps.
Brain extraction for pig MRI, where human tools such as FSL BET and AFNI fail. Each volume is cut into sagittal, coronal and axial 2.5D images so ImageNet pre-trained encoders can be fine-tuned, and the three views vote on every voxel.
Python, PyTorch, segmentation-models, NiBabel, FSL
Staffing for multi-week pig studies used to run on Excel sheets passed around by email. Lab members answer an availability link, the scheduler fills every chore and event day from their answers and plain-English rules, and an assistant that knows which page you are on can build, explain or edit the schedule on request.
JavaScript, Firebase (Hosting, Firestore, Cloud Functions), LLM tool use
Controls and monitors milk-replacer feeding for 48 cages across 8 stations. I built the operator software around the lab’s LabVIEW feeding controller: a web app for cage modes, group control, study feeding schedules, live monitoring and data export, a React cage monitor, and a test bench for the controller’s TCP API.
Node.js, Express, SQLite, React, TypeScript, Python, TCP/IP, LabVIEW
Turns a lab protocol into an interactive 3D training simulation. The pipeline builds a digital twin of our 24 by 13 m lab building from the university’s CAD floor plan and walkthrough video, with frames extracted and narration transcribed by Whisper, then runs a full allotment day on top of it: 25 young pigs and seven team members move through unloading, ear tags, weighing, dosing, allotment and first feeding. Trainees scrub the timeline, follow any person or open the cage map before they ever set foot in the barn.
three.js, JavaScript, Node.js, Python, Whisper, Playwright, agent simulation
An open pipeline from raw pig diffusion MRI to weighted structural connectivity matrices and reconstructions of five major white matter tracts, with tractography algorithms benchmarked for reproducibility.
MRtrix3, FSL, ANTs, DIPY, NetworkX
A VS Code extension that opens MRtrix .tck tractograms as interactive 3D streamlines, so a tractogram can be checked without leaving the editor.
TypeScript, three.js, VS Code API
Splits animals into treatment groups with matched mean body weight and variance, blocking by sex, and keeps every allotment plan on record.
JavaScript, Node.js, Express, SQLite, Firebase
An LLM review step for the lab’s controlled-document system, where SOPs and methods live as Markdown in git and go through PI approval.
Python, GitHub Actions, LLM agents
Peer-to-peer video calls with real-time speech-to-speech interpretation between Chinese and English, so two people who do not share a language can simply talk.
React, TypeScript, WebRTC (PeerJS), Node.js, WebSockets, speech-to-speech model API
A satirical alternate-history game for America’s 250th: draft three advisors from history (and a bald eagle), make ten decisions, and an LLM writes up the America you just created.
React, TypeScript, LLM generation
A free, real-time classroom quiz game in the style of Kahoot, built for the anatomy and physiology lab sections I teach.
JavaScript, Firebase
A Discord tutor for an anatomy and physiology course that answers only from approved course materials, declines exam-fishing, and holds student uploads for staff approval before indexing them.
Python, retrieval-augmented generation, Discord API, Docker
A two-stage computer vision pipeline for novel object recognition videos: a detector finds the pig, its head and the objects, then a classifier decides frame by frame whether the pig is investigating. Replaces slow manual scoring that varies between raters.
Python, YOLO (Ultralytics), OpenCV, PyTorch
An application for analyzing and filtering the frequency content of pig audio recordings.
Python, signal processing
Segment Anything adapted to segment pigs in farm images. Won first place at the 2023 AI Foundry for Agriculture Application Hackathon.
Python, PyTorch, Segment Anything
Image-processing scripts that count tissue sections on hippocampal histology slides, one image at a time or in bulk.
Python, Jupyter
A dashboard for the lab’s behavior cameras that handles connectivity, camera settings and recordings, written so new cameras can be added without rework.
React, Flask, Docker
I mentor undergraduate researchers in the lab, usually on projects that pair a biological question with a computational method.
Poster, 2026
Tested whether sow-reared and artificially reared young pigs differ in hippocampal myelin, using Gallyas silver staining, whole-slide imaging and quantitative myelin density.
Symposium abstract, 2026
A two-stage deep learning pipeline that detects pigs, heads and objects in video and classifies pig-object interactions, replacing manual annotation.
Ongoing
Comparing models for the decision stage of the scoring pipeline, the step that turns detections into investigation calls.
Ongoing
Synchronization work on the lab’s automated feeding system.
Away from the scanner I am usually outdoors, on a trail or a ski slope, or building something that probably did not need building.
For research collaborations, engineering roles, questions about the pig imaging tools, or anything else, write to me.
zimul3@illinois.edu