Building a Better Template
Our framework is built on the open-source ANTs toolkit and uses a novel approach to ensure accuracy.
Data-Driven
Created from 30 diverse low-dose CT scans from the NLST* cohort.
*We accessed de-identified data from NLST database under an approved data use agreement (CDAS Project Number: NLST-1175).
Texture-Based
Unlike other methods that use binary masks, we preserved the lung's internal texture (Hounsfield Units) to guide a more physically plausible registration.
Automated & Reproducible
The entire pipeline is automated, avoiding manual selection bias and ensuring others can replicate our work.
The process converged after 11 iterations, producing a sharp, low-variance template (DSC = 0.992) using a built-in iterative implementation of SyN registration algorithm:
Revealing Disease Patterns in Emphysema
We tested the template by registering 60 subjects with varying emphysema severity. By analyzing the Jacobian determinant, we quantified local tissue deformation.
Our analysis successfully identified distinct patterns of lung expansion linked to the disease:
Severe Emphysema
Showed marked expansion at the lung bases.
Typical Emphysema
Showed more diffuse changes throughout the lungs.
This demonstrates our template's power to link biomarkers to specific anatomical changes.