Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Registration of Longitudinal Spine CTs for Monitoring Lesion Growth

Malika Sanhinova, Nazim Haouchine, Steve D. Pieper, William M. Wells, Tracy A. Balboni, Alexander Spektor, Mai Anh Huynh, Jeffrey P. Guenette, Bryan Czajkowski, Sarah Caplan, Patrick Doyle, Heejoo Kang, David B. Hackney, Ron N. Alkalay

TL;DR

This work tackles tracking lesion progression in metastatic spine by automatic registration of longitudinal spine CTs. It introduces a two-step pipeline in which a deep learning model segments and labels vertebrae to produce 3D surfaces, followed by surface-based registration that aligns follow-up scans to a baseline using a Gaussian Mixture Model optimized with EM; the transformation between centroid locations is $T(\mathbf{y}_m; \mathbf{R}, \mathbf{t}, s) = s \mathbf{R} \mathbf{y}_m + \mathbf{t}$ and the objective is $E(\mathbf{R},\mathbf{t},s,\sigma^2) = \frac{1}{2\sigma^2} \sum_{m,n} P(\mathbf{y}_m|\mathbf{x}_n) \|\mathbf{x}_n - s\mathbf{R}\mathbf{y}_m - \mathbf{t}\|^2 + \frac{3N_P}{2}\log(\sigma^2)$. The method achieved Dice 0.92 and Hausdorff distance 0.65 mm across 111 registrations on 5 patients, demonstrating robust alignment despite large shape/intensity changes. A demonstrated correlation between segmentation accuracy and registration quality highlights the importance of precise vertebral delineation for reliable lesion-tracking. The work lays groundwork for automated lesion-growth analysis and motivates future expansion to larger datasets and lesion-type-aware models.

Abstract

Accurate and reliable registration of longitudinal spine images is essential for assessment of disease progression and surgical outcome. Implementing a fully automatic and robust registration is crucial for clinical use, however, it is challenging due to substantial change in shape and appearance due to lesions. In this paper we present a novel method to automatically align longitudinal spine CTs and accurately assess lesion progression. Our method follows a two-step pipeline where vertebrae are first automatically localized, labeled and 3D surfaces are generated using a deep learning model, then longitudinally aligned using a Gaussian mixture model surface registration. We tested our approach on 37 vertebrae, from 5 patients, with baseline CTs and 3, 6, and 12 months follow-ups leading to 111 registrations. Our experiment showed accurate registration with an average Hausdorff distance of 0.65 mm and average Dice score of 0.92.

Registration of Longitudinal Spine CTs for Monitoring Lesion Growth

TL;DR

This work tackles tracking lesion progression in metastatic spine by automatic registration of longitudinal spine CTs. It introduces a two-step pipeline in which a deep learning model segments and labels vertebrae to produce 3D surfaces, followed by surface-based registration that aligns follow-up scans to a baseline using a Gaussian Mixture Model optimized with EM; the transformation between centroid locations is and the objective is . The method achieved Dice 0.92 and Hausdorff distance 0.65 mm across 111 registrations on 5 patients, demonstrating robust alignment despite large shape/intensity changes. A demonstrated correlation between segmentation accuracy and registration quality highlights the importance of precise vertebral delineation for reliable lesion-tracking. The work lays groundwork for automated lesion-growth analysis and motivates future expansion to larger datasets and lesion-type-aware models.

Abstract

Accurate and reliable registration of longitudinal spine images is essential for assessment of disease progression and surgical outcome. Implementing a fully automatic and robust registration is crucial for clinical use, however, it is challenging due to substantial change in shape and appearance due to lesions. In this paper we present a novel method to automatically align longitudinal spine CTs and accurately assess lesion progression. Our method follows a two-step pipeline where vertebrae are first automatically localized, labeled and 3D surfaces are generated using a deep learning model, then longitudinally aligned using a Gaussian mixture model surface registration. We tested our approach on 37 vertebrae, from 5 patients, with baseline CTs and 3, 6, and 12 months follow-ups leading to 111 registrations. Our experiment showed accurate registration with an average Hausdorff distance of 0.65 mm and average Dice score of 0.92.
Paper Structure (6 sections, 2 equations, 2 figures)

This paper contains 6 sections, 2 equations, 2 figures.

Table of Contents

  1. METHODS
  2. Multi-class Segmentation of Lesioned Vertebrea from CT
  3. Surface-based Longitudinal Registration
  4. RESULTS
  5. CONCLUSION
  6. ACKNOWLEDGEMENT

Figures (2)

  • Figure 1: Our two-step registration pipeline first classifies and segment the vertebra to generate 3D surfaces using, then use a surface-based registration approach to align the longitudinal CTs.
  • Figure 2: Axial, coronal and sagittal planes showing the lesion progression in one case. The lesion is outlined over time (baseline in blue, 3 months in green, 6 months in red) and fused with the longitudinal CT set. Top to bottom: baseline, 3 and 6 months.