Khalid Saifullah, PhD

Khalid Saifullah, PhD

Research Fellow @MGH & Harvard Medical School
Neuropathology | Neuroimaging | Machine Learning

In Vivo MRI Tangles Marker

In vivo MRI Tangles Marker visual summary

What this figure shows

This visual summary illustrates my end-to-end workflow for building an in vivo MRI marker of neurofibrillary tangles. The central idea is to discover pathology-anchored imaging signatures in an autopsy cohort, train and refine models on ex vivo MRI (where pathology is known), and then translate those signatures to in vivo MRI so the marker can be used in living cohorts.

Pipeline (left → right)

  • Multimodal discovery: identify candidate imaging features and biomarker/pathology correlates to define hypothesis-driven signatures. [Volume and Shape patterns] [AD and LATE patterns]
  • Ex vivo ML model training: train models using ex vivo MRI features (e.g., morphometry/DTI/R2/WMH and related measures), with careful preprocessing and feature selection to predict tangles burden. [See this]
  • In vivo → ex vivo translation: generate and QC in vivo features and map them onto the ex vivo feature space to enable consistent scoring across modalities. [For Reference]
  • Marker deployment: package the pipeline so an in vivo MRI input yields a tangles score.
  • Testing: validate the marker in external and/or independent datasets (e.g., Rush / ADNI / NACC) and evaluate associations with cognition, blood biomarkers, and tau-PET.

Why it matters

A scalable, MRI-based tangles score can help stratify participants, track neurodegeneration, and support clinical trial enrichment—especially when PET or invasive measures are limited.

Resources

ISMRM 2022 · Abstract
AAIC 2022 · Abstract

For related work from our lab, see ARTS [Paper] [Software] (in vivo MRI marker for arteriolosclerosis) and MARBLE [Paper] (in vivo MRI marker for LATE).