Featured2025Neuroimaging researcher and data analyst

DTI Cancer Connectomics

Built a diffusion-MRI connectomics pipeline comparing DMN/ECN white-matter structure in 40 pediatric brain-tumor survivors and 36 healthy controls.

TL;DR

A reproducible neuroimaging workflow that preprocesses diffusion MRI, reconstructs whole-brain tractography, extracts structural connections in the default mode and executive control networks, and prepares de-identified matrices for group-level statistics.

Problem

Pediatric brain tumors and their treatment can disrupt developing white matter, but the relationship between DMN/ECN structural connectivity and cognitive outcomes is not well characterized.

TL;DR

I built a diffusion-MRI analysis pipeline for studying structural connectivity in the default mode and executive control networks of pediatric brain-tumor survivors. The work connects image preprocessing, tractography, tensor metrics, group statistics, and cognitive analysis while keeping participant scans and clinical data private.

Why it matters

Survivors can experience long-term working-memory and information-processing deficits even when treatment is successful. Network-specific connectomics offers a way to study how developing white matter and cognition differ between survivors and healthy children, while recognizing that DTI measures are indirect rather than literal measurements of axons or biological connectivity.

Dataset / inputs

The retrospective cohort contained 40 pediatric brain-tumor survivors and 36 healthy controls between 6 and 18 years old. Inputs included T1-weighted MRI, diffusion-weighted MRI, b-values, b-vectors, cognitive assessment scores, and an HCP-MMP1-derived cortical parcellation. Raw scans and clinical records are not published; the repository contains scripts and de-identified derived products.

Technical decisions

The workflow generated 10 million anatomically constrained streamlines and used SIFT to reduce them to a one-million-streamline whole-brain connectome. I then isolated intrahemispheric and interhemispheric connections among seven bilateral DMN/ECN regions and summarized each connection with FA, MD, AD, RD, and streamline count. Subject matrices were organized for Network-Based Statistic testing rather than treating individual edges as independent comparisons.

Challenges

The analysis crosses several tools and coordinate spaces, so registration and visual quality control are as important as the statistical code. The public scripts also preserve assumptions from the original environment; paths, phase- encoding direction, parcellation resources, and tractography settings all need to be reviewed before the workflow is applied to another acquisition.

Methods

Preprocessed diffusion MRI with denoising, motion/eddy-current correction, bias correction, and anatomical registration
Generated anatomically constrained whole-brain tractography and SIFT-filtered connectomes
Extracted intrahemispheric and interhemispheric DMN/ECN tracts and sampled FA, MD, AD, RD, and streamline count
Assembled subject-level matrices for Network-Based Statistic testing and BrainNet Viewer visualization
Compared cognitive tasks and latent factors with group tests and age correlations

Results

The report found significantly lower working-memory and information-processing-speed performance in survivors than in healthy controls
Exploratory factor analysis identified visual processing, reaction time, verbal/auditory memory, and visual memory factors
Several age-related cognitive correlations seen in healthy controls were reduced or absent in survivors
The public repository includes de-identified derived matrices, statistical designs, selected NBS outputs, and plotting tables

Lessons learned

A traceable preprocessing and quality-control chain is essential when many neuroimaging tools feed the same result
DTI measures and streamline counts are indirect proxies and require cautious biological interpretation

Limitations

Raw participant MRI and clinical data are excluded for privacy, so the public repository cannot reproduce the cohort analysis end to end
The scripts preserve a 2021 analysis environment and contain hard-coded paths and acquisition-specific settings that must be adapted before reuse
Observed associations do not establish a tumor- or treatment-specific biological mechanism

Next steps

Parameterize paths and acquisition settings in a portable workflow configuration
Account for motor deficits and diagnosis-to-assessment timing in follow-up cognitive analyses
Record complete NBS thresholds, contrasts, permutations, and component measures alongside every exported result

Related projects

Featured2025

Gap Junction Connectomics

Built a CNN-based pipeline that segments gap junctions in 3D electron-microscopy volumes and converts them into electrical-connectivity measurements.

Outcome: Created a reusable path from raw EM slices to 3D gap-junction predictions, per-neuron measurements, contactomes, and normalized electrical-connectivity matrices.