# Network-based disease fingerprinting with neuroinflammation PET imaging

## TL;DR

This study introduces a network-based method to analyze TSPO PET scans, revealing disease-specific neuroinflammatory patterns in disorders like multiple sclerosis and depression.

## Contribution

A novel network-based approach for TSPO PET imaging is developed to capture spatial pharmacokinetic similarities and disease-specific neuroinflammatory signatures.

## Key findings

- TSPO similarity patterns showed high reproducibility with strong test-retest correlations (mean Spearman’s ρ = 0.84).
- Disease classification exceeded chance performance by 23–89% across conditions using machine learning.
- Condition-specific regional hubs mirrored disease pathophysiology with minimal overlap in feature importance.

## Abstract

Neuroinflammation is a hallmark of numerous neurodegenerative, psychiatric, and chronic pain disorders and can be assessed in vivo with 18 kDa translocator protein (TSPO) positron emission tomography (PET). However, conventional quantification methods of TSPO PET are limited and often overlook the spatial relationships between regional signals. The application of network-based approaches to TSPO PET imaging may provide a novel framework to capture disease-specific neuroinflammatory patterns. To address this question, here we developed a data-driven, network-based approach to generate individual brain-wide TSPO PET matrices, employing Euclidean distance to quantify inter-regional pharmacokinetics similarity. We applied this approach to a large multicenter dataset of 528 PET scans utilizing three different TSPO tracers ([11C]-PBR28, [18F]-DPA714, [11C]-PK11195), including healthy controls and patients with different diseases such as multiple sclerosis, traumatic brain injury, schizophrenia, depression, and chronic low back pain. Statistical modelling and machine learning classifiers were applied to evaluate the impact of experimental and biological factors on TSPO similarity patterns and to investigate their potential for capturing disease-specific signatures. TSPO similarity patterns demonstrated high biological specificity and reproducibility, with strong test-retest correlations (mean Spearman’s ρ = 0.84). Average precision of disease classification exceeded chance performance by 23–89% across conditions and was driven by condition-specific regional hubs whose topological distributions closely mirrored disease pathophysiology. This specificity was further supported by minimal overlap in feature importance values across conditions. Altogether, our findings show that network-based analysis of human TSPO PET data can detect disease-specific neuroinflammatory signatures. Such methodologies underscore the biological significance of TSPO PET and enhance its translational value, supporting precision medicine strategies for neuroinflammatory disorders.